ianfullpac

Understanding Language Series Series Editors: Bernard Comrie and Greville Corbett
This page intentionally left blank
Understanding Morphology 2nd edition Martin Haspelmath Max Planck Institute for Evolutionary Anthropology
Andrea D. Sims The Ohio State University
First Edition Published 2002 This Edition Published 2010 Hodder Education, an Hachette UK Company, 338 Euston Road, London NW1 3BH www.hoddereducation.com Copyright © 2010 Martin Haspelmath and Andrea D. Sims All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher or under licence from the Copyright Licensing Agency Limited. Further details of such licences (for reprographic reproduction) may be obtained from the Copyright Licensing Agency Limited, of Saffron House, 6–10 Kirby Street, London EC1N 8TS. British Library Cataloguing in Publication Data A catalogue record for this title is available from the British Library ISBN: 9780340950012 Impression number Year
10╇ 9╇ 8╇ 7╇ 6╇ 5╇ 4╇ 3╇ 2╇ 1 2012, 2011, 2010, 2009 2008 2007
[Environmental statement to be inserted on all biblio pages and deleted by Production if using printers where statement is NOT true] Hachette UK’s policy is to use papers that are natural, renewable and recyclable products and made from wood grown in sustainable forests. The logging and manufacturing processes are expected to conform to the environmental regulations of the country of origin. Cover photo credit: © Image Source/Corbis Typeset by Phoenix Photosetting, Chatham, Kent Printed in Great Britain for Hodder Education, An Hachette UK Company, 338 Euston Road, London NW1 3BH by CPI Antony Rowe
Contents
Preface to the second edition Preface to the first edition Abbreviations
1 Introduction
1.1 What is morphology? 1.2 Morphology in different languages 1.3 The goals of morphological research 1.4 A brief user’s guide to this book Summary of Chapter 1 Further reading Comprehension exercises
xi xiii xv 1 1 4 6 9 11 11 12
2 Basic concepts
14
15 19 22 27 27 29 30
2.1 Lexemes and word-forms 2.2 Affixes, bases and roots 2.3 Morphemes and allomorphs Summary of Chapter 2 Appendix. Morpheme-by-morpheme glosses Comprehension exercises Exploratory exercise
3 Rules 3.1
33 Morphological patterns 3.1.1 Affixation and compounding 3.1.2 Base modification 3.1.3 Reduplication 3.1.4 Conversion 3.1.5 Outside the realm of morphology
34 34 35 38 39 40
viâ•…
CONTENTS
3.2 Two approaches to morphological rules 3.2.1 The morpheme-based model 3.2.2 The word-based model Summary of Chapter 3 Further reading Comprehension exercises Exploratory exercise
40 41 46 54 54 55 56
4 Lexicon
60
61 66 70 75 75 76 77
4.1 A morpheme lexicon? 4.2 A strict word-form lexicon? 4.3 Reconciling word-forms and morphemes Summary of Chapter 4 Further reading Comprehension exercises Exploratory exercise
5 Inflection and derivation 5.1 5.2 5.3 5.4 5.5
Inflectional values Derivational meanings 5.2.1 Derived nouns 5.2.2 Derived verbs 5.2.3 Derived adjectives Properties of inflection and derivation 5.3.1 Relevance to the syntax 5.3.2 Obligatoriness 5.3.3 Limitations on application 5.3.4 Same concept as base 5.3.5 Abstractness 5.3.6 Meaning compositionality 5.3.7 Position relative to base 5.3.8 Base allomorphy 5.3.9 Word-class change 5.3.10 Cumulative expression 5.3.11 Iteration Dichotomy or continuum? 5.4.1 Inherent and contextual inflection Inflection, derivation and the syntax-morphology interface 5.5.1 The dichotomy approach and split morphology 5.5.2 0007The continuum approach and single-component architecture Summary of Chapter 5 Appendix. Notation conventions for inflectional values
81 81 86 87 88 89 89 90 92 93 93 94 94 95 96 96 98 98 98 100 102 102 105 106 107
C O N T E N T S â•…
Further reading Comprehension exercises Exploratory exercise
vii
109 110 110
6 Productivity
114
6.1 Speakers’ knowledge of productivity 6.2 Productivity, creativity and gradience 6.3 Restrictions on word-formation rules 6.3.1 Phonological restrictions 6.3.2 Semantic restrictions 6.3.3 Pragmatic restrictions 6.3.4 Morphological restrictions 6.3.5 Borrowed vocabulary strata 6.4 Productivity and the lexicon 6.4.1 Processing restrictions 6.4.2 Synonymy blocking 6.4.3 Productivity and analogy 6.5 Measuring productivity Summary of Chapter 6 Further reading Comprehension exercises Exploratory exercise
114 116 117 118 119 120 120 121 122 123 125 127 129 131 132 133 134
7 Morphological trees
137
137 142 144 147 150 150 151 152
7.1 Compounding types 7.2 0007Hierarchical structure in compounds 7.3 0007Hierarchical structure in derived lexemes 7.4 Parallels between syntax and morphology? Summary of Chapter 7 Further reading Comprehension exercises Exploratory exercise
8 Inflectional paradigms
156
8.1 8.2 8.3 8.4 8.5
156 158 160 162 163 165 167 172
Syntagmatic and paradigmatic relations in morphology Inflection classes 8.2.1 Inflection class assignment 8.2.2 Relationship to gender 8.2.3 Inflection classes and productivity Paradigmatic relations and inflection class shift Inheritance hierarchies Stems and Priscianic formation
viiiâ•…
CONTENTS
8.6 Syncretism 8.6.1 Systematic versus accidental inflectional homonymy 8.6.2 Underspecification 8.6.3 Rules of referral 8.7 More form-meaning mismatches 8.7.1 Defectiveness 8.7.2 Deponency 8.8 Periphrasis Summary of Chapter 8 Further reading Comprehension exercises Exploratory exercise
174 174 176 179 180 180 182 183 184 184 185 187
9 Words and phrases
189
190 196 197 203 206 207 207 209
9.1 Compounds versus phrases 9.2 Free forms versus bound forms 9.3 Clitics versus affixes 9.4 Lexical integrity Summary of Chapter 9 Further reading Comprehension exercises Exploratory exercise
10 Morphophonology
211
211 217 220 222 228 231 231 232
10.1 Two types of alternations 10.2 The productivity of morphophonological alternations 10.3 The diachrony of morphophonological alternations 10.4 Morphophonology as phonology 10.5 Morphophonology as morphology Summary of Chapter 10 Further reading Comprehension exercises
11 Morphology and valence
234
11.1 Valence-changing operations 11.1.1 0007Semantic valence and syntactic valence (argument structure and function structure) 11.1.2 Agent-backgrounding operations 11.1.3 Patient-backgrounding operations 11.1.4 Agent-adding operations: causatives 11.1.5 Object-creating operations: applicatives 11.1.6 General properties of valence-changing operations
234 234 237 240 241 242 243
C O N T E N T S â•…
ix
11.2 Valence in compounding 11.2.1 Noun incorporation 11.2.2 V–V compound verbs 11.2.3 Synthetic nominal compounds 11.3 Transpositional derivation 11.3.1 Transposition and argument inheritance 11.3.2 Action nouns (V Æ N) 11.3.3 Agent nouns (V Æ N) and deverbal adjectives (V Æ A) 11.3.4 Deadjectival transposition (A Æ N, A Æ V) 11.4 Transpositional inflection Summary of Chapter 11 Further reading Comprehension exercises
245 245 247 249 253 253 254 255 256 257 262 263 263
12 Frequency effects in morphology
265
12.1 Asymmetries in inflectional values 12.1.1 Frequent and rare values 12.1.2 The correlation between frequency and shortness 12.1.3 The correlation between frequency and differentiation 12.1.4 Local frequency reversals 12.1.5 Explaining the correlations 12.2 The direction of analogical levelling 12.3 Frequency and irregularity Summary of Chapter 12 Further reading Comprehension exercises Exploratory exercise
265 265 267 268 270 272 273 274 276 277 277 278
Key to comprehension exercises
281
References
301
Glossary of technical terms
318
Language index
347
Subject index
357
This page intentionally left blank
Preface to the 2nd edition
Readers who are familiar with the first edition of Understanding Morphology (of which Martin Haspelmath was sole author) will find that the book’s fundamental character has not changed. This book provides an introduction to linguistic morphology, with a focus on demonstrating the diversity of morphological patterns in human language and elucidating broad issues that are the foundation upon which morphological theories are built. At the same time, the material in this book has been substantially restructured and some topics have been expanded. The goal was to bring foundational issues to the forefront. This was accomplished mostly by expanding existing chapter sections or creating new chapter sections to centralize and focus discussion that was previously spread throughout a chapter. In some cases, however, the restructuring has been more radical. Notably, Chapter 3 from the first edition (‘Lexicon and Rules’) has been divided into two chapters, with more attention given to the question of whether the lexicon is fundamentally morpheme-based or word-based. Also, the chapter ‘Word-based Rules’ (formerly Chapter 9) has been eliminated, with its material redistributed elsewhere, as relevant. There are also some new and expanded features: answers to each chapter’s comprehension exercises can now be found at the back of the book; the glossary has been significantly enlarged; Chapter 5 has a new appendix on notation conventions for inflectional values; and perhaps most notably, nine chapters now contain exploratory exercises. The exploratory exercises are larger in scope than the comprehension exercises and extend the themes of the chapters. They guide readers through research questions in an open-ended way, asking them to gather and analyze data from a variety of sources, such as descriptive grammars, corpora, and native speaker consultants. The exercises are broadly constructed so that they can be tailored to the needs and interests of particular individuals or groups. In a classroom setting, instructors can use them with different levels of
xiiâ•…
P R E FA C E T O T H E 2 N D E D I T I O N
students by adjusting their expectations regarding depth of analysis and methodological rigor. A number of people have helped to improve this new edition. First and foremost, we thank the series editors, Bernard Comrie and Greville Corbett, whose numerous suggestions and dedication to the project have greatly improved it. (Naturally, all errors remain the fault of the authors.) At Hodder we are also grateful to Bianca Knights and Tamsin Smith for their encouragement and deep well-springs of patience, and to Liz Wilson, for shepherding the project through production. In the end, textbooks are for students, and we would also like to thank Andrea Sims’s morphology students at Northwestern University and The Ohio State University for their feedback. We especially thank Christine Davis, Caitlin Ferrarell, Laura Garofalo, Alexander Obal, Zach Richards, Cenia Rodriguez, and Honglei Wang. They provided extensive, detailed, valuable, and sometimes unexpected perspectives on the first edition. Their critique of some aspects of the second edition (particularly, drafts of the exploratory exercises) also proved crucial. This second edition contains some new examples, and we thank the following people for their help in understanding the relevant languages and providing appropriate examples: Hope Dawson (Sanskrit), Maggie Gruszczynska (Polish), Jessie Labov (Hungarian), and Amanda Walling (Old English). Any errors remain the fault of the authors. We are indebted to the various scholars and teachers who wrote reviews of the first edition, or who have passed on their experiences in teaching with the book. We are happy that the book was, on the whole, warmly received. We have tried to improve that which was deemed in need of improvement. Finally, we thank our families, and especially our partners, Susanne Michaelis and Jason Packer, for all manner of help and support. Leipzig, Germany Columbus, Ohio, USA July 2010
Preface to the 1st edition
This book provides an introduction to the field of linguistic morphology. It gives an overview of the basic notions and the most important theoretical issues, emphasizing throughout the diversity of morphological patterns in human languages. Readers who are primarily interested in understanding English morphology should not be deterred by this, however, because an individual language can be understood in much greater depth when viewed against the cross-linguistic background. The focus of this book is on morphological phenomena and on broad issues that have occupied morphologists of various persuasions for a long time. No attempt is made to trace the history of linguists’ thinking about these issues, and references to the theoretical literature are mostly confined to the ‘Further reading’ sections. I have not adopted any particular theoretical framework, although I did have to opt for one particular descriptive format for morphological rules (see Section 3.2.2). Readers should be warned that this format is no more ‘standard’ than any other format, and not particularly widespread either. But I have found it useful, and the advanced student will soon realize how it can be translated into other formats. Although it is often said that beginning students are likely to be confused by the presentation of alternative views in textbooks, this book does not pretend that there is one single coherent and authoritative view of morphology. Debates and opposing viewpoints are so much part of science that omitting them completely from a textbook would convey a wrong impression of what linguistic research is like. And I did not intend to remain neutral in these debates, not only because it would have been virtually impossible anyway, but also because a text that argues for a particular view is invariably more interesting than one that just presents alternative views. A number of people have helped me in writing this book. My greatest thanks go to the series editors, Bernard Comrie and Greville Corbett, who provided countless suggestions for improving the book. I also thank Renate Raffelsiefen for her expert advice on phonological
xivâ•…
P R E FA C E T O T H E 1 S T E D I T I O N
questions, as well as Tomasz Bak and Agnieszka Reid for help with Polish examples, and Claudia Schmidt for help with the indexes. Finally, I thank Susanne Michaelis for all kinds of help, both in very specific and in very general ways. This book is dedicated to our son, Gabriel. Martin Haspelmath Leipzig December 2001
Abbreviations
These abbreviations are consistent with the Leipzig Glossing Rules (v. February 2008). abe abl abs acc act adj adv aff ag agr all antic antip aor appl art asp aux caus clf comp compl cond cont cvb dat decl
abessive ablative absolutive accusative active adjective adverb(ial) affirmative agent agreement allative anticausative antipassive aorist applicative article aspect auxiliary causative classifier complementizer completive conditional continuative converb dative declarative
def dem deobj desid det DO du dur ela erg ess excl f foc fut g gen hab hyp imp impf impv incl ind indf iness inf
definite demonstrative deobjective desiderative determiner direct object dual durative elative ergative essive exclusive feminine focus future gender (e.g. g1 = gender 1) genitive habitual hypothetical imperative imperfect(ive) imperative inclusive indicative indefinite inessive infinitive
xviâ•…
A B B R E V I AT I O N S
ins instrumental intf interfix intr/intr. intransitive iobj indirect object loc locative m masculine masd masdar N noun n neuter nec necessitative neg negation, negative nom nominative NP noun phrase obj object obl oblique OED 0007Oxford English Dictionary opt optative p patient par partitive part participle pass passive pfv perfective pl plural poss possessive pot potential
pP pred Pref pret prf prs priv prog propr pst ptcp purp recp refl rel rep sbj sbjv sg ss subord Suf top tr/tr. V VP
prepositional phrase predicate prefix preterite perfect present privative progressive proprietive past participle purposive reciprocal reflexive relative clause marker repetitive subject subjunctive singular same-subject subordinator suffix topic transitive verb verb phrase
1.1 WHAT IS MORPHOLOGY?
Introduction
1
1
1.1 What is morphology? Morphology is the study of the internal structure of words.1 Somewhat paradoxically, morphology is both the oldest and one of the youngest subdisciplines of grammar. It is the oldest because, as far as we know, the first linguists were primarily morphologists. The earliest extant grammatical texts are well-structured lists of morphological forms of Sumerian words, some of which are shown in (1.1). They are attested on clay tablets from Ancient Mesopotamia and date from around 1600 bce. (1.1) badu baduun bašidu bašiduun
‘he goes away’ ‘I go away’ ‘he goes away to him’ ‘I go away to him’
ing˜en ‘he went’ ing˜enen ‘I went’ inšig˜en ‘he went to him’ inšig˜enen ‘I went to him’ (Jacobsen 1974: 53–4)
Sumerian was the traditional literary language of Mesopotamia but, by the second millennium bce, it was no longer spoken as a medium of everyday communication (having been replaced by the Semitic language Akkadian), so it needed to be recorded in grammatical texts. Morphology was also prominent in the writings of the greatest grammarian of Antiquity, the Indian Paˉ n.ini (fifth century bce), and in the Greek and Roman grammatical tradition. Until the nineteenth century, Western linguists often thought of grammar as consisting primarily of word structure, perhaps because the
1
The reader should be aware that this sentence, while seemingly straightforward, conceals a controversy – there is no agreed upon definition of ‘word’. The relevant issues are addressed in Chapter 9, but here, and through most of the book, we will appeal to a loose, intuitive concept of ‘word’.
2 
C H A P T E R 1 â•… I N T R O D U C T I O N
classical languages Greek and Latin had fairly rich morphological patterns that were difficult for speakers of the modern European languages. This is also the reason why it was only in the second half of the nineteenth century that the term morphology was invented and became current. Earlier there was no need for a special term, because the term grammar mostly evoked word structure, i.e. morphology. The terms phonology (for sound structure) and syntax (for sentence structure) had existed for centuries when the term morphology was introduced. Thus, in this sense morphology is a young discipline. Our initial definition of morphology, as the study of the internal structure of words, needs some qualification, because words have internal structure in two very different senses. On the one hand, they are made up of sequences of sounds (or gestures in sign language), i.e. they have internal phonological structure. Thus, the English word nuts consists of the four sounds (or, as we will say, phonological segments) [nts]. In general, phonological segments such as [n] or [t] cannot be assigned a specific meaning – they have a purely contrastive value (so that, for instance, nuts can be distinguished from cuts, guts, shuts, from nets, notes, nights, and so on). But often formal variations in the shapes of words correlate systematically with semantic changes. For instance, the words nuts, nights, necks, backs, taps (and so on) share not only a phonological segment (the final [s]), but also a semantic component: they all refer to a multiplicity of entities from the same class. And, if the final [s] is lacking (nut, night, neck, back, tap), reference is made consistently to only one such entity. By contrast, the words blitz, box, lapse do not refer to a multiplicity of entities, and there are no semantically related words *blit, *bok, *lap.2 We will call words like nuts ‘(morphologically) complex words’. In a morphological analysis, we would say that the final [s] of nuts expresses plural meaning when it occurs at the end of a noun. But the final [s] in lapse does not have any meaning, and lapse does not have morphological structure. Thus, morphological structure exists if there are groups of words that show identical partial resemblances in both form and meaning. Morphology can be defined as in Definition 1. Definition 1: Morphology is the study of systematic covariation in the form and meaning of words. It is important that this form–meaning2 covariation occurs systematically in groups of words. When there are just two words with partial form– meaning resemblances, these may be merely accidental. Thus, one would
2
The asterisk symbol (*) is used to mark nonexistent or impossible expressions.
1 . 1 W H AT I S M O R P H O L O G Y ?
3
not say that the word hear is morphologically structured and related to ear. Conceivably, h could mean ‘use’, so h-ear would be ‘use one’s ear’, i.e. ‘hear’. But this is the only pair of words of this kind (there is no *heye ‘use one’s eye’, *helbow ‘use one’s elbow’, etc.), and everyone agrees that the resemblances are accidental in this case. Morphological analysis typically consists of the identification of parts of words, or, more technically, constituents of words. We can say that the word nuts consists of two constituents: the element nut and the element s. In accordance with a widespread typographical convention, we will often separate word constituents by a hyphen: nut-s. It is often suggested that morphological analysis primarily consists in breaking up words into their parts and establishing the rules that govern the co-occurrence of these parts. The smallest meaningful constituents of words that can be identified are called morphemes. In nut-s, both -s and nut are morphemes. Other examples of words consisting of two morphemes would be breaking, hope-less, re-write, cheese-board; words consisting of three morphemes are re-writ-ing, hope-less-ness, ear-plug-s; and so on. Thus, morphology could alternatively be defined as in Definition 2. Definition 2: Morphology is the study of the combination of morphemes to yield words. This definition looks simpler and more concrete than Definition 1. It would make morphology quite similar to syntax, which is usually defined as ‘the study of the combination of words to yield sentences’. However, we will see later that Definition 2 does not work in all cases, so we should stick to the somewhat more abstract Definition 1 (see especially Chapters 3 and 4). In addition to its main sense, where morphology refers to a subdiscipline of linguistics, it is also often used in a closely related sense, to denote a part of the language system. Thus, we can speak of ‘the morphology of Spanish’ (meaning Spanish word structures) or of ‘morphology in the 1980s’ (meaning a subdiscipline of linguistics). The term morphology shares this ambiguity with other terms such as syntax, phonology and grammar, which may also refer either to a part of the language or to the study of that part of the language. This book is about morphology in both senses. We hope that it will help the reader to understand morphology both as a part of the language system and as a part of linguistics. One important limitation of the present book should be mentioned right at the beginning: it deals only with spoken languages. Sign languages of course have morphology as well, and the only justification for leaving them out of consideration here is the authors’ limited competence. As more and more research is done on sign languages, it can be expected that these Â�
4 
C H A P T E R 1 â•… I N T R O D U C T I O N
studies will have a major impact on our views of morphology and language structure in general.
1.2 Morphology in different languages Morphology is not equally prominent in all (spoken) languages. What one language expresses morphologically may be expressed by a separate word or left implicit in another language. For example, English expresses the plural of nouns by means of morphology (nut/nuts, night/nights, and so on), but Yoruba uses a separate word for expressing the same meaning. Thus, o.kùnrin means ‘(the) man’, and the word àwo.n can be used to express the plural: àwo.n o.kùnrin ‘the men’. But in many cases where several entities are referred to, this word is not used and plurality is simply left implicit. Quite generally, we can say that English makes more use of morphology than Yoruba. But there are many languages that make more use of morphology than English. For instance, as we saw in (1.1), Sumerian uses morphology to distinguish between ‘he went’ and ‘I went’, and between ‘he went’ and ‘he went to him’, where English must use separate words. In Classical Greek, there is a dual form for referring to two items, e.g. adelphoˉ´ ‘two brothers’. In English it is possible to use the separate word ‘two’ to render this form, but it is also possible to simply use the plural form and leave the precise number of items implicit. Linguists sometimes use the terms analytic and synthetic to describe the degree to which morphology is made use of in a language. Languages like Yoruba, Vietnamese or English, where morphology plays a relatively modest role, are called analytic. Consider the following example sentences.3 (1.2) Yoruba Nwo.n ó maa gbà pó.nùn mé. waˇ ló.sò.ò.sè.. they fut prog get pound ten weekly ‘They will be getting £10 a week.’ (Rowlands 1969: 93) (1.3) Vietnamese Hai d-ú.a
bo? nhau
là ta.i
gia-d-ình thàng chông.
two individual leave each.other be because.of family
‘They divorced because of his family.’
3
guy
husband
(Nguyen 1997: 223)
For each example sentence from an unfamiliar language, not only an idiomatic translation is provided, but also a literal (‘morpheme-by-morpheme’) translation. The key for abbreviations is found on pp. xv–xvi, and further notational conventions are explained in the Appendix to Chapter 2.
1.2 MORPHOLOGY IN DIFFERENT LANGUAGES 
5
When a language has almost no morphology and thus exhibits an extreme degree of analyticity, it is also called isolating. Yoruba and Vietnamese, but not English, are usually qualified as isolating. Languages like Sumerian, Swahili or Lezgian, where morphology plays a more important role, would be called synthetic. Let us again look at two example sentences. (1.4) Swahili Ndovu wa-wili wa-ki-song-ana zi-umia-zo ni nyika. elephants pl-two 3pl-subord-jostle-recp 3sg-hurt-rel is grass ‘When two elephants jostle, what is hurt is the grass.’ (Ashton 1947: 114) (1.5) Lezgian Marf-adi wicˇ i-n qalin st’al-ra-ldi qaw gata-zwa-j. rain-erg self-gen dense drop-pl-ins roof hit-impf-pst ‘The rain was hitting the roof with its dense drops.’ (Haspelmath 1993: 140) When a language has an extraordinary amount of morphology and Â� perhaps many compound words, it is called polysynthetic. An example is West Greenlandic.4 (1.6) West Greenlandic Paasi-nngil-luinnar-para ilaa-juma-sutit. understand-not-completely-1sg.sbj.3sg.obj.ind come-want-2sg.ptcp ‘I didn’t understand at all that you wanted to come along.’ (Fortescue 1984: 36) The distinction between analytic and (poly)synthetic languages is not a bipartition or a tripartition, but a continuum, ranging from the most radically isolating to the most highly polysynthetic languages. We can determine the position of a language on this continuum by computing its degree of synthesis, i.e. the ratio of morphemes per word in a random text sample of the language. Table 1.1 gives the degree of synthesis for a small selection of languages.
4
There is another definition of polysynthetic in use among linguists, according to which a language is polysynthetic if single words in the language typically correspond to multiword sentences in other languages. In this book we will not use the term in this sense, but under such a definition, Swahili would be classified as a polysynthetic language.
6 
C H A P T E R 1 â•… I N T R O D U C T I O N
Language
Ratio of morphemes per word
West Greenlandic Sanskrit Swahili Old English Lezgian German Modern English Vietnamese
3.72 2.59 2.55 2.12 1.93 1.92 1.68 1.06
Table 1.1╇ The degree of synthesis of some languages Source: based on Greenberg (1959), except for Lezgian
Although English has much more morphology than isolating languages like Yoruba and Vietnamese, it still has a lot less than many other languages. For this reason, it will be necessary to refer extensively to languages other than English in this book.
1.3 The goals of morphological research Morphological research aims to describe and explain the morphological patterns of human languages. It is useful to distinguish four more specific sub-goals of this endeavour: elegant description, cognitively realistic description, system-external explanation and a restrictive architecture for description. (i) Elegant description. All linguists agree that morphological patterns (just like other linguistic patterns) should be described in an elegant and intuitively satisfactory way. Thus, morphological descriptions should contain a rule saying that English nouns form their plural by adding -s, rather than simply listing the plural forms for each noun in the dictionary (abbot, abbots; ability, abilities; abyss, abysses; accent, accents; …). In a computer program that simulates human language, it may in fact be more practical to adopt the listing solution, but linguists would find this inelegant. The main criterion for elegance is generality. Scientific descriptions should, of course, reflect generalizations in the data and should not merely list all known individual facts. But generalizations can be formulated in various ways, and linguists often disagree in their judgements of what is the most elegant description. It is therefore useful to have a further objective criterion that makes reference to the speakers’ knowledge of their language. (ii) Cognitively realistic description. Most linguists would say that their descriptions should not only be elegant and general, but they should
1.3 THE GOALS OF MORPHOLOGICAL RESEARCH 
7
also be cognitively realistic. In other words, they should express the same generalizations about grammatical systems that the speakers’ cognitive apparatus has unconsciously arrived at. We know that the speakers’ knowledge of English not only consists of lists of singulars and plurals, but comprises a general rule of the type ‘add -s to a singular form to get a plural noun’. Otherwise speakers would be unable to form the plural of nouns they have never encountered before. But they do have this ability: if you tell an English speaker that a certain musical instrument is called a duduk, they know that the plural is (or can be) duduks. The dumb computer program that contains only lists of singulars and plurals would fail miserably here. Of course, cognitively realistic description is a much more ambitious goal than merely elegant description, and we would really have to be able to look inside people’s heads for a full understanding of the cognitive machinery. Linguists sometimes reject proposed descriptions because they seem cognitively implausible, and sometimes they collaborate with psychologists and neurologists and take their research results into account. (iii) System-external explanation. Once a satisfactory description of Â� morphological patterns has been obtained, many linguists ask an even more ambitious question: why are the patterns the way they are? In other words, they ask for explanations. But we have to be careful: most facts about linguistic patterns are historical accidents and as such cannot be explained. The fact that the English plural is formed by adding -s is a good example of such a historical accident. There is nothing necessary about plural -s: Hungarian plurals are formed by adding -k, Swedish plurals add -r, Hebrew plurals add -im or -ot, and so on. A frequent way to pursue explanation in linguistics is to analyze universals of human language, since these are more likely to represent facts that are in need of explanation at a deep level. And as a first step, we must find out which morphological patterns are universal. Clearly, the s-plural is not universal, and, as we saw in the preceding section, not even the morphological expression of the plural is universal – Yoruba is an example of a language that lacks morphological plurals. So even the fact that English nouns have plurals is no more than a historical accident. But there is something about plurals that is not accidental: nouns denoting people are quite generally more likely to have plurals than nouns denoting things. For instance, in Tzutujil, only human nouns have regular morphological plural forms (Dayley 1985: 139). We can formulate the universal statement in (1.7). (1.7) A universal statement: If a language has morphological plural forms of nouns at all, it will have plurals of nouns denoting people. (Corbett 2000: ch. 3) Because of its ‘if … then’ form, this statement is true also of languages like English (where most nouns have plurals) and Yoruba (where nouns do not
C H A P T E R 1 â•… I N T R O D U C T I O N
8 
have a morphological plural). Since it is (apparently) true of all languages, it is in all likelihood not a historical accident, but reflects something deeper, a general property of human language that can perhaps be explained with reference to system-external considerations. For instance, one might propose that (1.7) is the case because, when the referents of nouns are people, it makes a greater difference how many they are than when the referents are things. Thus, plurals of people-denoting nouns are more useful, and languages across the world are thus more likely to have them. This explanation (whatever its merits) is an example of a system-external explanation in the sense that it refers to facts outside the language system: the usefulness of number distinctions in speech. (iv) A restrictive architecture for description. Many linguists see an important goal of grammatical research in formulating some general design principles of grammatical systems that all languages seem to adhere to. In other words, linguists try to construct an architecture for description (also called grammatical theory) that all language-particular descriptions must conform to. For instance, it has been observed that rules by which constituents are fronted to the beginning of a sentence can affect syntactic constituents (such as whole words or phrases), but not morphological constituents (i.e. morphemes that are parts of larger words). Thus, (1.8b) is a possible sentence (it can be derived from a structure like (1.8a)), but (1.9b) is impossible (it cannot be derived from (1.9a)). (The subscript line __ stands for the position that the question word what would occupy if it had not been moved to the front.) (1.8) a. We can buy cheese. b. What can we buy __ ? (1.9) a. We can buy a cheeseboard. b. *What can we buy a __ -board? This restriction on fronting (which seems to hold for all languages that have such a fronting rule) follows automatically if fronting rules (such as whatfronting) and morpheme-combination rules (such as compounding, which yields cheeseboard from cheese and board) are separated from each other in the descriptive architecture. A possible architecture for grammar is shown in Figure 1.1, where the boxes around the grammatical components Â� ‘syntax’, ‘morphology’ and ‘phonology’ symbolize the separateness of each of the components.
meanings
morphology
syntax
phonology
• morphemecombination rules
• fronting rules • word-combination rules
• pronunciation rules
Figure 1.1 A possible descriptive architecture for grammar
sound
1.4 A BRIEF USER’S GUIDE TO THIS BOOK 
9
This architecture is restrictive because it automatically disallows certain logically possible interactions of rules (see Section 9.4 for more discussion). Many linguists assume that the architecture of grammar is innate – it is the same for all languages because it is genetically fixed for the human species. The innate part of speakers’ grammatical knowledge is also called Universal Grammar. For these linguists, one goal of morphological research is to discover those principles of the innate Universal Grammar that are relevant for word structure. The goals (iii) and (iv) are similar in that both ask deeper, theoretical questions, and both exclusively concern universal aspects of morphology. And both are more ambitious than (i) and (ii) in that they involve explanation in some sense. Thus, one might ask questions such as ‘Why cannot constituents of words be fronted to the beginning of the sentence?’ and answer them from a Universal Grammar-oriented perspective with reference to a hypothesis about the innate architecture of grammar (‘Because fronting rules are part of the syntactic component, and morphemecombinations are part of morphology, and syntax and morphology are separate’). However, explanations of this kind are strictly system-internal, whereas explanations of the kind we saw earlier are even more general in that they link universal properties of grammars to general facts about human beings that are Â� external to the grammatical system. It is a curious observation on the sociology of science that currently most linguists seem to be concerned either with system-external explanation or with formulating an architecture for grammatical description, but not with both goals simultaneously. There are thus two primary orientations in contemporary theoretical morphological research: the functionalist Â� orientation, which aims at system-external explanation, and the generative (or formalist) orientation, which seeks to discover the principles of the innate grammatical architecture. However, it does not seem wise to divide the labour of morphological research in this way, because neither system-external factors nor innate principles can explain the whole range of morphological patterns. Accordingly, both goals will be simultaneously pursued in the more theoretically oriented parts of this book.
1.4 A brief user’s guide to this book Sources of data In this book we give examples from many different languages, and attributions for this data follow standard practice. For examples from less widely known languages, the reference is given after the example. However, when the examples are from well-known and widely studied languages such as Modern English, Russian, Standard Arabic or Old English, we
10 
C H A P T E R 1 â•… I N T R O D U C T I O N
do not give a reference because the data can easily be obtained from any standard reference book. Sources of ideas In this book, we focus on morphological data and problems of analysis, not on the history of thinking about these issues in linguistics. Thus, we rarely Â� mention names of particular authors in the text, and references to sources of ideas are given only in a few very specific cases (as in Table 1.1 and example (1.7)). In general, the reader is referred to the section ‘Further reading’, where important works on theoretical morphology are mentioned. Comprehension exercises Each chapter contains exercises designed to help the reader solidify understanding of the material. Answers to these exercises can be found at the end of the book. Exploratory exercises Many chapters also contain a longer exercise that extends the chapter material. These are exploratory in nature, so no answers are provided. Glossary The glossary contains the technical terms relating to morphology that are used in this book. In addition to giving a brief definition, the glossary also refers the reader to the most important places where the term is discussed in the text. These terms are printed in bold where they are first discussed in the text. Language index Many languages mentioned in this book will be unfamiliar to the reader. The language index serves to give information on each language, in particular its genealogical affiliation, the place where it is spoken, and its ISO 639-3 code. ISO 639-3 is an international standard that assigns a unique code to every language. The reader is encouraged to use these codes to find more information about the languages discussed in the book; the on-line language encyclopaedia Ethnologue (www.ethnologue.com) is particularly helpful in this regard. Spelling and transcription Morphology of spoken languages deals with spoken words, so ideally all the examples should be in phonetic transcription in this book. But since many languages have a conventional spelling that renders the pronunciation more or less faithfully, it was more practical and less confusing to adopt that spelling for the examples here. (Although English spelling is not particularly close to the pronunciation, English examples will usually be given in the
FURTHER READING 
11
spelling, because it is assumed that the readers know their pronunciation.) Examples cited in the spelling (or conventional transliteration) are always printed in italics, whereas examples cited in phonetic transcription are printed in ordinary typeface and are usually included in square brackets. Readers not familiar with phonetic transcription should consult any phonetics or phonology textbook. Abbreviations A list of abbreviations (especially abbreviations of grammatical terms) is found on pp. xv–xvi.
Summary of Chapter 1 Morphology is most simply defined as the study of the combination of morphemes to yield words, but a somewhat more abstract definition (as the study of systematic covariation in the form and meaning of words) will turn out to be more satisfactory. Different languages vary strikingly in the extent to which they make use of morphology. The goals of morphological research are (on the descriptive level) elegant and cognitively realistic description of morphological structures, plus (on the theoretical level) system-external explanation and the discovery of a restrictive architecture for description.
Further reading For an elementary introduction to morphology, see Coates (1999) or Katamba and Stonham (2006). Other morphology textbooks that are somewhat similar in scope to the present book are Bauer (2003), Bubenik (1999), and Plag (2003) (as well as Scalise (1994), in Italian, and Plungian (2000), in Russian). Spencer (1991) is a very Â� thorough introduction that concentrates on the generative orientation in morphology. Matthews (1991) puts particular emphasis on the definition of morphological concepts. Carstairs-McCarthy (1991) gives an excellent overview of the theoretical debates in the 1970s and 1980s. Booij (2007) devotes a chapter to the mental processing and storage of words. Aronoff and Fudeman (2005) is a source for techniques of morphological analysis. The most comprehensive work on morphology that has ever been written by a single author is Mel’cˇuk (1993–2000) (five volumes, in French). Although its style is somewhat unusual, it is very readable. Reference works that are devoted exclusively to morphology are Spencer and Zwicky (1998) and Booij, Lehmann and Mugdan (2000–2004). A
12 
C H A P T E R 1 â•… I N T R O D U C T I O N
biblioÂ�graphy is Beard and Szymanek (1988). Bauer (2004) is a glossary of morphological terms. The complementarity of the functionalist and the generative approaches to morphology is explained and emphasized in the introductory chapter of Hall (1992). An introduction to a sign language that also discusses morphology is Sutton-Spence and Woll (1999). A note on the history of the term morphology: in the biological sense (‘the study of the form of animals and plants’), the term was coined by Johann Wolfgang von Goethe (1749–1832), and, in the linguistic sense, it was first used by August Schleicher (1859).
Comprehension exercises 1. Which of the following English words are morphologically complex? For each complex word, list at least two other words that provide evidence for your decision (i.e. words that are both semantically and formally related to it).
nights, owl, playing, affordable, indecent, religion, indolent, bubble, during, searched, hopeful, redo
2. Identify the morphological constituents and describe their meanings in the following Mandarin Chinese nouns.
chàngcí chàngjıˉ chuánweˇi cíweˇi diànlì diànshì dòngwùxué dòngwùyóu dòngwùyuán fángdıˇng fángkè feˉichuán feˉijıˉ feˉiyú huaˉcheˉ huaˉyuán jıˉcheˉ jiaˇolì kèfáng
‘libretto’ ‘gramophone’ ‘stern’ ‘suffix’ ‘electric power’ ‘television’ ‘zoology’ ‘animal oil’ ‘zoo’ ‘roof’ ‘tenant’ ‘airship’ ‘aeroplane’ ‘flying fish’ ‘festooned vehicle’ ‘flower garden’ ‘locomotive’ ‘strength of one’s legs’ ‘guest house’
dıˇngdeˉng diàncheˉ diàndeˉng diànjıˉ qìcheˉ qìchuán shaˉndıˇng shìchàng shìlì shùcí shuıˇcheˉ shuıˇlì shùxué weˇideˉng weˇishuıˇ yóudeˉng yóuzhıˇ yúyóu zhıˇhuaˉ
‘top light’ ‘streetcar, tram’ ‘electric lamp’ ‘electrical machine’ ‘car’ ‘steamship’ ‘summit’ ‘sightseeing’ ‘eyesight’ ‘number word’ ‘watercart’ ‘waterpower’ ‘mathematics’ ‘tail light’ ‘tail water’ ‘oil lamp’ ‘oil paper’ ‘fish oil’ ‘paper flower’
COMPREHENSION EXERCISES 
13
3. Identify the morphological constituents and their meanings in the following Tzutujil verbs (Dayley 1985: 87) (A note on Tzutujil spelling: x is pronounced [∫], and 7 is pronounced [ʔ].)
xinwari neeli ne7eeli nixwari xateeli natwari
‘I slept’ ‘he or she leaves’ ‘they leave’ ‘you(pl) sleep’ ‘you(sg) left’ ‘you(sg) sleep’
How would you say ‘I left’, ‘he or she sleeps’, ‘we sleep’?
xoqeeli ninwari xixwari xe7eeli xwari
‘we left’ ‘I sleep’ ‘you(pl) slept’ ‘they left’ ‘he or she slept’
4. In the following list of Hebrew words, find at least three sets of word pairs whose two members covary formally and semantically, so that a morphological relationship can be assumed. For each set of word pairs, describe the formal and semantic differences.
kimut diber ˉhašav sagra ˉhašva kalat maklet kalta kimet
‘wrinkling’ ‘he spoke’ ‘he thought’ ‘she shut’ ‘she thought’ ‘he received’ ‘radio receiver’ ‘she received’ ‘he wrinkled’
mah ˉ šev masger dibra milmel kimta milmla sagar dibur
‘computer’ ‘lock’ ‘she spoke’ ‘he muttered’ ‘she wrinkled’ ‘she muttered’ ‘he shut’ ‘speech’
2
Basic concepts
W
e have seen that morphological structure exists if a group of words shows partial form-meaning resemblances. In most cases, the relation between form and meaning is quite straightforward: parts of words bear different meanings. Consider the examples in (2.1). (2.1) read wash write
read-s wash-es write-s
kind kind-ness happy happi-ness friendly friendli-ness
read-er wash-er writ-er
read-able wash-able writ-able
un-kind un-happy un-friendly
These words are easily segmented, i.e. broken up into individually meaningful parts: read + s, read + er, kind + ness, un + happy, and so on. These parts are called morphemes.1 Words may of course consist of more than two morphemes, e.g. un-happi-ness, read-abil-ity, un-friend-ly, un-friend-li-ness. Morphemes can be defined as the smallest meaningful constituents of a linguistic expression. When we have a sentence such as Camilla met an unfriendly chameleon, we can divide it into meaningful parts in various ways, e.g. Camilla/met an unfriendly chameleon, or Camilla/met/an/unfriendly/ chameleon, or Camilla/met/an/un/friend/ly/chameleon. But further division is not possible. When we try to divide chameleon further (e.g. cha/meleon), we do not obtain parts that can be said to be meaningful, either because they are not found in any other words (as seems to be the case with meleon), or because the other words in which they occur do not share any aspect of meaning with chameleon (cf. charisma, Canadian, caboodle, capacity, in which it would be theoretically possible to identify a word part cha/ca-). Thus, chameleon cannot 1
Some approaches question the usefulness of the notion ‘morpheme’. We will discuss these extensively in Chapters 3 and 4, but for the moment it is helpful to begin in this more conventional way.
2 . 1 ╇ L E X E M E S A N D W O R D - F O R M S
15
be segmented into several morphemes; it is monomorphemic. Morphemes are the ultimate elements of morphological analysis; they are, so to speak, morphological atoms. In this chapter we introduce some other fundamental concepts and their related terms, starting with lexemes and word-forms.
2.1╇ Lexemes and word-forms The most basic concept of morphology is of course the concept ‘word’. For the sake of convenience, let us assume for the moment that a word is whatever corresponds to a contiguous sequence of letters.2 Thus, in one sense the first sentence of this paragraph consists of twelve words, each separated by a blank space from the neighbouring word(s). And in another sense the sentence has nine words – there are nine different sequences of letters separated by spaces. But when a dictionary is made, not every sequence of letters is given its own entry. For instance, the words live, lives, lived and living are pronounced differently and are different words in that sense. But a dictionary would contain only a single entry live. The dictionary user is expected to know that live, lives, lived and living are different concrete instantiations of the ‘same’ word live. Thus, there are three rather different notions of ‘word’. When a word is used in some text or in speech, that occurrence of the word is sometimes referred to as a word token. In this sense the first sentence in the paragraph consists of twelve words. The other two senses of the term ‘word’ are not defined in reference to particular texts; they correspond to the ‘dictionary word’ and the ‘concrete word’. Since this distinction is central to morphology, we need special technical terms for the two notions, lexeme and word-form, respectively. A lexeme is a word in an abstract sense. live is a verb lexeme. It represents the core meaning shared by forms such as live, lives, lived and living. In most languages, dictionaries are organized according to lexemes, so it is usually reasonable to think of a lexeme as a ‘dictionary word’. Although we must assign names to lexemes to be able to talk about them, lexemes are abstract entities that have no phonological form of their own. live is therefore just a convenient label to talk about a particular lexeme; the sequence of sounds [lIv] is not the lexeme itself. Sometimes we will use the convention of writing lexemes in small capital letters. By contrast, a word-form is a word in a concrete sense. It is a sequence of sounds that expresses the combination of a lexeme (e.g. live) and a set
2
Of course, we should really define words in terms of sounds, since language is primarily a spoken (not written) medium, and there are other problems with this definition as well. But it is sufficient for the present purposes. A more sophisticated approach is deferred to Chapter 9.
16 
C H A P T E R 2 â•… B A S I C C O N C E P T S
of grammatical meanings (or grammatical functions) appropriate to that lexeme (e.g. third person singular present tense). Lives is a word-form. Thus, word-forms are concrete in that they can be pronounced. Lexemes can be thought of as sets of word-forms, and every word-form belongs to one lexeme. The word-forms live, lives, lived, and living all belong to the lexeme live. Word-forms belonging to the same lexeme express different grammatical functions, but the same core concept. When a wordform is used in a particular text or in speech, this instance of use is a word token. The first sentence of this paragraph thus has sixteen word tokens, fifteen word-forms (of is repeated), and thirteen lexemes (e.g. lexemes and lexeme both belong to lexeme). In the most interesting case, lexemes consist of a fair number of wordforms. The set of word-forms that belongs to a lexeme is often called a paradigm. The paradigm of the Modern Greek noun lexeme filos ‘friend’ is given in (2.2). (Earlier we saw a partial paradigm of two Sumerian verb lexemes (Section 1.1).) (2.2) The paradigm of filos singular nominative fílos accusative fílo genitive fílu
plural fíli fílus fílon
This paradigm contains six different word-forms and expresses notions of number (singular, plural) and case (nominative, accusative, genitive).3 By contrast, English nouns have no more than four word-forms (e.g. island: island, islands and perhaps island’s, islands’), but the notional distinction between lexemes and word-forms is no less important when the paradigm is small. In fact, for the sake of consistency we have to make the distinction even when a lexeme has just a single word-form, as in the case of many English adjectives (e.g. the adjective solid, which has only the word-form solid). It is not always immediately clear how many word-forms belong to a lexeme. This is shown by the paradigm of the Latin noun lexeme insula ‘island’ in (2.3). Are there ten word-forms in this lexeme’s paradigm, or seven? (2.3) The paradigm of insula singular nominative insula accusative insulam genitive insulae dative insulae ablative insulaˉ 3
plural insulae insulaˉs insulaˉrum insulıˉs insulıˉs
The meanings of the cases are discussed in Chapter 5. They are also given in the Glossary.
2 . 1 ╇ L E X E M E S A N D W O R D - F O R M S
17
Above we defined a word-form in terms of a lexeme and a set of grammatical functions. The importance of the latter part of the definition is seen in paradigms like insula. Although there are only seven different sequences of sounds in (2.3), we can still say that the paradigm of insula has ten wordforms, because ten different sets of grammatical functions are expressed (e.g. genitive singular and nominative plural are distinct, despite having the same form). Not all morphological relationships are of the type illustrated in (2.2) and (2.3). Different lexemes may also be related to each other, and a set of related lexemes is sometimes called a word family (though it should more properly be called a lexeme family): (2.4) Two English word families a. read, readable, unreadable, reader, readability, reread b. logic, logician, logical, illogical, illogicality Although everyone recognizes that these words are related, they are given their own dictionary entries. Thus, the difference between word-forms and lexemes, and between paradigms and word families, is well established in the practice of dictionary-makers, and thereby known to all educated language users. At this point we have to ask: why is it that dictionaries treat different morphological relationships in different ways? And why should linguists recognize the distinction between paradigms and word families? After all, linguists cannot base their theoretical decisions on the practice of Â� dictionary-makers – it ought to be the other way round: lexicographers ought to be informed by linguists’ analyzes. The nature of the difference between lexemes and word-forms will be the topic of Chapter 5, but the most important points will be anticipated here. (i) Complex lexemes (such as reader or logician) generally denote new concepts that are different from the concepts of the corresponding simple lexemes, whereas word-forms often exist primarily to satisfy a formal requirement of the syntactic machinery of the language. Thus, word-forms like reads or reading do not stand for concepts different from read, but they are needed in certain syntactic contexts (e.g. the girl reads a magazine; reading magazines is fun). (ii) Complex lexemes must be listed separately in dictionaries because they are less predictable than word-forms. For instance, one cannot predict that the lexeme illogicality exists, because by no means all adjectives have a corresponding -ity lexeme (cf. nonexistent words like *naturality, *logicality). It is impossible to predict that a specialist in logic should be called a logician (rather than, say, a *logicist), and the meaning of complex lexemes is often unpredictable, too: a reader can denote not just any person who reads, but also a specific academic position (in the
18 
C H A P T E R 2 â•… B A S I C C O N C E P T S
British system) or even a kind of book. By contrast, the properties of word-forms are mostly predictable and hence do not need to be listed separately for each lexeme. Thus, there are two rather different kinds of morphological relationship among words, for which two technical terms are commonly used: (2.5) Kinds of morphological relationship
inflection 0007(= inflectional morphology): the relationship between word-forms of a lexeme
derivation 0007(= derivational morphology): the relationship between lexemes of a word family
Morphologists also use the corresponding verbs inflect and derive. For instance, one would say that the Latin lexeme insula is inflected (or inflects) for case and number, and that the lexeme reader is derived from the lexeme read. A derived lexeme is also called a derivative. (Note that we are making a terminological simplification here: a lexeme is an abstract entity without phonological form so, strictly speaking, one lexeme cannot be derived from another. When morphologists talk about derived lexemes, they mean that form a (e.g. reader), corresponding to lexeme A (reader), is derived from form b (read), corresponding to lexeme B (read). However, since this phrasing becomes quite clumsy, morphologists commonly simplify the terminology. We will do the same in this book.) It is not always easy to tell how word-forms are grouped into lexemes. For instance, does the word-form nicely belong to the lexeme nice, or does it represent a lexeme of its own (nicely), which is in the same word family as nice? Issues of this sort will be discussed in some detail in Chapter 5. Whenever it is unclear or irrelevant whether two words are inflectionally or derivationally related, the term word will be used in this book instead of �lexeme or word-form. And for the same reason even the most technical writings on morphology often continue to use the term word. Some morphologically complex words belong to two (or more) word families simultaneously. For instance, the lexeme firewood belongs both in the family of fire and in the family of wood. Such relationships are called compounding, and lexemes like firewood are called compound lexemes, or just compounds, for short. Compounding is often grouped together with derivation under the category of word formation (i.e. lexeme formation). The various conceptual distinctions that we have seen so far are summarized in Figure 2.1.
2 . 2 ╇ A F F I X E S , B A S E S A N D R O O T S
19
morphological relationships inflection (‘word-form formation’)
paradigms: e.g. live, lives, living, . . . island, islands, . . .
word formation (‘lexeme formation’) derivation
compounding
word families: e.g. LOGIC, LOGICIAN, . . .
FIREWOOD
Figure 2.1╇ Subdivisions of morphology
2.2╇ Affixes, bases and roots In both inflection and derivation, morphemes have various kinds of meanings. Some meanings are very concrete and can be described easily (e.g. the meanings of the morphemes wash, logic, chameleon, un-), but other meanings are abstract and more difficult to describe. For instance, the morpheme -al in logic-al can perhaps be said to mean ‘relating to’ (cf. logical, mathematic-al, physic-al, natur-al), -able in read-able can be said to mean ‘capable of undergoing a process’, and the meaning of -ity is ‘quality’ (e.g. readability is ‘the quality of being readable’). Some meanings are so abstract that they can hardly be called meanings. For example, the Latin morpheme -m in insula-m (see (2.3)) serves to mark the direct object in a sentence, but it is difficult to say what its meaning is. And English -s in read-s is required when the subject is a third person singular noun phrase, but again it is unclear whether it can be said to have meaning. In such cases, linguists are more comfortable saying that these morphemes have certain grammatical functions. But, since the ultimate purpose of grammatical constructions is to express meaning, we will continue to say that morphemes bear meaning, even when that meaning is very abstract and can be identified only in the larger grammatical context. Word-forms in an inflectional paradigm generally share (at least) one longer morpheme with a concrete meaning and are distinguished from each other in that they additionally contain different shorter morphemes, called affixes. An affix attaches to a word or a main part of a word. It usually has an abstract meaning, and an affix cannot occur by itself. For instance, Russian nouns have different affixes in the paradigm in (2.6), which have case meaning (-a for nominative, -u for accusative, etc.), and Classical Nahuatl nouns have different affixes in the paradigm in (2.7) that indicate a possessor (no- for ‘my’, mo- for ‘your’, etc.).
20 
C H A P T E R 2 â•… B A S I C C O N C E P T S
(2.6) Russian case inflection (singular forms) nominative ruk-a ‘hand’ accusative ruk-u genitive ruk-i dative ruk-e locative ruk-e instrumental ruk-oj (2.7) Nahuatl possessor inflection 1sg no-cal ‘my house’ 2sg mo-cal ‘your (sg) house’ 3sg i-cal ‘his/her house’ 1pl to-cal ‘our house’ 2pl amo-cal ‘your (pl) house’ 3pl in-cal ‘their house’
(Sullivan 1988: 26)
Morphologists often use special terms for different kinds of affixes, depending on their position within the word. Affixes that follow the main part of the word are called suffixes (e.g. the Russian case suffixes in (2.6)), and affixes that precede it are called prefixes (e.g. the Classical Nahuatl possessor prefixes in (2.7)). The part of the word that an affix is attached to is called the base, e.g. ruk- in Russian, or -cal in Classical Nahuatl. Affixes and bases can, of course, be identified both in inflected word-forms and in derived lexemes. For instance, in read-er, read-able and re-read, read is the base, -er and -able are suffixes, and re- is a prefix. A base is also sometimes called a stem, especially if an inflectional (as opposed to derivational) affix attaches to it. There are still other kinds of affixes, besides prefixes and suffixes, which are briefly described and illustrated in Table 2.1. Types of affixes
Examples
suffix: follows the base Russian -a in ruk-a ‘hand’ English -ful in event-ful prefix: precedes the base 0007Classical Nahuatl no- in no-cal ‘my house’ English un- in unhappy infix: occurs inside the base 0007Arabic -t- in (i)š-t-ag∙ala ‘be occupied’ (base: šag∙ala) 0007Tagalog -um- in s-um-ulat ‘write’ (base: sulat) circumfix: occurs on both sides 0007German ge-…-en, e.g. ge-fahr-en ‘driven’ of the base (base: fahr) Table 2.1╇ Types of affixes
2 . 2 ╇ A F F I X E S , B A S E S A N D R O O T S
21
Bases or stems can be complex themselves. For instance, in activity, -ity is a suffix that combines with the base active, which itself consists of the suffix -ive and the base act. A base that cannot be analyzed any further into constituent morphemes is called a root. In readability, read is the root (and the base for readable), and readable is the base for readability, but it is not a root. Thus, the base is a relative notion that is defined with respect to the notion ‘affix’. (We will refine this definition of ‘base’ in the next chapter to account for words which are difficult to describe in terms of morphemes, but will keep the idea that bases are relative notions.) Affixes are similar to roots in that they cannot be further analyzed into component morphemes; they are primitive elements. A base may or may not be able to function as a word-form. For instance, in English, cat is both the base of the inflected form cats and itself a wordform (active is a word-form and the base for the derived form activity, etc.). However, in Italian word-form gatti (‘cats’) can be broken up into the suffix -i (‘plural’) and the Â�base gatt- (‘cat’), but gatt- is not a word-form. Italian nouns must inflect for number, and even in the singular, an affix is required to express this information (e.g. gatt-o ‘cat’, gatt-i ‘cats’). In this respect Italian differs from English. Bases that cannot also function as word-forms are called bound stems. Roots and affixes can generally be distinguished quite easily, but sometimes there are problems. For example, the Salishan language Bella Coola has a number of suffix-like elements that do not seem to have an abstract meaning (see 2.8). In (2.9), we see two examples of how these elements are used. (2.8) -us -an -uc -al˛ -ak (2.9) a.
‘face’ -lik ‘body’ ‘ear’ -altwa ‘sky, weather’ ‘mouth’ -lt ‘child’ ‘foot’ -lst ‘rock’ ‘hand’ -lx.s ‘nose’ quc´-a-ic wash-foot-I.him ‘I am going to wash his foot’ (lit.: ‘foot-wash him’)
b.
kma-lx.s-c hurt-nose-I ‘my nose hurts’ (lit.: ‘I nose-hurt’)
(Mithun 1998: 300–5)
In these cases, it is not immediately clear whether we are dealing with suffix– root combinations or with root–root combinations, i.e. compounds. The elements in (2.8) do not occur as lexemes by themselves but must always be combined with other roots. In this respect they have a property that is typical of affixes, and scholars of Salishan languages have generally regarded them as such. However, if affixes are defined as ‘short morphemes with an abstract meaning’, then these elements are very atypical affixes, to say the least.
22 
CHAPTER 2â•… BASIC CONCEPTS
English has a number of morphemes that are similarly difficult to classify as roots or affixes. Some examples are given in (2.10). (2.10) biogeography aristocrat bioethics autocrat bioengineering democrat biorhythm Eurocrat bioterrorism plutocrat biomedicine technocrat biochip theocrat The elements bio- and -crat could be regarded as affixes because they do not occur as independent lexemes, but their very concrete meaning and also their (not particularly short) form suggest that they should be regarded as bound stems that have the special property of occurring only in compounds.
2.3╇ Morphemes and allomorphs While the distinction between roots on the one hand and affixes on the other is by itself quite useful, these concepts turn out to be more complicated than the simple picture that we have seen so far. One of the most common complications is that morphemes may have different phonological shapes under different circumstances. For instance, the plural morpheme in English is sometimes pronounced [s] (as in cats [kæts]), sometimes [z] (as in dogs [dɒgz]), and sometimes [-əz] (as in faces [feisəz]). When a single affix has more than one shape, linguists use the term allomorph. Affixes very often have different Â� allomorphs – two further cases from other languages are given in (2.11). (2.11) a. Korean accusative suffix (marker of direct object): two allomorphs -ul ton ‘money’ ton-ul ‘money-acc’ chayk ‘book’ chayk-ul ‘book-acc’ -lul tali ‘leg’ tali-lul ‘leg-acc’ sakwa ‘apple’ sakwa-lul ‘apple-acc’ b. Turkish first person possessive suffix: five allomorphs -im ev ‘house’ ev-im ‘my house’ dil ‘language’ dil-im ‘my language’ -üm köy ‘village’ köy-üm ‘my village’ gün ‘day’ gün-üm ‘my day’ -um yol ‘way’ yol-um ‘my way’ tuz ‘salt’ tuz-um ‘my salt’ -ım 4 ad ‘name’ ad-ım ‘my name’ kız ‘girl’ kız-ım ‘my daughter’ -m baba ‘father’ baba-m ‘my father’ 4
The Turkish letter ı corresponds to IPA [ɯ] (high back unrounded vowel).
2.3╇ MORPHEMES AND ALLOMORPHS 
23
Not only affixes, but also roots and stems may have different allomorphs (or, as linguists often say, ‘exhibit allomorphy’). For instance, English verbs such as sleep, keep, deal, feel, mean, whose root has the long vowel [i:] in the present-tense forms, show a root allomorph with short [ε] in the past-tense forms (slept, kept, dealt, felt, meant). Cases of stem allomorphy from other languages are given in (2.12). (2.12) a. 00070007German: a voiced obstruent becomes voiceless in syllable-final position Tag [ta:k] ‘day’ Tage [ta:gə] ‘days’ Hund [hυnt] ‘dog’ Hunde [hυndə] ‘dogs’ Los [lo:s] ‘lot’ Lose [lo:zə] ‘lots’ b. 0007Russian: when the stem is followed by a vowel-initial suffix, the vowel o/e is often dropped if it is the last vowel in the stem zamok ‘castle’ zamk-i ‘castles’ kamen’ ‘stone’ kamn-i ‘stones’ nemec ‘German’ nemc-y ‘Germans’ nogot’ ‘nail’ nogt-i ‘nails’ The crucial properties which define the German stems [ta:k] and [ta:g] or the Korean suffixes [-ul] and [-lul] as being allomorphs are that they have the same meaning and occur in different environments in complementary distribution. Additionally, all our examples so far have shown only fairly small differences in the shapes of morphemes, which can by and large be regarded as mere Â�differences in pronunciation. Being phonologically similar is a common property of allomorphs, but is not a necessary one. Allomorphs that have this property are phonological allomorphs. The formal relation between two (or more) phonological allomorphs is called an alternation. Linguists often describe alternations with a special set of morphophonological rules, which were historically phonetically motivated, but affect morphology. Morphophonological rules and the difference between them will be discussed more extensively in Chapter 10, and we will consider them only briefly here. Metaphorically, it is often convenient to think about phonological allomorphy in terms of a single underlying representation that is manipulated by rules under certain conditions. The end result, i.e. what is actually pronounced, is the surface representation. For instance, the alternations in (2.12a, b) can be described by the underlying representations in the (a) examples below, and by the respective rules in the (b) examples. The surface representations (resulting word-forms) are given in (c). (2.13) a. underlying: [ta:g] ‘day.sg’ b. 0007rule: 0007a voiced obstruent becomes voiceless in syllable-final position (application: [ta:g] Æ [ta:k])5 c. surface: [ta:k] ‘day.sg’
5
In this (morpho)phonological context, the arrow (‘XÆY’) means that X turns into Y.
24 
C H A P T E R 2 â•… B A S I C C O N C E P T S
(2.14) a. underlying: [ta:g-ə] ‘day-pl’ b. rule: 0007a voiced obstruent becomes voiceless in syllable-final position (doesn’t apply) c. surface: [ta:g-ə] ‘day-pl’ (2.15) a. underlying: [zamok] ‘castle.sg’ b. 0007rule: 0007o/e in the final stem syllable disappears when the stem is followed by a vowel-initial suffix (doesn’t apply) c. surface: [zamok] ‘castle.sg’ (2.16) a. underlying: [zamok-i] ‘castle-pl’ b. 0007rule: 0007o/e in the final stem syllable disappears when the stem is followed by a vowel-initial suffix (application: [zamok-i] Æ [zamk-i]) c. surface: [zamk-i] ‘castle-pl’ Notice that for (2.13) and (2.14), the underlying representation (morpheme) meaning ‘day’ is the same, and the rule applies only when its conditions are met. The same is true for (2.15) and (2.16). That the alternation is produced by the morphophonological rule is made particularly clear in this way: the underlying representation shows no allomorphy at all. In many cases of phonological allomorphy, it is evident that the historical reason for the existence of the morphophonoÂ�logical rule and thus for the allomorphy is to facilitate pronunciation. For instance, if the English plural were uniformly [-z], words such as cats and faces would be almost unpronounceable (try to pronounce [kætz] and [feisz]!). Since this is a textbook on morphology, we cannot go into greater phonological detail here, but phonological allomorphs will be taken up again in Chapter 10. Overall, the main point here is that at some level, phonological allomorphs represent a single morpheme whose form varies slightly depending upon the phonological context created by combining morphemes. For this reason, it is common to think of the morpheme as the more abstract underlying representation, rather than the more concrete surface word-form. The underlying and surface representations may be the same, or they may differ as a result of the application of morphophonological rules. However, it is important to remember that the underlying representation is a tool used by linguists. It may or may not reflect the kinds of generalizations that language users make. There are examples where it seems unlikely that there is a single underlying representation in the minds of speakers; we see this in another type of allomorphy: suppletion. Besides phonological allomorphs, morphemes may also have allomorphs that are not at all similar in pronunciation. These are called suppletive allomorphs. For instance, the English verb go has the suppletive stem wen in the past tense (wen-t), and the English adjective good has the suppletive stem bett in the comparative degree (better). The Russian noun cˇelovek ‘human being’ has the suppletive stem ljud’ in the plural (ljud-i ‘people’).
2 . 3 ╇ M O R P H E M E S A N D A L L O M O R P H S
25
The Spanish verb ir ‘go’ has the suppletive stem va- in the present tense (vas ‘you go’, va ‘s/he goes’, vamos ‘we go’, etc.). The term suppletion is most often used to refer to stem shape (ir and va- are both verbal stems), and some linguists reserve the term for this use, but others also talk about affixes as being potentially suppletive (see (2.17) later in this chapter for an example from Persian). It is not always easy to decide whether an alternation is phonological or suppletive, because the categories are end points on a continuum of traits, rather than a clear-cut binary distinction. Some examples are therefore intermediary. For instance, what about English buy/bought, catch/caught, teach/taught? The root allomorphs of these verbs ([bai]/[bɔ], [kæt∫]/[kɔ:], [tit∫]/[tɔ]) are not as radically different as go/wen-t, but they are not similar enough to be described by phonological rules either. In such cases, linguists often speak of weak suppletion, as opposed to strong suppletion in cases like go/went, good/better. For both weak and strong suppletion, it is theoretically possible to posit an underlying representation from which suppletive allomorphs are derived by rule. However, considering that suppletive allomorphs share little or no form, the underlying representation would need to be very abstract, and the rules converting the underlying representation to surface representations could not exist to make pronunciation easier. There is no evidence that language users make such abstractions, so underlying representations are perhaps best treated as useful metaphors. Type of allomorphy Description
Example
Phonological allomorphy
Alternation could be described by a rule of pronunciation
English plural [-z], [-s],[-əz]; Russian zamok/zamk-
Weak suppletive allomorphy
Allomorphs exhibit some similarity, but this cannot be described by phonological rules
English buy/bough-, catch/ caugh-, etc.
Strong suppletive allomorphy
Allomorphs exhibit no English good/bettsimilarity at all
Table 2.2╇ Types of allomorphy: summary
When describing the allomorphy patterns of a language, another important dimension is the conditioning of the allomorphy, i.e. the conditions under which different allomorphs are selected. Phonological allomorphs typically have phonological conditioning. This means that the phonological context
26 
C H A P T E R 2 â•… B A S I C C O N C E P T S
determines the choice of allomorph. For instance, the English plural allomorphs [-z], [-s] and [-əz] are strictly phonologically conditioned: [-əz] appears after a sibilant (i.e. [s], [z], [ʃ], [Z], [tS] or [dZ], e.g. face-s, maze-s, bush-es, garage-s, church-es, badge-s), [-s] appears after a voiceless non-sibilant obstruent (e.g. cat-s, book-s, lip-s, cliff-s) and [-z] appears elsewhere (e.g. bag-s, bell-s, key-s). The Korean accusative allomorphs -ul/-lul (see (2.11a)) are also phonologically conditioned: -ul appears after a consonant, -lul after a vowel. By contrast, stem suppletion usually has morphological conditioning, meaning that the morphological context (usually, grammatical function) determines the choice of allomorph (e.g. Spanish ir ‘go’ in the infinitive and future tense, va- in the present and imperfective past tense and fu- in the perfective past tense).6 And, finally, we find lexical conditioning, where the choice of a suppletive affix allomorph is dependent on other properties of the base, for instance semantic properties as in (2.17). (2.17) Persian plural marking: human nouns -an, non-human nouns -ha -an mœrd ‘man’ mœrd-an ‘men’ geda ‘beggar’ geday-an ‘beggars’ -ha gorbe ‘cat’ gorbe-ha ‘cats’ ettefaq ‘incident’ ettefaq-ha ‘incidents’ (Mahootian 1997: 190) Lexical conditioning is also involved where the choice of allomorph cannot be derived from any general rule and must be learned individually for each word. This is the case for the English past participle suffix -en: speakers must simply learn which verbs take this suffix and not the more common suffix -ed. Type of conditioning
Description
Example
Phonological conditioning
Choice of allomorphs depends on phonological context
English plural depends on final sound in stem
Morphological conditioning
Choice of allomorphs depends on the morphological context Choice of allomorphs depends on the individual lexical item
Spanish ir, va- or fu-, depending on tense
Lexical conditioning
English past participle -en/-ed is unpredictable and depends on individual verbs
Table 2.3╇ Types of conditioning: summary 6
It is clear that phonological allomorphs can also have morphological conditioning. However, whether suppletive allomorphs can have phonological conditioning is subject to ongoing debate.
A P P E N D I X . M O R P H E M E - B Y- M O R P H E M E G L O S S E S
27
Summary of Chapter 2 This chapter introduced several concepts that are basic to morphology. Three different notions of word have to be distinguished: the word token (‘use of a word in a text or in speech’), the lexeme (‘abstract, dictionary word’) and the word-form (‘concrete word’). Inflectional morphology describes the relationship between the word-forms in a lexeme’s paradigm, and derivational morphology describes the relation between lexemes. Complex words can often be segmented into morphemes, which are called affixes when they are short, have an abstract meaning, and cannot stand alone, and roots when they are longer and have a more concrete meaning. When two or more morphemes express the same meaning and occur in complementary distribution, they are often considered allomorphs. Allomorphs come in two types, phonological and suppletive, depending on the degree to which they are similar in form. Suppletive allomorphs are further subdivided into examples of strong suppletion and weak suppletion. The distinction between strong suppletion, weak suppletion and non-suppletion is a continuum. Allomorphs may have phonological, morphological, or lexical conditioning.
Appendix. Morpheme-by-morpheme glosses When presenting longer examples (such as sentences or entire texts) from a language that the reader is unlikely to know, linguists usually add interÂ�linear morpheme-by-morpheme glosses to help the reader understand the structure of the examples. We saw instances of such glosses in (1.2)–(1.6), and we will see more examples later in this book. Interlinear morpheme-by-morpheme glosses are an important aspect of ‘applied morphology’, and they are needed in other areas of linguistics as well (e.g. by syntacticians and fieldworkers). We will therefore explain the most important principles involved. The conventions used in this book are based on the Leipzig Glossing Rules (www.eva.mpg.de/lingua/resources/glossing-rules.php; accessed July 2010). The Leipzig Glossing Rules are more detailed than the principles presented here, but include the following: (i) One-to-one correspondence. Each element of the object language is translated by one element of the metalanguage (in the present context, this is English). Hyphens separate both the word-internal morphemes in the object language and the gloss, e.g. Japanese Taroo ga hana o migotoni sak-ase-ta. Taro nom flower acc beautifully bloom-caus-pst ‘Taro made the flowers bloom beautifully.’ (Shibatani 1990: 309)
28 
C H A P T E R 2 â•… B A S I C C O N C E P T S
Object-language words and their glosses are aligned at their left edges. The interÂ�linear gloss is usually followed by an idiomatic translation in quotation marks. (ii) Grammatical-value abbreviations. Grammatical elements (both function words and inflectional affixes) are not translated directly, but are rendered by grammatical-value labels, generally in abbreviated form (see the list of abbreviations on pp. xv–xvi). To highlight the difference between the value labels and the ordinary English words, the value labels are usually printed in small capitals, as seen in the above example. (iii) Hyphens and periods. Hyphens are used to separate word-internal morphemes in object-language examples, and each hyphen in an example corresponds to a hyphen in the gloss. Periods are used in the gloss when two gloss elements correspond to one element in the example. This may be when a single example element corresponds to a multi-word expression in the gloss, e.g. Turkish çık-mak come.out-inf ‘to come out’ or it may be when a single example element corresponds to several inflectional meanings: Latin insul-aˉrum island-gen.pl ‘of the islands’ or it may be when an inflectional meaning is expressed in a way that cannot be segmented, e.g. Albanian fik fig.sg ‘fig’
fiq fig.pl ‘figs’
(The Albanian letter q corresponds here to IPA [c] (voiceless palatal stop), and k corresponds to [k] (voiceless velar stop).) The period is omitted when the two meanings are person and number, e.g. Tzutujil x-in-wari compl-1sg-sleep ‘I slept’
(Dayley 1985: 87)
COMPREHENSION EXERCISES 
29
Here ‘1sg’ is used instead of ‘1.sg’. (The period is felt to be redundant because person and number combine so frequently.) (iv) Possible simplifications. Sometimes the precise morpheme division is irrelevant or perhaps unknown. Authors may still want to give inforÂ� mation on the inflectional meanings, and again periods are used to separate these elements, e.g. Japanese sakaseta bloom.caus.pst ‘made to bloom’
Latin insulaˉrum island.gen.pl ‘of the islands’
Sometimes morpheme-by-morpheme glosses are used also when the example is not set off from the running text. In such cases the gloss is enclosed in square brackets, e.g. ‘the Japanese verb saka-se-ta [bloom-causpst] “made to bloom” ….’.
Comprehension exercises 1. Somali exhibits a great amount of allomorphy in the plural formation of its nouns. Four different allomorphs are represented in the following examples. Based on these examples, formulate a hypothesis about the phonological conditions for each of the plural allomorphs. (In actual fact, the conditions are more complex, but for this exercise, we have to limit ourselves to a subset of the data and generalizations.)
singular awowe baabaco beed buug cashar fure ilmo miis qado shabeel waraabe xidid
plural awowayaal baabacooyin beedad buugag casharro furayaal ilmooyin miisas qadooyin shabeello waraabayaal xididdo
Based on the generalizations found, form the plural of the following nouns:
tuulo tog
‘grandfather’ ‘palm’ ‘egg’ ‘book’ ‘lesson’ ‘key’ ‘tear’ ‘table’ ‘lunch’ ‘leopard’ ‘hyena’ ‘eagle’
‘village’ ‘river’
30 
C H A P T E R 2 â•… B A S I C C O N C E P T S
albaab buste
‘door’ ‘blanket’
(Berchem 1991: 98–117)
2. The English past participle suffix spelled -ed has three different alternants: [d], [t], and [əd]. Are these phonologically or morphologically conditioned? Try to describe the conditioning factors in an approximate way. 3. Italian inhabitant nouns (e.g. Anconetano ‘person from Ancona’) exhibit different degrees of similarity to the corresponding city names. Order the following pairs of city names and inhabitant names on a scale from clear suppletion in the base form to clear non-suppletion, depending on the number of segments in which the base for the inhabitant noun differs from the base for the city name (see Crocco-Galèas 1991). Assume that word-final vowels are suffixes in Italian; the base for Ancona would thus be Ancon-. Additionally, inhabitant nouns contain the suffixes -an, -in, or -es, so the base for Anconetano is Anconet-. city name Ancona Bologna Bressanone Domodossola Gubbio Ivrea Milano Napoli Palermo Palestrina Piacenza Savona Trento Treviso Venezia Volterra
inhabitant noun Anconetano Petroniano Brissinese Domese Eugubino Eporediese Milanese ‘Milan’ Partenopeo ‘Naples’ Palermitano Prenestino Piacentino Savonese Trentino Trevigiano Veneziano ‘Venice’ Volaterrano
Exploratory exercise This chapter introduced the idea that the set of word-forms belonging to the same lexeme is known as a paradigm. Readers may have noticed that a table-like format was used to list members of paradigms. The Modern Greek noun paradigm that we encountered in (2.2) is repeated below as (2.18). Here, the rows list cases and the columns list numbers. This format is sometimes called a grid. The grid format will be used elsewhere in the book, especially in Chapters 5 and 8, where inflectional morphology is discussed.
E X P L O R AT O R Y E X E R C I S E
31
(2.18) The paradigm of filos ‘friend’ singular plural nominative fílos fíli accusative fílo fílus genitive fílu fílon The grid format subtly implies that for a given lexeme there should be a word-form corresponding to each combination of case and number. The format makes sense because the expectation is usually fulfilled; in Greek, noun lexemes almost always have six word-forms corresponding to the six cells in the grid. There is thus some sense in which paradigms can be ‘complete’ or ‘incomplete’. In this exercise, you will explore whether the same notion applies to word families. Do word families usually have an equal number of members and parallel content? For instance, if the verb lexeme read has readable, unreadable, reader and reread in its word family, does every other verb lexeme X also have Xable, unXable, Xer and reX? Or do word families often have some lexemes, but not others that seem equally possible? Finally, does the notion of ‘completeness’ apply to word families? Is it reasonable to talk about a word family as being incomplete? We will address some of these questions in Chapters 5 and 6, but in this exercise you will anticipate that discussion with some exploratory analysis. English is used here for demonstration purposes because it is familiar to all readers, but you are encouraged to investigate your native language, whatever that might be. Instructions Step 1: Create a list of at least 20 adjectival (or nominal, verbal…) lexemes, e.g. clear, false, happy. For each one, list all of the lexemes belonging to its word family. Use a dictionary to prod your memory if needed, but do not rely on dictionary entries when they contradict your own judgements. For instance, the Oxford English Dictionary lists the following entries (among others) as being related to the adjective happy: happify, happiless, happily, happiness, happious, happy-go-lucky, happy-slappy, enhappy, mishappy, and trigger-happy. Some of these, like happiness, are quite normal, but others, like happify, happious, and enhappy, seem odd at best. For the authors, the word family of happy does not contain these three lexemes. For each word family in your data set, decide its content for yourself, according to your own usage and judgements about whether a given lexeme is possible. Step 2: Compare the sets. The lexeme clear is in the same set as clarify, and false is related in a parallel fashion to falsify, but the word family for happy does not contain happify (despite being in the dictionary). Do the word families in your data set mostly have parallel content, or mostly not? Step 3: Discuss the content of these word families in terms of the following questions:
32 
C H A P T E R 2 â•… B A S I C C O N C E P T S
1) In your data, was it ever hard to decide whether two lexemes belong to the same word family? If so, why? Discuss the issues surrounding any choices you had to make. 2) What kinds of meanings are expressed by the derivationally related lexemes? It is not important at this stage to be precise about terminology – describe them as best you can. How do these compare to the inflectional meanings that you have seen in this chapter? Can derivationally related lexemes be organized into grids in the way inflectionally related wordforms are? Why or why not? 3) Does it make sense to talk about word families as complete, or at least potentially complete? Are paradigms and word families similar or different in this respect? Explain your reasoning. (Both a ‘yes’ and a ‘no’ answer to the question is probably possible. The important part is that you explain and justify your answer.)
Rules
33
3
S
o far we have talked about morphological structure in mostly Â�static terms: words ‘have’ affixes or ‘share’ parts, they ‘exhibit’ resemblances and they ‘consist of’ a base and an affix. However, it is often convenient to describe complex words as if they were the result of a process or event. Thus, we said that affixes ‘are attached’ to the base or that they ‘combine’ with it. Linguists use such process terms very frequently. They talk about elements ‘being affixed’ to bases, or about a complex word ‘being derived from’ (i.e. built on the basis of) a simpler one.1 Most linguists agree that complex words need not be derived from simpler ones each time they are used. Instead, frequently used words are probably listed in the lexicon. The lexicon is the linguist’s term for the mental dictionary that language users must be equipped with, in addition to the grammatical rules of their language. When a linguist says that something is listed in the lexicon, this means that it must be stored in speakers’ memories. If a complex word has its own lexical entry (i.e. listing in the lexicon), it does not need to be actively derived from a simpler form; it can simply be retrieved from memory when needed. (The content of the lexicon will be discussed in Chapter 4.) Still, speakers have the capacity to create, and hearers can understand, an almost unlimited number of new words. The set of words in a language is never quite fixed. There must therefore be some processes by which new complex words are created. And even when a complex word is likely to be listed in the lexicon, it is useful to think of the relationship between it and its base in terms of these same processes. These processes, and how they can be formally described using morphological rules, are the topic of this chapter.
1
The term derive is somewhat confusing because it is also commonly applied to inflectional morphology, not just to derivational morphology. Thus, one would say that the comparative form warmer is derived from the positive form warm, or that the past-tense form played is derived from the present-tense form play.
34 
C H A P T E R 3 â•… R U L E S
3.1╇ Morphological patterns Morphological structure is much more various than simply affixes combining with bases. For example, in German, one way of forming the plural of a noun is by replacing a back vowel of the singular form (e.g. [υ], [a:], [ɔ]) by a front vowel (e.g. [y], [e:], [ø], spelled ü, ä, ö). Some examples are given in (3.1). (3.1)
singular Mutter Vater Tochter Garten Nagel
plural Mütter Väter Töchter Gärten Nägel
‘mother(s)’ ‘father(s)’ ‘daughter(s)’ ‘garden(s)’ ‘nail(s)’
Here, we have a clear-cut example of morphological structure in that a recurrent aspect of meaning (‘plural’) corresponds to a recurrent aspect of form (vowel quality), but the plural word-forms cannot be segmented into two morphemes. Intuitively, it is easier to think of the stem vowel having been changed, rather than a morpheme having been added. We will use the term morphological pattern to cover both examples in which morphological meaning can be associated with a segmentable part of the word, and examples where this is not possible. A morpheme, then, is a frequently occurring, special subtype of morphological pattern. We begin by examining a range of morphological patterns, both common and uncommon, from various languages.
3.1.1╇ Affixation and compounding Linguists often distinguish two basic types of morphological patterns: concatenative, which is when two morphemes are ordered one after the other, and non-concatenative, which is everything else. Most of the examples of morphologically complex words that we have seen so far can be neatly segmented into roots and affixes, and are therefore concatenative patterns. In process terms, these can be described as derived by affixation (subtypes suffixation, prefixation, etc.) and compounding. Affixation involves more than just combining two morphemes. A rule of affixation is also a statement about which types of morphemes may combine. This is the combinatory potential of the affix. (Other terms that are widely used are subcategorization frame and selectional restriction.) For example, un- and intelligent may combine via affixation to form unintelligent, but it is not the case that any affix and any base can combine. The suffix -able attaches only to verbs; intelligentable is not a potential word of English because intelligent is an adjective, not a verb. And un- can attach
3 . 1 ╇ M O R P H O L O G I C A L P A T T E R N S
35
to adjectives, but does not generally attach to nouns; ungrass is also not a possible word of English (although the soft drink company 7UP played on this restriction to grab attention with the slogan ‘7UP: the uncola’). The combinatory potential of an affix cannot be entirely predicted from its meaning. For example, the prefix non- is virtually identical in meaning to un-, but it commonly attaches to nouns (e.g., non-achiever) and less readily to adjectives (non-circular, but *non-kind, ??non-intelligent). Combinatory potential must therefore be specified along with other information about the affixation process. As with un-, non- and -able, the word-class of the base (noun, verb, adjective, etc.) is an important factor for combinatory potential. Linguists thus sometimes say that affixes ‘select’ a particular word-class to attach to. The combinatory potential of the prefix un- can be expressed with the notation in (3.2a), where ‘—’ stands for the affix and ‘A’ indicates both the word-class of the base and the position of the base relative to the affix. (3.2)
a. Combinatory potential of un- b. Combinatory potential of -able c. Combinatory potential of comparative -er d. Combinatory potential of -ful
[— A ] [V—] [A— ] [N—]
Affixation is thus a process that has a number of important parameters (we will encounter more later in the book), but which can nonetheless be described in a fairly straightforward way.
3.1.2╇ Base modification At the same time, a range of morphological patterns exists that cannot be straightforwardly segmented into two meaningful parts. As with the German example in (3.1), it is often easiest to describe non-concatenative patterns as results of processes or operations that apply to a base form. Some non-concatenative patterns exist in a wide variety of languages, including English, and will probably be familiar to the reader. Others may seem more ‘exotic’. Example (3.3) shows a non-concatenative pattern in Albanian. Here, a stem-final [k] in the singular becomes [c] in the plural, [g] becomes [J¯], and [] becomes [j]: (3.3)
singular armik […k] fik […k] frëng […g] murg […g] papagall […] portokall […]
plural armiq […c] ‘enemy/enemies’ fiq […c] ‘fig(s)’ frëngj […J¯] ‘Frenchman/-men’ murgj […J¯] ‘monk(s)’ papagaj […j] ‘parrot(s)’ portokaj […j] ‘orange(s)’ (Buchholz and Fiedler 1987: 264–5)
36 
C H A P T E R 3 â•… R U L E S
Non-concatenative patterns force us to revise our original definition of base (‘the part of a word that an affix is attached to’; Section 2.2). We want to say that armik is the base for armiq, but our earlier definition is adequate only in the context of concatenation. Thus, (3.4) is a better definition of base. (3.4) The base of a morphologically complex word is the element to which a morphological operation applies. This subsumes the earlier definition and allows us to also talk about nonconcatenative patterns in a satisfactory way. One important class of non-concatenative patterns is base modification (or stem modification/alternation). This is a collective term for morphological patterns in which the shape of the base is changed without adding segmentable material. A common type of base modification pattern results from changing place of articulation. For example, the Albanian example in (3.3) involves palatalization of the last consonant of the base (producing the sound at the palate), and the German example in (3.1) consisted of fronting of the stem vowel (changing the place of articulation so that the vowel is pronounced more towards the front of the mouth). We can also find examples of morphological patterns involving changed manner of articulation. In Scottish Gaelic, indefinite nouns undergo weakening of word-initial obstruent consonants in the genitive plural. Here, stop consonants become fricatives: [b] becomes [v], [kj] becomes [ç], [g] becomes [], and [th] becomes [h]. (Note that for some pairs, a change in place of articulation also occurs.) (3.5)
nom sg indf [b…] bard [kj…] ceann [g…] guth [th…] tuagh [b…] balach
gen pl indf [v…] bhàrd [ç…] cheann […] ghuth [h…] thuagh [v…] bhalach
‘bard’ ‘head’ ‘voice’ ‘axe’ ‘boy’
(Calder 1923: 81–93)
Standard Arabic, Quechua and Hindi/Urdu have morphological patterns involving length. In Standard Arabic, a causative verb is formed by gemination. This means that a consonant becomes lengthened, in this case the second consonant in the root (e.g. darasa ‘learn’ Æ darrasa ‘teach’, waqafa ‘stop (intr.)’ Æ waqqafa ‘stop (tr.)’, damara ‘perish’ Æ dammara ‘annihilate’).2 In Huallaga Quechua, the first person singular of verbs is formed in part by lengthening the final stem vowel (in phonology, long vowels are usually indicated by a colon). 2
The arrow symbol (Æ) is used to express a relationship of derivation: ‘A Æ B’ means that B is derived from A.
3 . 1 ╇ M O R P H O L O G I C A L P A T T E R N S
(3.6)
2nd singular aywa-nki ‘you go’ aywa-pti-ki ‘when you went’ aywa-shka-nki ‘you have gone’
37
1st singular aywa: ‘I go’ aywa-pti: ‘when I went’ aywa-shka: ‘I have gone’ (Weber 1989: 99, 118)
By contrast, in Hindi/Urdu, intransitive verbs are formed from transitive verbs by shortening the stem vowel (e.g. ma:r- ‘kill’ Æ mar- ‘die’, kho:l‘open (tr.)’ Æ khul- ‘open (intr.)’, phe:r- ‘turn (tr.)’ Æ phir- ‘turn (intr.)’). Base modification also commonly takes the form of a tonal change or stress shift. For example, in Chalcatongo Mixtec, adjectives are formed from nouns by changing the tone pattern of the base to a high–high pattern (indicated by two acute accents): (3.7)
noun káʔ ba ‘filth’ žuù ‘rock’ xaʔ à ‘foot’
adjective káʔ bá ‘dirty’ žúú ‘solid, hard’ xáʔ á ‘standing’ (Macaulay 1996: 64)
Somewhat similarly, English has verbs that differ from their corresponding nouns only by stress placement (e.g. díscount (noun) ´ discóunt (verb), ímport (noun) ´ impórt (verb), ínsult (noun) ´ insúlt (verb)). English also has a few cases where a verb is derived from a noun by a different operation – voicing the last consonant of the root (e.g. hou[s]e (noun) Æ hou[z]e (verb), thie[f] (noun) Æ thie[v]e (verb), wrea[q] (noun) Æ wrea[ð]e (verb)). Finally, two interesting but less commonly attested morphological patterns result from subtraction (the signalling of a morphological relationship by deleting one or more segments from the base), and metathesis (switching of two or more segments within the base). For example, one way of forming the plural in Murle is by subtracting the last consonant: (3.8)
singular nyoon wawoc onyiit rottin
plural nyoo wawo onyii rotti
‘lamb(s)’ ‘white heron(s)’ ‘rib(s)’ ‘warrior(s)’
(Arensen 1982: 40–1)
And Clallam marks a distinction between actual and non-actual events by metathesis of the first vowel with the preceding consonant. (Clallam verbs must have suffixes marking the agent, but only the stem is given here to keep the data simple.)
38 
C H A P T E R 3 â•… R U L E S
(3.9)
non-actual qq’í- pkw’´ə - t’c´ə - kw’s´ə -
actual qíq’- p´ə kw’- t’´ə c- kw’´ə s-
‘restrain’ ‘smoke’ ‘shatter’ ‘count’ (Thompson and Thompson 1969: 216)
The examples in this section show that while many base modification patterns may seem odd to the English speaker, there are few inherent restrictions on how morphological relationships can be signalled. In addition to adding segments, morphological patterns can be formed by deleting, rearranging, lengthening, shortening, weakening, palatalizing, etc. Also, non-concatenative morphological processes are similar to concatenative processes in having restrictions that are equivalent to combinatory potential. For example, the tonal change pattern shown in (3.7) applies only to nouns, and the resulting complex word is an adjective. This is comparable to -able applying only to verbs, with the resulting complex word being an adjective. We can therefore think of combinatory potential as a restriction that applies broadly to all morphological processes, and not only to affixation.
3.1.3╇ Reduplication A very common morphological operation is reduplication, whereby part of the base or the complete base is copied and attached to the base (either preceding or following it). In Malagasy, adjectives with stress on the first syllable copy the entire base. In the reduplicated form the meaning of the adjective is less intense. (3.10)
Reduplication of entire stem: Malagasy be ‘big, numerous’ be-be ‘fairly big, numerous’ fotsy ‘white’ fotsi-fotsy ‘whitish’ maimbo ‘stinky’ maimbo-maimbo ‘somewhat stinky’ hafa ‘different’ hafa-hafa ‘somewhat different’ (Keenan and Polinsky 1998: 571)
In Ponapean and Mangap-Mbula, only part of the base is copied. In Ponapean a consonant + vowel (CV) sequence is prefixed to the stem, whereas in Mangap-Mbula a vowel + consonant (VC) sequence is suffixed to the stem. (3.11)
Reduplication of a CV sequence before the base: Ponapean duhp ‘dive’ du-duhp ‘be diving’ mihk ‘suck’ mi-mihk ‘be sucking’ wehk ‘confess’ we-wehk ‘be confessing’ (Rehg 1981: 78)
3 . 1 ╇ M O R P H O L O G I C A L P A T T E R N S
(3.12)
39
Reduplication of a VC sequence after the base: Mangap-Mbula kuk ‘bark’ kuk-uk ‘be barking’ kel ‘dig’ kel-el ‘be digging’ kan ‘eat’ kan-an ‘be eating’ (Bugenhagen 1995: 53)
The element that is attached to the base often consists of both copied segments and fixed segments, so that a kind of mixture between affix and reduplicant results. Such elements may be called duplifixes. (3.13)
Plurals in Somali: duplifix -aC buug ‘book’ buug-ag fool ‘face’ fool-al koob ‘cup’ koob-ab jid ‘street’ jid-ad
‘books’ ‘faces’ ‘cups’ ‘streets’
(3.14) ‘Sort of’ adjectives in Tzutujil: duplifix -Coj saq ‘white’ saq-soj ‘whitish’ rax ‘green’ rax-roj ‘greenish’ q’eq ‘black’ q’eq-q’oj ‘blackish’ tz’iil ‘dirty’ tz’il-tz’oj ‘dirtyish’
(Berchem 1991: 102)
(Dayley 1985: 213)
Linguists often treat reduplication as affixation of a template and copying of the root as needed to fill out the segments of that template. For Ponapean, the template is CV-, where C and V stand for ‘empty’ slots that can be filled with a consonant and vowel, respectively. The prefixation of the template itself is easily understood as concatenation, and it is reasonable to think of the template as a kind of morpheme. However, it is less clear that the copying process is concatenative; it seems to have more in common with gemination or vowel lengthening.
3.1.4╇ Conversion Finally, the limiting case of a morphological pattern is conversion, in which the form of the base remains unaltered. A standard example is the relationship between some verbs and nouns in English: (3.15)
noun hammer plant ship walk drink
verb hammer plant ship walk drink
40 
C H A P T E R 3 â•… R U L E S
Since we defined morphological patterns as partial resemblances in form and meaning among groups of words, conversion can be regarded as morphological in nature only if this definition is relaxed somewhat, because the resemblance in form is total here. Conversion is generally invoked only for derivational morphology, and primarily for relating two lexemes that differ only in lexical class.
3.1.5╇ Outside the realm of morphology Sometimes a number of additional types are given under the heading of morphological operations, such as alphabet-based abbreviations (acronyms such as NATO, and alphabetisms such as CD (pronounced [si:’di:]), Ph.D. (pronounced [pi:eitʃ’di:])), clippings (e.g. fridge from Â�refrigerator, pram from perambulator) and blends (e.g. smog from smoke and fog, infotainment from information and entertainment). These are clearly operations that can be used to create new words. However, they do not fall under morphology, because the resulting new words do not have different meanings to the longer words from which they are formed. Thus, not all processes of wordcreation fall into the domain of morphological structure, and abbreviations and clippings will play no role in this book.
3.2╇ Two approaches to morphological rules Taking this range of observed phenomena, we can now turn our attention to analysis and ask what these morphological patterns indicate about morphological structure. The ultimate goal is to create a system of morphological rules that mimics speakers’ linguistic knowledge, but this is not always a straightforward process. In addition to accurately representing morphological generalizations, rules should be elegant and cognitively realistic (see Section 1.3 for discussion of these goals). Moreover, the generalizations that we consider important to describe with morphological rules depend in part on the kinds of explanations that we posit. For instance, concatenative patterns are more common in the world’s languages than non-concatenative patterns. Why is this? And is it important that this generalization be directly reflected by the system of morphological rules that we formulate? One possibility is that languages favour concatenative morphological patterns due to some inherent property of the language system, perhaps because morphological structure is fundamentally similar to syntactic structure. If so, our system of morphological rules should be structured in such a way that it places restrictions on non-concatenative patterns, and maximizes similarity to syntactic structure. Alternatively, concatenative patterns might be more common for reasons external to the morphological system, for example, having to do with how languages
3 . 2 ╇ T W O A P P R O A C H E S T O M O R P H O L O G I C A L R U L E S
41
change. This would suggest that morphological rules themselves need not be highly restrictive because the frequency of concatenative patterns does not result directly from the structure of the morphological system. The preponderance of concatenative patterns thus provides one example of how the type of explanation that we pursue, system-external explanation versus inherently restrictive architecture, affects our approach to morphological rules. There are many such issues. On the whole, the emphasis in this book is on questions of substance rather than questions of formal description. But in this section, two representative formalisms for morphological rules will be presented and contrasted, to help bring two major approaches into clear focus. One emphasizes commonalities between morphology and syntax and favours a restrictive architecture of description. The other tends to minimize the importance of parallels between syntax and morphology and invests in system-external explanations. As such, it sees a restrictive formal architecture as less important. We will call these the morpheme-based model and the wordbased model, respectively. The morpheme-based model is associated with the morpheme-combination approach to morphology: ‘Morphology is the study of the combination of morphemes to yield words’ (Section 1.1). The word-based model represents a view of morphology consistent with the following definition: ‘Morphology is the study of systematic covariation in the form and meaning of words’ (also from Section 1.1). As we shall see, each approach has its own advantages and disadvantages.
3.2.1╇ The morpheme-based model In the morpheme-based model, morphological rules are thought of as combining morphemes in much the same way that syntactic rules combine words. To see how this works, we can use the syntactic phrase structure rules in (3.16) to create a sentence by replacing elements on the left of the ‘=’ by elements on the right. In this notation, elements in parentheses are optional; curly brackets and commas represent a choice between alternative options. (3.16) Phrase-structure rules in syntax a. sentence = noun phrase + verb phrase b. noun phrase = (i) ╇ ╇determiner (+ adjective) + noun (ii) sentence╇╇╇╇ ╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇ c. verb phrase = verb (+ noun phrase) d. determiner = the, a, some, … e. noun = cat, rat, bat, … f. verb = chased, thought, slept, … g. adjective = big, grey, …
{
}
To produce the sentence A big cat chased the bat, we need the following individual steps, utilizing the rules in (3.16) (‘X Æ Y’ means ‘insert Y for X’).
42 
C H A P T E R 3 â•… R U L E S
(3.17) sentence Æ noun phrase + verb phrase noun phrase Æ determiner + adjective + noun verb phrase Æ verb + noun phrase noun phrase Æ determiner + noun determiner + adjective + noun Æ a big cat verb Æ chased determiner + noun Æ the bat sentence: A big cat chased the bat.
(by 3.16a) (by 3.16b) (by 3.16c) (by 3.16b) (by 3.16d, g, e) (by 3.16f) (by 3.16d, e)
Likewise, in order to describe the structure of English words like cheeseboard, bags, unhappier, eventfulness, one could make use of the wordstructure rules in (3.18), which are analogous to the syntactic phrase structure rules above. (3.18) Word-structure rules a. word-form = stem (+ inflectional suffix) b. stem = (i) (deriv. prefix +) root (+ deriv. suffix) (ii) stem + stem╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇ c. inflectional suffix = -s, -er, … d. derivational prefix = un-, … e. root = bag, event, cheese, board, happy, … f. derivational suffix = -ful, -ness, …
{
}
We can use these word-structure rules to create complex words. In the following, we see the individual steps by which the words bags, unhappier and cheeseboard can be created using the rules in (3.18). (3.19)
word-form Æ stem + inflectional suffix stem Æ root Æ bag inflectional suffix Æ -s word-form: bag-s
(by 3.18a) (by 3.18bi, 3.18e) (by 3.18c)
(3.20)
word-form Æ stem + inflectional suffix stem Æ derivational prefix + root derivational prefix Æ un- root Æ happy inflectional suffix Æ -er stem: un-happy word-form: un-happi-er
(by 3.18a) (by 3.18bi) (by 3.18d) (by 3.18e) (by 3.18c)
(3.21)
word-form Æ stem stem Æ stem + stem stem Æ root root Æ cheese root Æ board stem: cheese-board word-form: cheese-board
(by 3.18a) (by 3.18bii) (by 3.18bi) (by 3.18e) (by 3.18e)
3 . 2 ╇ T W O A P P R O A C H E S T O M O R P H O L O G I C A L R U L E S
43
There are several ways to modify this general approach. For example, many linguists argue that we should dispense with word-structure rules in (3.18) and put all the relevant information, including combinatory potential, into lexical entries. This parallels an argument within syntactic theory. Many syntacticians have called into question the need for phrase-Â�structure rules like (3.16a–c), on the grounds that the same information is already contained in words’ lexical entries, making the general rules redundant. In line with this approach, an alternative formalism to (3.18) is illustrated in (3.22). These lexical entries contain information on the pronunciation, properties and meaning of the morpheme. The pronunciation is given between slashes in phonetic transcription, the properties consist of the word-class (for roots) or the combinatory potential (for affixes), and a rough indication of the meaning is given in quotation marks. (Naturally, a lot more needs to be said on the semantics of morphemes, but the details can be ignored for present purposes; see Section 11.1.1 for some aspects of the semantics that are relevant to morphology.) (3.22) proposed lexical entries for some morphemes: a. bag b. -s c. happy d. un ╇ /bæg/ ╇ /z/ ╇ /hæpi/ ╇ /n/ ╇ N ╇ N — ╇ A ╇ — A ‘bag’╇╇ ╇ ‘plural’ ‘happy’ ‘not’╇
[ ]
[ ] [ ]
[ ]
When lexical entries of roots and affixes are enriched in this way, morphological description seems to reduce largely to the description of the lexical entries of morphemes. Concatenation becomes a property of the lexical entry itself, all but removing the distinction between rules and morphemes. Despite this difference, the core principle is the same in both (3.18) and (3.22): morphology consists of one basic type of lexical entry (morphemes) and one type of process that operates on those entries (concatenation). And by doing so, the morpheme-based model maximizes the formal similarity between morphology and syntax. The morpheme-based model raises two questions. First, what are the advantages and disadvantages, both empirical and theoretical, of reducing morphological structure to morpheme concatenation? Second, can this model account for base modification, reduplication and conversion, and if so, how? There are at least two good reasons to favour a theory of purely concatenative rules. First, as noted above, morpheme concatenation is the most common kind of morphological pattern cross-linguistically. By treating concatenation as the fundamental (or only) type of morphological rule, the morpheme-based model provides a natural explanation for this fact. Second, so far we have compared morphology to syntax by noting that the morpheme-based model treats morphological structure as a string
44 
C H A P T E R 3 â•… R U L E S
of morphemes in much the same way that syntax consists of a string of words. However, morphology and syntax are similar in other respects as well. Much of the data is too complicated to consider here, but we can focus on one basic principle: hierarchical structure. Consider the sentence We need more intelligent leaders. It consists of a string of words, but it also has internal structure that is not necessarily reflected in the linear order of words. We know this is true because this sentence can have two possible interpretations, either ‘We need more leaders who are intelligent’, or ‘We need leaders who are more intelligent.’ Each interpretation corresponds to a different hierarchical structure. In the first interpretation, intelligent leaders forms a subgroup (called a syntactic constituent), which is modified by more. In the second interpretation, more intelligent forms a constituent, and collectively modifies leaders. This hierarchical structure is represented using the (simplified) formalism of syntactic tree diagrams in Figure 3.1.
Interpretation 1
Interpretation 2
we
we need
need more intelligent leaders
more intelligent leaders
3.1 Hierarchical structure in syntax Figure 3.1Figure Hierarchical structure in syntax
Not all sentences have more than one possible interpretation, but all have hierarchical structure. Important here is that concatenative morphology also (arguably) has hierarchical structure, and words like undoable also have two possible interpretations: ‘unable to be done’ (do and -able form a constituent), or ‘able to be undone’ (un- and do are a constituent). These are represented in Figure 3.2 using simplified morphological tree diagrams. Inasmuch as morphology and syntax exhibit fundamental similarities of this type, we can hypothesize that morphology and syntax operate according to shared principles due to some innate property of language architecture, and that this is perhaps the reason that concatenative patterns are more commonly found. If so, it is desirable to posit a model that
3 . 2 ╇ T W O A P P R O A C H E S T O M O R P H O L O G I C A L R U L E S
Interpretation 1
45
Interpretation 2
undo
-able
un-
do
-able
3.2 Hierarchical structure morphology Figure 3.2Figure Hierarchical structure in in morphology
maximizes the formal similarity between morphology and syntax. For these and other reasons, the morpheme-based, concatenation-only approach to morphological analysis has been popular. (See Chapter 7 for a more thorough and critical discussion of hierarchical structure in morphology and its relation to syntactic structure.) Still, there are some disadvantages to positing that concatenation is the only rule type. Most notably, base modification and conversion are difficult to accommodate. Consider again Albanian plural nouns, e.g. armik ‘enemy’, armiq ‘enemies’. In principle, a morpheme-based model can account for this pattern by positing lexical entries as in (3.23): (3.23) a. armik b. ‘plural’ ╇ /armik/ ╇ / Ø / ╇ N ╇ N — ‘enemy’╇ ‘plural’
[
] [ ]
The entry in (3.23b) means that there is a suffix that contributes the meaning ‘plural’, but which has no phonological form. This is a zero affix (or zero expression). To get the right phonological form, the morphemebased model must assume that the zero affix has some property that triggers palatalization on the final consonant of the root. This amounts to treating palatalization as morphologically conditioned allomorphy: the allomorph armiq is selected by the zero plural affix. This analysis allows us to avoid violating the fundamental principles of the morpheme-based model. Formally, it is the suffix in (3.23b), not the process of palatalization, that expresses plurality. While this is a possible analysis, it is not a very satisfactory one. Why should the zero plural affix trigger palatalization? There is no motivation for this. Moreover, in a language with several different base modification patterns, there would be several different zero affixes in the lexicon, each of which would trigger allomorphy differently. In short, zero morphemes
46 
C H A P T E R 3 â•… R U L E S
are ad hoc devices that are posited for no purpose other than to save the principle of a concatenation-only model. Inasmuch as base modification and conversion represent common morphological patterns, this is a significant flaw of the morpheme-based model.
3.2.2╇ The word-based model In the word-based model, the fundamental significance of the word is emphasized and the relationship between complex words is captured not by splitting them up into parts and positing a rule of concatenation, but by formulating word-schemas that represent the features common to morphologically related words. For instance, the similarities among the English words bags, keys, gods, ribs, bones, gems (and of course many others) can be expressed in the wordschema in (3.24c). (3.24) a. Words: bags, keys, gods, ribs, bones, gems, … b. Lexical entries for words ╇/bægz/n ╇/kijz/n ╇/gɑdz/n ╇/rıbz/n ‘bags’╇╇╇╇ ‘keys’╇╇╇ ‘gods’╇╇╇ ‘ribs’╇╇╇
[
] [
] [
]
[
]
c. Word-schema ╇/Xz/n ‘plurality of xs’
[
]
A word-schema is like the lexical entries in (3.24b) in that it contains information on pronunciation, syntactic properties and meaning. But a word-schema may additionally contain variables. In this way, it abstracts away from the differences between the related words and just expresses the common features. The schema in (3.24c) expresses the fact that all words in (3.24a,b) end in /z/, that they all denote a plurality of things and that they are all nouns (indicated by subscript N after the phonological representation). The phonological string preceding the /z/ is quite diverse and is thus replaced by the Â�variable /X/. Likewise, semantically these words share nothing besides the plurality component, so again the semantic part of the schema contains a variable (‘x’). We will use the terms match and subsume for the relation between concrete words and the abstract schema: words match a schema, and a schema subsumes words (for example, the schema in (3.24c) subsumes the nouns in (3.24a, b) and many others, but not all English plural nouns match it; for instance, the plural feet does not match its phonological part). Crucially, a word-schema stands for complete words, not for individual morphemes in the sense of the morpheme-based model. The word-schema in (3.24c) is a generalization based on the lexical entries in (3.24b), which are themselves word-forms, not morphemes. When the word-schema contains
3 . 2 ╇ T W O A P P R O A C H E S T O M O R P H O L O G I C A L R U L E S
47
both a segmentable piece of sound and a corresponding meaning, as is true in (3.24c), we can call this a ‘morpheme’. However, it is important to remember that in a word-based model a morpheme is just a convenient term. It is simply one kind of schema among many, and has no special status. Now what makes word-schemas really significant for morphology is the fact that closely related schemas are connected to each other. A schema that subsumes a very similar set of words is given in (3.25c). (3.25) a. Words: bag, key, god, rib, bone, gem, … b. Lexical entries ╇/bæg/n ╇/kij/n ╇/gɑd/n ╇/rıb/n ‘bag’╇╇╇ ‘key’╇╇╇ ‘god’╇╇╇ ‘rib’╇╇╇
[
]
[
]
[
]
[
]
c. Word-schema ╇/X/n ‘x’╇╇╇
[
]
The morphological relationship between these sets of words can now be represented in the morphological correspondence in (3.26).
[
]
[
]
(3.26) ╇/X/n ´ ╇/Xz/n ‘x’╇╇╇ ‘plurality of xs’
The double arrow means that, for some word matching the schema on the left, there is a corresponding word matching the schema on the right. Example (3.26) thus shows what a morphological rule looks like in the wordbased model. The rule in (3.26) is the word-based equivalent of (3.22b). It says that plural nouns can be formed from singular nouns by suffixing /z/. Note that combinatory potential is represented here as well. In the correspondence, the left schema is marked as a noun. This indicates that the rule forming plurals with [z] requires a base that is a noun. In this way too, the word-based formalism of (3.26) is equivalent to the morphemebased formalism in (3.22b), in which the lexical entry for the morpheme /-z/ contains the restriction that the suffix must attach to a noun. Unlike the morpheme-based model, the word-based model has no way of dispensing with morphological rules, and while the correspondence in (3.26) represents a rule of suffixation, there is nothing in the model that necessarily restricts morphological rules to concatenation. As with the morpheme-based model, this raises the question: What are the advantages and disadvantages of an approach that allows a wide variety of rules, and gives no special status to rules of concatenation? There are at least three significant advantages of the word-based model. First, the most striking is that non-concatenative patterns can be described with it quite naturally, whereas such phenomena are difficult to accommodate in morpheme-based models. As an example, (3.27b) shows
48 
C H A P T E R 3 â•… R U L E S
the correspondence for English noun–verb conversion of nouns denoting instruments. (3.27) a. hammern/hammerv, sawn/sawv, spoonn/spoonv, funneln/funnelv, …
[
]
[
]
b. ╇/X/n ´ ╇/X/v ‘x (= an instrument)’ ‘use x (= an instrument)’
Here the word-schema on the right differs from the schema on the left in word-class and meaning, but not in phonological form. Processes of base modification can also be easily described by elaborating the phonological variable somewhat. For instance, shortening in Hindi/Urdu can be represented as in (3.28b), where /V:/ stands for any long vowel. (3.28) a. ma:r- ‘kill’, mar- ‘die’
[
]
[
b. ╇/XV:Y/v ´ ╇/XVY/v ‘A causes B to happen’ ‘B happens’
[
Reduplication is described by copying part of the phonological string in one of the word-schemas. (3.29b) shows the rule for the Somali duplifix -aC that we saw in (3.13) (here /C/ stands for any arbitrary consonant). (3.29) a. buug/buugag ‘book(s)’, fool/foolal ‘face(s)’, koob/koobab ‘cup(s)’, …
[
]
[
]
b. ╇/XC1/n ´ ╇/XC1aC1/n ‘x’╇╇╇╇╇╇ ‘plurality of xs’
A second advantage is that the word-based model can explain how backformations (like to babysit, which is historically derived from babysitter) are possible. In the morpheme-based model, it is quite puzzling that speakers should be able to create a verb babysit. English has many compounds denoting agents in which the first lexeme in the compound is interpreted as the object of an implicit action (babysitter, truck driver, window washer, etc.). English does not, however, have a productive rule combining a noun and a verb to create compounds like to babysit (*to bookread, *to fishcatch, *to truckdrive, *to windowwash). This causes a problem for the morpheme-based model because compounds can only be built up from component lexemes in this approach, and not derived directly from other compounds. But in the word-based model, the fact that babysit came from babysitter can be readily described. The noun babysitter matches two word-schemas simultaneously. First, it matches the nominal compound schema in (3.30), and everyone agrees that it was first created using this rule (i.e. as baby + sitter).
[ ]
[ ]
[
]
(3.30) /X/n & /Y/n /XY/n ‘x’ ‘y’ ´ ‘a y that has to do with x’╇╇╇
(Note that for compounds, the left-hand side of the correspondence must consist of two word-schemas.) And, second, it matches the word-schema
3 . 2 ╇ T W O A P P R O A C H E S T O M O R P H O L O G I C A L R U L E S
49
of non-compound agent nouns given on the right in (3.31). (Here ‘dox’ represents a variable action meaning.)
[ ]
[
(3.31) /X/v /Xər/n ‘dox’ ´ ‘a person who (habitually) doesx’
]
Crucially, the bidirectional arrow indicates that the correspondence in (3.31) is not directed. In addition to the creation of -er nouns from verbs (like bake Æ bak-er, write Æ writ-er, sin Æ sinn-er, etc.), this rule allows the creation of verbs from nouns containing -er that denote an agent of some sort (babysitter Æ babysit). This is a fundamental difference between the word-based model and the morpheme-based model. Clearly, the rule in (3.31) is much more productive from left to right than from right to left (e.g. one cannot form a verb to butch from butcher), but under what sort of circumstances a rule is productive or unproductive is a separate question that we return to in Chapter 6. Once a back-formed word has become a normal word of the language, it is synchronically indistinguishable from a non-derived word (thus, only historical linguists, but not other speakers of English, know that edit was back-formed from editor). From the perspective of the morpheme-based model, we might therefore argue that back-formations are solely a problem for language change, and that our theory can treat babysitter and editor as synchronically derived from to babysit and to edit, rather than the reverse. While this is true, it does not solve the problem because it still fails to explain how back-formation could ever arise in the first place. The wordbased model has the advantage of being able to naturally explain this kind of historical development. Third, even some concatenative patterns cause problems for morphemebased models but are easily described within a word-based model. Crossformations are good examples of this. Consider the three sets of English words in (3.32). (3.32)
attract suggest prohibit elude insert discuss — —
attraction suggestion prohibition — insertion discussion illusion aggression
attractive suggestive prohibitive elusive — — illusive aggressive
In order to describe the relations between these three sets, we minimally need the two correspondences in (3.33a–b). (For the sake of simplicity, we use the spelling rather than the pronunciation in representing the affixes -ion and -ive here.)
50 
C H A P T E R 3 â•… R U L E S
[ ] [
] [ ] [
]
(3.33) a. /X/v ´ /Xion/n b. /X/v ´ /Xive/a ‘dox’ ‘action of doingx’ ‘dox’ ‘prone to doingx’
But these two rules do not suffice, because there are pairs like illusion/illusive, aggression/aggressive that lack a corresponding verb (*aggress, *illude). This means that within the word-based model we also need the rule (3.33c).
[
] [
]
(3.33) c. /Xion/n ´ /Xive/a ‘action of doingx’ ‘prone to doingx’
Rules of the type in (3.33c) are cross-formations: a morphological rule in which both word-schemas in the correspondence exhibit a constant phonological element (Becker 1993a). Cross-formations are in no way unusual or uncommon. Consider the pairs of words in (3.34), which demonstrate cross-formation in compounding. (3.34)
seasick sealane seafare seaborne seamanship seaworthy seaman
airsick airlane airfare airborne airmanship airworthy airman
(Becker 1993a: 13–14)
For the first few words one can still imagine that the sea and air compounds were created independently of each other – i.e. airsick from air + sick, without direct relation to the older word seasick, or airlane from air + lane, without direct relation to sealane. But for some of the others this seems very unlikely because the meaning is noncompositional: the meaning of the word-form is more than the sum of the meanings of the parts. A seaman is a low-ranking navy member, not any man with some relation to the sea, and similarly an airman is a low-ranking air force member. Thus, we are probably dealing with a rule as in (3.35).
[
]
[
]
(3.35) /seaX/ /airX/ ‘an x having to do ´ ‘an x having to do with sea travel’╇╇ with air travel’╇╇
We will see more examples of cross-formation when we discuss inflectional morphology in Chapter 8. Despite being concatenative, cross-formation patterns cannot be described so easily in a purely morpheme-based model. Since the morpheme-based model usually assumes that complex words are not stored in the lexicon, it must posit that illusion and illusive are each derived from a root illude, with the suffixes having the lexical entries [/-ion/; N; V— ] and [/-ive/; A; V— ]. The problem lies in the fact that once illude is posited as the root, the morpheme-based model then faces difficulty explaining why the verb
3 . 2 ╇ T W O A P P R O A C H E S T O M O R P H O L O G I C A L R U L E S
51
illude does not exist. By allowing for a direct relationship between illusion and illusive, the word-based model avoids this problem. So in summary, the word-based model, by virtue of allowing a wide variety of rules, provides more satisfactory analyses of non-concatenative patterns, some results of language change such as back-formation, and even some concatenative patterns such as cross-formations. Of course, the word-based model has its own potential disadvantages. One of the most common criticisms of the word-based model is that it is not restrictive. Restrictiveness is an important feature of a morphological model because it allows us to make generalizations about what is possible and impossible in language. The word-based model allows morphological rules of virtually any type – including many that are not known to exist in any language. For example, in the well-known language game Pig Latin the basic principle is that if a word begins with one or more consonants, its Pig Latin equivalent moves those consonants to the end of the word and adds ‘ay’ ([ej]). So, English book Æ Pig Latin ookbay, star Æ arstay, etc. And in one version of the rules, if the word begins with a vowel, the Pig Latin word adds ‘say’: orange Æ orangesay, apple Æ applesay, etc. We must remember that Pig Latin is a language game, and thus different in many respects from normal language. Morphological patterns in which sounds are moved from the beginning to the end of the word do not seem to exist in natural languages. The morphological pattern of metathesis does involve switching the order of sounds (see (3.9)), but it always involves ‘local’ rearranging – the sounds must be next to each other. No known natural languages have the Pig-Latin type of ‘long distance’ rearranging. Many linguists argue that a good model of morphology will preclude this kind of long-distance rearranging of sounds. The morpheme-based model automatically excludes the Pig Latin pattern (local rearranging is also excluded), but nothing about the word-based model necessarily does. If re-arranging of sounds is generally allowed in the word-based model (e.g. in order to account for metathesis), some extra theoretical device would be needed to restrict long-distance rearranging (e.g. an ad hoc principle that the rearranging must be local). Inasmuch as this makes the word-based model more complicated than the morpheme-based model, many linguists would argue that non-restrictiveness is a significant disadvantage of the word-based model. A related concern is how frequently morphological patterns involve concatenation. As discussed above, one of the advantages of the morphemebased model is that it can easily account for the fact that concatenative patterns are very common in the world’s languages, since the model reduces the formal architecture to morphemes and concatenation. By contrast, the word-based model posits morphological rules that effect a wide variety of changes on the base, including concatenation, metathesis,
52 
C H A P T E R 3 â•… R U L E S
vowel lengthening, etc., and treats them as equal in type. This means that the word-based model loses all morphology-internal explanation for the dominance of concatenative patterns. Here, however, it may be possible to find an explanation outside of the morphological system. Concatenative patterns are more common than non-concatenative patterns in part because the language structures that are the historical sources of concatenative patterns are more common than the language structures that are the historical sources of many nonconcatenative ones. By far the most common way in which new morphological patterns, and particularly concatenative patterns, arise is by coalescence of several formerly free syntactic elements. When the two elements that coalesce are content words, the result is a compound. When one of the coalescing elements is a semantically abstract element that mostly serves grammatical functions in the sentence, the result of the coalescence is an affixed word. Let us consider an example from Spanish, which has a future tense that is formed by adding the suffix -r to the stem, followed by a series of special person–number suffixes: (3.36) 1sg 2sg 3sg 1pl 2pl 3pl
present tense cant-o ‘I sing’ canta-s canta canta-mos cantá-is canta-n
future tense canta-r-é ‘I will sing’ canta-r-ás canta-r-á canta-r-émos canta-r-éis canta-r-án
Originally this was a syntactic pattern, involving the auxiliary verb habere ‘have’ (Modern Spanish haber), which was combined with the infinitive to express obligation: habeo cantare or cantare habeo ‘I have to sing’. Then the meaning shifted from obligation to future, and the verb haber lost its freedom of position and came to occur only immediately after the main verb. As a result of phonological reduction, the infinitive lost its final -e (cantare became cantar) and the forms of the verb haber were shortened (he, has, ha, habemos, habéis, han). Finally, the infinitive and the forms of haber were fused together to form a set of single complex words: (3.37)
cantar cantar cantar cantar cantar cantar
he has ha (hab)emos (hab)éis han
> > > > > >
cantaré cantarás cantará cantarémos cantaréis cantarán
Such grammaticalization changes are extremely common in languages, and the vast majority of all (non-compound) concatenative patterns seem ultimately to go back to such syntactic phrases with auxiliary words.
3 . 2 ╇ T W O A P P R O A C H E S T O M O R P H O L O G I C A L R U L E S
53
By contrast, many non-concatenative patterns historically began as phonological patterns. (The reader may have noticed that most of the base modification patterns in Section 3.1.2 were described using terms from phonology.) For example, the English verb house used to have a voiceless [s], like the noun house, but until Middle English times, all forms of the verb had a suffix that began with a vowel (hous[ε]n, hous[ε]th, hous[ε]d, etc.). In the noun the [s] was in word-final position already in Old English. At some point, [s] between vowels came be pronounced as [z] (a phonological rule of voicing). This, of course, affected the stem-final consonant in the verb, but did not affect the noun. Later, the suffixes disappeared from most of the verbal forms, causing the noun and the verb to be identical – except for the voicing of the final consonant. In this way, voicing became the (only) morphological marker distinguishing the verb house from the noun house (verb wreathe from noun wreath, thieve from thief, etc.). The point here is that voicing is a common phonological process, so there are many opportunities for voicing to develop into a morphological pattern, but still not nearly as many opportunities as there are for free words to become morphemes. There are even fewer phonological patterns that involve switching the order of two sounds. As a result, there are historically very few opportunities for patterns like morphological metathesis to develop (Janda 1984). Thus, one possibility is to explain the frequency of concatenative patterns by positing a theory of mostly or solely concatenative rules, as the morpheme-based model does. But another possibility is to explain the frequency of concatenation as the indirect result of some historical changes being more common than others (Bybee and Newman 1995). This is the preferred explanation within a word-based approach. Moreover, the same logic can be applied to the non-restrictiveness of the word-based model. While the word-based model can describe many morphological patterns that do not exist in natural languages, including Pig Latin-type long distance movement of sounds, perhaps this is not a problem if those patterns would never arise by historical processes. Since system-external explanation is a major goal of linguistic research, lack of restrictiveness is not necessarily a disadvantage of the word-based model. In the end, in these two approaches to rules we find a classic conflict between different goals of morphological research. The morpheme-based model is restrictive and captures similarities between morphology and syntax, but at the cost of empirical adequacy. The word-based model is more empirically adequate, but at a cost of lost restrictiveness. It is therefore not the case that one approach is inherently superior to the other. Morphologists (and morphology students) must decide for themselves which goals of research are most important. But inasmuch as the problem of non-restrictiveness is minimized by system-external factors, whereas the problem of empirical inadequacy is not, the word-based approach should perhaps be evaluated more favourably.
54 
C H A P T E R 3 â•… R U L E S
Summary of Chapter 3 In addition to concatenative patterns (affixation and compounding), morphology includes a wide variety of non-concatenative patterns. These include conversion, reduplication, and base modification (palatalization, weakening, gemination, lengthening, shortening, tonal change, stress shift, voicing, subtraction, metathesis, etc.). Given these patterns, it is difficult to create a formal analysis of morphological structure that posits only concatenative rules similar to those found in syntax (the morpheme-based model). Any such theory would have to posit extensive zero affixes and unmotivated rules of allomorphy. The opposite view, that rules represent morphological correspondences between word-schemas (the word-based model), allows for a more straightforward explanation of both nonconcatenative patterns and issues of analysis, such as back-formation and cross-formation. The word-based model is thus more empirically satisfactory. At the same time, one consequence is that the word-based model is capable of describing many kinds of morphological patterns that are not found in the world’s languages. The morpheme-based model is more restrictive. Some linguists consider this a significant advantage of the morpheme-based model. However, a counterargument is that the word-based model does not need to be highly restrictive if unattested patterns are very unlikely to occur for reasons external to the morphological system (e.g. because of conditions in language change). The same argument helps explain why concatenative patterns are much more common than non-concatenative ones.
Further reading An early version of the morpheme-based model was advocated by Bloomfield (1933). More recent morpheme-based models are Word Syntax (Selkirk 1982; Di Sciullo and Williams 1987; and Lieber 1992) and Distributed Morphology (Halle and Marantz 1993, 1994; and Harley and Noyer 1998). Versions of the word-based model, grouped together under the name of Word and Paradigm models, are advocated most strongly by Matthews (1972), Bybee (1985), Becker (1990), Anderson (1992), Bochner (1993), Aronoff (1994), and Stump (2001a), and in Dasgupta et al. (2000) and Singh and Starosta (2003). See also Aronoff (2007). The word-schema formalism used in this chapter is based especially on Becker (1990, 1993a,b) and Bochner (1993).
COMPREHENSION EXERCISES 
55
Psycholinguists have debated whether both morpheme-based rules and schema-like correspondences (often called associative processes) exist side-by-side. See Bybee and Slobin (1982), Skousen (1989), and Alegre and Gordon (1999). Derwing (1990) is a summary. Janda (1982, 1984) and Bybee and Newman (1995) argue for a role for historical explanation in morphological theory. Joseph (1998) provides an overview of the historical sources of morphological patterns, and Haspelmath (1995) considers morphological reanalysis.
Comprehension exercises 1. Which formal operation (or combination of operations) is involved in the following morphological patterns? a. Mbay (v` = low tone, v´ = high tone, vˉ = mid tone) tétə` ‘break’ tétaˉ ‘break several times’ ìndaˉ ‘wrap’ índaˉ ‘wrap several times’ rıˉyaˉ ‘split’ ríyaˉ ‘split several times’ (Keegan 1997: 40) b. Yimas manpa kika yaka
‘crocodile’ ‘rat’ ‘black possum’
manpawi kikawi yakawi
‘crocodiles’ ‘rats’ ‘black possums’ (Foley 1991: 129)
c. Coptic koˉt hoˉp toˉm
‘build’ ‘hide’ ‘shut’
keˉt heˉp teˉm
‘be built’ ‘be hidden’ ‘be shut’ (Layton 2000: 129)
d. Hausa (v` = low tone, v´ = high tone) búgàa ‘beat’ búbbùgáa ‘beat many times’ táakàa ‘step on’ táttàakáa ‘trample’ dánnèe ‘oppress’ dáddànnée ‘oppress (many (times))’ (Newman 2000: 424) e. Tagalog ibigay ‘give’ ipaglaba ‘wash (for)’ ipambili ‘buy (with)’
ibinigay ‘gave’ ipinaglaba ‘washed (for)’ ipinambili ‘bought (with)’
f.
gefunden gesungen gebunden
German finden ‘find’ singen ‘sing’ binden ‘tie’
‘found’ ‘sung’ ‘tied’
56 
C H A P T E R 3 â•… R U L E S
2. What would be the lexical entries of the following English morphemes (using the formalism of (3.22))?
hear, -ing (as in he is playing, she is dancing, etc.), re- (as in to replay, to rewrite, etc.), good, -s (as in sells, knows, etc.)
3. Formulate the morphological rule in the word-based format of (3.26) (i.e. as a correspondence between word-schemas) for the following pairs of words (each standing for a large set of such pairs): warm – warmer happy – unhappy play – replay happy – happily 4. Formulate pairs: búhay gútom tákot hábaʔ gálit
the morphological rule for the following Tagalog lexeme ‘life’ ‘hunger’ ‘fear’ ‘length’ ‘anger’
buháy gutóm takót habáʔ galít
‘alive’ ‘hungry’ ‘afraid’ ‘long’ ‘angry’
5. The following pairs of English lexemes are related by cross-formation. Formulate the rule for them, analogous to (3.33c). astronomy astronomer philosophy philosopher ethnography ethnographer 6. For French adjectives, linguists have often advocated an analysis in terms of subtraction: the masculine form is formed from the feminine form by subtracting the final consonant (Bloomfield 1933: 217): plat/platte ‘flat’ [pla/plat] laid/laide ‘ugly’ [lε/lεd] long/longue ‘long’ [lõ/lõg] soul/soule ‘drunk’ [su/sul] gris/grise ‘grey’ [gri/griz]
Why is this an attractive analysis?
Exploratory exercise One issue that was not addressed in this chapter is whether morphological rules ever fail to apply. We implied that morphological rules apply to all and only the bases that meet the conditions for a given rule, and linguists generally strive to formulate rules for which this is true. But is it always possible to write rules that are this efficient? In other words, how common are
E X P L O R AT O R Y E X E R C I S E
57
exceptions? The goals of this exercise are to practise writing morphological rules using the formalism of word-schemas and morphological correspondences, and to consider some of the issues that exceptions pose for an analysis of morphological structure. The methodology consists of classic morphological analysis. An optional last step adds a simple experimental component. Croatian is used for demonstration purposes, but the reader could choose to investigate virtually any language with significant inflectional morphology. Instructions Step 1: Select a language and morphological relationship to study. The best choices will be pairs of word-forms belonging to the same lexeme. One of the forms should also be more ‘basic’ than the other. For example, consider the following words from Croatian. (3.38) a. singular plural b. singular plural blog blogovi ‘blog(s)’ album albumi ‘album(s)’ džep džepovi ‘pocket(s)’ biskup biskupi ‘bishop(s)’ film filmovi ‘film(s)’ dokument dokumenti ‘document(s)’ grad gradovi ‘city(ies)’ kamen kameni ‘stone(s)’ park parkovi ‘park(s)’ papir papiri ‘paper(s)’ vlak vlakovi ‘train(s)’ razgovor razgovori ‘conversation(s)’ Here we have chosen singular and plural nouns in the nominative form. (This data represents only masculine nouns with ‘hard’ stems. Croatian has other types of nouns as well, but they are not relevant here.) These meet both criteria: the word-forms are inflectionally related (e.g. blog and blogovi belong to the same lexeme), and the form of the singular in these examples is more basic than that of the plural. This indicates that it is reasonable to set up a morphological correspondence that derives plurals from singulars. It is also best to pick a morphological relationship that exhibits multiple morphological patterns. Here there are two patterns: the nouns in (3.38a) form the plural with –ovi, whereas the nouns in (3.38b) have –i in the plural. For reasons that will be apparent below, the more morphological patterns there are, the more interesting the data will be to analyze. Step 2: Gather examples. Build a long list of word-form pairs that express the chosen inflectional relationship (e.g. singular-plural). Be sure to include all of the relevant morphological patterns. (A good way to do this is to consult a dictionary that gives inflectional information, and record all of the relevant examples on every 10th (20th, 50th, etc.) page.) Step 3: Write morphological rules. Sort the data into groups according to morphological pattern. Write rules to describe each group. For example, the major generalization for Croatian is that words with only one syllable in the nominative singular (monosyllabic nouns) form the nominative plural with -ovi, whereas words with more than one syllable (polysyllabic nouns)
58 
C H A P T E R 3 â•… R U L E S
form nominative plural with -i. In (3.39), the symbol ‘’ is used to represent a syllable, so ‘X’ means any string of sounds that is one syllable in length. ‘X+’ means any string that is at least two syllables in length. (3.39) Croatian rule for plural formation (masculine hard-stem nouns)
[ [
] ]
[ [
] ]
a. /X/n Æ /Xovi/n ‘x (nom sg)’ ‘x (nom pl)’ b. /X+/n Æ /Xi/n ‘x (nom sg)’ ‘x (nom pl)’
Be as specific as possible about the relevant factors. The goal is for the schemas to subsume as many word-forms as possible that follow the rule, while excluding as many as possible that do not. For example, the rule in (3.39a) excludes all polysyllabic nouns, because only monosyllabic nouns match the left schema (linguists say that monosyllabic nouns ‘meet the conditions’ for the rule of plural formation). Formulating rules may not be easy. For example, a large data set of Croatian would also include the nouns in (3.40). (3.40) a. singular plural b. singular plural golub golubovi ‘pigeon(s)’ cent centi ‘cent(s)’ jastreb jastrebovi ‘hawk(s)’ dan dani ‘day(s)’ pramen pramenovi ‘tuft(s), hair gost gosti ‘guest(s)’ lock(s)’ The problem is apparent: the singular word-forms in (3.40a) meet the conditions for the rule in (3.39b), but seem to undergo rule (3.39a). The examples in (3.40b) have the same problem with regard to the rule in (3.39a). Does this mean that the rules in (3.39) are incorrect? This is where type frequency can help. More than 93 per cent of monosyllabic nouns followed the pattern in (3.38a); fewer than 7 per cent followed the pattern in (3.40b). Polysyllabic nouns are similarly likely to follow the pattern in (3.38b), rather than (3.40a). This suggests that the generalization formalized by the rules in (3.39) is fundamentally correct. The examples in (3.40) are exceptions. Look for similar issues in your data. You may need to rewrite some rules, or write new rules, to cover all of the examples in your data set. There is no rigid formula for morphological analysis – linguists develop a feel for good analysis through practice. Step 4: Ponder the implications of any exceptions. Consider the following questions in the context of your analysis. 1) How many exceptions can accrue before they no longer seem ‘exceptional’, and instead seem to be a rule-governed pattern? There is no absolute right or wrong answer here. Consider your data, the Croatian data, or other examples. Identify factors that would lead you to conclude
E X P L O R AT O R Y E X E R C I S E
59
that a given example is rule-governed, and factors that would lead you to conclude that it is truly idiosyncratic, and needs to be learned on an itemby-item basis. Explain your reasoning. 2) Think about whether additional data might help decide whether a particular group of exceptions is, or is not, represented by a morphological rule. For instance, if exceptions do not follow any rules (i.e. if the relevant word-forms are truly anomalous), how would you expect new words to behave? If a series of new monosyllabic words were to enter Croatian, would you expect 93 per cent of these new words to form plurals with -ovi, and 7 per cent with -i? This would be a result that matches the type frequency of similar existing nouns. Would you expect all of the new words to have plurals formed with -ovi? Or would you expect something else? Would your predictions change if the exceptions were described by a rule, albeit one that does not apply very widely? Do new words in a language (e.g. ones borrowed from another language) necessarily follow the same morphological patterns as existing native words? Apply the same questions to your data. Try to explain why you expect one or the other result. Step 5 (optional): Test your predictions about new forms with native speakers. Make up a variety of hypothetical words of the language (nonwords). Non-words follow the phonological rules of a language, and sound like they could be words, but are not. Your non-words should match the schema for the ‘basic’ form. Present native speakers with the non-words and ask them to produce the relevant derived form. For example, we might ask Croatian speakers to decide what the plural form would be, given the hypothetical singular nouns brag, glik, adret, bakral, mokilar, etc. Are your predictions correct? And in general, do speakers behave in the ways that your rules would suggest?
Lexicon
4
I
n this chapter we look more closely at morphemes, focusing on the following fundamental issue: Do speakers memorize entire complex word-forms (readable, reads, washable), their component morphemes (read, wash, -able, -s), or both? Another way to ask the same question is: What is the content of the lexicon? Remember that the lexicon is the linguist’s term for the language user’s mental dictionary. When a linguist says that something is listed in the lexicon, this means that it must be stored in speakers’ memories (but linguists generally prefer the more abstract, less psychological-sounding terminology).1 The content of the lexicon is an important issue for any theory of morphology because lexical items are the fundamental building blocks of morphological structure. They are the bases to which morphological rules apply. As such, our view of the lexicon affects our analysis of morphological structure in broad ways. If evidence points to the lexicon consisting primarily of morphemes, the rules that we write will operate on morphemebased structures. And correspondingly, if evidence suggests that the lexicon consists primarily of words, the rules that we posit will be fundamentally word-based. The material in this chapter is thus complementary to the discussion in Chapter 3. All linguists agree that the lexicon must contain at least all the information that is not predictable from general rules. For instance, an English speaker’s lexicon must contain the monomorphemic English verbs arrive, refuse, deny, and words showing extreme semantic peculiarities (e.g. awful, which is not 1
A distinction is sometimes made between a lexicon and a mental lexicon, where the lexicon is a purely abstract tool of linguists to describe roots and affixes that does not necessarily correspond in any way to speakers’ mental knowledge. The term mental lexicon is then used for the more psychological concept of a speaker’s mental dictionary. However, we follow the view that linguists should strive to analyze language in ways that are plausible representations of speakers’ knowledge, so we will continue to talk about the lexicon in terms of a (hypothetical) speaker, and not distinguish between these terms.
4 . 1 ╇ A M O R P H E M E L E X I C O N ?
61
at all the same as ‘full of awe’). But does the lexicon additionally contain predictable information? For example, does an English speaker’s lexicon contain the complex word-form helpful, despite the fact that this word is easily segmented into the morphemes help and -ful, and fully predictable from the meaning of these parts? Here, there is disagreement. We can identify three major positions. One possibility is that no regular complex words (like helpful) are stored in the lexicon. On this view, the lexicon contains, to the extent possible, just simple, monomorphemic elements, i.e. roots and affixes. Idiosyncratic complex words are also lexical entries, but virtually all complex words are created by rule, rather than being listed. This is a morpheme lexicon. It corresponds to the morphemebased model. Another position takes exactly the opposite view: not just some, but all complex word-forms are included in the lexicon, whether they are predictable or idiosyncratic. This is a strict word-form lexicon. The third position is intermediary – it posits that word-forms, morphemes and derived stems are all potentially listed in the lexicon. Whether any particular word-form is listed depends on a variety of factors. Since wordforms still play the primary role in this approach, we call it a moderate word-form lexicon. Both the strict word-form lexicon and the moderate word-form lexicon are consistent with the word-based model from Chapter 3. In the following sections we evaluate these hypotheses.
4.1╇ A morpheme lexicon? Based on the apparent parallelism between sentences, morphemes and phonemes shown in Figure 4.1, we might assume that the basic units of the lexicon are morphemes. Just as language users do not memorize each sentence that they use, we can also hypothesize that language users do not generally memorize complex words. syntax sentences consist of words morphology words consist of morphemes phonology morphemes consist of phonemes
Camilla
met
an
unfriendly
chameleon
Camilla
met
an
unfriendly
chameleon
un - friend - ly f-r-ε-n-d
Figure 4.1: A simple picture
This is an appealing hypothesis not only because of the parallels to syntax and phonology, but also because linguists seek to provide an elegant description of language structure.
62 
C H A P T E R 4 â•… L E X I C O N
In the context of the lexicon, elegance is often measured in terms of economy, and a morpheme lexicon is maximally economical. To see how this works, consider again the word-forms belonging to the Modern Greek noun lexeme filos ‘friend’ (repeated from (2.2)): (4.1) The paradigm of filos singular nominative fílos accusative fílo genitive fílu
plural fíli fílus fílon
Combined, these six word-forms contain seven unique morphemes (fil-, -os, -o, -u, -i, -us, and -on). It might therefore seem more efficient to store the individual complex word-forms: six versus seven lexical entries. However, there are of course other lexemes whose word-forms follow the same morphological patterns as filos, including kosmos ‘world’ (kósm-os, kósm-o, kósm-u, kósm-i, kósm-us, kósm-on), fovos ‘fear’ (fóv-os, …), gamos ‘marriage’ (gám-os, …), and skilos ‘dog’ (skíl-os, …), just to list a few. If all word-forms are directly listed in the lexicon, there are thirty lexical entries corresponding to just these five lexemes. However, if only the morphemes are listed, there are eleven corresponding lexical entries, because the suffixes do not need to be repeated: fíl-, kósm-, fóv-, gám-, skíl-, -os, -o, -u, -i, -us and -on. Multiplied by thousands of Greek verb lexemes, the morpheme approach becomes much more economical than listing each individual word-form. Note that economical representation is not restricted to inflectionally related forms. The English word-forms read, reader and readable belong to the same word family (i.e. they are derivationally related), and the wordforms write, writer and writeable, and many other sets of words, are related in a parallel fashion. In principle, it is thus possible to posit a lexicon that is quite economical – one lexical entry per morpheme. Unfortunately, a morpheme lexicon is not usually as elegant in practice as it is in principle. There are several kinds of complications that force even a theory seeking a minimal lexicon to posit some word-forms as lexical entries. In fact, we have already anticipated two potential problems in preceding chapters – unpredictability of meaning and lack of morpheme segmentability. First, if the lexicon consists primarily of separate morphemes that are combined together to form words, the meaning of a complex word should be equal to the sum of the meanings of its component morphemes. Stated differently, the word should exhibit compositional meaning. But as we have already seen, this kind of direct relationship between form and meaning does not always occur, and derivational morphology presents a particular problem in this regard. A reader is not just any person who reads, but also a kind of textbook and the title of an academic job (in the British system). These last two meanings are not predictable from the meanings of read and -er individually; the meaning is non-compositional. This indicates
4 . 1 ╇ A M O R P H E M E L E X I C O N ?
63
that reader (textbook) and Reader (British academic title) are probably represented in the lexicon as complex words, rather than according to the component morphemes. The hypothesis that the lexicon consists (almost) exclusively of morphemes thus faces the same practical problem that has led dictionary-makers to give one entry to each lexeme – the meaning of a derived lexeme is often more than the sum of the meanings of the component parts. Since many languages have a large number of derived lexemes with unpredictable meaning, there is correspondingly a significant problem for the hypothesis of a morpheme lexicon. Another potential problem for a morpheme lexicon is lack of morpheme segmentability. Here we briefly describe four types: base modification, cumulative expression, zero expression and empty morphs. We have already encountered examples of base modification in Chapter 3. A familiar example, plurals in German, is repeated as (4.2). Some German plurals are formed by replacing a back vowel of the singular form by a front vowel ([υ], [a:], [ɔ] are replaced by [y], [e:], [ø], spelled ü, ä, ö). Although this is clearly a morphological pattern, because an aspect of form corresponds to an aspect of meaning, it is not possible to segment a proper morpheme meaning ‘plural’. (4.2)
singular plural Mutter Mütter Vater Väter Tochter Töchter Nagel Nägel
‘mother(s)’ ‘father(s)’ ‘daughter(s)’ ‘nail(s)’
Second, when an affix expresses two different morphological meanings simultaneously, we have cumulative expression (also called fusion). For example, the Serbian noun ovca ‘sheep’ has the number and case forms shown in (4.3). (4.3) nominative accusative genitive dative instrumental vocative
singular ovc-a ovc-u ovc-e ovc-i ovc-om ovc-o
plural ovc-e ovc-e ovac-a ovc-ama ovc-ama ovc-e
Clearly, it is not possible to isolate separate singular or plural or nominative or accusative (etc.) morphemes. The suffixes that follow the stem ov(a)c- express number and case simultaneously, or, in the technical term of morphology, cumulatively. Cumulative or fused expression is most often illustrated with different inflectional meanings, but it is also possible for an inflectional meaning and a derivational meaning to be expressed cumuÂ� latively. In Krongo, a language of Sudan, the derivational meaning ‘agent’
64 
C H A P T E R 4 â•… L E X I C O N
and the inflectional meanings ‘singular’ and ‘plural’ are expressed in a single affix: cà-/cò- denotes ‘agent/singular’, and kà-/kò- denotes ‘agent/plural’. (4.4) màlìŋ mòtò
‘theft’ cà-màlìŋ ‘thief’ kà-màlìŋ ‘thieves’ ‘work’ cò-mòtò ‘worker’ kò-mòtò ‘workers’ (Reh 1985: 157)
A suppletive stem may also simultaneously express the base meaning and the grammatical meaning. Thus, English worse expresses the lexeme meaning ‘bad’ and the inflectional meaning ‘comparative’ in a cumulative way. Affixes and stems that cumulatively express two meanings that would be expected to be expressed separately are also called portmanteau morphs. A particularly important phenomenon that causes problems for segmentation is the existence of words in which a morphological meaning corresponds to no overt form, i.e. a zero affix (or zero expression). Two examples are given in (4.5) and (4.6). (4.5)
Coptic cˇcoˉ-i cˇcoˉ-k cˇcoˉ cˇcoˉ-f cˇcoˉ-s
‘my head’ ‘your (m) head’ ‘your (f) head’ ‘his head’ ‘her head’
(4.6)
Finnish oli-n oli-t oli oli-mme oli-tte oli-vat
‘I was’ ‘you were’ ‘he/she was’ ‘we were’ ‘you(pl) were’ ‘they were’
(Layton 2000: 69, 103)
Some morphologists have worked with the requirement that the segmentation of words into morphemes must be exhaustive and all meanings must be assigned to a morpheme. If we adopt this requirement, then we are forced to posit zero morphemes here that have a meaning, but no form (so Finnish oli would really have the structure oli-Ø, where the morpheme Ø stands for the third person singular, and Coptic cˇoˉ would formally have the structure cˇoˉ-Ø). But the requirement is not necessary, and alternatively one could say, for instance, that Finnish has no marker for the third person singular in verbs. To be sure, the practical difference between the affixation of an unpronounced element and no affixation at all is not great, but at a conceptual level the two approaches are substantially different. And it does seem to be the case that the latter is less far-fetched and cognitively more plausible. Finally, the opposite of zero affixes can also be found: apparent cases of morphemes that have form but no meaning (called empty morphs). For example, in Lezgian all nominal case-forms except for the absolutive case (i.e.
4 . 1 ╇ A M O R P H E M E L E X I C O N ?
65
the most basic case) arguably contain a suffix that follows the noun stem and precedes the case suffix. In (4.7), four of Lezgian’s sixteen cases are shown. (4.7) absolutive genitive dative subessive
sew sew-re-n sew-re-z sew-re-k ‘bear’
fil fil-di-n fil-di-z fil-di-k ‘elephant’
Rahim Rahim-a-n Rahim-a-z Rahim-a-k (male name) (Haspelmath 1993: 74–5)
Under this analysis, the suffixes -re, -di and -a have no meaning, but they must be posited if we want to have a maximally general description. With the notion of an empty morph we can say that different nouns select different suppletive stem suffixes, but that the actual case suffixes that are affixed to the stem are uniform for all nouns. The alternative would be to say that the genitive suffix has several different suppletive allomorphs (-ren, -din, -an), the dative case has several different allomorphs (-rez, -diz, -az), and so on. But such a description would be Â� inelegant, missing the obvious and exceptionless generalization that the non-absolutive case suffixes share an element. In all four of these examples we find a similar problem for morphological segmentation. We can identify a morphological pattern which applies to the word, but it is difficult or impossible to segment a proper morpheme. This has undesirable consequences for the hypothesis of a morpheme lexicon in at least two ways. First, in Section 1.1 we defined a morpheme as the smallest meaningful part of a linguistic expression that can be identified by segmentation. However, empty morphs have no meaning, and zero affixes have no form and therefore cannot be segmented. So positing zero affixes and empty morphs allows for a maximally economical lexicon, but if we incorporate these devices, we must expand the idea of what counts as a ‘morpheme’ lexical entry. And allowing lexical entries that can have form but no meaning, or the reverse, greatly reduces the restrictiveness of the morpheme-based model. Second, non-segmentable morphological patterns may force the morphological system to become more complicated in other ways. For instance, in Tiv, some classes of words express the imperative with a high tone on the final syllable. This is thus a kind of base modification. (In this analysis, the low tones on non-final syllables are filled in by default.) (4.8)
root kimbi kəŋgəsə de gba vá
imperative kìmbí kə`ŋgə` sə´ dé gbá vá
gloss ‘pay’ ‘chew cud’ ‘leave’ ‘fall’ ‘come’ (based on Abraham 1940: 29)
Since this pattern is completely regular, it is possible here to argue that the lexicon contains only the root, and that the imperative is derived by
C H A P T E R 4 â•… L E X I C O N
66 
a rule of tone assignment. However, note that under such an analysis, the imperative meaning is carried by the rule, not by an affixal lexical entry. So some inflectional meanings are associated with lexical entries, and some with rules. It seems undesirable to complicate our description of the morphological system in this way, but this problem cannot be easily resolved if the primary goal is to maintain a maximally economical lexicon. In short, a morpheme lexicon seeks to minimize the information in the lexicon by subsuming as much information as possible under general principles of grammar, and including in the lexicon only that information that is unpredictable. This approach has the advantage of being highly elegant if (and only if) it is empirically adequate, and if a minimal lexicon does not lead to complications elsewhere in the morphological system. However, a morpheme lexicon often runs into one or both kinds of problems, depending upon the morphological patterns of a given language. There are thus quite a few problems that are faced by any attempt to make morphemes (in the sense of minimal morphological constituents) the cornerstone of morphological analysis and the basic unit of the lexicon. The major issues are summarized in Table 4.1. Problems for a morpheme lexicon
Example
Unpredictable meaning of derived lexemes
Reader (British academic) does not mean read + -er
Lack of morpheme segmentability Base modification Cumulative expression Zero expression Empty morphs
German plurals, e.g. Mütter Serbian noun paradigm, e.g. ovc-a Finnish third person singular, e.g. oli Lezgian non-absolutive noun paradigm, e.g., sew-re-n
Table 4.1╇ Problems for a morpheme lexicon: summary
4.2╇ A strict word-form lexicon? An alternative hypothesis is that the lexicon consists entirely of wordforms, both simple and complex. This approach is free of the problems with a morpheme lexicon that we have identified. For example, since meaning is associated in the lexicon with word-forms, not with morphemes, the meaning of a word-form need not be equal to the combined meanings of its morphemes. And morpheme segmentability becomes a significantly lesser problem if morphemes are not the basic units of the lexicon.
4 . 2 ╇ A S T R I C T W O R D - F O R M L E X I C O N ?
67
A word-form lexicon also has a number of other advantages. For example, it helps to explain traits unique to morphology, such as lack of productivity. Morphological patterns that can be used to create new words are called productive. Both derivational and inflectional patterns are often productive. Thus, the German plural suffix -en (e.g. Fahrt ‘trip’, plural Fahrt-en ‘trips’) can create new words when it is applied to new bases such as loanwords (e.g. Box ‘loudspeaker unit’, borrowed from English box, in German receives the plural Box-en). True novel words are far less common than novel sentences, and most of the time we use words that we have used many times before. But, in principle, morphology is like syntax in that it may be productive. From this perspective, what is really remarkable about morphology is that morphological patterns may also be unproductive. For example, there are a number of English action nouns containing -al (some of which are listed in (4.9a)). As the hypothetical but unacceptable forms in (4.9b) show, there are many verbs to which this suffix cannot be applied. (4.9) a. refusal, revival, dismissal, upheaval, arrival, bestowal, denial, betrayal b. *repairal, *ignoral, *amusal, *belial, *debuggal But the crucial point is one that cannot be made by giving examples: the suffix -al cannot be used at all to form novel lexemes in English. The list of nouns formed with -al is fixed (it contains 35 nouns according to the OED), and no new nouns can be added to this list. The reason why languages may have unproductive morphological patterns is that complex words, like simple words, are stored in the lexicon. Since it cannot be predicted that these verbs have an action noun in -al, the lexicon contains the nouns arrival, refusal, denial, in addition to arrive, refuse, and deny. When English speakers use a noun like arrival, in all likelihood they simply retrieve it from their lexicon rather than constructing it on the fly. A variety of complex words must therefore be listed in the lexicon. At the very least, the list includes complex words for which a suffix is unproductive and thus unpredictable (like arrival), and those for which the meaning is unpredictable (like Reader in the British academic sense). Faced with these ‘exceptions’, we could conclude that there is overwhelming evidence in favour of word-based structure, that the word-based model is therefore superior to the morpheme-based model and that we do not need morphemes at all in morphology. This is the essence of a strict word-form lexicon, and a number of morphologists have drawn this conclusion. But at this stage a few words of caution seem in order. There are some apparent, and some real, problems with the hypothesis of a strict word-form lexicon. First, should we worry about the inherent lack of elegance in a theory that lists all words? The answer depends upon which goal(s) of morphological research we consider most important. In addition to elegant description,
68 
C H A P T E R 4 â•… L E X I C O N
in our morphological description we aim for cognitive realism, systemexternal explanation and restrictive architecture. There are indications that a word-form lexicon is more cognitively realistic than a morpheme lexicon. Speakers remember a word not only if it is unpredictable, but also if it is very frequent. This is a general feature of animal (including human) cognition: the more often something is encountered, the more easily it is remembered (for instance, the more often a pianist plays a piece, the sooner she or he will be able to play it by heart). This applies to words, whether predictable or unpredictable, as to anything else. This is thus a classic case in which different goals of morphological research lead to a conflict. And to the degree that cognitive realism is given greater priority than elegance, we need not be very concerned about the lack of elegance of a word- form lexicon. Of course, if we prioritize elegance, we would not find a word-form lexicon to be very satisfactory. Second, a common argument against a strict word-form lexicon relates to agglutinative languages such as Turkish. In Turkish, words can be quite long; see (4.10). According to one count, 20% of Turkish words have at least five morphemes (Hankamer 1989), and it is possible (though certainly not common) for Turkish verbs to contain ten or more inflectional morphemes. (4.10) a. oku-r-sa-m read-aor-cond-1sg ‘If I read…’ b. oku-malı-y-mıs¸-ız read-nec-be-rep.pst-1pl ‘They say that we have to read.’ c. okú-ya-ma-yabil-ir-im read-pot-neg-pot-aor-1sg ‘I might not be able to read.’
(Kornfilt 1997: 367–75)
Moreover, the inflectional system contains dozens of verbal affixes. Not all can co-occur (e.g. a word can have only one subject agreement marker). Still, the combinatory possibilities entail that every verb root can appear in a very large number of word-forms – at least 2,000. Multiplied by thousands of verbs, it seems completely impossible to memorize all forms of all verbs that a speaker might want to use (Hankamer 1989). Certainly agglutinative languages present a challenge to claims of a strict word-form lexicon, but it is not clear whether languages like Turkish are really problematic. It may be possible to assume a weaker version of the word-form lexicon, according to which a speaker memorizes all wordforms that they have heard, or that they have heard a certain number of times. This issue remains to be investigated. A more significant problem is that there is some evidence that speakers themselves see words as consisting of morphemes. Some linguists have
4 . 2 ╇ A S T R I C T W O R D - F O R M L E X I C O N ?
69
claimed that morphological rules never make reference to word-internal structure, so there is no need to assume that words ‘have structure’ once they have been formed according to the rules. But this does not seem quite right. For one thing, allomorphy is often conditioned by the morphological structure of the base. For example, Dutch past participles are marked by the prefix ge- (e.g. spreken ‘speak’, Â� ge-sproken ‘spoken’) unless the verb bears a derivational prefix such as be- (e.g. be-spreken ‘discuss’, be-sproken ‘discussed’, not *ge-be-sproken). The Sanskrit converb is formed by the suffix -tvaˉ if the verb has no prefix (e.g. ga-tvaˉ ‘having gone’, nıˉ-tvaˉ ‘having led’), but by the suffix -ya if the verb has a prefix (e.g. aˉ-gam-ya ‘having come’, not *aˉ-ga-tvaˉ; *pari-n.ˉı -tvaˉ) (Carstairs-McCarthy 1993). This pattern would be very difficult to describe if we think that speakers have no knowledge of which stems have prefixes. Finally, morphemes also seem to have relevance for phonology. For example, many languages have phonological morpheme structure conditions – i.e. restrictions on the co-occurrence of sounds within a morpheme. For example, English allows combinations such as [tθ] and [dθ] in complex words like eighth and width, but not within a single morpheme. German allows syllable-final consonant clusters such as [rpsts] as in Herbst-s (genitive of Herbst ‘autumn’), but within a single morpheme four consonants (e.g. [rpst]) are the maximum. In addition, alternations may be sensitive to morpheme boundaries. Standard Northern Italian has an alternation in the pronunciation of s between [s] and [z], whereby [z] is chosen if the s occurs between vowels (e.g. casa [-z-] ‘house’) and [s] is chosen elsewhere (e.g. santo [s-] ‘saint’). However, if the s is morpheme-initial, it is pronounced [s] even if it occurs between vowels (e.g. asimmetrico [-s-] ‘asymmetric’, Â�risocializzare ‘resocialize’) (Baroni 2001). These phenomena, too, seem to require that we recognize morphemes as real entities. A strict word-form lexicon is thus faced with a number of problems, which are summarized in Table 4.2. These facts make the strict word-form lexicon hypothesis less than fully satisfactory. Problems for a strict word-form lexicon
Example
Some morphological patterns necessarily refer to morpheme-based structures
Conditions on Dutch past participle prefix ge-
Some phonological patterns necessarily refer to morpheme-based structures
Italian asimmetrico pronounced with [s], not [z], because of morpheme boundary
Speakers are unlikely to memorize all word-forms in rich inflectional languages
Turkish 
Table 4.2╇ Problems for a strict word-form lexicon: summary
C H A P T E R 4 â•… L E X I C O N
70 
4.3╇ Reconciling word-forms and morphemes Thankfully, the evidence for morphemes as real entities is compatible with a lexicon in which most lexical entries consist of (complex) words. This is because it is possible to treat a morpheme as a generalization based on word-forms in the lexicon. For instance, according to the hypothesis of a moderate word-form lexicon, both the complex lexemes in (4.11) and the morphemes in (4.12) can be lexical entries. The latter are descriptions of patterns found among words in the lexicon. (4.11) word lexical entries (Russian) a. /ruka/N b. /ruku/N ‘hand.nom.sg’ ‘hand.acc.sg’
[ [ [
]
]
[ [ [
]
]
c. /rba/N d. /rbu/N ‘fish.nom.sg’ ‘fish.acc.sg’
]
]
e. /sestra/N f. /sestru/N ‘sister.nom.sg’ ‘sister.acc.sg’ (4.12) word-schema lexical entries (Russian) a. suffixes /Xa/N /Xu/N ‘x.nom.sg’ ‘x.acc.sg’
[
]
[
]
b. roots /rukX/N /rbX/N /sestrX/N ‘hand’╇╇ ‘fish’╇╇╇ ‘sister’╇╇╇
[
]
[
]
[
]
Note that in (4.11), the lexical entries are presented using the wordschema formalism from Chapter 3. This formalism is useful because under the hypothesis of a moderate word-form lexicon, it is more accurate to say that a morphological pattern, rather than a morpheme, can be a lexical entry. In other words, ‘morpheme’ lexical entries need not be restricted to roots or affixes in a moderate word-form lexicon. We will often continue to talk about the contents of the lexicon in terms of morphemes for the sake of convenience, but it is important to remember that the same principles also apply to morphological patterns that are not as easily described in terms of morphemes. The primary difference between this approach and the strict word-form lexicon lies in the status of morphological patterns. In a moderate word-form lexicon, word-forms are still primary, but morphological patterns, including ones that we can identify as ‘morphemes’, are allowed a secondary role and they can be lexical entries. Under this view, many complex words will be listed in the lexicon, but some will be composed on the fly from component parts when needed. From here on, the terms word-form lexicon and word-
4 . 3 ╇ R E C O N C I L I N G W O R D - F O R M S A N D M O R P H E M E S
71
based model will be used to refer to this approach, rather than the more rigid word-form-only approach to the lexicon discussed above. In Section 3.2.2, where the word-based model was first discussed, it was taken for granted that both predictable complex word-forms and morphological patterns are lexical entries. In other words, that discussion already assumed the hypothesis of the moderate word-form lexicon, without calling it that or justifying it. We now look at evidence for this position. A moderate word-form lexicon faces an immediate challenge. If the lexicon consists of both word-forms and morphemes, in our description we must determine which words are directly stored in the lexicon, and which are composed on the fly from morphemes. While most morphologists agree that all simple and at least some complex words are listed in speakers’ lexicons, it is difficult to say which complex words are listed and which ones are not. Part of the issue has to do with the fact that any language contains both words that are familiar to most speakers (such as mis-represent and global-ize in English) and words that are novel and were perhaps never used before (such as mis-transliterate and bagel-ize, two words that we have just made up). Morphologists refer to these two types of words as actual words and Â�possible words (or usual and potential words). Thus, the set of words in a language is never quite fixed. Speakers have the capacity to create, and hearers can understand, an almost unlimited number of new words. Dictionaries can record only the actual words, but at any time a speaker may use a possible (but non-actual) word, and, if it is picked up by other speakers, it may join the set of actual words (thus, if the number of bagel shops in Europe continues to grow, people will perhaps start saying that Europe is being bagelized). Attested novel lexemes that were not observed before in the language are called neologisms, and neologisms that do not really catch on and are restricted to occasional occurrences are called occasionalisms (or nonce formations). Most occasionalisms are probably never recorded, and, even among those that are recorded, many disappear soon afterwards. For instance, in 1943 the new word deglamorize was observed and recorded by a linguist, perhaps because it was used repeatedly around that time (Algeo 1991). But it seems that the word has not caught on and has not really become part of the English lexicon (even though the OED records it). Around the same time, the word decolonize arose. This word was more successful, and most English speakers nowadays know it. It has thus become a truly actual word of English. On a practical level, then, it is simply not possible to document which rare words have been encountered (and sufficiently well remembered) by which speakers. With a word-form lexicon (strict or moderate), we would be forced to decide how established a word must be in the language to be memorized by speakers. Additionally, and more importantly, many linguists argue that the hypothesis of a moderate word-form lexicon also faces a complex challenge
72 
C H A P T E R 4 â•… L E X I C O N
related to how speakers recall words stored in the lexicon. Imagine that the language user has just heard the word insane. If both morphemes and wordforms are stored in the lexicon, that person can retrieve the meaning of the word from his mental lexicon in two different ways – either by breaking the word into its component morphemes and looking each up (in-, sane), or by looking up the word-form directly (insane), assuming it is stored. The process of looking up a word in the lexicon is known as lexical access. When lexical access occurs by breaking up words into morphemes, this is the (morphological) decomposition route; retrieving complex word-forms without decomposition is the direct route. Which method is a language user more likely to employ? Some linguists argue that the answer is, in some sense, both. There is evidence that when speakers need to retrieve a word from the lexicon, they try both routes simultaneously. The ‘winner’ is whichever method is faster in accessing the information. Lexical access is thus a kind of race which exists to make the mental work of processing language more efficient. This postulation is represented in Figure 4.2. The solid lines indicate lexical access via the decomposition route. The dashed line indicates lexical access via the direct route. (We will return to the thickness of the circles in the discussion below.)
insane
in
sane
“... insane...”
Figure 4.2:╇ A schematized dual-route model of lexical access (Hay 2001: 1045)
Figure 4.2 A schematized dual-route model of lexical access (Hay 2001)
Given that the decomposition route and the direct route compete to retrieve words, a natural question is: What determines which route wins? There seems to be a variety of relevant factors; here we discuss three:
4 . 3 ╇ R E C O N C I L I N G W O R D - F O R M S A N D M O R P H E M E S
73
frequency, morpheme segmentability and allomorphy. These factors may all support the same route, or they may conflict. Lexical access, and by extension also the contents of the lexicon, are thus quite complicated.2 First, there is a general consensus among linguists that speakers have detailed (subconscious) knowledge of how frequently a word is used. Metaphorically, this can be thought of as a representation of how well a person remembers a word. In a basic sense, lexical entries (both wordforms and morphemes) that are used more frequently – i.e. items that have a higher token frequency – are more firmly established in a person’s memory, and have a stronger representation in the lexicon. We will say that these words have greater memory strength. Those that are used less have less memory strength. In Figure 4.2 above, the thickness of the circles represents token frequency (thicker = more frequently used = greater memory strength). Frequently used words can be accessed more quickly. As a consequence, token frequency strongly influences the representation of complex words in the lexicon: if the word-form has a higher frequency than its base (all other factors being equal), it is more likely to be accessed via the direct route (Hay 2001). This is the scenario represented in Figure 4.2. In this case, the complex word insane is more common in English than the word sane. According to one count, the word-form sane is used about 8 times in every 1 million words of written text. Insane is used 14 times per million words (CELEX English database, Baayen et al. 1995). This difference leads insane to have greater memory strength, and the direct route – represented by the dashed line – wins. Conversely, whenever the root is more common (again, all else equal), decomposition is more likely. The second factor affecting whether a complex word is stored in the lexicon is segmentability. Complex words with segmentable affixes are intuitively more likely to be stored according to these affixes. Words that display morphological patterns which are less segmentable, such as base modification, are more likely to be stored as word-forms. Third, affixes that induce allomorphy in the base to which the affix attaches are less likely to be decomposed into morphemes than are affixes which do not cause allomorphy in the base. Examples of phonological allomorphy induced by the English suffix -ity are given in (4.13a). The final [k] in electric changes to an [s] in electricity, and the stress changes syllables. In divinity, the vowel changes quality. By contrast in the (b) examples, adding -ship to a base does not change its phonological shape.
2
It should be noted that lexical access is different from lexical content. This means that how a lexical entry is accessed is not exactly the same as what is listed in the lexicon. In this chapter we are interested in the latter issue. Nonetheless, the two are closely related, so we will assume that the same factors are important for both.
74 
C H A P T E R 4 â•… L E X I C O N
(4.13) a. 0007base allomorphy induced by –ity: electric – electricity, divine – divinity b. 0007no base allomorphy induced by –ship: ambassador – ambassadorship, representative – representativeship We can perhaps think of these last two issues in terms of the saliency of the internal morphemic structure of a complex word-form. When morphemes are segmentable and do not have multiple allomorphs, the internal structure may be more salient because the boundaries between morphemes are easier to identify. A summary of all three factors is given in Table 4.3. Factors
Direction of influence
Relative token frequency of word-form Segmentability Allomorphy
More frequent Æ word-form storage Less segmentable Æ word-form storage More effect on base Æ word-form storage
Table 4.3╇ Three factors influencing lexical storage: summary
Finally, an important issue is whether decomposition entails full decomposition. The discussion above considered only words with two morphemes, but of course, words can have more than two morphemes, e.g. insanely (in+sane+ly). It is unlikely that insanely is directly listed in the lexicon because insanely is used much less frequently than is insane (less than one time in every one million words), and all three morphemes are easily segmented. But it is not entirely clear how insanely is decomposed. Is it stored in the lexicon with three separate lexical entries (in-, sane, -ly), or as two (insane, -ly)? There is some evidence that the latter is probably correct in this case, based primarily on the relative frequency of sane and insane. Decomposition thus does not necessarily entail full decomposition into component morphemes. Ultimately, the structure of the lexicon is a very active area of research, but the evidence increasingly suggests that while both morphemes and words are listed, the balance is easily tipped in favour of word-based storage. This suggests that the lexicon, and thus morphology more generally, are fundamentally word-based. While a moderate word-form lexicon is not very economical, research to date suggests that it is the most cognitively realistic of the three proposals.
FURTHER READING 
75
Summary of Chapter 4 A language user’s mental dictionary is the lexicon. A major theoretical issue is whether the lexicon consists of morphemes or word-forms. This is important because lexical entries are the fundamental building blocks of morphological structure. There are several problems with the hypothesis of a morpheme lexicon, including the non-compositional meaning of many derived lexemes, and problems breaking words into morphemes, including base modification patterns, cumulative expression, zero affixes and empty morphs. Still, there are also problems with the strict word-form lexicon hypothesis, for example, some types of morphological rules seem to rely on the concept of the morpheme, and morphemes may have a real status for speakers. Combined with factors (e.g. frequency) that seem to promote decompositional route (morpheme-based) lexical access under some conditions and direct route (word-based) lexical access under others, the best conclusion is that morphological structure is fundamentally word-based, but morphemes (or more properly, morphological patterns) represent secondary generalizations. Word-forms, morphemes and derived stems are all stored in the lexicon.
Further reading Classic works on the lexicon and lexical access include Butterworth (1983) and Taft and Forster (1975). The papers in Feldman (1995) and Baayen and Schreuder (2003) discuss a wide variety of issues related to morphological processing, and the papers in Jarema and Libben (2007) provide current perspectives on the structure of the lexicon from a psycholinguistic perspective. For details of a parallel dual-route model, see Schreuder and Baayen (1999). Also, Hay (2003) demonstrates the role of relative frequency, Jarvikivi et al. (2006) look at allomorphy, and Bertram et al. (2000) discuss more generally the complex interaction of factors which promote or inhibit morphological decomposition. Pinker (1991) and Clahsen et al. (1997) represent an opposing view that words with irregular inflection are accessed as whole word-forms, but regularly inflected words are always decomposed. Similarly, Caramazza et al. (1988) and Taft (1994) give greater weight to morphological decomposition. Non-psycholinguistic models of morphology are also divided over whether morphemes, stems, or whole words are the fundamental units of morphological structure. See the Further reading section in Chapter 3 for references.
76 
C H A P T E R 4 â•… L E X I C O N
Comprehension exercises 1. Which of the following English words are actual, possible and impossible?
replay, libertarian, itinerance, reknow, fraternitarian, penchance, rebagelize, abundance, happytarian
2. Which of the following words have compositional meaning, and which have non-compositional meaning?
a. ability, popularity, community, morality, authority b. materialize, modernize, legalize, vaporize, specialize
3. Look again at the words in question 2. According to the hypothesis of a moderate word-form lexicon, the words in (a) are more likely to be stored as whole word-forms than the words in (b). Why? What is the most relevant factor that distinguishes the two groups? (Ignore frequency, since the necessary information is not provided.) 4. For each of the following languages, determine whether the examples exhibit cumulative expression, empty morphs or zero expression. (Some may exhibit more than one of these features.) Explain your answers. a. Finnish pronouns (partial paradigm) 1st p. pl 2nd p. pl 3rd p. pl nom me ‘we’ te ‘you’ he ‘they’ gen meidän teidän heidän par meitä teitä heitä ess meinä teinä heinä iness meissä teissä heissä ela meistä teistä heistä b.
Ndebele imperative verbs root imperative gloss lim- lima ‘cultivate!’ nambith- nambitha ‘taste!’ dl- yidla ‘eat!’ m- yima ‘stand!’ z- yiza ‘come!’ lw- yilwa ‘fight!’ (Inkelas and Zoll 2000: 5)
c. 0007Serbian present tense verbs: govoriti ‘to speak, say’ and tresti ‘to shake’ singular plural 1st person govorim govorimo 2nd person govoriš govorite 3rd person govori govore
E X P L O R AT O R Y E X E R C I S E
1st person 2nd person 3rd person
singular tresem treseš trese
77
plural tresemo tresete tresu
Exploratory exercise In this chapter we argued that frequency has an effect on whether a complex word is directly given in the lexicon. But we did not ask two important methodological questions: if it is possible for a word-form like insane to be stored in the lexicon either via morphemes or directly, how can we, as researchers, know how the word is stored? And how can we know that frequency is an important factor? After all, we cannot directly observe the contents of the lexicon. There are a handful of methods for testing the content of the lexicon, and factors related to it. This research exercise has two purposes: first, to introduce one of these methods, and second, to test the hypothesized relationship between frequency and lexical storage that was presented in this chapter. The basic methodology involves asking speakers to judge how related two words are to each other in meaning. For target items (i.e. the ones that are the focus of interest), one of the words is complex and the other is its base. The assumption underpinning this task is that speakers should judge complex words that are stored as morphemes as being closely related in meaning to their bases because the base and the complex word share a lexical entry. Thus, for example, if insane is stored in the lexicon according to the morphemes in- and sane, the meaning of the entire word should depend closely on the meaning of sane. By contrast, a complex word that is stored as its own lexical entry may be judged as less close in meaning to its root because the two are formally distinct in the lexicon. By manipulating the frequency of the complex word, we can test for a correlation between frequency and meaning closeness. And under the assumption that meaning closeness reflects lexical storage, a correlation (or lack thereof) should indicate whether frequency influences lexical storage. The instructions below use English for demonstration purposes because it is a language that all readers are familiar with. However, this exercise could be conducted using virtually any language. Instructions Step 1: Choose an affix to study. The best choices will be ones that (a) frequently attach to monomorphemic bases, as opposed to only attaching to already-derived stems, and (b) create a new lexeme rather than a wordform of the same lexeme (see Section 2.1 for this distinction). For example, for English we might choose to study the suffix -ity (obscurity, immensity, etc.).
78 
C H A P T E R 4 â•… L E X I C O N
Step 2: Create words to be used in the experiment. Make a list of words containing the affix that you chose in Step 1. Also use a frequency dictionary (i.e. a dictionary that gives counts of how frequently a particular word-form is used) or a written or spoken corpus of the language to gather information about the token frequency of these words. Then, sort them into ‘frequent’ and ‘infrequent’ groups. (You may want to remove words of intermediate frequency, to create two maximally distinct frequency groups.) A few examples for -ity are given below. Of course, you will need a longer list of words. Complex word
Freq/Infreq
Root
obscurity acidity modernity opportunity priority security
infrequent infrequent infrequent frequent frequent frequent
obscure acid modern opportune prior secure
Table 4.4╇ Exploratory exercise: possible stimuli
Tips for finding words: For some languages, online dictionaries can be searched to find all words that contain some sequence of letters in a particular word position (e.g. the letters ity at the end of the word). Look for ones that allow wildcard searching. Some languages also have freely available online corpora that can be used in the same way. Also, if you have chosen a suffix rather than a prefix, find out whether the language has a reverse dictionary. In a reverse dictionary, entries are alphabetized from the end of the word to the beginning, rather than the usual beginning-to-end method. This groups all words with a given suffix together (assuming the word ends with the suffix), making examples easier to find. Next, add a variety of ‘filler’ words. The fillers should be derived lexemes that range in semantic similarity to their roots, but do not contain the affix that you chose in Step 1. For example, at one end of the scale might be words like helpful, which is very similar in meaning to help, at the other end of the scale words like awful, which is not at all similar to awe, and in the middle words like artful. You might even want to include ‘false’ derived lexemes. For example, defend looks like it can be broken into two morphemes (de-fend) because de- is sometimes a prefix (e.g. de-regulate) and fend is a real word, but this would be a false segmentation. The fillers are included primarily to create a wide range of pairs of words. They also help to distract the speaker from the purpose of the study, namely studying the effects of frequency. (People do strange things when they think they know what you want them to say!)
E X P L O R AT O R Y E X E R C I S E
79
Write each of the target words and its root on an index card, with the complex word first. For example, one card might be obscurity-obscure. Do the same for the filler words. Alternative procedure: Hay (2001) argues that relative frequency is more important for lexical access than absolute frequency. In other words, as far as the lexicon is concerned, a frequent complex lexeme is one that is more frequent than its base, regardless of how often it is used in absolute terms. An infrequent complex lexeme is one that is less frequent than its base. As an alternative in Step 3, choose words that are frequent or infrequent in this relative sense. (Warning: This is much more difficult!) Step 3: Review the discussion in Section 4.3 about frequency. This discussion represents the hypothesis about the relationship between frequency and word-form storage in the lexicon. Based on this discussion, develop specific predictions. Predictions are expectations about how the data will turn out if the hypothesis is correct. Based on what you have read, how do you expect frequency of the complex word to be related to speakers’ judgements about semantic relatedness of a complex word and its root? For example, do you expect the words to be judged to be further apart in meaning when the complex word is frequent? Do you expect the opposite? Or no relationship between the frequency and meaning closeness? Explain the rationale behind your prediction. Remember to also consider what data you would expect to find if the hypothesis is not true. Step 4: Run the study. Find native speakers of the language who are willing to participate in the project. Several participants is ideal, but even a few people can produce interesting results. Present them with each card. Ask them to ‘Rate how much the meaning of the first word is related to the meaning of the second word on a scale from 1 to 7 in which 7 means “very related” and 1 means ”not at all related”.’ Write down the rating for each card and each speaker. Hint: The target word cards should be presented in random order, with fillers mixed in. However, it helps if the first 5–10 cards are filler pairs representing different points on the scale. This allows the subject to practise and calibrate her/his judgements. Step 5: Analyze the data. Determine whether ratings differ depending upon the frequency of the complex word. Was your prediction upheld? At the same time, also consider factors that are not directly related to the research questions but may have affected the results. For example, did all speakers give similar ratings? Did all frequent (or infrequent) target words produce similar ratings? If not, this may suggest other issues that the study did not take into consideration. Also think about the impact of the methodology. For example, is semantic transparency a good measure of word-based vs. morpheme-based listing? Why or why not? Will a complex word that has its own lexical entry always lead speakers to assign lower semantic relatedness ratings? Why or why not? Step 6: Draw conclusions. If your prediction was upheld, does this
80 
C H A P T E R 4 â•… L E X I C O N
suggest that the hypothesis is correct? Why or why not? If your prediction was not upheld, what does this suggest? In short, what do the data indicate about the relationship between frequency, semantic transparency and the content of the lexicon?
5.1 INFLECTIONAL VALUES 
Inflection and derivation
81
5
I
n this chapter, we discuss inflection and derivation in greater depth. As we saw in Section 2.1, this conceptual distinction is quite basic to most morphological theorizing and terminology, though it is not always easy to determine the relation between two word-forms: does nicely belong to a separate lexeme from nice, or are both word-forms in the paradigm of nice? In other words, is the suffix -ly that is attached to nice to form nicely a derivational suffix or an inflectional suffix? We survey inflectional functions in Section 5.1 and derivational meanings in Section 5.2. In Section 5.3 we examine a range of properties that have been proposed as distinguishing between inflection and derivation, and between two subtypes of inflection. Section 5.4 gives an overview of the ways in which the relation between inflection and derivation has been conceptualized by morphologists. The two most important views are the dichotomy approach, which argues that complex words can be neatly divided into two disjoint classes, and the continuum approach, which claims that morphological patterns are best understood as lying on a continuum ranging from the most clearly inflectional patterns to the most clearly derivational patterns. Finally, in Section 5.5 we briefly show some implications of these views for how linguists model the relationship between morphology and syntax.
5.1 Inflectional values Morphologists usually talk in quite different terms about inflection and derivation. For instance, the different inflectional formations are referred to as expressing inflectional values (or inflectional feature values), so we say, for instance, that English verbs express the inflectional values ‘present’ (e.g. (he/she) walks) and ‘past’ (e.g. (he/she) walked). But for derived lexemes like walker we would not normally say that it represents a ‘derivational value’
82 
C H A P T E R 5 â•… I N F L E C T I O N A N D D E R I VAT I O N
(‘agent noun’) – instead we simply talk about derivational meanings. One reason for this distinction is that inflectional values often do not have a clearly identifiable meaning, only a syntactic function. For example, walks differs from walk in that walks is used when the subject is third person singular (she, he, it) and walk is used with other subjects (I, you, we, they), but many linguists feel uncomfortable calling this a difference of meaning because it is quite abstract. Different languages vary quite dramatically in the amount of inflectional complexity that their words exhibit. Some languages, such as Vietnamese and Igbo, a language of Nigeria, have no (or virtually no) inflectional values, and others have inflection for more than a dozen values (though it is unÂ�common for a single word-form to be inflected for more than half a dozen values). However, despite all this diversity, the types of inflectional values that we find across languages are surprisingly uniform. Perhaps more than two-thirds of all inflectional values fall into one of the classes of Table 5.1. On nouns, pronouns
On verbs
On adjectives, demonstratives, relative pronouns, adpositions
number (singular, plural,…)
number (singular, plural,…)
number (singular, plural,…)
case (nominative, accusative,…)
person (1st, 2nd, 3rd)
case (nominative, accusative,…)
gender (masculine, feminine,…)
tense (present, future, past, …)
gender (masculine, feminine,…)
person (1st, 2nd, 3rd)
aspect (perfective, imperfective, habitual, …)
person (1st, 2nd, 3rd)
mood (indicative, subjunctive, imperative,…) Table 5.1╇ Common inflectional features and values
As the organization of Table 5.1 suggests, inflectional values are often naturally grouped together into super�-categories that we will call
5 . 1 I N F L E C T I O N A L VA L U E S
83
inflectional features.1 Two values belong to the same feature if they share a semantic (or more generally, functional) property and are mutually exclusive. For instance, the English present and past tenses both have to do with when an event happens, relative to the moment of speaking (so they share a semantic property), and they cannot occur together in the same verb (they are mutually exclusive). Thus, they are values of the feature ‘tense’. We have already seen number and case inflection of nouns, e.g. in Latin (repeated as Figure 5.1).2 The number feature is self-evident; it indicates quantity. Case indicates the semantic and syntactic role of a noun in a sentence. A given case may express many roles, but one is usually considered basic. Among the Latin cases, nominative canonically marks subjects and is the citation form, accusative marks direct objects, and dative marks indirect objects. Genitive canonically indicates the possessor (‘s in student’s book is a genitive marker and one of the few case values in English), and ablative means ‘movement away from’.
number
case
SINGULAR
PLURAL
NOMINATIVE
insula
insulae
ACCUSATIVE
insulam
insulaˉs
GENITIVE
insulae
insula¯rum
DATIVE
insulae
insulı¯s
ABLATIVE
insula¯
insulı¯s
Figure 5.1╇ Case and number in Latin
Latin is a fairly typical language in terms of number: most languages mark singular and plural on nouns. Fewer distinguish a dual number, and even fewer a paucal number (paucal means ‘a few’). Languages vary in the number of morphological cases they express; Latin has five cases, but many languages have no case distinctions at all, and a few have more than ten different cases. Typically, inflectional values of number and inflectional values of case combine freely, as shown.
1
2
Some linguists use the term inflectional category for our inflectional feature, and inflectional property for our inflectional (feature) value. See the Appendix of this chapter for notation conventions related to inflectional values.
84 
C H A P T E R 5 â•… I N F L E C T I O N A N D D E R I VAT I O N
Person distinctions are widely marked on verbs in the world’s languages, but only in a limited way in English (and not even all dialects of English). The only verb to fully mark person values is be; there are distinct forms of be depending upon whether the subject refers to the speaker (first person), the addressee (second person) or a third party (third person), i.e. (I) am, (you) are, (he/she/it) is. In some languages, the verb marks person according to the value of the object, rather than the subject, or according to both (see (5.6) in Section 5.3.1 for an example from Yimas). The features tense, aspect and mood exist to some extent in virtually all languages that have any inflection at all. Tense indicates the temporal location of a verb’s action (past, present, future). Aspect has to do with the internal temporal constituency of an event, for example, whether the action is viewed as completed (perfective), non-completed (imperfective), habitual, etc. Finally, mood denotes the certainty, desirability or conditionality of an event. It subsumes a wider range of inflectional values, including the imperative (commands), subjunctive (non-realized events) and indicative (events viewed as objective facts). The three feature names ‘tense’, ‘aspect’ and ‘mood’ suggest that values from these different features can be combined in the same way that case and number, or person and number, can be freely combined. Indeed, this is sometimes possible, for instance, in Latin, which has three tense values (present, past, future), two aspect values (infectum and perfectum; the latter is similar to the English perfect) and two mood values (indicative and subjunctive). See Figure 5.2.
INDICATIVE INFECTUM PRESENT PAST FUTURE
SUBJUNCTIVE
PERFECTUM
canta-t canta-v-it canta-ba-t canta-v-era-t canta-bi-t canta-v-eri-t
INFECTUM PRESENT PAST FUTURE
PERFECTUM
cant-e-t canta-v-eri-t canta-re-t canta-v-isse-t — —
Figure 5.2╇ Latin tense, aspect and mood forms (third person singular)
However, the Latin system does not have all possible combinations: there are no future subjunctive forms. Moreover, this system is quite atypical in being as symmetrical as it is. In most languages, different inflectional values for tense, aspect and mood are difficult to combine. A language that contrasts with Latin in this respect is Swahili, where tense, aspect and mood are expressed by inflectional prefixes. In Figure 5.3, forms with the prefix n(i)- (first person singular) are given. Here, there are no obvious formal reasons for setting up such a paradigm with two mood values, three tense
5 . 1 I N F L E C T I O N A L VA L U E S
85
INDICATIVE
NORMAL PROGRESSIVE PERFECT
PRESENT
PAST
FUTURE
n-a-fanya ni-na-fanya ni-me-fanya
ni-li-fanya — —
ni-ta-fanya — —
PRESENT
PAST
HYPOTHETICAL NORMAL PROGRESSIVE PERFECT
ni-n ge-fanya ni-n gali-fanya — — — —
FUTURE
— — —
Figure 5.3╇ Swahili tense, aspect and mood forms (first person singular, -fanya ‘do’)
values and three aspect values because there are no word-forms to express most of the combinations. From a formal point of view, positing just a single feature (‘tense/aspect/mood’) with seven values is simpler and does not seem to miss crucial generalizations. Thus, many linguists nowadays work with a single feature ‘tense/aspect/mood’. The explanation for the different behaviour of the combinations ‘case + number’ and ‘tense + aspect + mood’ lies in their semantics. All combinations of different cases and numbers are roughly equally plausible because the syntactic role of a noun in a sentence is logically independent of whether the noun refers to one or many entities. By contrast, certain combinations of aspect, tense and mood are unusual or downright exotic. For instance, perfective aspect (which implies that an event is viewed in its totality) does not go together well with present tense (which implies that the speaker is still in the middle of the event). Even more obviously, the imperative mood (which expresses a command) does not combine with the past tense. It is not surprising that most languages lack straightforward inflectional means for these combinations. Besides the inflectional values that we have seen up to now, there are quite a few others that are less easy to generalize about, but that are also less widespread. In English, adjectives have inflectional markers of comparative and superlative degree (big, bigger, biggest), but this kind of inflection is not common in the world’s languages – it seems to be largely confined to the languages of Europe and south-western Asia. In verbs, some languages have passive voice inflection, which indicates an unusual association of semantic roles and syntactic functions: the semantic patient is the syntactic subject (e.g. Swedish kasta ‘throw’, kasta-s ‘be thrown’). This is the opposite of active voice, in which the semantic agent is the syntactic subject. (For more on passives, see Section 11.1.2.) And
86 
C H A P T E R 5 â•… I N F L E C T I O N A N D D E R I VAT I O N
many languages have inflectional expression of polarity (i.e. affirmative versus negative, e.g. Japanese kir-u [cut-prs] ‘cuts’, kir-ana-i [cut-neg-prs] ‘doesn’t cut’). Finally, the most important kind of inflection that we have not already discussed is the group of dependent verb forms. Many languages have special verb forms that are confined to dependent clauses. Although the terminology is not uniform, a rough generalization says that verb forms marking relative clauses are called participles, verb forms marking adverbial clauses are called converbs and verb forms marking complement clauses are called infinitives or masdars. Examples of a participle, a converb and an infinitive are given in (5.1)–(5.3). (5.1) Korean participle Hankwuk-ul pangmwunha-nun salam-i nul-ko iss-ta. Korea-acc visit-ptcp person-nom increase-ing be-decl ‘Those who visit Korea are increasing.’ (S.-J. Chang 1996: 148) (5.2) Hindi/Urdu converb Banie ke bet.e ne cit.t.hii likh-kor d.aak me˜ d.aal-ii. grocer poss son erg letter(f).sg write-cvb box in put.pst-f.sg ‘The grocer’s son wrote and posted a letter.’ (lit. ‘having written a letter, posted (it).’) (5.3) Mparntwe Arrernte infinitive Re lhe-tyeke ahentyene-ke. she go-inf want-pst ‘She wanted to go.’
(Wilkins 1989: 451)
5.2 Derivational meanings Derivational meanings are much more diverse than inflectional values. Besides cross-linguistically widespread meanings such as agent noun (e.g. drinkv Æ drink-ern), quality noun (e.g. kinda Æ kind-nessn) and facilitative adjective (e.g. readv Æ read-ablea), we also find highly specific meanings that are confined to a few languages. For instance, Big Nambas, a language of the South Pacific island of Vanuatu, has a suffix -et that derives reverential terms from ordinary nouns (e.g. dui ‘man’ Æ dui-et ‘sacred man’, navanel ‘road’ Æ navanel-et ‘sacred road’ (Fox 1979)). And French has a suffix -ier that derives words for fruit trees from the corresponding fruit nouns (e.g. pomme ‘apple’ Æ pomm-ier ‘apple tree’, poire ‘pear’ Æ poir-ier ‘pear tree’, prune ‘plum’ Æ prun-ier ‘plum tree’).
5 . 2 D E R I VAT I O N A L M E A N I N G S
87
There are too many types of derivational meaning to present here, but it is worth discussing one frequent characteristic of derivation. Derivational patterns commonly change the word-class of the base lexeme – i.e. nouns can be derived from verbs, adjectives from nouns, and so on. For such cases, the terms denominal (‘derived from a noun’), Â�deverbal (‘derived from a verb’) and deadjectival (‘derived from an adjective’) are in general use.
5.2.1 Derived nouns Since creating new words for new concepts is one of the chief functions of derivational morphology, and since we have a greater need for naming diverse nominal concepts, languages generally have more means for deriving nouns than for deriving verbs and adjectives (Bauer 2002). Some common meanings with examples from various languages are listed in Table 5.2. III. Deverbal nouns (V Æ N) agent noun3 English Arabic patient noun English instrument noun Spanish action noun Russian
drinkv Æ ˉh amalav Æ ‘carry’ invitev Æ picarv Æ ‘mince’ otkry-t’v Æ ‘discover’
drink-ern ˉh ammaaln ‘carrier’ invit-een pica-doran ‘meat grinder’ otkry-tien ‘discovery’
III. Deadjectival nouns (A Æ N) quality noun Japanese person noun Russian
atarasi-ia Æ ‘new’ umn-yja Æ ‘smart, clever’
atarasi-san ‘newness’ umn-ikn ‘clever guy’
III. Denominal nouns (N Æ N) diminutive noun Spanish augmentative noun Russian status noun English inhabitant noun Arabic female noun German
gat-o Æ ‘cat’ borod-a Æ ‘beard’ child Æ Mis.r Æ ‘Egypt’ König Æ ‘king’
gat-it-o ‘little cat’ borod-išcˇ a ‘huge beard’ child-hood mis.r-iyyu ‘Egyptian’ König-in ‘queen’
Table 5.2╇ Common derivational meanings of nouns
3
The glossary gives definitions of the derivational meanings in Tables 5.2–5.4.
88 
C H A P T E R 5 â•… I N F L E C T I O N A N D D E R I VAT I O N
Besides these widespread derivational meanings, many more specific derivational meanings are found in languages, but usually these are restricted to a few languages each. Thus, Russian has a suffix for nouns denoting kinds of meat (e.g. kon’ ‘horse’, kon-ina ‘horse meat’). Tagalog has a pattern for nouns meaning vendors (e.g. kandila ‘candle’, magkakandila ‘candle vendor’ (Schachter and Otanes 1972: 103)). Various sciences have developed terminological conventions for creating new technical terms by suffixation (e.g. -itis as a suffix for inflammatory diseases, -ite as a suffix for minerals, -ide and -ate as suffixes for certain kinds of chemicals, and so on).
5.2.2 Derived verbs Verb-deriving patterns are generally less numerous and diverse. Most commonly, verbs are derived from other verbs. Denominal and deadjectival verbs are much less widespread than deverbal verbs (Bauer 2002). Again, some typical examples are given in Table 5.3. III. Deverbal verbs (V Æ V) causative verb Korean ╇╇ (see Section 11.1.4) applicative verb German ╇╇ (see Section 11.1.5) anticausative verb Swedish ╇╇ (see Section 11.1.2) desiderative verb Greenlandic ╇ repetitive verb English reversive verb Swahili
cwuk- Æ ‘die’ laden Æ ‘load’ öppna Æ ‘open (tr.)’ sini- Æ ‘sleep’ write Æ chom-a Æ ‘stick in’
cwuk-i‘kill’ be-laden ‘load onto’ öppna-s ‘open (intr.)’ sini-kkuma‘want to sleep’ re-write chom-o-a ‘pull out’
III. Denominal verbs (N Æ V) ‘act like N’ Spanish ‘put into N’ English ‘cover with N’ Russian
pirat-a Æ ‘pirate’ bottlen Æ sol’ Æ ‘salt’
pirat-ear ‘pirate’ bottlev sol-it’ ‘salt’
III. Deadjectival verbs (A Æ V) factitive Russian inchoative Spanish
cˇ ern-yj Æ ‘black’ verde Æ ‘green’
cˇ ern-it’ ‘make black’ verde-ar ‘become green’
Table 5.3╇ Common derivational meanings of verbs
5 . 3 P R O P E R T I E S O F I N F L E C T I O N A N D D E R I VAT I O N
89
5.2.3 Derived adjectives Derived adjectives are even less common than derived verbs, because adjectives are used more rarely than verbs, let alone nouns. Moreover, the semantic class of adjectives that is the most developed in a number of European languages, denominal relational adjectives (of the type government Æ governmental), seems to be quite rare in other areas of the world. Typical examples of derived adjectives are shown in Table 5.4. III. Deverbal adjectives (V Æ A) facilitative Basque agentive Spanish III. Denominal adjectives (N Æ A) relational Russian ╇╇ (= ‘related to N’) proprietive Ponapean ╇╇ (= ‘having N’) privative Russian ╇╇ (= ‘lacking N’) material German III. Deadjectival adjectives (A Æ A) attenuative Tzutujil intensive Turkish negative German
jan Æ ‘eat’ habla-r Æ ‘talk’
jan-garri ‘edible’ habla-dor ‘talkative’
korol’ Æ ‘king’ pihl Æ ‘water’ vod-a Æ ‘water’ Kupfer Æ ‘copper’
korol-evskij ‘royal’ pil-en ‘watery’ bez-vod-nyj ‘waterless’ kupfer-n ‘made of copper’
kaq Æ ‘red’ yeni Æ ‘new’ schön Æ ‘beautiful’
kaq-koj ‘reddish’ yep-yeni ‘brand new’ un-schön ‘ugly’
Table 5.4╇ Common derivational meanings of adjectives
5.3 Properties of inflection and derivation Let us now look at the properties of inflectional and derivational morphology. The ultimate goal is to determine whether inflection and derivation have sufficiently different traits as to suggest that they represent two distinct subsystems in morphological architecture. We will call the hypothesis of a formal distinction the dichotomy approach. The other possibility is that we should model inflection and derivation as a continuum, with canonical inflection at one extreme, and canonical derivation at the other, but many
C H A P T E R 5 â•… I N F L E C T I O N A N D D E R I VAT I O N
90 
intermediary types. We call this the continuum approach. This choice has broad consequences for the relationship between morphology and syntax (discussed in Section 5.5 below), but we start by looking at some of the empirical facts underlying this debate. Table 5.5 gives an overview of properties that differentiate inflection and derivation. Some of these are all-or-nothing properties, and others are relative, i.e. a complex word may have the property to a greater or lesser extent. We discuss these in turn below.
Inflection
Derivation
vii(i) relevant to the syntax not relevant to the syntax vi(ii) obligatory expression of feature not obligatory expression v(iii) unlimited applicability possibly limited applicability i (iv) same concept as base new concept ii (v) relatively abstract meaning relatively concrete meaning i (vi) compositional meaning possibly non-compositional meaning (vii) expression at word periphery expression close to the base (viii) less base allomorphy more base allomorphy i (ix) no change of word-class sometimes changes word-class ii (x) cumulative expression possible no cumulative expression i (xi) not iterable possibly iterable Table 5.5╇ A list of properties of inflection and derivation
5.3.1 Relevance to the syntax (i)
Inflection is relevant to the syntax; derivation is not relevant to the syntax.
For the most part, ‘relevant to the syntax’ means that the grammatical function or meaning expressed by a morphological pattern is involved in syntactic agreement or syntactic government. In syntactic government, one word requires another word or phrase to have a particular inflectional value. For instance, Polish verbs that are negated often require a direct object in the genitive case (5.4). Verbs that are not negated require a direct object in the accusative case (5.5). Since the presence or absence of negation leads to a difference of case marking on the object, case must be relevant to the syntax. It is therefore possible to conclude that case is inflectional in Polish according to criterion (i).
5 . 3 P R O P E R T I E S O F I N F L E C T I O N A N D D E R I VAT I O N
91
(5.4) Tomek nie czytał gazet-y tomek.m.nom.sg not read.3sg.m.pst newspaper-gen.sg ‘Tomek was not reading a newspaper.’ (5.5) Tomek czytał tomek.m.nom.sg read.3sg.m.pst ‘Tomek was reading a newspaper.’
gazet-e˛ newspaper-acc.sg
Agreement is a kind of syntactic relation in which the inflectional value of a word or phrase (the target) must be the same as the inflectional value of another word of phrase in the sentence (the controller) to which it is closely related. For instance, in [the boy]np [walk-s]v and the [girl-s]np [walk]v, the target verb walk(s) agrees with the subject NP in number.4 And in this girl and these boys, the target demonstrative this/these agrees with its head noun (girl/boys) in number. Looking back at Table 5.1, we can notice that the most common inflectional features for nouns and pronouns are the same as the most common inflectional features for adjectives, demonstratives, relative pronouns and adpositions. This is because in agreement relations, the controller is almost always a noun, pronoun or noun phrase. Adjectives, demonstratives, etc. are typical targets for noun agreement. Verbs are also frequent targets for number, person, and sometimes gender agreement. A word-form may agree with a controller for multiple features, and/or agree with multiple controllers. Examples of agreement are shown in (5.6)– (5.8). (5.6) 0007Agreement of verb with subject and object in person, number and gender (Yimas) Krayŋ narmaŋ k-n-tay. frog.sg(g6) woman.sg(g2) 3sg.g6.p-3sg.g2.ag-see ‘The woman saw the frog.’ (Foley 1991: 194) (5.7) 0007Agreement of preposition with complement NP in person and number (Classical Nahuatl) i-pan noyac 3sg-on my.nose ‘on my nose’ (Sullivan 1988: 108)
4
NP stands for noun phrase.
C H A P T E R 5 â•… I N F L E C T I O N A N D D E R I VAT I O N
92 
(5.8) 0007Agreement of demonstrative and adjective in number and gender (Swahili) wa-le wa-tu wa-refu pl.g2-that pl-person(g2) pl.g2-tall ‘those tall people’ Note that agreement features are sometimes overtly marked only on the target. For instance, in Italian, determiners and adjectives agree with nouns for gender. But while Italian nouns are all lexically associated with one of the two genders, they do not have morphological marking for gender. Examples like il poeta ‘the poet’, la casa ‘the house’, la mano ‘the hand’, il cuoco ‘the cook’, la chiave ‘the key’, il fiume ‘the river’ (il = masculine article, la = feminine article) show that -a does not in general mean ‘feminine’, and -o does not mean ‘masculine’ (despite this commonly being taught in language classrooms). In Italian, only the determiners and adjectives have morphological gender marking. The criterion of syntactic relevance covers most of the features listed in Section 5.1, but there is one problematic area: tense/aspect/mood patterns are not obviously relevant to the syntax. Tense, and even more so aspect, hardly ever occur in an agreement-like relationship, and are not otherwise assigned by the syntax. Do we have to consider tense/aspect/mood to be derivational by this criterion, rather than inflectional? A slightly modified interpretation of ‘relevant to the syntax’ resolves this problem. Specifically, it turns out that certain syntactic rules seem to require reference to tense and aspect; this allows us to include tense/aspect/mood under the rubric of inflection according to this criterion.
5.3.2 Obligatoriness (ii)
Inflectional features are obligatorily expressed on all applicable word-forms. Derivational meanings are not obligatorily expressed.
This can be illustrated by Latin: the lexeme insula ‘island’ has ten wordforms in its paradigm, and each word-form expresses (and must express) one value from each of the features ‘number’ and ‘case’. The Latin speaker thus had no choice about whether to use a noun with or without case and number features – omitting these features was impossible. Number and case are thus inflectional features in Latin according to this criterion. (Note that inflection need not be expressed via an overt suffix. For instance, the paradigm of the Spanish verb camina ‘walk’ contains the form camina ‘s/he walks’, with no affix directly corresponding to the third person singular meaning. But here the absence of an affix is meaningful in itself; this is not an uninflected form, but an inflected form with zero expression.)
5 . 3 P R O P E R T I E S O F I N F L E C T I O N A N D D E R I VAT I O N
93
By contrast, expression of a given derivational meaning is not obligatory. The English suffix -er applies to verbs to derive nouns with the meaning of ‘agent’, e.g. drinker. But it is not the case that all nouns must express an agentive meaning.
5.3.3 Limitations on application (iii) Inflectional values can be applied to their base without arbitrary limitations; derivational formations may be limited in an arbitrary way. Generally speaking, a lexeme’s paradigm contains a full set of inflected forms: verbal paradigms have word-forms expressing all the tense–aspect–mood values that are relevant to the language, noun paradigms have word-forms expressing all relevant case-number combinations, adjectival paradigms have all relevant comparative forms, etc. This is because a lexeme that does not have a full set of forms cannot function in every syntactic context. And notably, when exceptions do occur, this can usually be explained easily by the incompatibility of the inflectional meaning and the base meaning, i.e. the problematic syntactic context never arises in the first place. For instance, stative verbs may not have certain aspectual forms (e.g. English *She is knowing me), collective nouns may have only singular or only plural forms (e.g. English information, *informations), and non-gradable adjectives do not have comparative forms (e.g. *Mammoths are deader than Neanderthals). Incomplete paradigms whose gaps are not semantically motivated are very rare (see Section 8.7 for discussion of these exceptions). In comparison, arbitrary derivational gaps are quite common. Conceivable derived lexemes may be lacking without any obvious semantic explanation. For instance, English has female nouns in -ess such as authoress, heiress, priestess, but it is not possible to say *professoress ‘female professor’, *presidentess ‘female president’, and so on, although these make perfect sense semantically. The Spanish inchoative formation in -ear (see Table 5.3) occurs with colour adjectives (verde Æ verdear ‘become green’, negro Æ negrear ‘become black’, etc.), but it cannot be used freely with other adjectives where a ‘become’ sense would be just as appropriate and useful (e.g. caro Æ *carear ‘become expensive’ – this word does not exist).
5.3.4 Same concept as base Some properties are best discussed in terms of canonical inflectional traits and canonical derivational traits. (iv) Canonical inflected word-forms express the same concept as the base; canonical derived lexemes express a new concept.
C H A P T E R 5 â•… I N F L E C T I O N A N D D E R I VAT I O N
94 
While everyone would probably agree that the same concept is expressed in go and goes, or in Latin insula (‘island.nom.sg’) and insulae (‘island.gen.sg’), this is less clear with singular–plural pairs in nouns. For instance, at one point in the history of English the plural of brother was brethren. But at a later stage, brethren took on the specialized meaning of members of a Christian fellowship, and came to be interpreted as a separate lexeme. A new plural (brothers) was created to pair with brother in the meaning of male sibling. This kind of split into separate lexemes shows that the singular brother and plural brethren expressed somewhat different concepts. Number in nouns is thus sometimes more similar to derivation according to criterion (iv) than to canonical inflection. And on the other side, derivation does not always lead to an obviously new concept. Although ‘baker’ is clearly a different concept from ‘bake’, in what sense is ‘kindness’ a different concept from ‘kind’? This example also seems to fall into the middle ground between inflection and derivation.
5.3.5 Abstractness (v)
Inflectional values express a relatively abstract Derivational meanings are relatively concrete.
meaning;
The abstractness criterion works quite well for inflectional meanings, because all of them are highly abstract (in some intuitive sense). And many derivational meanings are quite concrete (e.g. French -ier, which denotes a kind of tree). But there are also derivational meanings that are just as abstract as inflectional meanings (e.g. the meaning ‘status’ of -hood in Â�childhood). So -hood is neither canonically derivational, nor canonically inflectional.
5.3.6 Meaning compositionality i(vi) Canonical inflected word-forms have compositional meaning; canonical derived lexemes have non-compositional meaning. While inflectional values usually make a predictable semantic contribution to their base, derived lexemes are often semantically idiosyncratic. For instance, the Russian derivational suffix -nik means ‘thing associated with (base concept)’, and this meaning is clearly present in dnev-nik ‘diary’ (dn-ev- ‘day’), nocˇ-nik ‘night lamp; night worker’ (nocˇ’ ‘night’). However, the meaning of dnevnik is not exhausted by that of dnev- and -nik: a diary is indeed a kind of thing associated with days (or daily activities), but the additional meaning components ‘notebook’ and ‘used for writing’ cannot
5 . 3 P R O P E R T I E S O F I N F L E C T I O N A N D D E R I VAT I O N
95
be predicted on the basis of the meaning of the two constituent morphemes and must be associated with the lexeme as a whole. Even clearer examples are ignorance and reparation; their meanings are probably only historically related to ignore and repair. But of course, some derivational formations exhibit compositional meaning. For instance, the meaning of German female nouns with -in (König-in ‘queen’, Professor-in ‘female professor’) is very regular. The suffix -in is by most criteria clearly derivational, but in this respect it shows a property that is more typical of inflection.
5.3.7 Position relative to base (vii) Canonical inflection is expressed at the periphery of words; canonical derivation is expressed close to the root. This property can be used as distinguishing criteria only in special circumstances because it is a relative property, and not an absolute one. The first is best illustrated by words that have one derivational affix and one inflectional affix on the same side of the root. In such cases, the derivational affix almost always occurs between the root and the inflectional affix: (5.9) a. English b. English c. English d. Turkish e. Arabic
king-dom-s root – status (D) – plural (I) real-ize-d root – factitive (D) – past tense (I) luck-i-er root – proprietive (D) – comparative (I) iç-ir-iyor root – causative (D) – imperfective aspect (I) [drink-caus-impf.3sg] ‘makes (somebody) drink’ na-ta-labbasa 1st plural subject (I) – reflexive (D) – root [1pl-refl-clothe.prf] ‘we clothed ourselves’
When there are more than two affixes, normally all the derivational affixes occur closer to the root than the inflectional affixes (e.g. German nation-al-isier-te-n ‘(they) nationalized’: root – relational adjective (D) – factitive verb (D) – past tense (I) – third person plural subject agreement (I)). Yet here again, it is possible to find examples of inflection and derivation that do not exhibit the canonical traits. For example, German has deadjectival factitive verbs that are based on the inflectional comparative form (e.g. schön ‘beautiful’ Æ schön-er ‘more beautiful’ Æ ver-schön-er-n ‘make more beautiful’); the inflectional comparative affix -er is closer to the root than the derivational affix -n. And English allows plurals inside many compounds (e.g. publications list, New York Jets fan).
C H A P T E R 5 â•… I N F L E C T I O N A N D D E R I VAT I O N
96 
5.3.8 Base allomorphy (viii) Inflection induces less base allomorphy; derivation induces more base allomorphy. Base allomorphy is also a relative property. It can be demonstrated with roots that show allomorphy in derived lexemes, but not in inflected wordforms: (5.10) a. English b. English c. German d. Latin e. Italian f. Arabic
root destroy broad Erde ‘earth’ honor ‘honour’ dialogo [-g-] ‘dialogue’ kataba ‘he wrote’
inflected form destroy-ed broad-er Erde-n ‘earths (pl)’ honor-is ‘honour-gen’ dialogh-i [-g-] ‘dialogue-s’ katab-tu ‘I wrote’
derived lexeme destruc-tion bread-th ird-isch ‘earthly’ hones-tus ‘honest’ dialogico [-dʒ-] ‘dialogical’ kitaab ‘book’
But the opposite pattern can also be found, as in the following examples from Serbian: (5.11)
root junak ‘hero (m)’ pesnik ‘poet (m)’ psiholog ‘psychologist (m)’ monah ‘monk’
inflected form junac-i ‘heroes’ pesnic-i ‘poets’ psiholoz-i ‘psychologists’ monas-i ‘monks’
derived lexeme junak-inja ‘heroine’ pesnik-inja ‘poet (f)’ psiholog-inja ‘psychologist (f)’ monah-inja ‘nun’
Base allomorphy is thus yet another tendency, according to which a morphological pattern may be more typically inflectional, or more typically derivational.
5.3.9 Word-class change (ix) Canonical inflection does not change the word-class of the base; derivational affixes may change the word-class of the base.
5 . 3 P R O P E R T I E S O F I N F L E C T I O N A N D D E R I VAT I O N
97
It is often claimed that derivational formations may change the word-class of the base, but inflectional ones never do. While the first part of this claim is true (e.g. many of the examples in Section 5.2 consisted of word-classchanging derivation), the second part is questionable. There seem to be two types of word-class-changing morphological patterns – one that is typical of derivational patterns, and one that is associated with patterns that are, to some degree, inflectional. In most instances, when word-class-changing affixes are added to a base, the grammatical properties of the base are no longer relevant for purposes of agreement. In (5.12), the Russian adjective otkrytoe ‘open’ agrees for gender (and number and case) with the noun okno ‘window’, but when the denominal adjective okonnaja ‘window (adj)’ is derived from okno, the nominal stem can no longer serve as the controller for agreement. This is a typical consequence of (derivational) word-class-changing operations, and the reason (5.12b) is ungrammatical. (5.12) Russian a. otkryt-oe open-n.sg.nom ‘open window’
okno window.n.sg.nom
b. *otkryt-oe okon-naja rama open-n.sg.nom window-f.sg.nom frame.f.sg.nom ‘open window frame’ (i.e. ‘frame of an open window’) However, some languages have affixes that can occur in structures parallel to (5.12b). Consider the example in (5.13), from another Slavic language: Upper Sorbian. Here, mojeho ‘my’ agrees for gender with the masculine noun muž ‘husband’, despite this being the root of the denominal adjective mužowa. (5.13) moj-eho muž-ow-a my-m.sg.gen husband-poss-f.sg.nom ‘my husband’s sister’
sotra sister.f.sg.nom (Corbett 1987: 303)
Thus, -ow seems to have the property that it is word-class changing, but in a way that allows the properties of its base to still control agreement by a modifying adjective. Moreover, and crucially, -ow meets some of the criteria for inflection, e.g. it has fully compositional meaning. It is also highly productive (a trait not discussed in detail here, but one that is typical of inflection). Perhaps, then, no clear-cut binary distinction can be made between (derivational) affixes that change word-class, and (inflectional) affixes that do not (Corbett
C H A P T E R 5 â•… I N F L E C T I O N A N D D E R I VAT I O N
98 
1987; Haspelmath 1996). It is better to say that canonical inflection does not change word-class. Word-class-changing inflection is discussed further in Section 11.4.
5.3.10 Cumulative expression (x)
Inflectional values may be expressed cumulatively; derivational meanings are not expressed cumulatively.
This criterion applies only to a small subset of cases, but is nevertheless interesting. We saw above that several inflectional values may be expressed by a single affix, as in Latin insulaˉrum ‘of the islands’, where the suffix -aˉrum expresses both ‘genitive’ and ‘plural’. Such cases of cumulation seem to be very rare in derivational formations, but a possible example is Dutch -ster ‘agent’ and ‘female’.
5.3.11 Iteration (xi) Inflectional values cannot be iterated; derivational meanings can sometimes be iterated. Inflection is more restricted in that inflectional affixes cannot be iterated. Thus, although it would make sense logically to have an iterated plural (e.g. *cat-s-es ‘sets of cats’), such double plurals are virtually unattested. Or one could imagine a past-tense affix to be repeated to give a sense of remote past (e.g. *didded ‘had done’). With derivational formations, iteration is not common either, but it is possible, for instance, with diminutives in Afrikaans (kind-jie-tjie ‘a little little child’), and with various prefixes in English (post-post-modern) and German (Ur-ur-ur-großvater ‘great-greatgreat-grandfather’). Another instance is the double causative, as we find it in Huallaga Quechua: wañu- ‘die’, wañu-chi- ‘kill’, wañu-chi-chi- ‘cause to kill’ (Weber 1989: 164). This property also applies to only a small number of morphological patterns, those for which iteration would be semantically plausible.
5.4 Dichotomy or continuum? Do these facts indicate that the difference between inflection and derivation is a dichotomy, or a continuum? It turns out that there is less disagreement about the facts themselves, and more disagreement about the importance of some facts. Proponents of the dichotomy approach tend to consider the first
5.4 DICHOTOMY OR CONTINUUM? 
99
three properties in Section 5.3 (relevance to the syntax, obligatoriness and limitations on applicability) to be the most important, especially relevance to the syntax. And inasmuch as these three criteria are logically independent of each other, but nonetheless tend to categorize a given morphological meaning in the same way, either as derivation or as inflection, proponents of the dichotomy approach have argued that these traits are indicative of a distinction between inflection and derivation in the formal architecture of the morphological system. By contrast, the reason some morphologists prefer the continuum approach is that they want to avoid making an arbitrary choice from the criteria in Table 5.5. Proponents of the continuum approach thus tend to consider the properties as a collective whole. If all these criteria are taken seriously, then the continuum approach is almost inevitable, because different criteria may point to different conclusions. But what is particularly interesting is that the mismatches between the criteria are not random, but present a surprisingly orderly picture. As an example, let us look at Table 5.6. It gives a sample list of six morphological formations, which are evaluated by five of the eleven criteria.
Language Formation
Example
cum obl new unl cm
English English Spanish English English English
walk/walks song/songs gato/gatito write/rewrite poet/poetess resent/resentment
I D D D D D
3rd singular noun plural diminutive repetitive female noun action noun
I I D D D D
I I I D D D
I I I I D D
I I I I I D
Note: cum= cumulative expression; obl = obligatory; new = new concept; unl = unlimited applicability; cm = compositional meaning.
Table 5.6╇ A continuum from inflection to derivation
Table 5.6 is a simplification in various respects (e.g. in that it ignores the difficulties in applying some of the criteria), but it suffices to illustrate the continuum approach. The English third person singular suffix -s cumulatively expresses person/number and present tense; the other formations show no cumulation. Both verbal agreement and nominal number are arguably present in any verb and noun form, so these two are obligatory, whereas this is not the case for the other formations. Diminutives are like classical inflected forms in that they do not (necessarily) denote a new concept – Spanish gatito often refers to the same kind of cat as gato, but occurs only under special pragmatic circumstances. Only the English female suffix -ess
100 
C H A P T E R 5 â•… I N F L E C T I O N A N D D E R I VAT I O N
and the action-noun suffix -ment are limited in applicability, and only -ment is semantically irregular (as we can see in govern/government, which shows a different semantic relation from resent/resentment). On such a continuum view, agreement morphology such as -s (in walks) is canonical inflection and English action nouns in -ment are canonical derivation, but they are merely extremes of a continuum on which many intermediate items are found as well. In the end, linguists must decide the relative merit of these eleven criteria as diagnostics of inflectional/derivational morphology. If we consider (i)–(iii) to be the most important diagnostics, this naturally leads to the conclusion that inflection and derivation are dichotomous categories, and as a result, that inflectional and derivational rules (potentially) operate in the architecture of the language system according to distinct principles. However, if all traits are considered equally important indicators, the best conclusion is that canonical derivation and inflection are end-points along a continuum of morphological properties, and no sharp division can be made between them.
5.4.1 Inherent and contextual inflection Before considering the formal architecture of inflection and derivation, we should consider one other approach. Specifically, one popular modification of the dichotomy approach divides inflectional morphology into two subtypes: inherent inflection and contextual inflection. A tri-partition takes better account of the full range of characteristics described above, while maintaining sharp boundaries between types. Inherent inflection comprises features that are relevant to the syntax but convey a certain amount of independent information. These include a verb’s tense and aspect values, and the number values for nouns. For example, the number value of a noun is mostly dictated by the real-world entity that the noun refers to (the referent) – if the referent is one, the noun is singular. If it is more than one, the noun is usually plural (assuming the language has only two inflectional values for number). In this sense, nominal number contributes independent information to syntactic structure. Some grammatical cases can also be inherent, for instance, locative (e.g. Turkish ev-de [house-loc] ‘in the house’), ablative (e.g. Huallaga Quechua mayu-pita [river-abl] ‘from the river’) and instrumental (e.g. Russian nož-om [knifeins] ‘with a knife’), which similarly make their own semantic contribution and are mostly not syntactically determined. By contrast, contextual inflection comprises values which are assigned to a word because of the syntactic context in which it appears. Included here are structural cases – i.e. cases like nominative, accusative and genitive, which are typically required by syntactic agreement or government but
5.4 DICHOTOMY OR CONTINUUM? 
101
express largely redundant information. Note that inflectional features may be inherent for one word-class and contextual for another. Number is inherent for nouns because it expresses independent information, but contextual for adjectives, verbs, etc. where they agree with the noun for this feature. A distinction between inherent and contextual inflection is appealing in part because inherent inflection often shares properties with derivation, whereas contextual inflection does not to the same degree. Here we highlight three examples. First, although both contextual and inherent inflection often have compositional meaning, on the relatively rare occasion that inflection has an unpredictable, idiosyncratic meaning, the relevant examples come from inherent inflection. Some examples from Dutch are given in (5.14). (5.14) value comparative plural past participle present participle infinitive
inflected word ouder vaders bezeten ontzettend eten
compositional meaning ‘older’ ‘fathers’ ‘possessed’ ‘appalling’ ‘(to) eat’
idiosyncratic meaning ‘parent’ ‘forefathers’ ‘mad’ ‘very’ ‘food’ (Booij 1993)
Plural forms of verbs and other examples of contextual inflection do not seem to exhibit this kind of idiosyncrasy. Second, just as derivational patterns tend to be closer to the root than inflectional patterns, inherent inflectional patterns tend to be closer to the root than contextual ones. Moreover, those exceptional cases in which an inflectional affix is closer to the root than a derivational affix, and those in which an inflectional affix occurs on a first compound member, generally involve inherent inflection (see the discussion following (5.9) for an example). Finally, inherent inflection is more likely to induce base allomorphy than is contextual inflection. A few examples are given in (5.15). (5.15) English German Arabic
contextual inflection sing/sings (person/number agreement) warm-er/warm-e (gender agreement) ‘warm-masc/warm-fem’ kitaab-un/kitaab-in (structural case) ‘book-nom/book-gen’
inherent inflection sing/sang (past tense) warm/wärmer (comparative) ‘warm/warmer’ kitaab-/kutub (plural) ‘book.sg/book.pl’
102 
C H A P T E R 5 â•… I N F L E C T I O N A N D D E R I VAT I O N
Thus, the conceptual distinction between contextual and inherent inflection is useful because there are a number of points on which the two kinds of inflection behave rather differently. Proponents of the continuum approach take this as continued evidence that canonical inflection and canonical derivation are end-points on a continuum of traits, but proponents of the dichotomy approach sometimes work with a tri-partition, rather than a strict dichotomy.
5.5 Inflection, derivation and the syntax-morphology interface Perhaps the most interesting thing about the debate between the dichotomy and continuum approaches is that these perspectives have significant consequences for our formal description of morphological architecture. In particular, morphologists have debated whether inflectional and derivational rules are collected together into a single morphological component of the grammar, or split between two different components. The continuum approach allows for only the former argument. The dichotomy approach is consistent with either, but perhaps more naturally aligned with the proposal that inflection and derivation occupy separate areas of the grammar. The dichotomy and continuum perspectives thus suggest a very different formal relationship between inflection, derivation, and syntax.
5.5.1 The dichotomy approach and split morphology As we have seen, morphologists who adopt the dichotomy approach think of inflection and derivation as having fundamentally different properties, and they usually take the first property of Table 5.5 – relevance to the syntax – as the crucial criterion for distinguishing inflection from derivation. Given this perspective, it is logical to conclude that inflectional and derivational patterns are governed, at least in part, by distinct grammatical principles. In essence, there is not one coherent morphological system, but rather two systems or subsystems – one for derivation and compounding, and one for inflection. What is the relationship of these two systems to each other, and to syntax? One proposal argues that rules of derivation and compounding (i.e. all of word-formation) operate in a component of the grammar that feeds into the syntax, and that inflectional rules apply only after the syntactic rules have applied. In other words, word-formation is pre-syntactic, inflection is postsyntactic. This is referred to as the split-morphology hypothesis, and the architecture of the grammar is shown schematically in Figure 5.4.
5.5 INFLECTION, DERIVATION AND THE SYNTAX-MORPHOLOGY INTERFACE
lexicon simple lexemes word-formation
syntax
inflection
103
phonology, pronunciation
complex lexemes
Figure 5.4╇ Word-formation and inflection in a split-morphology architecture
Let us look at a concrete example of how this works. Our example sentence is (5.16) from Latin. (5.16) Imperator saluta-v-it emperor(nom.sg) greet-prf-3sg ‘The emperor greeted the people.’
popul-um. people-acc.pl
The Latin lexicon contains simple lexemes such as imperare ‘command’, salus ‘health’ and populus ‘people’. The word-formation rules create complex derived lexemes such as imperator ‘commander, emperor’ and salutare ‘greet’. Word-formation is said to operate ‘in the lexicon’ (i.e. in this approach, the lexicon contains both a list and rules), so both simple lexemes and derived lexemes are the output of the lexicon. The syntax contains phrase structure rules (e.g. S Æ NP VP, VP Æ V NP) (‘VP’ is an abbreviation of verb phrase), case-assignment rules, which among other things ensure that the direct object gets accusative case ([vp V NPacc]), and agreement rules, which ensure that the inflectional values on the target match those on the controller. The syntactic rules might thus generate an abstract representation, as in (5.17) (here, the subject NP is the controller for person and number agreement, and the verb is the target). The featurevalue notations at the end of the tree are often called morphosyntactic representations. All of this is of course greatly simplified, but is sufficient for our present purposes. S
(5.17)
NP
VP
case: nom number: sg person: 3rd
[
]
V
[
number: person: tense:
NP
]
sg 3rd prf
[
case: number: person:
acc sg
3rd
]
104 
C H A P T E R 5 â•… I N F L E C T I O N A N D D E R I VAT I O N
The lexemes from the lexicon are inserted into these abstract syntactic representations, yielding a lexically specified syntactic representation as shown in (5.18) (the representation with labelled brackets and subscripts is equivalent to the tree representation in (5.17) and saves space). (5.18) [s [np imperatornom/sg/3rd] [vp [v salutaresg/3rd/prf] [np populus-acc/sg/3rd] ] ] Now the rules of inflection operate and create the correct word-forms from the lexemes with their feature specification: imperatornom/ sg/3rd becomes Â�imperator, salutaresg/3rd/prf becomes salutavit, and populusacc/sg/3rd becomes Â�populum. This gives us the correct output (once phonological rules of pronunciation have applied): Imperator salutavit populum. In addition to its intuitive plausibility, this architecture of the formal grammar is often claimed to have one significant advantage: it explains the fact that derivation is generally ‘inside’ inflection, i.e. it occurs closer to the root. If affixes are always attached peripherally by morphological rules, then the affix order of king-dom-s automatically follows from the order of application of the rules in Figure 5.4. The lexicon creates kingdom from the simple lexeme king, and the inflection -s is added after the syntactic component. There is no way a form like *king-s-dom could ever arise, because inflected forms like king-s cannot be the input to wordformation rules. Inflected forms should also not occur inside compounds, because compounding is a lexeme-forming rule in the lexicon. Thus, the impossibility of *trees plantation in English follows from this as well (the correct form is tree plantation, where the first part is uninflected, despite the plural meaning). This seems to lend further support to the split-morphology claim. At the same time, however, split morphology encounters some empirical problems. First and most obviously, exceptions in which inflection occurs inside derivation are occasionally observed; we have already encountered the German example ver-schön-er-n ‘make more beautiful’, in which the (inflectional) suffix -er is closer to the root than the (derivational) suffix -n. Such exceptions cannot obviously be accommodated under the splitmorphology proposal. Even more importantly, as we saw in Section 5.4.1, inherent and contextual inflection sometimes behave quite differently, with inherent inflection having characteristics typical of derivation. This presents an intrinsic problem for split morphology, which groups inherent and contextual inflection together. Attentive readers might suggest that perhaps the solution is to maintain the split-morphology hypothesis, but divide the two grammatical components such that derivation, compounding and inherent inflection apply presyntactically, and only contextual inflection applies post-syntactically.
5.5 INFLECTION, DERIVATION AND THE SYNTAX-MORPHOLOGY INTERFACE
105
Based on the evidence presented so far, this would seem to be reasonable. However, further evidence shows that this is not a satisfactory solution either. The most important data here is the fact that the same morphological pattern may express both inherent and contextual inflection. For instance, in Latin nouns, a single morpheme marks both number (inherent) and case (often contextual). This cumulative expression indicates an interaction of inherent and contextual inflection that is extremely difficult to account for if they are located in separate grammatical components. Note that the dichotomy approach does not require us to accept the split-morphology hypothesis. Another possibility is that derivation and inflection operate according to partially different principles, but without them being split into pre- and post-syntactic components of the grammar. Still, split morphology is the most extreme logical interpretation of the dichotomy perspective.
5.5.2. The continuum approach and single-component architecture According to the continuum approach, no firm distinction between inflection and derivation can be made, so inflectional and derivational rules must operate in the same grammatical component according to fundamentally similar grammatical principles. This, of course, has consequences for the relationship between morphology and syntax: the continuum approach is not consistent with the split-morphology hypothesis. Instead, it is consistent with the idea that syntax generates abstract structures containing morphosyntactic representations but no lexical information, as in (5.17), but instead of inserting a (derived) lexeme at each syntactic node and then generating an inflected form, a word-form is inserted that is consistent with the syntactic information. This does not require derivation and inflection to be split between pre- and post-syntactic operations. This grammar architecture is represented in Figure 5.5. In this way, the job of the morphological component is to provide wordforms whose inflectional values match (or at least, do not contradict) those of the morphosyntactic representation.
morphosyntactic representation
word-formation + inflection
phonology, pronunciation
Figure 5.5╇ Word-formation and inflection in a single-component architecture
106 
C H A P T E R 5 â•… I N F L E C T I O N A N D D E R I VAT I O N
The dichotomy approach is also consistent with the idea that all of morphology is collected into one component of the grammatical architecture, but does not align with a single-component hypothesis as naturally as the continuum approach does. The single-component hypothesis cannot be used to explain why derivational affixes occur closer to the root and inflectional affixes occur more peripherally, but perhaps this should not worry us. The ordering of inflectional affixes with respect to derivational affixes is not the only generalization that can be made. Even within inflectional affixes and within derivational affixes, some orders are strongly preferred, and others are strongly dispreferred. For instance, the diminutive suffix in Spanish is always outside other derivational suffixes (e.g. the female noun suffix -es(a): baron-es-ita ‘little baroness’, not *baronitesa). And case suffixes almost always follow number suffixes, rather than vice versa (e.g. Turkish ev-ler-in [house-pl-gen] ‘of the houses’, not *ev-in-ler). There may be a system-external explanation for these affix ordering patterns. It has recently been proposed that the ability of two affixes to combine, and the order in which they do so, depend in part on constraints created by how words are mentally processed (Hay and Plag 2004). Affixes that are more likely to be accessed in the lexicon via the decomposition route (as opposed to the whole-word route) occur outside of affixes that are less likely to be decomposed (see Chapter 4 for discussion of lexical access). Thus, restrictions on the order of affixes may not be something that needs to be captured directly by our description of morphological architecture at all. The possibility of a system-external explanation suggests that the inability of a single-component architecture to explain affix order is not necessarily a disadvantage of the model.
Summary of Chapter 5 Morphologists use different terminology for talking about inflection and derivation. Inflection is described in terms of values grouped into features; derivation is described in terms of individual morphological patterns and their meanings. The range of inflectional meanings found in languages is severely restricted; most of them fall under the general headings of number, gender, case, person, tense, aspect and mood. Derivational meanings are more varied, but many recurrent types can be identified as well. A persistent question in morphological theory is whether differences between inflection and derivation fall along a continuum, or sharply
A P P E N D I X . N O TAT I O N C O N V E N T I O N S F O R I N F L E C T I O N A L VA L U E S
107
divide inflection and derivation into distinct types. Linguists adopting the dichotomy approach have usually emphasized criteria such as relevance to the syntax and obligatoriness, whereas linguists favouring the continuum approach have considered a whole range of criteria, including compositional meaning, cumulative expression and closeness to the root. Within inflection, a distinction between (more derivation-like) inherent inflection and contextual inflection can be made. Formally, the dichotomy approach is consistent with a theory that splits derivational and inflectional rules into separate components that apply pre- and post-syntactically, respectively (the split-morphology hypothesis). By contrast, some formalizations of the dichotomy approach and all instantiations of the continuum approach treat inflection and derivation as comprising a single-morphological area (the single-component hypothesis). While there is little consensus about how distinct inflection and derivation are, a number of empirical issues argue against split morphology.
Appendix. Notation conventions for inflectional values There are three widespread conventions used to represent inflectional values. The format of Figure 5.6, sometimes called a grid, is the standard way to represent the set of word-forms (i.e. the paradigm) of a lexeme.
English PRESENT TENSE
tense
Spanish
PAST TENSE
walk(-s) walked
number
SINGULAR
person
PLURAL
1ST camin-o camina-mos 2ND camina-s caminá-is 3RD camina-Ø camina-n
Figure 5.6╇ Inflectional features and values
In a grid, values of the same feature are shown in columns or rows that are usually labelled with the name of the feature. Each combination of inflectional values defines a cell. In Figure 5.6, values are printed in small capitals and features are enclosed in ovals. The first paradigm is from English, where verbs primarily inflect for tense. The second paradigm is
108 
C H A P T E R 5 â•… I N F L E C T I O N A N D D E R I VAT I O N
from Spanish, where verbs inflect for two values of the feature ‘number’ (singular and plural) and three values of the feature ‘person’ (first, second and third) (they also inflect for tense – see below). When a lexeme inflects for three features simultaneously, a twodimensional representation is no longer sufficient, and we would need a three-dimensional grid. Figure 5.7 is an attempt at drawing one. For practical purposes, three-dimensional (and especially n-dimensional, for n > 3) paradigms are mostly shown in two spatial dimensions as well. Thus, Figure 5.7 is generally replaced by Figure 5.8.
Spanish
number
SINGULAR
person
PLURAL
1ST camin-o 2ND camina-s 3RD camina-Ø Present
camina-mos caminá-is camina-n
1ST camina-ba-Ø 2ND camina-ba-s Past 3RD camina-ba-Ø
tense
caminá-ba-mos camina-ba-is camina-ba-n
Figure 5.7╇ A three-dimensional representation of a three-dimensional paradigm
PRESENT TENSE SINGULAR
1ST camin-o 2ND camina-s 3RD camina-Ø
PLURAL
camina-mos caminá-is camina-n
PAST TENSE SINGULAR
1ST camina-ba-Ø 2ND camina-ba-s 3RD camina-ba-Ø
PLURAL
caminá-ba-mos camina-ba-is camina-ba-n
Figure 5.8╇ A two-dimensional representation of a three-dimensional paradigm
FURTHER READING 
109
The inflectional information contained in a word-form can also be represented in a feature-value notation, as in the examples in (5.19). Feature-value notation indicates the inflectional feature on the left side of the colon, and the inflectional value on the right, with all ‘feature: value’ pairs enclosed in square brackets. This notation is most commonly used when writing lexical entries or word-schemas. Thus, (5.20) is a more informative and formal way to represent (5.21). (5.19) a. Spanish caminábamos tense: past ‘we were walking’ person: 1st number: plural
[
]
b. Sanskrit daˉtr̥n.oh. number: dual ‘of two givers’ gender: neuter case: genitive
[
[
]
[
]
(5.20) /X/n /Xz/n ‘x’ ´ ‘x’ [ number: singular ] [ number: plural ]
[
]
(5.21) /X/n ‘x’
[
´ /Xz/n ‘plurality of xs’
]
]
Finally, for practical purposes, the inflectional values may also be written as subscripts of word-forms, e.g. caminábamos1pl.pst, daˉtr̥n.oh. du.n.gen.
Further reading A useful survey of the kinds of meanings that are expressed by derivational morphology is found in Bauer (2002). The dichotomy and split-morphology approach to inflection and derivation is represented by works such as Scalise (1988a), Perlmutter (1988) and Anderson (1992), and it is implicit in much further work. The continuum approach is defended by Stephany (1982), Bybee (1985), Dressler (1989) and Plank (1994) (and see Wurzel (1996)). The tripartition between contextual inflection, inherent inflection and derivation was proposed by Booij (1993, 1996). In this chapter we assumed that morphology operates largely separately from syntax, and focused on the question of whether there is one morphological component or two. But not all theories make this assumption, including Word Syntax and Distributed Morphology: see
110 
C H A P T E R 5 â•… I N F L E C T I O N A N D D E R I VAT I O N
Toman (1998) and Embick and Noyer (2007), respectively, for overviews of the syntax-morphology interface in these theories. Also, see the discussion in Chapter 9.
Comprehension exercises 1. Give the inflectional information of the following word-forms in feature-value notation (see (5.19)):
Spanish Latin Latin English Serbian Classical Nahuatl English
caminabas insulam cantabit books ovci incal bigger
(3 features, see Figure 5.8) (2 features, see Figure 5.1) (5 features, see Figure 5.2) (1 feature) (2 features, see (4.3)) (2 features, see (2.7)) (1 feature)
2. Lezgian verbs have suffixes for aspect (-zawa imperfective, -nawa perfect, -da habitual), followed by suffixes for polarity (-Ø affirmative, -cˇ negative), followed by suffixes for tense (-Ø present, -j/-ir past; -ir is chosen after -cˇ ). For instance katzawajimpf.aff.pst ‘was running’, katdacˇirhab.neg.pst ‘would not run’. Give the whole three-dimensional paradigm in a two-dimensional representation (as in Figure 5.8), using the verb kat- ‘run’ (i.e. a grid with 3 × 2 × 2 = 12 cells) (Haspelmath 1993). 3. Consider the meanings of the following denominal and deadjectival verbs of English and classify them using the categories of Table 5.3. For some of them, you need to set up new categories not represented in that table.
butter, flatten, categorize, peel, legalize, phone, blacken, cannibalize, unionize, skate, modernize, terrorize, ski
4. At the beginning of this chapter, we asked whether the English deadjectival adverb-forming pattern (nice Æ nicely) is inflectional or derivational. Apply the criteria of Section 5.3 and try to form an opinion on this question.
Exploratory exercise In language research, data sometimes point to opposite conclusions regarding the structure of the linguistic system. And when this happens,
E X P L O R AT O R Y E X E R C I S E
111
linguists do not always agree on which data is most important and convincing. The following is primarily a thought exercise designed to help you develop your own arguments about what constitutes (un)convincing data. It also allows you to practise applying the criteria in Table 5.5. You will use the results to decide for yourself which approach – continuum or dichotomy – seems to be the best way to describe the differences between inflection and derivation. Instructions Step 1: Pick a variety of morphological patterns in a language of your choice. You might want to choose several morphological patterns that express a similar meaning/function, for instance, all of the patterns expressing plurality in nouns, and/or all of the patterns that turn verbs into abstract nouns. English examples for these functions are given in Table 5.7 and Table 5.8 below, but you need not be limited by these (see Sections 5.1 and 5.2 for ideas). In fact, you do not need to work on English at all, but it is helpful to work on a language that you know quite well. Step 2: Using a good descriptive grammar of the language and/or your own knowledge, apply the eleven criteria for distinguishing inflection from derivation. For easy reference, we recommend organizing your analysis as shown in Tables 5.7 and 5.8 (for reasons of space we give only two of the criteria).
Pattern
Example
Obligatory Base allomorphy
-s -s -es ([-iz]) vowel change vowel change vowel change -us Æ -i -on Æ -a no change -en
cat/cats house/houses index/indices goose/geese man/men crisis/crises alumnus/alumni criterion/criteria sheep/sheep ox/oxen
I I I I I I I I I I
I D D ? ? ? I (cross-formation, see (3.33)) I (cross-formation, see (3.33)) I I
Table 5.7╇ Morphological patterns for English plural nouns
Note that it might be difficult or impossible to apply some criteria. For example, it is not clear how the base allomorphy criterion should be applied to morphological patterns that consist of base modification, rather than affixation.
112 
C H A P T E R 5 â•… I N F L E C T I O N A N D D E R I VAT I O N
Pattern
Example
Obligatory
Base allomorphy
-ance/-ence -al -age -ion -ation -ication -ment -y no change
resist/resistance maintain/maintenance deny/denial cover/coverage destroy/destruction hyphenate/hyphenation consult/consultation magnify/magnification endorse/endorsement assemble/assembly walk/walk
D D D D D D D D D D D
I D I I D D D D I I D
Table 5.8╇ Morphological patterns for deverbal abstract nouns in English
Determine whether each morphological pattern is inflectional or derivational. For instance, the pattern represented by cat-cats has no base allomorphy, and number is obligatorily expressed. Both of these are indicative of inflectional patterns, so the criteria lead us to categorize plural –s as inflectional. Conversely, deverbal –ion does not express an obligatory meaning, and does induce base allomorphy, so both criteria indicate that this pattern is derivational. The criteria need not agree, of course (see, e.g. endorse/endorsement and house/houses); use a ‘majority wins’ principle to categorize each morphological pattern as inflectional or derivational. Step 3: Consider the following questions. 1) Are some criteria more reliable than others? In other words, are there any criteria that always indicate that a given pattern is inflectional when the overall majority also categorize it as inflectional (and the same for derivation)? And if some criteria are more reliable than others in this sense, does this mean that these criteria are more valuable/important for distinguishing between inflection and derivation? Why or why not? 2) Can you think of another way to evaluate the merits of each criterion? 3) Are some morphological patterns ‘more inflectional’ or ‘more derivational’ than other patterns that express the same feature/meaning? If yes, which criteria are primarily responsible for this result? How important do you think these properties are as data, relative to the others? Why? 4) Return to the definitions of inherent and contextual inflection from Section 5.4.1: value making independent semantic contribution (inherent) vs. value assigned by agreement or government (contextual). Divide the inflectional patterns in your data according to this definition. Now look at the other criteria from Step 2. Are the inherent inflectional patterns more like derivation than the contextual inflectional patterns? In other words, is
E X P L O R AT O R Y E X E R C I S E
113
the division into inherent/contextual inflection independently supported by the other criteria? 5) Based on the results of Step 2 and your answers to the previous questions, do you view inflection and derivation as dichotomous classes, or end-points along a continuum of traits? Support your conclusion by explaining your reasoning.
Productivity
6
A
morphological rule or pattern is said to be productive if (and to the extent that) it can be applied to new bases and new words can be formed with it. The notion of productivity is in principle applicable both to word-formation and to inflection, but in this chapter we focus more on productivity in word-formation. A variety of questions arise when considering productivity: Is the productivity of a rule part of speakers’ implicit knowledge of their language? What makes one rule more productive than another, i.e. what factors determine the likelihood of a given rule being used to create a new word? Also, are rules categorically productive or unproductive, or is productivity gradient? How can the productivity of a rule be measured, and how can the productivity of two rules be compared? In this chapter we explore possible answers to these questions.
6.1╇ Speakers’ knowledge of productivity One might begin by asking why productivity should be such a big issue in morphology. After all, syntactic rules are productive as well, but few syntacticians worry much about how to define and determine their productivity (and no syntax textbook devotes an entire chapter to productivity). In syntax, linguists study possible sentences, and they do not care much whether these are actual sentences in some sense or not. Indeed, we could carry over this procedure to morphology and say that linguists who are interested in the morphological systems of languages should study possible words, regardless of whether these words happen to be actual words in common use or not. In other words, according to one view, morphological competence (speakers’ knowledge of the words and rules of the language) and morphological performance (the actual use of that knowledge for communication and other tasks) are conceptually quite
6 . 1 ╇ S P E A K E R S ’ K N O W L E D G E O F P R O D U C T I V I T Y
115
distinct and should therefore be studied separately. Linguists who adhere to this position have often attempted to define away the whole issue of productivity by treating it as purely related to performance. One view that denies the relevance of productivity to the study of competence says that productivity is exclusively a diachronic phenomenon, i.e. one having only to do with the way in which morphological patterns change over the centuries. When a neologism is coined, and especially when it is accepted by the other speakers and becomes a usual word, this means that a new word enters the language and the language thereby changes. Thus, some linguists argue that when a strictly synchronic point of view is adopted, the issue of productivity does not arise. But this does not seem quite right. It fails to explain why some morphological patterns are more likely than others to be used to coin new words. Hypothetical words like *helpnessful (with the wrong order of the suffixes -ful and -ness) or *frownity (where the suffix -ity attaches to a verb) are clearly ungrammatical, as every speaker will agree (Aronoff 1980). Since such judgements are otherwise routinely used to study linguistic competence, this suggests that the productivity of a rule should also be considered part of speakers’ (synchronic) knowledge of their language. The relevance of productivity has also been denied by trying to equate the productivity of a rule with its restrictedness. The set of bases to which a rule could apply in principle is called its domain. Whenever the domain is less than the entire word-class, we say that there are systematic selectional restrictions on the rule. On this view, all morphological rules are equally productive, but they are not equally restricted. Some are quite unrestricted (like English -ness, which attaches to almost any kind of adjective), whereas others are heavily restricted (like English deadjectival -en in blacken, redden, etc., which attaches only to monosyllabic adjectives, among other restrictions). However, it is also quite unlikely that this view is correct. There are simply too many rules that are not obviously heavily restricted and yet their productivity is limited. For example, the English diminutive suffix -let (e.g. streamlet, piglet, booklet) could in principle combine with any monosyllabic concrete noun, but in fact it is very rarely used for new words. It is, of course, possible that such unproductive rules are subject to restrictions that have not been discovered yet, but we must regard it as more plausible that there is no such direct relation between the degree of productivity and unrestrictedness of a morphological rule. Thus, we have to accept that speakers’ knowledge of a language comprises not only words and rules, but also the probability of a given rule applying to create a new word, i.e. productivity. While we do not fully understand what causes a morphological rule to be productive or unproductive, research suggests a complex interaction of grammatical, social/pragmatic and processing restrictions (some of these are discussed in Sections 6.3–6.4 below), and speakers may even acquire some of their
116 
C H A P T E R 6 â•… P R O D U C T I V I T Y
knowledge of productivity by observing how morphological patterns are used (and not used) in the community. It is thus a phenomenon that cannot be explained away, or ignored, and most linguists consider productivity an important topic for morphological study.
6.2╇ Productivity, creativity and gradience When a rule is very productive, neologisms formed by that rule are hardly noticed – by speakers, hearers and lexicographers. For instance, English adjectives with the suffix -less can be formed quite freely (childless, joyless, shoeless, and so on), and if a speaker or writer creates a new word with -less (e.g. commaless: the poet writes in long, commaless sentences), this does not strike hearers or readers as particularly innovative. The author may not have noticed herself that she was using a new word. Some linguists have proposed that the unconscious nature of the formation of new words is not merely a typical property of highly productive rules, but should be a necessary criterion for regarding a rule as productive. According to this view, there is a sharp distinction between productivity and creativity. A productive rule allows speakers to form new words unconsciously and unintentionally, whereas creative neologisms are always intentional formations that follow an unproductive pattern. An example of a creative neologism would be the word mentalese (‘the mental language of our thoughts’), because new words with the suffix -ese (such as motherese, computerese, translationese) are probably always coined intentionally, and they immediately strike hearers and readers as new and unusual. (The word mentalese must have been coined by a philosopher in the middle of the twentieth century.) However, the proposed distinction between productivity and creativity has both a methodological and an empirical problem. The methodological problem is that it defines productive rule application as unconscious or unintentional, but we have no way of knowing what speakers’ intentions and state of consciousness are when they form a new word. Moreover, we can distinguish consciousness and intentionality at several levels. When the philosopher coined the word mentalese, he or she probably intended to create a catchy single-word expression for a highly abstract concept that would make that concept more popular. At this level the coinage was no doubt conscious. But why did he or she not choose thoughtese or mindese, two words that would have made perfect sense to describe the language of our thoughts in the mind? It so happens that English words with the suffix -ese have a strong preference for a stress pattern strong–weak–strong (e.g. compùterése, mòtherése, translàtionése, and also Jàpanése, not *Japànése, Viètnamése, not *Vietnàmése), and the words thòughtése and mìndése would not conform to this pattern (Raffelsiefen 1996). It seems unlikely that the
6 . 3 ╇ R E S T R I C T I O N S O N W O R D - F O R M AT I O N R U L E S
117
philosopher was aware of this phonological regularity, and in this sense the choice of mèntalése instead of thoughtese or mindese was probably unintentional. The empirical problem is that there are many rules that yield neologisms that are neither totally unremarkable nor immediately noticed. The English verb-deriving suffix -ize, for example, often forms new words, so it would be very odd to say that it is unproductive, but it may well be that quite a few of these new words are conscious creations (e.g. technical scientific terms such as pronominalize, transistorize, multimerize). Perhaps the term creativity is most appropriate when it is applied to violations of ordinary language norms (this is called poetic licence). Poetic licence manifests itself as the creation of novel words by unproductive rules. In English, verb + noun compounds of the type killjoy are unproductive, yet J. Thurber used kissgranny, and G. M. Hopkins coined daredeath. In Russian, the denominal suffix -acˇ (e.g. trubacˇ ‘trumpeter’, from truba ‘trumpet’) is unproductive, but V. Mayakovsky created stixacˇ ‘verse-maker’ (from stix ‘verse’), and V. Khlebnikov used smexacˇ ‘laugher’ (from smex ‘laughter’) (Dressler 1981). These cases should not be completely dismissed as abnormal use of language by a few exceptional individuals, because their poetry is intended for a (reasonably) wide audience, and readers must be expected at least to understand the neologisms. Thus, they provide interesting evidence that speakers are able to recognize the structure of unproductively formed words, and that the rules, even if unproductive by ordinary standards, at least exist. But the vast majority of newly formed words are not due to poetic licence. It seems more realistic to arrange rules on a continuous scale of productivity than to divide them into classes of ‘productive’ and ‘creative’ rules, or ‘productive’ and ‘unproductive’ rules. In this book we will therefore say that morphological rules can be gradiently productive, and the less productive a rule is, the greater the chance a neologism will be noticed. The suffix -ese is less productive than the suffix -less, so we expect -ese neologisms to be more striking than -less neologisms.
6.3╇ Restrictions on word-formation rules So what makes a rule (relatively) productive or unproductive? In many cases, we can give specific reasons why a word-formation rule does not give rise to words that it might be expected to permit. For Â� example, the German female-noun suffix -in (as in König-in ‘queen’, Löw-in ‘lioness’) does not generally combine with the names of lower animals (examples like Käfer-in ‘female beetle’, Würm-in ‘female worm’ can be found in small numbers on the internet, but speakers are less likely to accept them, as compared to Königin and Löwin); and the English suffix -ity systematically
118 
C H A P T E R 6 â•… P R O D U C T I V I T Y
fails to combine with adjectives ending in -ish, -y and -ful (*hopefulity). In other words, the rules of -in and -ity have restricted domains. There is disagreement about whether such restrictions on a rule should be considered issues of productivity. Some linguists posit that a rule may be considered very productive if it routinely creates new words from bases within the rule’s domain (bases not within the domain are irrelevant). Others define productivity without regard to domain. Under this definition, a rule with an unrestricted domain may be productive or unproductive, but a rule with a very restricted domain cannot be highly productive because its restrictions keep it from contributing a large number of new words to the language. Which stance we take will affect how we measure productivity (see Section 6.5 for several productivity measures), but whichever definition we employ, it is clear that restrictions on the domain of a rule significantly limit the coining of new words. The kinds of restrictions that can be observed are discussed in this section.
6.3.1╇ Phonological restrictions Phonological restrictions on the domain of a word-formation rule are particularly common with derivational suffixes, much less so with prefixes and compounding. In some cases, there is a straightforward reason for the restriction: certain complex words are impossible because they would create difficulties for phonetic processing (i.e. pronunciation or perception). A common restriction rules out the repetition of identical features, e.g. the repetition of the phoneme // (spelled ll) in Spanish (which reduces the domain of the diminutive suffix -illo, (see (6.1)), or the repetition of the vowel /i(:)/ in English (which reduces the domain of the suffix -ee (see (6.2))). (6.1)
Spanish diminutive suffix -illo mesa mesilla ‘(little) table’ grupo grupillo ‘(little) group’ gallo *gallillo ‘(little) rooster’ camello *camellillo ‘(little) camel’
(6.2)
English patient-noun suffix -ee draw drawee pay payee free *freeee accompany *accompanyee
(Rainer 1993: 18)
(Raffelsiefen 1999a: 246)
Somewhat similar is the requirement that the derived word must have an alternating rhythm (strong–weak–strong). As a result, the English suffix -ize
6 . 3 ╇ R E S T R I C T I O N S O N W O R D - F O R M AT I O N R U L E S
119
freely attaches to bases with a strong–weak rhythm, but does not attach to bases that end in a strong (i.e. stressed) syllable. (The suffix -ese behaves similarly, as we saw in the previous section.) (6.3)
English verbalizing suffix -ize prívate prívatìze glóbal glóbalìze corrúpt *corrúptìze secúre *secúrìze
(Raffelsiefen 1996; Plag 1999: ch. 6)
Phonological restrictions can also be purely random. As noted above in Section 6.1, the English suffix -en (e.g. blacken, redden, tighten) attaches only to monosyllabic bases. This is an example of a phonological restriction, but not one with an (obvious) phonetic motivation.
6.3.2╇ Semantic restrictions In many cases, the meaning of an affix automatically restricts the domain of a word-formation rule, because some base–affix combinations simply make no sense. For example, it would be nonsensical to add the German female-noun suffix -in to a noun like Baum ‘tree’ (*Bäum-in), because we do not conceive of trees as having gender distinctions. Similarly, the English reversive prefix de- (as in de-escalate, decolonize) can be combined only with verbal bases that denote a potentially reversible process. Combinations such as deassassinate or deincinerate are hard to interpret, except perhaps in a science-fiction context. However, word-formation rules may also have semantic restrictions that seem quite arbitrary. For example, the Russian quality-noun suffix -stvo combines with adjectives that denote properties of human beings, not with adjectives denoting physical properties of objects. (6.4)
Russian quality-noun suffix -stvo bogatyj ‘rich’ bogat-stvo znakomyj ‘acquainted’ znakom-stvo udaloj ‘bold’ udal’-stvo lukavyj ‘wily’ lukav-stvo vjalyj ‘withered’ *vjal’-stvo priemlemyj ‘acceptable’ *priemlem-stvo
‘richness’ ‘acquaintance’ ‘boldness’ ‘cunning’ (Švedova 1980: 179)
Here there is no intrinsic reason why the suffix -stvo should not combine with other adjectives.
120 
C H A P T E R 6 â•… P R O D U C T I V I T Y
6.3.3╇ Pragmatic restrictions In addition to being phonologically and semantically well-formed, a neologism must also be useful – this is what is meant by pragmatic restrictions. We noted at the beginning of this section that German do not generally accept female nouns in -in denoting lower animals (Käferin ‘female beetle’, Würmin ‘female worm’). It seems clear that these gaps in the German lexicon are due to a pragmatic restriction: for animals like beetles and worms, it is simply not particularly useful to distinguish between males and females. Perhaps one should regard these derivations as potential German words, because it is not all that difficult to imagine a situation in which they might become useful (e.g. entomologists’ specialized publications, or fairy tales). But ordinary speakers react to Käferin in much the same way as they would to Bäumin, and it is not easy to argue that the former is a possible word, while the latter is impossible.
6.3.4╇ Morphological restrictions Some morphological patterns require special morphological properties in the base. For example, Modern Hebrew has a pattern for action nouns (CiC(C)uC) that is applied only to verbs of one particular inflection class (CiC(C)eC). Verbs of other inflection classes (CaCaC, hiCCiC, etc.) cannot form their action nouns in this way. (6.5)
Modern Hebrew action-noun pattern CiC(C)uC diber ‘speak’ dibur ‘talk’ kibec ‘gather’ kibuc ‘gathering; kibbutz’ liked ‘unite’ likud ‘union; Likud’ tixnet ‘program’ tixnut ‘programming’ katav ‘write’ *kituv amad ‘desire’ *imud hiškiv ‘put to bed’ *hiškuv
In Russian, the female-noun suffix -ja combines only with bases that are themselves derived by the suffix -un (see (6.6)). All other nouns must use some other female-noun suffix (-ka, -ša, -inja, -isa). (6.6)
Russian female-noun suffix -ja govor-it’ ‘talk’ govor-un beg-at’ ‘run’ beg-un pljas-at’ ‘dance’ pljas-un lg-at’ ‘lie’ lg-un
‘talker’ ‘runner’ ‘dancer’ ‘liar’
govor-un’-ja beg-un’-ja pljas-un’-ja lg-un’-ja (Švedova 1980: 203)
It appears that, with such nouns, the suffix -ja is 100 per cent possible, but since the suffix -un is not particularly common and not particularly productive, nouns in -ja are very rare.
6 . 3 ╇ R E S T R I C T I O N S O N W O R D - F O R M AT I O N R U L E S
121
6.3.5╇ Borrowed vocabulary strata In some languages, a large part of the lexicon consists of loanwords from another language that is (or traditionally has been) well known to many speakers, at least in some influential section of the population. These loanwords usually include many morphologically complex words. If an isolated complex word is borrowed into another language, its morphological structure inevitably gets lost (thus, the English word orangutan is monomorphemic, although this is a compound noun in the source language, Malay: orang ‘man’, utan ‘forest’). But when a language borrows many morphologically complex words from the same language, their morphological structure may be preserved, and their patterns may become productive in the target language. For example, Japanese Â� borrowed many verb–noun compounds from Chinese – e.g. those in (6.7). (6.7)
Japanese V + N compounds (borrowed from Chinese) doku-syo ‘reading a book’ kyuu-sui ‘supplying water’ satu-zin ‘killing a man’ noo-zei ‘paying tax’ tuu-gaku ‘going to school’ tai-kyoo ‘staying in Tokyo’ hoo-bei ‘visiting the United States’ (Kageyama 1982: 221–31)
In some cases, the Chinese simple words were borrowed as well, but, in many others, these noun and verb stems exist only in compounds (e.g. Â�bei- ‘US’ occurs only in compounds such as bei-koku [US-country] ‘United States’). The pattern of Chinese compounds is quite different from that of the Â� corresponding native Japanese compounds, which take the form N + V (e.g. hito-dasuke [person-help] ‘helping people’, yama-nobori [mountain-climb] ‘mountain climbing’). Thus, if Japanese had just borrowed a few compounds of the type in (6.7), they would have lost their morphological structure, but since they were borrowed in large quantities, these compounds are analyzable by Japanese speakers, and in effect Japanese borrowed the V + N pattern along with the compounds from Chinese. The pattern is productive in modern Japanese, and new compounds can be formed with it. However, and this is crucial in the present context, only stems borrowed from Chinese can be used in this compounding pattern. For example, the noun amerika (used with the same meaning as bei-koku) cannot be a second compound member (*hoo-amerika ‘visiting America’). Thus, the Chinese– Japanese morphological pattern is still restricted to the vocabulary stratum of Chinese–Japanese words. A similar situation can be found elsewhere. Many languages of India have borrowed heavily from the classical language Sanskrit and thus have many
122 
C H A P T E R 6 â•… P R O D U C T I V I T Y
derived lexemes of Sanskrit origin. In Kannada (a Dravidian language that is not genealogically related to Sanskrit), many Sanskrit affixes are used extensively, but mostly with bases that are themselves Sanskrit loanwords. For instance, the quality-noun suffix -te can be used freely as in (6.8), but it does not combine with non-Sanskrit bases such as kul.l.a ‘short’. (6.8)
khacita bhadra ghana kul.l.a
‘certain’ ‘safe’ ‘weighty’ ‘short’
khacitate ‘certainty’ bhadrate ‘safety’ ghanate ‘dignity’ *kul.l.ate (Sridhar 1990: 270, 278)
In many European languages, we find an analogous situation with loanwords from Latin. English has borrowed particularly extensively from Latin, and suffixes like -ive, -ity, -ous and adjectival -al (as in parental, �dialectal) are mostly restricted to bases of Latin origin (these are often called Latinate bases, as contrasted with Germanic bases). (6.9)
act brutal monster parent
active brutality monstrous parental
fight brittle spinster mother
*fightive *brittality *spinstrous *motheral (cf. maternal)
Now the question arises how speakers could learn whether a stem belongs to the native or to the borrowed stratum – after all, speakers do not acquire the historical information of etymological dictionaries during their normal process of language acquisition. In many cases, the phonological peculiarities of the borrowed stratum are probably of some help. Thus, in Kannada only Sanskrit loans have aspirated consonants (kh, bh, gh), and, in Japanese, Chinese loan morphemes never have more than two syllables. But otherwise the only way to infer that a word belongs to the borrowed stratum is by observing that it combines (or fails to combine) with certain affixes. Perhaps because of this, the restriction of a word-formation Â� pattern to a borrowed stratum is often unstable. Thus, English -ous has also been applied to non-Latinate bases (e.g. murderous, thunderous), and the Kannada Sanskrit-derived suffix -maya (e.g. haasya ‘humour’, haasya-maya ‘humorous’) has also been applied to non-Sanskrit words (e.g. lanca ‘bribe’, lanca-maya ‘corrupt’; influuyens ‘influence’, influuyens-maya ‘influential’ (Sridhar 1990: 282)). The English suffixes -able, -ize, -ify, -ism seem to have lost their restriction to Latinate bases almost entirely.
6.4╇ Productivity and the lexicon While these kinds of restrictions explain why some words fail to exist that might otherwise be expected, we are still left with the problem that
6 . 4 ╇ P R O D U C T I V I T Y A N D T H E L E X I C O N
123
morphological patterns that are largely unrestricted may nonetheless differ in degree of productivity. Why is this? This is a complicated issue, and one that is not yet fully understood. However, one hypothesis is that the productivity of a rule depends in large part on how the words that exemplify it are structured in the lexicon. (This also helps explain why productivity is more important to morphology than to syntax – both morphological and syntactic rules have restrictions, but only morphological structure is closely tied to the lexicon.) The main ways in which the lexicon is argued to influence morphological productivity are reviewed in this section.
6.4.1╇ Processing restrictions There is growing evidence that productivity is tied to how words are mentally processed, specifically, the extent to which complex words are stored in the lexicon. Remember from Section 4.3 that under the hypothesis of a moderate word-form lexicon, many complex words (and all simple words) are stored in the lexicon, but some complex words are decomposed and only their component morphemes are stored. This difference can affect the memory strength of an affix – when an affix occurs often in words of the second type, the lexical entry for the affix will be frequently accessed, and as a consequence will have greater memory strength. However, an affix that occurs mostly in words stored in the lexicon will have relatively weak memory strength because mentally accessing a stored complex word bypasses the lexical entry for the affix. The hypothesis is that high memory strength makes an affix more readily available to speakers for use in coining new words. All else being equal, morphological patterns thus tend to be more productive if they predominantly occur in words that are decomposed, rather than stored (Hay and Baayen 2002). Of course, many factors influence whether a word is decomposed; see Section 4.3 for an overview. Here we review just one issue: relative frequency. When a complex word is less frequent than the base that it contains, it will tend not to be stored in the lexicon, all else being equal. For instance, modernity has a lower token frequency than modern, so modernity is likely to be decomposed – the high frequency of modern makes it efficient to mentally process modernity via the lexical entries modern and -ity. When the complex word is more common than its base (e.g. security has a higher token frequency than secure), the complex word will tend to be stored. With regard to productivity, the key question then, is: What percentage of words with -ity are like modernity, and what percentage like security? This is the parsing ratio – the proportion of words with a given affix (or other morphological pattern) that are estimated to be stored in the lexicon according to their component parts. It turns out that complex words containing -ity are likely to be stored. This suffix has a parsing ratio of 0.17, meaning that about 17 per cent of unique
124 
C H A P T E R 6 â•… P R O D U C T I V I T Y
complex words containing -ity are decomposed (the modernity type), and correspondingly, about 83 per cent of words containing -ity are stored in the lexicon (the security type). Parsing ratios for a sample of English suffixes are given in (6.10). (6.10)
affix -ence -ity -ate -dom -ness -ish -like -proof -less
parsing ratio productivity (100  P) 0.1 ╇ 0.0 0.17 ╇ 0.1 0.31 ╇ 0.3 0.5 ╇ 0.2 0.51 ╇ 0.8 0.58 ╇ 0.5 0.68 38.1 0.8 ╇ 5.5 0.86 ╇ 1.7 (Hay and Baayen 2002: 233–5)
A higher parsing ratio means that the affix is being activated in the lexicon relatively more often. Thus, the higher the parsing ratio, the more productive we would expect that affix to be. How to calculate the productivity measure P is described in Section 6.5(v) below, but here it is sufficient to note that larger numbers indicate greater productivity. While there is not a perfect correlation, the overall pattern is as expected (e.g. -ity is not particularly productive). We should be cautious in our conclusions about the influence of morphological processing on productivity because most of the relevant research has been conducted on English. The relationship between processing and productivity in other languages remains to be investigated. But this kind of processing constraint may help explain a long-noted but imperfect correlation between semantic and phonological regularity of words created by a rule, and that rule’s productivity. It is intuitively not surprising that the rule of -th suffixation in English is unproductive, because many of the words containing -th are irregular phonologically (depth, breadth, length, youth) or semantically (wealth is not just ‘being well’). But some completely regular rules are unproductive (e.g. the female-noun suffix -ess in English: poetess, authoress, princess), and some highly productive patterns have a fairly large number of irregular existing words (e.g. German -chen diminutives, as seen in idiomatized words like Brötchen ‘bread roll’, not ‘little bread’, Teilchen ‘particle’, not ‘little part’, Weibchen ‘animal female’, not ‘little woman’, Zäpfchen ‘uvula’, not ‘little cone’). So it does not seem quite right to say that regularity directly determines productivity. However, semantic and phonological regularity are thought to influence whether complex words are stored in the lexicon, so there is likely to be an indirect relationship between regularity and productivity.
6 . 4 ╇ P R O D U C T I V I T Y A N D T H E L E X I C O N
125
6.4.2╇ Synonymy blocking Very often an otherwise productive derivational rule does not apply because it is pre-empted by an existing word that has the meaning of the potential neologism. For example, the verb to broom does not exist because the verb to sweep exists instead, which means the same as to broom would if it existed. Morphologists say that there is synonymy blocking (or just blocking, for short) under such circumstances. Apparently languages prefer not to have several words that mean exactly the same. Some other English examples are given in (6.11). (6.11)
base broom to type gymnastics good
blocked word *to broom *typer *gymnastician *goodly
blocking word to sweep typist gymnast well
related pair hammer/to hammer to write/writer statistics/statistician bad/badly
As the examples show, it is immaterial whether the blocking word is morphologically related to the blocked word or not. A puzzling fact about blocking is that it has many exceptions. For instance, English has synonymous pairs like piety/piousness, curiosity/ curiousness, accuracy/accurateness, etc. (Plank 1981: 175–80), in which one would expect the second member to be blocked by the first one. Here, one relevant factor may be the token frequency of the blocking word: the more frequent the blocking word is, the greater is its blocking strength (Plank 1981: 182; Rainer 1988). Since the effect of frequency is relative, it is best to compare a range of cases that are structurally identical but differ in token frequency. We will look at quality nouns in Italian and German (Rainer 1988: 167–71). The Italian quality noun suffix -ità is generally productive with adjectives ending in -oso such as furioso ‘furious’, furiosità ‘furiousness’. However, when the adjective in -oso is itself derived from a non-derived quality noun, this noun has the same meaning as the (potential) derivative in -ità, and is thus potentially subject to synonymy blocking. For example, the adjective bisognoso ‘needful’ (derived from bisogno ‘need’) does not form a quality noun *bisognosità, because this would have the same meaning as bisogno and is thus blocked by it. However, the blocking effect is not always observed. For instance, malizioso ‘malicious’ forms maliziosità ‘maliciousness’, although its base malizia ‘malice’ has the same meaning. When we look at the frequencies of a range of cases, we see that only the more frequent words have the blocking effect. In (6.12), the last column gives the frequency of the blocking word as determined by a frequency dictionary. (The frequency 0 means that the corpus is not large enough to contain a token of the word, not that the word does not exist.)
126 
C H A P T E R 6 â•… P R O D U C T I V I T Y
(6.12) base coraggioso pietoso desideroso fiducioso orgoglioso armonioso rigoroso malizioso acrimonioso parsimonioso ignominioso
potentially blocked word *coraggiosità *pietosità *desiderosità *fiduciosità *orgogliosità armoniosità rigorosità maliziosità acrimoniosità parsimoniosità ignominiosità
blocking word coraggio ‘courage’ pietà ‘pity’ desiderio ‘desire’ fiducia ‘confidence’ orgoglio ‘pride’ armonia ‘harmony’ rigore ‘rigour’ malizia ‘malice’ acrimonia ‘acrimony’ parsimonia ‘parsimony’ ignominia ‘ignominy’
its frequency 52.70 34.04 31.92 30.79 10.64 4.13 3.42 0 0 0 0
The frequency effect on blocking strength can also be observed when a productive quality noun rule competes with unproductive quality noun formations. In German, the suffix -heit ‘-ness’ has all monosyllabic adjectives in its domain, but it is blocked when a different quality noun is available, e.g. *Reichheit ‘richness’ from reich ‘rich’ is blocked by Reichtum ‘wealth’, which uses the unproductive suffix -tum. Again, the frequency of the blocking word is decisive, as shown in (6.13). (6.13) base potentially blocking its frequency blocked word word alt ‘old’ *Altheit Alter ‘(old) age’ 1400 groß ‘big’ *Großheit Größe ‘size’ 1301 tief ‘deep’ *Tiefheit’ Tiefe ‘depth’ 613 warm ‘warm’ *Warmheit Wärme ‘warmth’ 520 frisch ‘fresh’ *Frischheit Frische ‘freshness’ 107 eng ‘narrow’ *Engheit Enge ‘narrowness’ 67 blass ‘pale’ *Blassheit Blässe ‘paleness’ 23 schnell ‘quick’ *Schnellheit Schnelle ‘quickness’ 23 The explanation for the frequency effect on blocking strength is that frequent words have greater memory strength and are therefore retrieved faster from memory than rare words. When a German speaker wants to say ‘warmth’, she has two options: applying the productive rule of -heit suffixation or retrieving an existing word with that meaning – i.e. Wärme. Since Wärme is very frequent and thus easy to retrieve from the lexicon, it will win out in this case. When the existing word is rare, it has a relatively weak memory strength, and the process of forming a new word may be faster, so no blocking is observed (see Anshen and Aronoff 1988). Blocking thus represents both a semantic restriction on productivity (the desire of the language not to have two words with identical meaning), and also a constraint imposed by the lexicon (words with stronger representations in
6 . 4 ╇ P R O D U C T I V I T Y A N D T H E L E X I C O N
127
the lexicon create stronger blocking effects because they are more easily accessed).
6.4.3╇ Productivity and analogy Finally, this book is primarily about synchronic morphology, the nature of morphological patterns as they function in a particular language at a given time. But, in order to understand synchronic patterns better, it is sometimes useful to consider the diachronic aspect of morphology, in particular here, analogical change. An analogical change is said to occur when speakers form a new word on the model of (or by analogy with) another word. For instance, the English verb fling used to have a past tense formed with the suffix -ed (flinged), but at some point the past-tense form flung was created, clearly on the model of verbs like sting/stung. In order to show clearly what happens in analogical change, linguists often use proportional equations as in (6.14). The two terms on the left-hand side of the equation represent the model, and the X on the right-hand side represents the target word that was newly created by analogy. (6.14) sting : stung = fling╇ : X X = flung This is an example of analogical extension, which involves an existing morphological pattern being extended to a new lexeme. The i/u pattern was extended to the lexeme fling, which was ‘new’ in the sense that it did not previously exhibit this pattern. Another example is the Polish plural suffix -owie. Originally this suffix occurred only with a few nouns (those belonging to the u-declension), e.g. syn ‘son’, plural synowie ‘sons’. But later it was extended to quite a few other nouns denoting male humans, e.g. pan ‘lord, sir’, plural panowie (earlier plural form pany). (6.15) syn : synowie = pan╇ : X X = panowie Analogical extension also occurs in derivational morphology. For instance, on the model of pairs of French loanwords such as changev, changeablea, adjectives in -able were formed from native English words like wash: (6.16) change : changeable = wash╇ : X X = washable Analogical extension can thus create a new lexeme, thereby enriching the lexicon, or it can lead an existing lexeme to switch from one inflectional pattern to another. One way to look at analogical change is as the diachronic result of synchronic productivity.
128 
C H A P T E R 6 â•… P R O D U C T I V I T Y
The notation in terms of proportional equations suggests that a single word pair served as the model for the change, but in fact, there is no particular reason to assume that. The left-hand part of a proportional equation is best understood as a general pattern, a word-schema, rather than as a single word. If that is the case, then a formula such as sting/stung becomes virtually indistinguishable from word-based rules of the kind we have already encountered many times. In fact, thinking of the pattern (and all of the words that exemplify it) as the analogical model helps explain why patterns with low type frequency can be productive. Reconsider fling/flung. At first glance, a switch from the -ed pattern to the i/u pattern may seem odd. Why should a lexeme be drawn away from a pattern that is quite productive and has high type frequency, and begin to exhibit a pattern that applies to fewer than two dozen words? The key observation is that the i/u pattern is dominant among words that are phonologically like the target; see (6.17). (6.17) /(s)C(C)ıŋ/ 0007cling/clung, fling/flung, sling/slung, string/strung, â•… wring/wrung, sting/stung, swing/swung /(s)C(C)ıŋk/ slink/slunk, shrink/shrunk (or shrank), stink/stunk /(s)C(C)ın/ spin/spun, win/won /(s)C(C)ıg/ dig/dug /(s)C(C)æŋ/ hang/hung /(s)C(C)ık/ strike/struck /(s)C(C)ık/ stick/stuck /(s)C(C)ik/ sneak/snuck (or sneaked) These words are all highly similar in phonological form, and all share a morphological pattern. Linguists call such a group of words a lexical gang. Lexical gangs often have fuzzy boundaries; the correspondence in (6.18) describes the prototypical members of the i/u gang. Peripheral members of the gang differ from this prototype to varying degrees.




(6.18) ╇ /(s)C(C)ıŋ/v ╇ /(s)C(C)ŋ/v ╇ ‘x’ Æ â•‡ ‘x’ tense: present tense: past╇╇╇
According to a reverse dictionary of Modern English, seven of thirteen verbs matching the left schema in (6.18) have the i/u pattern (sometimes alongside another pattern). Thus, while ding/dinged (and sing/sang, and bring/brought) are all possible analogical models, they are relatively unlikely because the i/u pattern has a high type frequency among verbs with the shape /(s)C(C)ıŋ/. And as a result, the i/u pattern is a strong analogical model in this particular phonological context, despite being of much lower type frequency overall than the -ed pattern (Bybee and Moder 1983). We can conclude from this that the term analogical change emphasizes a diachronic outcome, whereas the term productivity emphasizes synchronic
6 . 5 ╇ M E A S U R I N G P R O D U C T I V I T Y
129
structure, but the solution of an analogical equation is in practical terms the same as the application of a (productive!) word-based rule to a novel lexeme. Moreover, the analogical perspective highlights that productivity is directly tied to words in the lexicon, because these words help determine the strength of the pattern as a model for analogical extension.
6.5╇ Measuring productivity We have seen that productivity is best regarded as a gradable property of morphological rules. Thus, for each rule we may want to ask how productive it is – i.e. we want to measure the degree of productivity of word-formation rules. In syntax, acceptability judgements are widely taken to be direct reflections of speakers’ competence, and are used to distinguish possible and impossible sentences. However, speakers tend to be more reluctant to accept new words than to accept new sentences, maybe because they do not encounter new words very often in ordinary life. Consider the set of adjectives bearded (having a beard), winged (having wings), pimpled (having pimples), eyed (having eyes). The last word in this set, eyed, seems odd, and speakers may judge it unacceptable. But does that mean that it is truly ungrammatical – i.e. not allowed by the morphological system? A straightforward explanation of the difference in the acceptability of bearded, winged and pimpled, on the one hand, and eyed, on the other, is that not all creatures have beards, wings and pimples, but virtually all have eyes, so one would rarely describe a person or an animal as eyed. But consider a context in which cave-dwelling bugs, worms and other lowly creatures are discussed, and the focus is on whether they have eyes or not. In such a context, the use of eyed suddenly becomes much more plausible, and, confronted with this context, speakers would perhaps reverse their acceptability judgements. So just what these judgements mean is not always obvious in morphology. For this reason, most morphologists are interested in actual words when measuring productivity, in addition to speakers’ acceptability judgements of hypothetical words. Various measures of productivity have been proposed, but it turns out that they measure rather different things. (i) The number of actual words formed according to a certain pattern (also called degree of generalization, profitability of a pattern or type frequency). This is an interesting concept, and it is fairly easy to measure by examining a comprehensive dictionary (though, of course, this works only to the extent that the dictionary faithfully records all and only the actual words of the language). However, type frequency is not the same as productivity: according to this measure, the English suffix -ment has a high type frequency (English has hundreds of words like investment, harassment,Â�
130 
C H A P T E R 6 â•… P R O D U C T I V I T Y
fulfilment), but it is not productive – only four neologisms with -ment are attested in the OED for the twentieth century. Conversely, there are not many usual words with the suffix -ese (as in journalese), but this can be used freely to coin new words denoting a special language or jargon. (ii) The number of possible words that can be formed according to a certain pattern. This concept is much more difficult to measure, because it requires that we correctly identify all the restrictions on the pattern. But even then it is unlikely that the set of possible words equals the likelihood that a new word can be coined. There are simply too many cases of (more or less) unproductive rules that do not seem to be restricted in any general way. For instance, en-/em- prefixation in English should be possible with any noun that denotes a container-like object (e.g. entomb, ensnare, embody), but the rule is simply not productive (cf. *embox ‘put into a box’, *encar ‘put in a car’). (iii) The ratio of actual words to possible words (also called the degree of exhaustion) (Aronoff 1976). Again, this requires that we be able to count the number of possible words, so it is not very practical. Moreover, when the possible bases include complex words that are themselves formed productively, the set of possible words becomes open-ended, and computing the ratio of actual to possible words is not really meaningful. For example, English or German N + N compounds can be formed freely without restrictions, and the compound members may be compounds themselves (see Section 7.1). Thus, the set of possible N + N compounds is staggeringly large (in principle, infinite), so the degree of exhaustion for N + N compounds is necessarily quite low (even though there are plenty of actual N + N compounds, and the pattern is highly productive). (iv) The number of neologisms attested over a certain period of time (also called diachronic productivity). This measure can be determined if a good historical dictionary is available (such as the OED), but again only to the extent that the dictionary is reliable. And we saw earlier that if a pattern is very productive, lexicographers are likely to overlook new words with this pattern. Another technique uses large text corpora. By looking at a newspaper corpus of recent decades, it should be possible, for instance, to observe how the English semi-suffix -gate (as in Watergate, Irangate, etc.) gained (and perhaps lost) productivity over the years. (v) The ratio of hapax legomena with a given pattern to total token frequency of words with that pattern (called theP measure, or the categoryconditioned degree of productivity) (Baayen and Lieber 199; Baayen 1993). This measure stems from the observation that productive morphological rules are likely to produce occasionalisms, whereas unproductive rules are not likely to do so. Occasionalisms are thus particularly important for determining productivity. A hapax legomenon is a word that occurs only once in a corpus; in a very large corpus hapax legomena can be assumed to be occasionalisms. The category-conditioned degree of productivity is
S U M M A RY O F C H A P T E R 6
131
calculated by dividing the total number of hapax legomena in the corpus exhibiting a given morphological pattern by the total number of word tokens in the corpus that exhibit that same pattern: (6.19) Category-conditioned degree of productivity P = V1,m / Nm Here, V1,m is the number of hapax legomena in a corpus that exhibit morphological pattern m. Nm is the collected token frequency of all words exhibiting pattern m. The ratio captures the likelihood that a word randomly drawn from the corpus and exhibiting the relevant pattern will be an occasionalism. (vi) The ratio of hapax legomena with a given pattern to all hapax legomena (called the P* measure, or the hapax-conditioned degree of productivity) (Baayen and Lieber 1991; Baayen 1993). This measure also relies on the tendency of productive rules to create occasionalisms, but P* measures the extent to which a given morphological pattern contributes to the total growth rate of the vocabulary. In this measure, the number of hapax legomena exhibiting a given morphological pattern is divided by the total number of hapax legomena in the corpus (with all morphological patterns): (6.20) Hapax-conditioned degree of productivity P * = V1,m / V1 V1 is the total number of hapax legomena in the corpus. This is thus similar to method (iv), but has the advantage that it measures productivity at a given moment in time, and a good historical dictionary is not needed. All of these measures are most useful for determining the relative productivity of patterns that serve similar functions, for example, -ness vs. -ity in English. It is often not clear how productivity measures should be interpreted when taken in isolation, or when comparing unlike morphological patterns (e.g. one that creates an adverb and one that creates a noun).
Summary of Chapter 6 The productivity of a morphological rule is a measure of the extent to which it can be used to create new words. Productivity is often regarded as a phenomenon that exclusively concerns language use (performance) or language change, but, in the view defended here, productivity is one part of speakers’ knowledge of language (competence). Productivity itself must be an object of morphological study.
132 
C H A P T E R 6 â•… P R O D U C T I V I T Y
Morphological � patterns can be arranged on a scale from totally unproductive to highly productive. A rigid dichotomy between creativity and productivity does not seem to be very useful, because there are always intermediate cases. The productivity of a pattern may be limited in various ways: phonologically, semantically, pragmatically and morphologically. Sometimes a pattern is productive only within a borrowed vocabulary stratum. The structure of the lexicon is also a likely factor. Patterns with high memory strength (by virtue of being frequently accessed in the lexicon) tend also to be more productive. High frequency words can block newly coined words with the same meaning. Finally, lexical gangs and other patterns with low token frequency can be (quasi-) productive because they are phonologically densely clustered in the lexicon. Various quantitative measures of productivity have been proposed; the corpus measures based on hapax legomena (P and P*) are increasingly used.
Further reading Excellent recent discussions of issues surrounding productivity are found in Plag (1999) and Bauer (2001b) (see also Kastovsky (1986) and Dressler and Ladányi (2000)). Bauer (2005) looks particularly at the history of theories of productivity. The view that competence and performance should be strictly separated is expressed in Di Sciullo and Williams (1987). Aronoff (1980) argues that productivity is relevant to competence and to the synchronic study of morphology more generally. The distinction between productivity and creativity is proposed in the classical paper Schultink (1961) (see also van Marle 1985). On productivity as a scalar notion, see Bauer (1992). The non-distinctness of analogy and morphological rules is pointed out in Becker (1990) and Blevins and Blevins (2009); see also Bybee (1985, 1988). Perhaps the most detailed analogical approach to morphological rules and productivity is Skousen (1989, 1992). The logical extreme of the analogical approach is connectionism. Classic articles in this framework are Rumelhart and McClelland (1986), MacWhinney and Leinbach (1991) and Daugherty and Seidenberg (1994). Blocking and its relation to frequency are discussed by Rainer (1988) and Anshen and Aronoff (1988). Aronoff (2007) shows that semantic regularity is not necessary for the creation of neologisms (which further implies that regularity and productivity are not quite the same thing).
COMPREHENSION EXERCISES 
133
For better or worse, a disproportionate amount of work on productivity has focused on English past tense inflection. In addition to the connectionist papers listed above, see in particular Bybee and Slobin (1982), Bybee and Moder (1983), Pinker and Prince (1988), Derwing and Skousen (1994) and Albright and Hayes (2002).
Comprehension exercises 1. The productivity of the suffix -ity in English is heavily restricted (see the examples below). What might be the nature of the restriction, and into which of the categories of Section 6.3 does it fall? electric electricity bountiful *bountifulity probable probability sonorant *sonorantity captive captivity aimless *aimlessity curious curiosity darkish *darkishity abnormal abnormality fearsome *fearsomity 2. Recall Exercise 1 of Chapter 4. Of the words listed there, you have probably characterized reknow and happytarian as impossible words in English, although the affixes re- and -(t)arian are widely attested and productive in English. What is it about the nature of these affixes that makes them unsuitable for these bases? (In other words, in what way is their productivity restricted?) 3. Modern Greek has two action-noun suffixes, -simo and -ma, which are both productive, but in different, complementary domains. Try to extract a generalization from the following examples that predicts when -simo occurs and when -ma is used. (Note that the phonological stem alternations are irrelevant.) verb meaning action noun meaning dˉjavázo ‘I read’ dˉjávasma ‘reading’ kóvo ‘I cut’ kópsimo ‘cutting’ lúzo ‘I bathe’ lúsimo ‘bathing’ mangóno ‘I squeeze’ mángoma ‘squeezing’ pjáno ‘I seize’ pjásimo ‘seizing’ skondáfto ‘I stumble’ skóndama ‘stumbling’ tinázo ‘I shake’ tínama ‘shaking’ tréxo ‘I run’ tréksimo ‘running’ 4. Which of the following words are impossible in the given meaning because of synonymy blocking? For words that cannot be explained as blocking, what is the reason for their impossibility? *musting (e.g. I hate musting get up every morning.)
134 
C H A P T E R 6 â•… P R O D U C T I V I T Y
*foots (e.g. Bobby played outside and has dirty foots now.) *cooker (e.g. This meal is superb. The cooker is a real artist.) *bishopdom (e.g. The bishop often travels through his bishopdom.) *teacheress (e.g. Our teacheress is a very competent woman.) *slickize (e.g. The Zambonis slickize the ice before the skaters compete.) *certainness (e.g. Nowadays there is less certainness about church teachings.) *sisterlily (e.g. She embraced her sisterlily.)
5. How did the suffixes (or perhaps bound roots) -(er)ati and -scape come into being? Consider the following examples: literati, glitterati, liberati, chatterati, soccerati, digiterati (Kemmer 2003) landscape, seascape, cloudscape, skyscape, waterscape, winterscape (Aldrich 1966).
Exploratory exercise Acceptability judgements are used less often in morphology than in syntax. Morphologists tend to study actual words, rather than potential ones, and the measures of productivity introduced in Section 6.5 reflect that. But acceptability judgements are still important to morphology and are, for example, cited in many places in this book, so we might wonder about the relationship between acceptability judgements and productivity measures. Do acceptability judgements and productivity scores measure the same thing? In this exercise, you will explore and compare the two types of measures. The instructions below use Russian for demonstration purposes, but this exercise could be conducted using any language for which a good electronic corpus or frequency list is available. Readers with basic programming skills will find it easier and more accurate to calculate productivity based directly on a corpus that is available electronically.1 However, in the following instructions we assume that existing frequency lists or frequency dictionaries (i.e. a list/dictionary of words found in a corpus, with counts of how often each word occurs) will be used. Instructions Step 1: For the language you will investigate, find a frequency list or frequency dictionary, preferably one generated from a large corpus that includes a wide variety of textual and/or spoken data. The Russian examples in (6.21) and (6.22) were taken from a frequency list containing approximately 32,000 unique lexemes, which in turn was generated from a moderately sized corpus of modern Russian (about 16 million word tokens), from a variety of sources (Sharoff 2002). 1
The Linguistic Data Consortium is a major distributor of corpora: www.ldc.upenn.edu.
E X P L O R AT O R Y E X E R C I S E
135
Step 2: Pick 2–5 morphological patterns to study. The best choices will be derivational patterns that express the same meaning, or a very similar meaning, and intuitively seem to differ in productivity. (Derivational patterns are desirable, as opposed to inflectional ones, because inflectional patterns are more uniformly productive.) Two Russian patterns that meet these criteria are shown in (6.21).2 (6.21) pattern -stvo
abstract noun base adjective bogat-stvo ‘wealth’ p’jan-stvo ‘drunkenness’
-ost’
naivn-ost’ hrabr-ost’
‘naiveté’ ‘bravery’
bogat-yj p’jan-yj
‘wealthy’ ‘drunk’
naivn-yj hrabr-yj
‘naive’ ‘brave’
Step 3: Calculate the hapax-conditioned degree of productivity (P *). This will require you to search your list or dictionary for all hapax legomena, regardless of morphological pattern, and record the total number that occur. (Most frequency lists are sorted from most frequent to least frequent, making this a relatively simple task.) Then, within the list of hapaxes, search for all words containing the chosen morphological pattern. For instance, of the 5,133 least frequent words in the Frequency Dictionary of Russian, 44 words contained -stvo and 118 contained -ost’. Some examples of these hapaxes are listed in (6.22). (6.22) a. b.
bescˇinstvo svjatotatstvo provorstvo dissidentstvo
‘excess, enormity’ ‘sacrilege’ ‘adroitness, dexterity’ ‘nonconformism (rel.)’
prazdnost’ brennost’ neperenosimost’ krasivost’
‘idleness’ ‘mortality’ ‘unbearability’ ‘beauty’
P*(stvo) = 44/5133 = 0.0086, and P*(ost) = 118/5133 = 0.023. Thus, -ost is approximately two and a half times more productive than -stvo according to the hapax-conditioned degree of productivity. Note: The accuracy of hapax-based productivity scores depends in part on the size of the corpus. Logically, a word that occurs only once in a 2 million-word corpus might occur about 10 times in a 20 million-word corpus (and thus it would not be counted as a hapax in the larger corpus), and a word that is too rare to be attested in a 2 million-word corpus might
2
It is helpful if the sequence of letters in the chosen affixes is not likely to produce many false hits. For instance, the English suffix -y creates abstract nouns from verbs (e.g. assemble – assembly), but it would be a poor choice because many adjectives and other words also end in the letter y (e.g. happy). By contrast, in English words ending in the letters ation, this is almost always a suffix (e.g. consult – consultation).
136 
C H A P T E R 6 â•… P R O D U C T I V I T Y
be a hapax in a larger corpus. And so presumably, the larger the corpus, the more likely it is that hapaxes will be true occasionalisms that are produced by the productive rule. For practical reasons, many frequency lists give only the words that occur at least a certain number of times in the source corpus. But assuming that such a list is large enough to include many words that are rarely used, it should be possible to treat the most infrequent words on the list as if they were ‘true’ hapaxes, even if they occur more than once. This sacrifices some degree of accuracy, but in many respects it is equivalent to using a smaller corpus. Step 4: Based on the results of Step 3, develop predictions about speakers’ acceptability judgements. Do you expect speakers to be more likely to accept possible words when the affix is more productive? Less likely? No relation? Do you expect speakers to give potential words exhibiting the more productive patterns higher ratings on an acceptability scale? Why or why not? Explain your reasoning. Step 5: Test your predictions with native speaker informants. Make a list of at least 10 possible words for each of the affixes. These test words should be built on real bases, and should obey the restrictions on the domain of the rule, but not be actual words. (See Section 6.3 for discussion of domain restrictions.) For instance, Russian has the adjectives matovyj ‘matted, suffuse (of light)’ and laskovyj ‘tender, gentle’, but the derived abstract nouns matovost’ and laskovost’ are not actual words. They can be considered possible words. Devise a ratings scale and ask at least three native speakers to judge the acceptability of these test words. Step 6: Analyze the data. Determine whether there is a correlation between the productivity scores and speakers’ acceptability judgements. Do speakers’ judgements match your predictions? If not, develop hypotheses about factors that might influence one measure, but not the other. Also consider the possible impact of your data collection methodology on your results, for both types of measures. Finally, think about the relationship between speakers’ acceptability judgements and hapax-conditioned productivity. Do they measure the same thing, or different things? Explain your reasoning.
7.1 COMPOUNDING TYPES 
Morphological trees
137
7
I
n this chapter we will see that various kinds of morphologically complex words can be thought of as having hierarchical structure. In this respect, morphological structure resembles syntactic structure, and the ways in which morphological and syntactic structure are similar and different are important. At issue is the degree to which syntactic principles govern wordlevel structure. Until now, we have largely assumed that morphological rules exist in a separate area of the grammar from syntactic rules, with each area subject to its own principles. However, strong commonalities between morphological and syntactic structure might suggest that the grammatical system is not divided in this way. In this chapter we assess the evidence related to hierarchical morphological structure. Hierarchical structure is quite evident in compound words, and less so in derivationally derived words. Thus, we will start by examining compounds in some detail.
7.1 Compounding types A compound is a complex lexeme that can be thought of as consisting of two or more base lexemes. In the simplest case, a compound consists of two lexemes that are joined together (called compound members). Some examples from English are given in (7.1). English allows several types of combinations of different word-classes (N: noun, A: adjective, V: verb), but not all such combinations are possible. 1
(7.1) English compounds: some examples
1
Note that the spelling of English compounds is inconsistent: often they are written as a � single word, but in many other cases (especially with N + N compounds), the constituents of the compound are separated by a space, like syntactic phrases (e.g. sugar plantation, morpheme lexicon). These spelling differences are irrelevant in the present context and should be ignored.
138 
C H A P T E R 7 â•… M O R P H O L O G I C A L T R E E S
N + N A + N V + N N + V N + A A + A
lipstick hardware drawbridge babysit leadfree bitter-sweet
(lipn + stickn) (harda + waren) (drawv + bridgen) (babyn + sitv) (leadn + freea) (bittera + sweeta)
Compounding rules may differ in productivity. In English, the N + N pattern is extremely productive, so novel compounds are created all the time and are hardly noticed. By contrast, the V + N pattern is unproductive and limited to a few lexically listed items, and the N + V pattern is not really productive either. For instance, one cannot say to hair-wash ‘wash one’s hair’, and the small handful of examples like babysit from Section 3.2.2 are mostly backformations from nouns, and are not produced directly by N + V compounding rules. However, there are many languages (especially morphologically rich, polysynthetic languages) that do allow compounds in which the notional object and the verb form a compound. Such compounding processes are called noun incorporation (metaphorically we say that the object is incorporated into the verb). An example from Alutor is given in (7.2). (For more on noun incorporation, see Section 11.2.1.) (7.2) gəmmə tə-məng-ilgətav-ək I 1sg-hand-wash-1sg ‘I washed (my) hands.’ (Lit.: ‘I hand-washed.’) (Koptjevskaja-Tamm and Muravyova 1993: 298) In a compound that consists of two lexemes, it is really the lexeme stems that are combined, not inflected forms. In this respect compounding is no different from derivational affixes, which attach to stems. Thus, we get English compounds such as lipstick (not *lipsstick), although it is used for both lips, and child support (not *children support), even if several children are supported. While we have already seen some examples in which the first lexeme of a compound is inflected (e.g. publications list from Section 5.3.7), this is not common. That the first compound member is almost always a stem, not an inflected word-form, can be seen most clearly in languages with richer inflection, such as Sanskrit. In Sanskrit, the first compound member in N + N/A compounds shows a vowel-final (or -r-final) form that does not occur as a member of the inflectional paradigm – this can thus be regarded as the pure stem. (7.3) deva-sena¯- pitr-bandhu- pati-jus.t.a-
‘army of gods’ ‘paternal relation’ ‘dear to the spouse’
(devah. ‘god’) (pita¯ ‘father’) (patih. ‘spouse’)
7.1 COMPOUNDING TYPES 
139
In German, many compounds even have a special semantically empty suffix (sometimes called an interfix) on the first compound member, which forms the stem that is appropriate for compounding. Some examples are shown in (7.4). (7.4)
German compounds with interfixes Volk-s-wagen lit. ‘people’s car’ (Volk ‘people’ + Wagen ‘car’) Liebe-s-brief ‘love letter’ (Liebe ‘love’ + Brief ‘letter’) Schwan-en-gesang ‘swansong’ (Schwan ‘swan’ + Gesang ‘song’)
That the first member of a compound is a stem rather than a particular word-form is also clearly seen in German V + N compounds, as in (7.5). (7.5) German V + N compounds Wasch-maschine ‘washing machine’ (wasch-en ‘wash’ + Schreib-tisch ‘(writing) desk’ (schreib-en ‘write’ + Saug-pumpe ‘suction pump’ (saug-en ‘suck’ +
Maschine ‘machine’) Tisch ‘desk, table’) Pumpe ‘pump’)
The elements wasch-, schreib- and saug- must be pure stems, because almost all word-forms of verbs have special suffixes (the suffix -en in (7.5) is the infinitive (and citation-form) suffix). The only suffixless word-form is the imperative, but it would not make sense semantically to claim that wasch in Waschmaschine is the imperative form of the lexeme waschen. From the point of view of semantics, not much needs to be said about the compounds that we have seen so far. The first compound member generally serves to modify and narrow the meaning of the second compound member. Thus, a lipstick is a special kind of stick (not a special kind of lip), a drawbridge is a special kind of bridge and a love letter is a special kind of letter. Since semantically the second member is in this sense more important, it is referred to as the semantic head of the compound, and the modifying element is called the dependent. Examples like lipstick belong to the endocentric type of compound (the term endocentric means that the semantic head (or centre) of the compound is ‘inside’ (endo-) the compound). In endocentric compounds, the meaning of the entire word is a subset of the meaning of the lexeme that serves as the head. In English, the semantic head of an endocentric compound is always the second member, but in other languages such as Spanish, the head is the first Â� member. (7.6) hombre-rana año luz pez espada
‘frogman’ (hombre ‘man’ + rana ‘frog’) ‘light year’ (año ‘year’ + luz ‘light’) ‘swordfish’ (pez ‘fish’ + espada ‘sword’)
The semantic relations that obtain between the head and the dependent in compounds are quite diverse: purpose (writing desk, lipstick), appearance
140 
C H A P T E R 7 â•… M O R P H O L O G I C A L T R E E S
(hardware, swordfish), location (garden chair, sea bird), event participant (e.g. agent: swansong, patient: flower-seller), and so on. There thus seem to be almost no restrictions on the kinds of semantic relations that may hold between the dependent and the head in compounds (at least in the languages in which compound meanings have been studied extensively). It is our knowledge of the world that tells us that a flower-seller is someone who sells flowers, and that a street-seller is someone who sells something on the street. But it is easy to imagine a world (say, a fable about commercially active bees) in which selling goes on on flowers, and even easier to imagine a world in which people specialize in selling entire streets. English morphology does not seem to say more than that the dependent must be in some kind of pragmatically sensible relation to the head. Not all compounds are of the endocentric type. Compounds may also be exocentric (i.e. their semantic head is ‘outside’ (exo-) the compound). Exocentric compounds can be illustrated with Â� examples from Ancient Greek. (7.7) kakó-bios polu-phármakos he¯dúoinos megaló-psukhos
‘having a bad life’ (kakós ‘bad’ + bíos ‘life’) ‘having many medicinal herbs’ (polús ‘much’ + phármakon ‘herb’) ‘having sweet wine’ (he¯dús ‘sweet’ + oı˜ınos ‘wine’) ‘having a large mind, i.e. magnanimous’ (mégas ‘large’ + psukhe¯ ‘mind’)
A compound such as he¯dúoinos refers to someone who has sweet (he¯dú-) wine (oino-), so it denotes a kind of person, not a kind of ‘sweet’ nor a kind of ‘wine’. The semantic head is ‘outside’ the compound: the reference to ‘someone’ must be inferred from the structure as a whole – there is no morpheme that refers to a person or to ownership. English has a few exocentric A + N Â� compounds of this semantic type (redhead ‘someone who has red hair’, Â�highbrow, lazybones), but this pattern is hardly productive in English. Another type of exocentric compound is illustrated by the Italian examples in (7.8). (7.8) portabagagli lavapiatti asciugacapelli
‘trunk’ (portare ‘carry’ + bagagli ‘luggage’) ‘dishwasher’ (lavare ‘wash’ + piatti ‘dishes’) ‘hair dryer’ (asciugare ‘dry’ + capelli ‘hairs’)
Here the ‘external’ semantic head is an instrument for carrying out an action on an object. Again, English has a few exocentric V + N compounds as well (referring to people rather than instruments: pickpocket, cutthroat, killjoy), but this pattern is totally unproductive in English. Using our word-based notation of Section 3.2.2, the rules that yield these exocentric compounds can easily be represented formally.
7.1 COMPOUNDING TYPES 
141
(7.9) Rule for Italian exocentric compounds of (7.8)
[
]
[
]
[
/Xre/v.inf & /Y/n.pl ´ /XY/n.sg ‘dox’╇╇╇╇╇╇ ‘ys’╇╇╇╇╇ ‘instrument for doingx ys’
]
Here the compound word-schema on the right contains the additional meaning element ‘instrument for’, which is not associated with a particular element of phonological form, but with the pattern as a whole (cf. the rule in (3.30), which is similar in this respect). Besides endocentric and exocentric compounds, there are also compounds that have more than one semantic head. In these compounds, each member has a separate referent. Both members are on an equal footing, and they can be paraphrased with ‘and’, so they are called coordinative compounds. Some examples from Korean are in (7.10). (7.10)
elun-ai ma-so non-path o-nwui son-pal
‘adult and child’ ‘horses and cattle’ ‘farm’ ‘brother and sister’ ‘hand and foot’
(elun ‘adult’ + ai ‘child’) (ma ‘horse’ + so ‘cow’) (non ‘rice field’ + path ‘dry field’) (o ‘brother’ + nwui ‘sister’) (son ‘hand’ + pal ‘foot’) (Sohn 1994: 416–7)
Coordinative compounds are widespread in the world’s languages, but they happen to be rare in European languages, including English. Another, more familiar type of compound is represented by examples such as (7.11) from Spanish, where both compound members have the same referent. Such compounds are called appositional compounds. (7.11) poeta-pintor ‘poet who is also a painter’ actor-bailarín ‘actor who is also a dancer’ compositor-director ‘composer who is also a director’ English also has some compounds of this kind (student worker, MarxismLeninism), and adjective compounds such as bitter-sweet and deaf-mute can be subsumed under this type as well. The last type of compound to be mentioned here is again exocentric, but it shares with coordinative compounds the feature of semantic equality of both compound members. A few examples from Classical Tibetan are given in (7.12). (7.12)
rgan-gžon yag-ñes mtho-dman srab-mthug
‘age’ ‘quality’ ‘height’ ‘density’
(rgan ‘old’ + gžon ‘young’) (yag ‘good’ + ñes ‘bad’) (mtho ‘high’ + dman ‘low’) (srab ‘thin’ + mthug ‘thick’) (Beyer 1992: 105)
142 
C H A P T E R 7 â•… M O R P H O L O G I C A L T R E E S
The semantic head of these compounds is something like ‘property’, so Â�rgan-gžon is literally ‘property (in the feature) of old and young’, i.e. ‘age’.
7.2 Hierarchical structure in compounds As we saw in the preceding section, the concept ‘semantic head’ is useful for talking about the kinds of semantic relations that may obtain between the members of a compound. In this section, we will see that not only a semantic notion of ‘head’, but also a formal notion of ‘head’ can play a role in morphology. Let us look at a number of examples of compounds and their tree diagrams.
NPL
NMASC.PL
V3SG.PRES
NMASC.WEAK
N
NPL
NMASC.PL
NFEM
N
V3SG.PRES
NFEM
NMASC.WEAK
lip
sticks
años
luz
baby
sits
Luft
pirat
English lipsticks
Spanish años luz ‘light years’
English babysits
German Luftpirat ‘air pirate, hijacker’
Figure 7.1╇ Compound trees: two compound members
Tree diagrams indicate hierarchical structure. In Figure 7.1 this is rather unexciting, because the compounds consist of only two lexemes. However, tree representations of compounds are particularly useful when a compound consists of members that are compounds themselves, because in that case several different hierarchical structures are possible. Two possibilities for three-term compounds are shown in Figure 7.2, and Figure 7.3 shows two possibilities for compounds with four terms. In the compound Berkeley Linguistics Society, for instance, the tree diagram shows that the second and third lexemes form a compound inside the larger structure. This hierarchical structure corresponds to the semantics: the Berkeley Linguistics Society denotes a kind of linguistics society, namely one established at the University of California, Berkeley. By contrast, in particle physics conference, the first two lexemes are grouped into a compound because it is a conference about particle physics.
7.2 HIERARCHICAL STRUCTURE IN COMPOUNDS 
N
143
N N
N
N
N
N
N
N
N
Berkeley
Linguistics
Society
particle
physics
conference
Figure 7.2╇ Compound trees: three compound members
N
N
N
N N Indiana
N N
N
University Linguistics
N N Club
N
N
University Cashiers
N
N
Office employee
Figure 7.3╇ Compound trees: four compound members
Sometimes a compound with more than two nouns may allow two � hierarchical structures simultaneously. For example, a compound like nuclear power station can be grouped as [[nuclear power][station]] or as [[nuclear][power station]] with equal justification, because both make sense semantically, and both the compounds nuclear power and power station exist in English. Tree diagrams can also be used to indicate the formal head of a compound. The notion of a formal head is mostly relevant to the endocentric type, for which the formal head and the semantic head coincide. In the trees in Figures 7.1 to 7.3, the formal head of each compound is symbolized by a double line connecting the head and the next higher node in the tree. This is straightforward for compounds with two members, as in Figure 7.1, but note that in larger compounds, there is a formal head for each component compound, e.g., conference is the formal head of particle physics conference, but physics is the head of the compound particle physics. We can identify at least two characteristics of formal heads in endocentric compounds. First, we see that in the noun lipsticks in Figure 7.1, sticks contains the number marking that characterizes the whole compound. The formal head of a compound is thus the morphosyntactic locus of the compound, in that it is the place where the morphosyntactic features of the compound are expressed.
144 
C H A P T E R 7 â•… M O R P H O L O G I C A L T R E E S
An astute reader might object to analyzing lipsticks as [[lip][stick-s]], in which plurality is a property of the head. An alternative proposal is [[lip] [stick]]s, where the plural suffix attaches to the complete compound word rather than to the head. This alternative works for this particular example, but attachment to the complete compound does not work for a case like Spanish años luz ‘light years’ (singular: año luz ‘light year’). As we saw in (7.6), Spanish has left-headed N + N compounds, and if plurality is a property of the entire compound, not the head, we expect año luz-es, rather than the Â� correct Spanish form años luz. At least in Spanish, then, the locus of plurality is the morphological head, and it is simpler to make the same analysis in ambiguous cases like lipsticks as well. Second, the formal head determines for the entire compound characteristics such as word-class, gender and inflection class. For instance, the English word babysits is a verb, just like its head sits, but unlike the nonhead baby. In Spanish años luz and German Luftpirat ‘air pirate, hijacker’, the nonheads luz ‘light’ and Luft ‘air’ are feminine, but the compound nouns are masculine, just like their heads. And in German Luftpirat we also see that the inflection class of the head is shared by the compound: both Pirat and Luftpirat are ‘weak’ nouns – i.e. their genitive singular suffix is -en rather than the more common -s. This can also be illustrated from English: the plural of church mouse is church mice, not *church mouses – i.e. the head determines the way the plural of the compound is formed. As we would expect, compounds that are not semantically endocentric do not necessarily behave like formally headed compounds. Thus, in coordinative compounds we often find double plural marking (e.g. Spanish actores-bailarines, compositores-directores, etc.). Also, the English exocentric compound sabretooth (‘a tiger whose teeth are like sabres’, not ‘a tooth that is like a sabre’) forms the plural sabretooths (not sabreteeth), and switchfoot (‘a surfer who can ride with either the right or left foot forward’) forms the plural switchfoots. (But by contrast, Blackfoot (‘a person belonging to the Blackfoot Native American tribe’) most commonly has the plural Blackfeet, showing that there is some variation in this regard.)
7.3 Hierarchical structure in derived lexemes Complex lexemes formed by derivational affixes are not unlike compounds in several respects, and many morphologists use tree representations to show the relations between the base and affixes. As with compounds, hierarchical tree structures are capable of showing semantic relations in derived lexemes in a salient way. For example, the two trees in Figure 7.4 distinguish the two different meanings of undoable very clearly. Undoable1
7.3 HIERARCHICALSTRUCTURE STRUCTUREININDERIVED DERIVED LEXEMES LEXEMES 7.3 HIERARCHICAL
145 145
(‘cannot be done’) is derived from doable with the negative prefix un-, and undoable2 (‘can be undone’) is derived from undo with the suffix -able. The tree structures in Figure 7.4 show these derivational origins quite directly. undoable 1 A
undoable 2 A A
V
Suf
Pref
V
Suf
Pref
V
un
do
able
un
do
‘which cannot be done’
able
‘which can be undone’
Figure 7.4╇ Two meanings and two structures of undoable
Sometimes different orderings of affixes yield significantly different meanings, and then hierarchical structure can be posited as well. Consider (7.13) from Capanahua. (7.13) a. pi-catsih-ma-hue eat-desid-caus-impv ‘Make him hungry.’ (Lit.: ‘Make him want to eat.’) b. pi-ma-catsihqu-i eat-caus-desid-prs ‘He wants to feed it.’ (Lit.: ‘He wants to make it eat.’) (Payne 1990: 228; data from Eugene Loos) Both of these example words contain desiderative (‘want to do something’) and causative (‘make someone do something’) suffixes, yet they have different meanings (even setting aside differences represented by the imperative and present tense suffixes). We can posit that the difference of meaning reflects a difference of internal structure. In particular, in the first example -ma [caus] attaches to the base pi-catsih- [eat-desid], resulting in the hierarchical structure [[pi-catsih]-ma]. Semantically, this means that -ma does not modify the root pi- ‘eat’, but rather the entire base, meaning ‘want to eat’. The result is the meaning ‘make want to eat.’ (Linguists say that -ma has semantic scope over picatsih-.) Likewise, in the (b) example, -catsihqu [desid] attaches to and has semantic scope over the base pi-ma- [eat-caus] ‘make eat’, resulting in the meaning ‘want to make eat’. The different orderings are therefore associated with different semantic scope, so two very different readings arise. This is just like syntax, and a tree-like representation as in syntax captures the properties of these affixes quite well.
146 
C H A P T E R 7 â•… M O R P H O L O G I C A L T R E E S
In addition to showing semantic relations, tree representations have also been used in morphology for expressing certain formal properties of derived lexemes. Some examples of such representations are given in Figure 7.5. English readable
Russian car-stvo ‘czardom’
Polish awans-owac´ ‘promote’
A
NNEUT
V
English delouse
V
V
Asuf
NMASC
Nsuf
N
Vsuf
Vpref
N
read
-able
carʹ
-stvo
awans
-owac´
de-
louse
Figure 7.5╇ Tree representations of derived lexemes
The English suffix -able is mostly attached to verbs (and occasionally to nouns, as in fashionable), turning them into adjectives. As we saw in Chapter 5, it is quite typical of derivational affixes that they change the word-class of their base lexeme. This can be expressed by saying that the derivational affixes belong to a word-class (noun, verb, adjective) just like full lexemes and stems, and that they may be the heads of the corresponding derived lexemes. Since the formal head determines word-class (as we saw for compounds in the preceding section), the word-class of the resulting lexeme is that of the derivational affix. Thus, read-able is an adjective because -able belongs to the word-class of adjectives, Russian carstvo ‘czardom’ is a noun because of –stvon, and Polish awans-owac´ Â�‘promote’ is a verb because of -owac´v. And as in compounds, derivational affixes also typically determine the gender of nouns (as is illustrated by Russian -stvo, which derives neuter nouns) and the inflection class of the derived lexeme (-stvo derives nouns of the o-declension, and Polish -owac´ derives verbs of the -owa/-uj conjugation). Not all derivational affixes are heads. Many derivational affixes do not determine the word-class and other properties of their derived lexemes. In the European languages, this is true in particular of prefixes and diminutive suffixes. Three such non-head affixes from three languages are listed in (7.14). (7.14) N A V
English co- co-author co-extensive co-exist
Spanish pre- Italian -ino pre-historia ‘prehistory’ tavol-ino ‘little table’ pre-bélico ‘pre-war’ giall-ino ‘yellowish’ pre-ver ‘foresee’ (Adv) ben-ino ‘rather well’
7 . 4 P A R A L L E L S B E T W E E N S Y N TA X A N D M O R P H O L O G Y ?
147
However, derivational affixes often behave like heads of compounds, and this may be regarded as a sufficient reason for treating them as heads as in Figure 7.5.
7.4 Parallels between syntax and morphology? In this book we have already looked at the architecture of the language system in terms of whether inflection and derivation are split between two different components of the grammar (the so-called split-morphology hypothesis; see Section 5.5). However, there is another, equally important question about the architecture of the system: Is morphology formally distinct from syntax? In other words, do the same principles apply to both word formation and sentence formation? There is a variety of evidence that bears on this question, but in this chapter we restrict ourselves to the evidence that is related to formal heads. (Other considerations will be added in Chapter 9 in the context of something called the Lexical Integrity Principle. See also the Further reading section of this chapter.) Some morphologists argue that compounds and derived lexemes have internal hierarchical structure and formal heads because syntactic structure is hierarchical and consists of heads and dependents. In other words, syntax and morphology are similar because syntactic principles govern word-internal structure. Others posit that compounding and derivation operate largely independently of syntax. In this perspective, it is possible for morphological structure to be similar to syntactic structure, but this need not be true. Empirically, then, strong parallels between (hierarchical) morphological structure and syntactic structure would suggest a language architecture in which there is no distinct morphological component, since this provides a more natural explanation for similarities. Strong differences would be inconsistent with a syntactic approach to morphology, and support an architecture in which there are distinct morphological and syntactic components. The question to be answered, then, is: To what extent are morphological heads and syntactic heads similar? Endocentric compounds and syntactic phrases share a semantic trait – in both cases, the dependent member narrows the meaning of the head. For instance, in the compound doghouse the dependent member dog specifies a particular kind of house. And in the syntactic phrase house for a dog, the dependent for a dog has the same kind of semantic relationship to the head noun house. In this limited sense, then, heads in endocentric compounds and in syntactic phrases are similar. However, in syntax, there are also three other purely formal properties that heads share:
148 
C H A P T E R 7 â•… M O R P H O L O G I C A L T R E E S
(7.16) Syntactic head properties a. The head is the morphosyntactic locus, i.e. it bears inflectional markers that belong to the whole phrase. b. The head may govern the form of its dependents. c. A dependent may agree in person/number with its head. As introduced in Section 5.3.1, government is a kind of syntactic relation in which one word requires another word or phrase to have a particular inflectional value. The latter is thus dependent upon the former for its properties. Similarly, in agreement, the inflectional value of a dependent word or phrase must be the same as the inflectional value of another word or phrase in the sentence. These properties can be illustrated with the sentence in (7.17) (a Russian example is chosen because the inflectional properties are less salient in English). In Figure 7.6, a tree diagram for this sentence is given. (7.17) Student-y pomaga-l-i zavedujušcˇ-ej kafedr-oj. student-pl help-pst-pl chairwoman-dat department-ins ‘The students helped the chairwoman of the department.’ SPAST.PL NPNOM.PL
VPAST.PL
NNOM.PL
NPDAT.SG NDAT.SG
NPINSTR.SG NINSTR.SG
Student-y student-PL
pomaga-l-i help-PAST-PL
zavedujusc-ej chairwoman-DAT
kafedr-oj department-INSTR
Figure 7.6╇ A tree diagram for (7.17)
In the tree in Figure 7.6, the head of each phrase is symbolized by a double line between the head and the next higher node in the tree, called the phrasal node. We see that the verb pomagali is the morphosyntactic locus of the sentence; it bears the tense marking that characterizes the whole clause. Likewise, nouns are the morphosyntactic locus of their noun phrases (NPs); they bear the case and number markers that ultimately belong to the NP. We also see two examples of government: The verb pomagali requires its dependent object NP to have dative case, and the noun zavedujušcˇej requires its dependent complement NP to have instrumental case. Finally, pomagali agrees with the subject NP in number.
7 . 4 P A R A L L E L S B E T W E E N S Y N TA X A N D M O R P H O L O G Y ?
149
Of these three syntactic head properties, only morphosyntactic locus applies to compounds as it does to syntactic phrases. The remaining two syntactic head properties cannot be observed because the dependent in compounds does not in general bear inflectional features. As we saw in Section 7.1, the dependent member in compounds is an uninflected stem whose inflectional form cannot be governed and which cannot be the target of agreement. Moreover, similarities between syntactic heads and derivational heads are even weaker. The semantic criterion does not apply here, for obvious reasons: reality is not a kind of -ity, something that is yellowish is not kind of -ish, and so on. So the similarity to the syntactic notion of head is tenuous, and many morphologists have expressed scepticism about the usefulness of carrying over this notion to affixes. It must also be kept in mind that not all derivational patterns involve affixes. It may be possible to describe English carri-er as a headed structure, but Arabic ¯hammaal ‘carrier’ (from ¯hamala ‘carry’) cannot be so described. So for derivational patterns that involve base modification or other non-concatenative operations, it is not clear that even morphosyntactic locus is relevant. Head characteristic
Syntactic Compound Derivational
Dependent narrows meaning
+
+
–
Morphosyntactic locus
+
+
(+)
Government
+
–
–
Agreement
+
–
–
Table 7.1╇ Properties of formal heads in syntax, compounding and derivation
In short, hierarchical structure at the word level is superficially similar to hierarchical structure at the level of the sentence, but parallels between syntactic and formal morphological heads are only partial for compounds, and even less strong for derivation. Clearly, there might be some underlying principle of language that causes both syntax and compounding (and maybe derivation) to exhibit hierarchical structure and head properties, but the principles governing syntactic structure and the principles governing morphological structure are not identical. The degree of formal separation between morphology and syntax is a topic of ongoing debate.
150 
C H A P T E R 7 â•… M O R P H O L O G I C A L T R E E S
Summary of Chapter 7 There are many types of nominal compounds: endocentric � compounds, exocentric compounds, and various kinds of compounds with more than one semantic head (e.g. coordinative compounds and appositional compounds). Like syntactic phrases, (endocentric) compounds are often conveniently described as having hierarchical structure and formal heads. These are represented using tree diagrams. Such hierarchical structures are often also applied to derived lexemes, and derivational suffixes are often described as the heads of their words. However, derivational suffixes share only a few properties with the heads of syntactic phrases, and even compounds do not exhibit the syntactic head properties of government and agreement. Parallels between syntactic and morphological structure are thus not particularly strong, indicating that principles of word formation are, at least to some degree, distinct from principles of sentence formation.
Further reading For a cross-linguistic survey of compounding, see Bauer (2001a) and Guevara and Scalise (2009). For noun incorporation, see Mithun (1984). For coordinative compounds, see Olsen (2001). Investigation of compounds from psycholinguistic and neurolinguistic perspectives has only recently begun in earnest, but see Libben and Jarema (2006). The approach that uses hierarchical structures is most prominently represented by works such as Selkirk (1982), Di Sciullo and Williams (1987), Lieber (1992) and Embick and Noyer (2007). On heads in morphology, see in particular Williams (1981a), Scalise (1988b) and Haspelmath (1992), and, for some sceptical voices, see Reis (1983) and Bauer (1990). A recent defence is found in Štekauer (2000). Regarding the broader argument of whether the morphological component can be collapsed into the syntactic one, classic arguments in favour are Baker (1985) and Pesetsky (1985). Specific attacks on the syntactic approach to morphological structure include Spencer (1997) and Smirniotopoulos and Joseph (1998). The typological evidence for compounding as a syntactic process is assessed in Sadock (1998) and Baker (1998), to opposite conclusions.
COMPREHENSION EXERCISES 
151
Comprehension exercises 1. Draw tree diagrams (analogous to those in Figures 7.1–7.3) for the following English compounds: family planning adviser, undersea cable repair team, fixed-line phone system, mad cow disease hysteria, World Trade Center rescue worker, credit cardÂ� agreement form, major league baseball game
(For some of these, two different solutions may be correct.)
2. Consider the following Sanskrit compound stems and, judging by their meaning, determine the type of compound in each case. (Note that in Sanskrit, morphophonological alternations often slightly alter the shape of roots at morpheme boundaries.) as´vakovida- ‘knowledgeable about horses’ bahuvrı¯hi- ‘having a lot of rice’ divyaru¯pa- ‘having divine shape’ grhapati- ‘houseowner’ maha¯ra¯ja- ‘great king’ maha¯tman- ‘having a big soul, i.e. magnanimous’ priyasakhı¯- ‘dear friend’ ra¯jars.i- ‘poet who is also royalty’ s´uklakrs.n.a- ‘bright and dark’ sukhaduh.kha- ‘joy and pain’
A list of relevant Sanskrit nominal and adjectival roots: a¯tman- ‘soul’ gr.ha- ‘house’ as´va- ‘horse’ kovida- ‘knowledgeable’ bahu- ‘much’ krs.n.a- ‘dark’ divya- ‘divine’ mahat- ‘big’ duh.kha- ‘pain’ pati- ‘lord’ priya- ‘dear’ s´ukla- ‘bright’ r s.i- ‘seer, poet’ sakhı¯- ‘female friend’ ra¯jan- ‘king’ sukha- ‘joy’ ru¯pa- ‘shape’ vrı¯hi- ‘rice’
3. In Spanish, there are two homophonous adjectives inmo´vilizable: inmo Â� ´vilizable1 ‘unmobilizable’ and inmovilizable2 ‘immobilizable’. The morphological structure of these words corresponds closely to the structure of the corresponding English words (prefix in- ‘un-’, suffix -able ‘-able’, suffix -iz ‘-ize’, móvil ‘mobile’). Draw the constituent Â� structure trees of these two words. 4. Russian has a productive class of exocentric A + N compounds comparable to the Ancient Greek compounds in (7.7):
152 
C H A P T E R 7 â•… M O R P H O L O G I C A L T R E E S
dlinnorukij krasnoborodyj cˇernokožij tolstonogij
Formulate the word-based rule for these compounds.
(Note: -yj and -ij are phonological allomorphs. Russian consonants can be phonologically ‘hard’ or ‘soft’. The adjectival allomorph -yj is used when the preceding consonant is hard, and -ij when it is soft. Since the difference results from phonology, the conditions for -yj vs. -ij do not need to be captured by your rule.)
‘long-armed’ ‘red-bearded’ ‘black-skinned’ ‘thick-legged’
dlinnyj krasnyj cˇernyj tolstyj
‘long’ ‘red’ ‘black’ ‘thick’
ruka boroda koža noga
‘arm’ ‘beard’ ‘skin’ ‘leg’
Exploratory exercise In this chapter, we saw that compounds and maybe derived lexemes can be described as having hierarchical structure. In some respects, word structure thus has much in common with sentence structure. We have also asked whether these similarities are sufficient to posit that the same or closely related principles govern both syntax and morphology. In this task, you will delve deeper into the parallels (or lack thereof) between sentence structure and word structure by conducting a typological survey of word order and the order of affixes and stems. Typologists have demonstrated that among the world’s languages, some grammatical structures systematically co-occur with others. For example, dominant word order is closely connected to the order of prepositions/ postpositions relative to nouns. (A postposition is like a preposition, except that it comes after the noun (post-), rather than before it (pre-). They are collectively referred to as adpositions.) In some languages, the dominant order for a sentence with both a subject and an object is subject-verbobject (abbreviated SVO). English is such a language: John (=subject) read (=verb) the letter (=object). In other languages, both the subject and the object normally come before the verb (SOV order). Japanese is this kind of language, as shown in (7.18). Other languages may have yet other dominant word orders. (7.18) John ga tegami o yon-da John sbj letter obj read-pst ‘John read the letter.’ Interestingly, languages with SVO order are very likely to have prepositions, whereas languages with SOV order are very likely to have postpositions. Table 7.2 shows the number of languages with each combination of word order and adposition type, based on a sample of 750 languages.
E X P L O R AT O R Y E X E R C I S E
Basic Order
Prep
Post
SOV SVO VSO VOS OVS OSV
╇╇ 8 278 ╇ 69 ╇ 20 ╇╇ 2 ╇╇ 0
328 ╇ 29 ╇╇ 6 ╇╇ 0 ╇╇ 7 ╇╇ 3
153
Table 7.2╇ Correlation between dominant word order and prepositions vs. postpositions (Dryer 2008a, b)
Linguists thus say that the order of subjects and objects relative to verbs and the order of adpositions relative to nouns exhibit an implicational relationship. Knowing the dominant word order of a language greatly increases the chance of correctly predicting whether adpositions come before or after nouns. (Of course, implicational relationships can have exceptions. Finnish and Estonian, for example, are predominantly SVO languages, but they have postpositions. Implicational relationships thus represent likely co-occurrence rather than an absolute correlation.) Linguists are interested in this implicational relationship because it suggests that the generalization governing the order of subjects/objects and verbs may be closely related to the generalization governing the order of adpositions and nouns. In particular, the generalization seems to be that most languages are either predominantly head-initial or predominantly head-final, meaning that syntactic heads occur at the beginning or end of phrases, respectively. Adpositions are the heads of their phrases, and verbs are the heads of VPs. Thus, if a language has both SOV word order and postpositions, it is predominantly head-final. The purpose of this research exercise is to determine whether there is an implicational relationship between word order and the order of affixes relative to stems, and to explore the importance of this typological data for the following research question: Are the principles governing word order and morpheme order closely related? If sentence structure and word structure follow the same basic principles, we might expect an implicational relationship to hold between them. Instructions Step 1: Develop hypotheses and predictions. Remember that a hypothesis is a formal guess about the relationship between syntactic and morphological rules (i.e. are they different?). A prediction is the result that you expect in your data, based on your hypothesis. So, for instance, if a language is
154 
C H A P T E R 7 â•… M O R P H O L O G I C A L T R E E S
syntactically head-initial, do you expect to find that morphological heads (i.e. affixes) will also precede their dependents (stems)? What would be predicted by a theory in which morphological and syntactic rules are not distinct? By a theory in which they are distinct? Step 2: Collect data using World Atlas of Language Structures (WALS). The atlas is available both as a book accompanied by CD-ROM, and also via the web (http://wals.info). Identify languages for which inflection is primarily expressed with prefixes or primarily with suffixes (Dryer 2008c). Then look at word order for these same languages (Dryer 2008a, d). Organize your data – including language family and where in the world the language is spoken – as shown in Table 7.3. Obviously, you will need a much bigger sample of languages than is shown in the table. Language
Language family Where spoken
Affixes
Basic Order
Paiute (Northern)
Uto-Aztecan, Numic
N. America
Mostly suffixes
SOV
Evenki
Altaic, Tungusic
C. Asia
Mostly suffixes
SOV
Arrernte (Mparntwe)
Australian, Pama- Australia Nyungan
Mostly suffixes
SOV
Albanian
Indo-European, Albanian
E. Europe
Mostly suffixes
SVO
Kikuyu
Niger-Congo, Benue-Congo
E. Africa
Mostly prefixes
SVO
Mixtec (Chalcatongo)
Oto-Manguean, Mixtecan
C. America
Mostly prefixes
VSO
Table 7.3╇ A sample of languages according to inflectional prefixation vs. suffixation and word order
Give some thought to how you select the languages in your sample. Do you want to pick randomly from among those for which information is available? Or do you want your sample to be representative according to some criterion (e.g. geographically representative, or evenly distributed among language families)? Try to anticipate how your choice of methodology might affect your results. Hint: The interactive tool available as part of the CD-ROM and web versions of WALS is the most efficient way to look for correlations, because it allows the combination of two features to be mapped (called the ‘Composer’ feature on the CD-ROM and ‘Feature Combination’ in the web version). The book and web versions contain articles with more detailed information
E X P L O R AT O R Y E X E R C I S E
155
about individual features; these articles are useful for understanding how the data set was constructed, and for overviews of typological patterns. Step 3: Analyze the data. Was your prediction upheld? Remember to consider any issues, including the following: 1) Is there is an implicational relationship between the order of words and the order of stems and affixes? If so, how strong is it? 2) Is any implicational relationship of the type that you predicted? For example, if you predicted that heads and dependents are in the same order at both the sentence level and the word level, is this the pattern found in your data? 3) Do some patterns occur predominantly in one language family, or in one area of the world? If so, is this what you would expect, given your hypothesis? What explanations could there be for grouping? 4) This chapter was primarily about hierarchical structure in compounds and derivational affixes, but in this study affix-stem order refers to inflectional affixes. How important is this discrepancy? Or stated differently, do morphologists ever claim that inflectional affixes are heads, and does this change your analysis/conclusions at all? (A good answer to this question will require you to read more about heads in morphology. See the Further reading section.) 5) Finally, also consider the possible impact of your data collection methodology on your results. Step 4: Draw conclusions. Consider whether your data supports only one answer to the research question, or whether multiple answers are consistent with your data. Extension: Explore other kinds of word orders (e.g. the order of adpositions and noun phrases, the order of adjectives and nouns, and so on), and possible correlations between these and the order of affixes and stems.
8
Inflectional paradigms
8.1╇ Syntagmatic and paradigmatic relations in morphology The relations between linguistic units are of two broad kinds: syntagmatic relations between units that (potentially) follow each other in speech, and paradigmatic relations between units that (potentially) occur in the same slot. In other words, syntagmatic relations have to do with items ordered one after the other, while paradigmatic relations have to do with items that stand in contrast to one another. We can think about syntagmatic and paradigmatic relations at the sentence level, for instance, in (8.1), where the horizontal dimension shows syntagmatically related units, and the vertical dimension shows paradigmatically related units. Parentheses show optionally occurring linguistic units, curly brackets show choices among units, and asterisks show impossible units.
{ }
  
{
}
(8.1) In the beginning ╇ God ╇ created the heaven (and the earth) (*not). *Ø â•‡ Allah ╇ made heavens ╇ he ╇ *create *why *rested Morphology can likewise be looked at from both a syntagmatic and a paradigmatic point of view. Bases are syntagmatically related to affixes that attach to them, whereas word-forms belonging to the same lexeme are paradigmatically related because they form a set of contrasting instantiations of the lexeme (to take a simple example, the English word bag is identifiable as having singular number exactly because it contrasts with the plural form bags, and because singular and plural forms generally form such a contrast in English).
8 . 1 ╇ S Y N TA G M AT I C A N D PA R A D I G M AT I C R E L AT I O N S I N M O R P H O L O G Y
157
{ }
(8.2) bag ╇ -Ø â•‡ -s *-en (as in children) Now, in developing a description of language architecture, we might ask whether we need formal mechanisms that encode both syntagmatic and paradigmatic dimensions of structure. To be sure, some linguists have posited models that describe morphological patterns in purely syntagmatic terms. The formalism in (8.3) (which we have used in this book to represent the morpheme-based model; see (3.22)), represents bags as the linear combination of the morphemes bag and -s. (8.3) bag -s ╇ /bæg/ ╇ /z/ ╇ N + ╇ N — ‘bag’╇ ‘plural’
{ } { }
Æ
bags
Crucially, (8.3) does not include the paradigmatic perspective at all, which is to say that there are no direct relations between word-forms belonging to the same lexeme. (Bag is the root, not a word-form with singular number. Formally, it is purely coincidental that the root has the same form as the singular.) The same pattern may also be described with emphasis on the paradigmatic dimension. Example (8.4), our standard formalism for the word-based model, also represents the syntagmatic aspect of the structure (the order in which the stem and plural marker appear in the word), but additionally it draws a direct relationship between singular and plural forms.
{
}
{
´
(8.4) 1 /X/n ‘x’ number: sing ´
/Xz/n ‘x’ number: plur
}
So it is clear that morphological rules need to capture the syntagmatic dimension since affixes occur in particular positions relative to their bases, but the real question is whether we need to also incorporate paradigmatic rules into our formal description of language structure. Logically, if our formal description could make do with only syntagmatic description and
1
This correspondence is set up vertically to emphasize that this is a paradigmatic relation and to be consistent in this respect with (8.1)–(8.2). But of course, the spatial layout is just a convenience. We will return to the normal, space-saving horizontal representation below.
158 
C H A P T E R 8 â•… I N F L E C T I O N A L P A R A D I G M S
still be empirically adequate, then this should be simpler, and therefore preferable. The answer depends on whether there are phenomena in language that can be adequately described only as a direct relationship between inflected forms. If so, then morphological structure would seem to include both syntagmatic and paradigmatic dimensions. If not, then a purely syntagmatic model would be sufficient. In this chapter, we show that some inflectional patterns do indeed seem to require a paradigmatic approach, indicating that paradigmatic relations are part of the architecture of the (inflectional) morphological system. But to make this discussion possible, we must first introduce the idea of the inflection class.
8.2╇ Inflection classes Perhaps the most important challenge for an insightful description of inflection is the widespread existence of allomorphy in many languages. Phonological and morphophonological allomorphy will be the topic of Chapter 10; in this section we focus on suppletive allomorphy. We saw some examples of suppletive inflectional affixes in Section 2.3, and two more are given in (8.5)–(8.6). (8.5)
Irish nom.sg. focal muc corón
(8.6)
Old English infinitive de¯m-an ‘to deem’ luf-ian ‘to love’
gen.pl focail muic-e corón-ach
‘word’ ‘pig’ ‘crown’ 3rd sg present de¯m-ð ‘deemeth’ luf-að ‘loveth’
3rd sg past de¯m-de ‘deemed’ luf-ode ‘loved’
In (8.5), for instance, all three Irish words show zero expression in the nominative singular, but in the genitive plural each word has a different inflectional marker: zero, -e or -ach. The genitive plural thus exhibits suppletion.2 When different lexemes show different suppletive inflectional allomorphs, morphologists say that they belong to different inflection 2
Remember from Section 2.3 that some linguists use the term suppletion only to refer to non-phonological allomorphy in stems. Under such a definition, we would say that the words in (8.5) and (8.6) exhibit different affixal morphemes (rather than different allomorphs). This is primarily a terminological issue, however, and it has no real impact on how inflection classes are defined.
8 . 2 ╇ I N F L E C T I O N C L A S S E S
159
classes.3 An inflection class is thus the set of paradigms that exhibit the same inflectional pattern. Inflection classes may be very large and contain hundreds or thousands of lexemes, or they may be small and contain only a handful of � lexemes. The limiting case would be an inflection class with just a single lexeme (for most purposes, this would amount to saying that the inflection of that lexeme is irregular). Languages differ in the number of inflection classes that they exhibit. In (8.5), three different nominal inflection classes (or declensions) are illustrated, and, in (8.6), two verbal inflection classes (or conjugations) are shown. The existence of different inflection classes is a hallmark of IndoEuropean languages, so many examples in this chapter will come from the Indo-European language family. Of course, the phenomenon is not restricted to Indo-European, but there are many languages with fairly complex morphological systems in which suppletion of this kind is not found or is at least much less prominent (for instance, Turkish, Korean, Quechua and Tamil). Thus, not all languages have multiple inflection classes. In inflection classes, the various suppletive allomorphs are grouped into sets. This can be seen by looking at the complete paradigms of the two Latin words: (8.7) sg nom acc gen dat abl pl nom acc gen dat abl
o-declension hort-us hort-um hort-ı¯ hort-o¯ hort-o¯ hort-ı¯ hort-o¯s hort-o¯rum hort-ı¯s hort-ı¯s
u-declension grad-us grad-um grad-u¯s grad-uı¯ grad-u¯ grad-u¯s grad-u¯s grad-uum grad-ibus grad-ibus
We can say that a Latin noun in -us (like hortus ‘garden’, gradus ‘step’) has a genitive plural in -orum if its genitive singular is -ı¯, and a Â� genitive plural in -uum if its genitive singular is -u¯s. If the distribution were arbitrary, we might expect that some nouns in Latin would have the genitive singular -ı¯, the ablative singular -u¯, the accusative plural -o¯s, and the dative plural -ibus, for instance. But, in fact, a noun can only choose a complete package of suffixes, either the package of hortus (generally called the o-declension) or the package of gradus (generally called the u-declension). Thus, one form can be used to predict another. Of course, in practice some of these dependencies may be more useful than others. For example, the nominative and accusative singular are identical in both classes and therefore have little
3
The term inflection class is not generally used for phonological allo�morphy.
160 
C H A P T E R 8 â•… I N F L E C T I O N A L P A R A D I G M S
or no predictive value. Less obviously, learners of Latin probably hear the genitive singular of a new word more often than its genitive plural, so the ability to predict the genitive plural from the genitive singular is probably more relevant than the ability to make the reverse prediction.
8.2.1╇ Inflection class assignment Words are assigned to inflection classes based on phonological, semantic or morphological criteria. Some examples of phonological class assignment are given in (8.8). (8.8) a. 0007Lezgian aorist participle: -j(i) after low vowel (a, e), -r after high vowel (u, ü, i) aorist finite awu-na t’ü-na fe-na ata-na aorist participle awu-r t’ü-r fe-ji ata-j ‘did/ ‘ate/ ‘went/ ‘came/ done’ eaten’ gone’ come’ (Haspelmath 1993: 131)
b. 0007Eastern Armenian plural: -er with monosyllabic bases, -ner with polysyllabic bases . sg jerk’ yug erexa tari . pl jerk’-er yug-er erexa-ner tari-ner ‘hand(s)’ ‘oil(s)’ ‘child(ren)’ ‘oil(s)’
c. 0007Standard Arabic plural: If the singular has the phonological shape CVCCVC, then the plural has the form CaCaaCiC. If the singular is CVCCVVC, the plural is CaCaaCiiC. sg qays.ar daftar dirham dustuur quft.aan pl qayaas.ir dafaatir daraahim dasaatiir qafaat.iin ‘emperor’ ‘notebook’ ‘drachma’ ‘statute’ ‘caftan’
Among semantic criteria, animacy distinctions are particularly widespread. In German, only animate nouns belong to the masculine n-declension ending in -e in the nominative singular (Hase ‘hare’, Affe ‘ape’, Junge ‘boy’). In Tamil, the locative suffix is -il with non-human nouns (e.g. na¯t.t.-il ‘in the country’), but -it.am with human nouns (e.g. manitan-it.am ‘in the man’) (Annamalai and Steever 1998: 105). Welsh has a special plural suffix for nouns denoting animals, -od (e.g. cath/cathod ‘cat(s)’, draenog/draenogod ‘hedgehog(s)’, Â�eliffant/eliffantod ‘elephant(s)’) (King 1993: 59). Lezgian has a special oblique-stem marker that is used with all consonant-final proper names, -a (e.g. Farid-a ‘Farid’, Talibov-a ‘Talibov’). Lezgian also illustrates the potential relevance of the mass-count distinction: mass nouns tend to have the oblique-stem suffix -adi/-edi (e.g. naq’w-adi ‘soil’, kf-adi ‘foam’, hüm-edi ‘haze’) (Haspelmath 1993: 75–6). In verbs, transitivity often plays a role. For
8 . 2 ╇ I N F L E C T I O N C L A S S E S
161
example, in Ossetic intransitive and transitive verbs show different agreement inflection in the past tense. The singular forms of the intransitive verb xuyssy‘sleep’ and of the transitive verb dzur-/dzyr- ‘say’ are given in (8.9). (8.9) 1sg 2sg 3sg
intransitive pattern xuyssy-d-œn ‘I slept’ xuyssy-d-œ xuyssy-d
transitive pattern dzyr-d-on ‘I said’ dzyr-d-aj dzyr-d-a (Isaev 1966: 247)
When there is morphological assignment, most typical is that the derivational pattern of a derived lexeme determines its inflectional behaviour. For example, Welsh has about a dozen different plural patterns, which are often unpredictably associated with individual nouns. However, when a noun has a derivational suffix, it is mostly predictable which plural affix the noun takes:4 (8.10) Derivational Plural Base suffix suffix -og (person) -ion swydd ‘job’ march ‘horse’
Derived form swydd-og(-ion) ‘official(s)’ march-og(-ion) ‘horseman/men’
-es (female)
-au
tywysog ‘prince’ tywysog-es(-au) ‘princess(es)’ Sais ‘Englishman’ Saesn-es(-au) ‘Englishwoman/ -women'
-ur (agent)
-iaid
pechu cachu
‘sin’ ‘shit’
pechad-ur(-iaid) ‘sinner(s)’ cachad-ur(-iaid) ‘coward(s)’
(King 1993: 53–61) Another example of derivational patterns determining inflectional behaviour comes from Tagalog. In this language most verbs have a derivational affix that indicates in some way the transitivity or voice of the verb (e.g. actor voice -um-, ma-, patient voice -in, -an). The perfective form of the verb can be formed in four different ways: (i) zero (when the voice affix is -um-); (ii) m- becomes n- (e.g. when the voice affix is ma-); (iii) infix -in- (e.g. when the voice affix is -an); and (iv) infix -in- and subtraction of -in (when the voice affix is -in): (8.11)
4
root takbo tulog hugas basah
basic form with voice affix tumakbo matulog hugasan basahin
perfective form tumakbo natulog hinugasan binasah
gloss ‘run’ ‘sleep’ ‘wash’ ‘read’
As we saw in Section 7.3, this is one of the reasons why some morphologists regard derivational affixes as heads of their lexemes – the derivational affix determines the inflectional pattern of the entire word-form.
162 
C H A P T E R 8 â•… I N F L E C T I O N A L P A R A D I G M S
Again, this illustrates the dependence of inflection-class membership on a morphological property of the lexeme (its derivational pattern). Of course, the derivational pattern need not be characterized by an affix. In Arabic, nouns derived by the pattern C1aaC2iC3 tend to have the plural C1uC2C2aaC3 (e.g. kaafir ‘infidel’, plural kuffaar; kaatib ‘writer’, plural kuttaab; zaahid ‘ascetic’, plural zuhhaad).
8.2.2╇ Relationship to gender Inflection classes are often linked to gender, but this relationship can be complicated. For instance, in the Bantu languages (e.g. Zulu), a link between inflection class and gender is evident: the markers that reflect the gender on agreement targets are formally similar to the inflectional affixes on controller nouns. Zulu has the four inflection classes illustrated in (8.12), among �others. (8.12)
sg prefix pl prefix example um- aba- umfazi/abafazi um- imi- umfula/imifula i- ama- itafula/amatafula isi- izi- isicathulo/ izicathulo
meaning ‘woman/-men’ ‘river(s)’ ‘table(s)’ ‘shoe(s)’
agr prefixes u-/bau-/ili-/asi/zi-
The agreement prefixes for the genders corresponding to the four inflection classes are given in the last column in (8.12). Two examples of their use as subject prefixes on verbs are given in (8.13). (8.13) a. Aba-fazi ba-biza aba-fana pl.g2-woman 3pl.g2.sbj-call pl.g2-boy ‘The women call the boys.’ b. Isi-hambi si-buza pl.g8-traveller 3pl.g8-ask ‘The traveller asks the road.’
um-gwaqo. sg.g3-road (Ziervogel et al. 1981: 34, 46)
The similarities between aba- and isi- as inflectional markers on the nouns, and ba- and si- as gender agreement markers on the verbs, are striking. Clearly, there is a close correspondence between the gender classes and inflection classes in Zulu. Nonetheless, we need to make a principled distinction between inflection class and gender. Consider the Italian examples (8.14)–(8.15). (8.14)
Two Italian inflection classes sg suffix pl suffix example -o -i giardino/giardini -a -e casa/case -o -i mano/mani -a -i poeta/poeti
meaning ‘garden(s)’ ‘house’ ‘hand’ ‘poet’
agr suffixes -o/-i (masc.) -a/-e (fem.) -a/-e (fem.) -o/-i (masc.)
8 . 2 ╇ I N F L E C T I O N C L A S S E S
(8.15)
163
Italian gender agreement (adjectives agree with nouns) a. il giardin-o nuov-o ‘the new garden’ b. la cas-a nuov-a ‘the new house’ c. la man-o rugos-a ‘the wrinkled hand’ d. il poet-a mort-o ‘the dead poet’
In Italian we clearly need to distinguish between inflection classes and genders because there are nouns that have the singular suffix -o but are feminine (e.g. mano ‘hand’) and nouns that have the singular suffix -a but are masculine (e.g. poeta ‘poet’). Such nouns are much rarer than the opposite pattern, but they exist. Thus, we must make a principled distinction between a noun’s inflection class (which determines the set of inflected forms of the noun lexeme), and its gender (an agreement feature). Formally, semantic factors and the inflection class of a noun together determine the noun’s gender (Corbett 1982, 1991). For instance, most Russian nouns belonging to inflection class II (indicated by a nominative singular form ending in -a, e.g. kniga ‘book’) are feminine, but a small subset are masculine (sluga ‘male servant’). These latter nouns take the same endings as class II feminine nouns, but we know that they are masculine because they trigger masculine syntactic agreement (xorošaja kniga ‘good(fem.) book’; xorošij sluga ‘good(masc.) servant’). If gender determined inflection class, we would expect these words to fall into class I, which contains the vast majority of masculine nouns. However, if inflection class and semantic factors together determine gender, the observed pattern is easier to explain. Importantly, all of the nouns like sluga refer to sexdifferentiable male beings. We must therefore simply assume that gender assignment in Russian is determined by natural sex when the word refers to a sex-differentiable animal or human, and by inflection class otherwise. (There are also languages in which gender is purely semantically based, or in which it depends on a combination of semantic and phonological criteria (Corbett 1991), but these are not directly relevant here.)
8.2.3╇ Inflection classes and productivity Like word-formation patterns, inflection classes can differ in productivity. Specifically, they can differ in their ability to apply to novel lexemes that come into the language, either as loanwords or as neologisms formed by productive word-formation rules, and they can also differ in their ability to attract new members by inflection class shift. A class shift is a diachronic change by which a lexeme changes its inflection class. (We have already seen an example of inflection class shift in Section 6.4.3: fling/flinged Æ fling/flung.) It is convenient to identify three degrees of inflection-class productivity on the basis of these criteria (Dressler 1997), as summarized in Table 8.1.
164 
C H A P T E R 8 â•… I N F L E C T I O N A L P A R A D I G M S
Criteria and examples
Highly productive Classes with classes intermediate productivity
Unproductive classes
Apply to loanwords
YES
NO
NO
Attract classshifting lexemes
YES
NO
NO
Can form neologisms
YES
YES
NO
Criteria
Examples Welsh plurals
-au, -iaid
-oedd
-edd
Italian nouns
-o/i
-a/i, -e/i
—
Russian nouns
-C (I), -a(II)
-o (IV)
-ja
Table 8.1╇ Three degrees of inflection-class productivity
Only highly productive classes are able to accommodate loanwords and attract lexemes from other, unstable classes. Productively formed neologisms, by contrast, often go into classes with intermediate productivity. Completely unproductive classes do not get new members at all, and, since they inevitably lose some members (e.g. when a word becomes obsolete), they are ultimately doomed to disintegration unless the productivity of the class changes. For exemplification, let us go back to Welsh plurals (King 1993: 52–64). As shown in (8.10), Welsh has several highly productive plural classes that can accommodate loanwords from English – for instance, the suffix -au, which is the most common Welsh plural suffix (e.g. siop/siopau ‘shop(s)’, trên/ trenau ‘train(s)’), or -iaid, which is often used with nouns denoting persons (e.g. doctor/doctoriaid ‘doctor(s)’, biwrocrat/biwrocratiaid ‘bureaucrat(s)’). Both these classes also apply to regularly formed neologisms. Thus, -au is always used with quality nouns in -deb (e.g. ffurfioldeb ‘formality’, ffurfioldebau ‘formalities’), and -iaid is always used with agent nouns in -dur (e.g. pechadur ‘sinner’, pechaduriaid ‘sinners’). The class in -au also shows its productivity in attracting members of other classes – e.g. from the class of plurals in -oedd. For example, amser ‘time’ has an older plural amseroedd and a newer plural amserau, and cylch ‘circle’ has an older plural cylchoedd and a newer plural cylchau. The plural class in -oedd is thus losing members, but it has at least intermediate productivity in that productively formed place-nouns in -fa have -oedd plurals (e.g. meithrinfa ‘nursery’, meithrinfaoedd ‘nurseries’). Completely unproductive is, for
8 . 3 ╇ P A R A D I G M AT I C R E L AT I O N S A N D I N F L E C T I O N C L A S S S H I F T
165
instance, the plural suffix -edd of bys/bysedd ‘finger(s)’, as well as the various classes of vowel-changing plurals (e.g. fford/ffyrd ‘road(s)’, asgell/esgyll ‘wing(s)’). In Italian nouns, the -o/-i and -a/-e inflection classes (see (8.14)) are highly productive: they are used with loanwords (e.g. il chimono, plural i chimoni ‘kimono(s)’, la giungla, plural le giungle ‘jungle(s)’), and occasionally they attract members from other inflection classes in non-standard varieties of Italian (e.g. standard language il pane ‘bread’ becomes il pano ‘bread’, la moglie becomes la moglia ‘wife’). The class in -a/-i (poeta/poeti ‘poet(s)’) cannot be used with loanwords: the noun lama ‘Tibetan monk’ does not get the plural -i (*i lami ‘lamas’) but remains unchanged in the plural (i.e. it joins the class of indeclinables, like all consonant-final loanwords). However, the -a/-i class is not totally unproductive, as it is used with the productive suffix -ista (e.g. leghista ‘follower of the Lega’, plural leghisti). There is no real unproductive class in Italian, unless one regards the few irregular nouns (uomo/uomini ‘man/men’, bue/buoi ‘ox(en)’, etc.) as classes of their own. In Russian, two highly productive classes have been absorbing many lexemes from other inflection classes over the past millennium. Loanwords become class I if they end in a ‘hard’ consonant (komp´juter), or class II if they end in -a (disketta). But by contrast, class IV (consisting almost entirely of neuter nouns ending in -o or -e) is not highly productive – even loanwords ending in -o (such as pal´to ‘coat’ from French paletot) do not follow this class but are indeclinable. However, the class still gets new members through productive suffixes like -stvo, which creates abstract nouns (e.g. professorstvo ‘professorship’). There is a small class of neuters in -ja (e.g. vremja ‘time’) that is totally unproductive.
8.3╇ Paradigmatic relations and inflection class shift In the rest of this chapter we explore why it can be useful, and in some cases necessary, to look at inflection classes from a paradigmatic perspective. We begin by using the word-based model from Section 3.2.2 to develop a formal description. In a word-based description, the relation between the inflected forms of a lexeme can be seen as parallel to the relation between two derivationally related lexemes. Thus, the relation between hortı¯ ‘garden, nom.pl’ and horto¯rum ‘garden, gen.pl’ can be characterized by the rule in (8.16). The full form of the rule is given in (8.16a), and (8.16b) shows an equivalent abbreviated notation.




(8.16) a. ╇ /Xı¯/n /Xo¯ rum/n ╇ ‘x’ ´ ‘x’ ╇ case: nominative case: genitive number: plural╇╇╇╇╇ number: plural
b. [/Xı¯/nom.pl] ´ [/Xo¯ rum/gen.pl]
166 
C H A P T E R 8 â•… I N F L E C T I O N A L P A R A D I G M S
The fact that there is no Latin noun with a nominative plural in -ı¯ and a genitive plural in -uum is thus expressed by the non-existence of a rule that would link these two suffixes. Thus, a correct genitive plural form can be created on the basis of other word-forms in the paradigm, and in fact every form can be created on the basis of every other form. Since there are ten forms in the paradigm, we can posit 45 pairwise rules like (8.16). Now recall that, even in derivational morphology, there is sometimes reason for positing rules that involve more than two word-schemas. In Section 3.2.2 we saw that, in addition to the English rules in (8.17a,b), we also need the rule in (8.18) because there are word families that contain two derived forms, but not the root (e.g. illusion, illusive, *illude).
[ ] [
] [ ] [ ]
]
(8.17) a. /X/v ´ /Xion/n b. /X/v ´ /Xive/a ‘dox’ ‘action of doingx’ ‘dox’ ‘prone to doingx’
[
]
[
(8.18) /Xion/n ´ /Xive/a ‘action of doingx’ ‘prone to doingx’
These three rules are more properly described as a single rule relating three word-schemas. This can be represented using a further notational convention: sets of corresponding word-schemas are enclosed in curly brackets and separated by commas. Thus, (8.19) is a convenient notation for the combination of (8.17) and (8.18).
{[
] [
]
[
]}
(8.19) /X/v , /Xion/n , /Xive/a ‘dox’ ‘action of doingx’ ‘prone to doingx’
If we adopt this formalism, we can formulate the rule in (8.20), which contains 10 corresponding word-schemas, to describe the Latin paradigm.5 (8.20) 0007{ [/Xus/nom.sg], [/Xı¯/gen.sg], [/Xo¯/dat.sg], [/Xum/acc.sg], [/Xo¯/abl.sg], [/Xı¯/nom.pl], [/Xo¯ rum/gen.pl], [/Xı¯s/dat.pl], [/Xo¯ s/acc.pl], [/Xı¯s/abl.pl] } In what follows, we will call rules like (8.20) paradigm rules. Such wordbased rules capture the generalization that inflectional markers come in packages. This approach is useful as a way to explain class shifts. For example, in later Latin quite a few nouns of the u-declension shifted to the o-declension – e.g. sena¯tus ‘senate’ (older genitive form sena¯tu¯s, newer genitive sena¯tı¯), exercitus ‘army’, fru¯ctus ‘fruit’. How can class shift be explained? First,
5
This word-based rule is just a notational variant of the paradigms found in Latin school grammars. Latin school grammars usually give a concrete lexeme like hortus, but everyone understands that hortus is just an example and really stands for /Xus/. Thus, the wordbased description is just a somewhat more explicit variant of what school grammars have long been doing.
8 . 4 ╇ I N H E R I TA N C E H I E R A R C H I E S
167
we noted earlier that not all word-forms are equally good indicators of inflection class membership. The nominative singular and genitive singular together can uniquely identify the class that a Latin noun belongs to, but the nominative singular by itself cannot. So to explain this shift, we need assume only that the innovating speakers did not remember the genitive form of these nouns for some reason (perhaps because it is infrequent). Now if they remember only the nominative form, the word matches two paradigm rules – i.e. it could belong either to the o-declension or to the u-declension. In such situations of choice, speakers tend to opt for those rules that generalize over more items. Latin always had many more o-declension nouns than u-declension nouns, so that the o-declension rule was stronger. This explains why shifts from the u-declension to the o-declension were common in Latin, but shifts in the opposite direction did not occur (see Wurzel 1987: 79). A key observation for explaining inflection class shift is thus that the form of a word is indicative of its inflection class membership, which is the same as saying that speakers use one word-form in a paradigm to predict another. This is a paradigmatic relation. A purely syntagmatic approach (such as the morpheme-based model) would have trouble describing this generalization, but word-based rules are naturally well suited to accounting for this data because they incorporate and emphasize the paradigmatic dimension of (inflectional) morphological structure.
8.4╇ Inheritance hierarchies The value of the paradigmatic perspective also emerges when we examine similarities across inflection classes. From what we have said so far, one might get the impression that inflection classes may differ arbitrarily in the kinds of markers that they exhibit. But in fact different inflection classes often show great similarities, to the point where it is unclear whether a separate inflection class needs to be set up. Let us consider seven important inflection classes of Modern Greek nouns, shown in the traditional way in (8.21). (To simplify the presentation, stress is ignored here, even though it is relevant to establishing inflection classes in Greek.) (8.21) sg nom acc gen pl nom acc gen
os-declension nomos nomo nomu nomi nomus nomon ‘law (masc.)’
as-declension pateras patera patera pateres pateres pateron ‘father (masc.)’
us-declension papus papu papu papuðes papuðes papuðon ‘grandfather (masc.)’
168 
C H A P T E R 8 â•… I N F L E C T I O N A L P A R A D I G M S
sg nom acc gen pl nom acc gen
a-declension imera imera imeras imeres imeres imeron ‘day (fem.)’
i1-declension texni texni texnis texnes texnes texnon ‘art, skill (fem.)’
i2-declension poli poli poleos polis polis poleon ‘town (fem.)’
u-declension maimu maimu maimus maimuðes maimuðes maimuðon ‘monkey (fem.)’
In the more abstract notation of our paradigm rules, these could be written as (8.22). (8.22) a. Paradigm rule for the os-declension { [/Xos/nom.sg], [/Xo/acc.sg], [/Xu/gen.sg], [/Xi/nom.pl], [/Xus/acc.pl], [/Xon/gen.pl] } b. Paradigm rule for the as-declension { [/Xas/nom.sg], [/Xa/acc.sg], [/Xa/gen.sg], [/Xes/nom.pl], [/Xes/acc.pl], [/Xon/gen.pl] }
and so on.6
None of the seven classes in (8.21) is completely identical to any other class, but the similarities among them are evident. Theoretically, given seven different declensions and six cells in the paradigm, we could have (6 × 7 =) 42 totally different suffixes. In reality we have almost the opposite: the declensions seem to differ only slightly from each other. One might even propose that some of them could be lumped together, especially the a-declension and the i1-declension. In order to express this generalization, we will introduce one additional descriptive device: the rule-schema, which generalizes over rules in much the same way as word-schemas generalize over words. Thus, given the paradigm rules for the a-declension and the i1-declension in (8.23), we can formulate the rule-schema in (8.24), which subsumes both rules. In addition to the stem variable X, this also contains the variable V for the vowel, which may be instantiated by a or i. (8.23) a. Paradigm rule for the a-declension { [/Xa/nom.sg], [/Xa/acc.sg], [/Xas/gen.sg], [/Xes/nom.pl], [/Xes/acc.pl], [/Xon/gen.pl] } b. Paradigm rule for the i1-declension { [/Xi/nom.sg], [/Xi/acc.sg], [/Xis/gen.sg], [/Xes/nom.pl], [/Xes/acc.pl], [/Xon/gen.pl] } 6
There is no point in rewriting all the paradigms of (8.19) in this format, because the tabular format is more perspicuous than the format with brackets and subscripts.
8 . 4 ╇ I N H E R I TA N C E H I E R A R C H I E S
169
(8.24) Rule schema for (8.23a–b) { [/XV/nom.sg], [/XV/acc.sg], [/XVs/gen.sg], [/Xes/nom.pl], [/Xes/acc.pl], [/Xon/gen.pl] } To make the notation more reader-friendly, let us introduce the formalism in Figure 8.1, where the slashes for the phonological representation and the inflectional values are omitted for the sake of simplicity. In this figure, the two declensions and the rule schema are shown in a tree format, the standard format for representing taxonomic hierarchies. In effect, the a-declension and the i1-declension are subtypes of the declension described by the rule-schema of (8.24), in much the same way as, say, a violin and a cello are subtypes of stringed instruments, and these are again a subtype of musical instrument. The taxonomic hierarchy of declension classes is completely parallel to hierarchies of this familiar kind, and is called an inheritance hierarchy. XV XV XVs
Xa Xa Xas
Xes Xes Xon
Xes Xes Xon
Xi Xi Xis
Xes Xes Xon
Figure 8.1╇ A hierarchy of declension classes
In this formalism, those pieces of information that are identical in the rule schema and in the individual rule need not be specified twice. They can be specified once in the rule-schema, and the individual paradigm rule can inherit the information from the superordinate node in the hierarchy (hence the name inheritance hierarchy). This is symbolized by the use of boldface and normal print in Figure 8.1: boldface information is necessary, and normal-print information is redundant and could in principle be inherited from the superordinate node. (If we wanted a completely redundancy-free representation of � grammatical information, normal-print material could simply be omitted. However, as we saw earlier in the discussion of word storage (Section 4.3), lack of redundancy does not seem to be a priority for human memory.) We can also extend this hierarchy to subsume the other Modern Greek inflection classes that we saw earlier. Figure 8.2 shows an attempt to draw a single inheritance hierarchy for the seven classes of (8.21) that has four different levels of abstractness.
170 
C H A P T E R 8 â•… I N F L E C T I O N A L P A R A D I G M S
XVZ XV XVZ XVs XV XV Xos Xo Xu
XV XV XVs
XVZ XVs Xon
Xi Xus Xon Xas Xa Xa
XVZ XVs Xon
XVs XV XV Xes Xes Xon
Xes Xes Xon Xus Xu Xu
XV XV XVs Xuðes Xuðes Xuðon
Xa Xa Xas
Xes Xes Xon
Xes Xes Xon
XVs XVs Xon Xi Xi Xeos
Xi Xi Xis
Xes Xes Xon
Xis Xis Xeon Xu Xu Xus
Xuðes Xuðes Xuðon
Figure 8.2╇ An inheritance hierarchy for seven Modern Greek declension classes
The top-level rule-schema is so abstract that it consists almost exclusively of variables (X for the stem, V for a vowel following the stem and Z for anything else, including zero, that follows that vowel). The only concrete elements that all classes share are the genitive plural suffix -on and the last consonant of the accusative plural suffix (-s). The major split is between the masculine classes (-os, -as, -us), on the one hand, and the feminine classes (-a, -i1, -i2, -u), on the other: all masculines are characterized by an -s in the nominative singular, and all feminines are characterized by an -s in the genitive singular. The inheritance network allows us a flexible and sophisticated answer to the question of how many different inflection classes should be set up for the Modern Greek data in (8.21). At the lowest level, there are seven classes, and we may call these microclasses. At an intermediate level, we might say that there are four classes (some of them with subclasses), and at a higher level, we could say that it has just two macroclasses, the masculine and feminine declension types. Now, the hierarchy in Figure 8.1 is just a single tree with no crossclassification, but in reality such cross-classifications are necessary, and examples are easy to find. This is again parallel to other domains of knowledge. One could cross-classify musical instruments into classical instruments (violin, cello, flute) and modern instruments (saxophone, electric guitar). One obvious generalization that is missed by Figure 8.2 but that is certainly not lost on speakers of Greek is the similarity between the us-class and the u-class, so for this, cross-classification is required.
8 . 4 ╇ I N H E R I TA N C E H I E R A R C H I E S
171
In the hierarchy of Figure 8.2 there is also never a conflict between a lower and a higher node; higher nodes are merely less specific. Now it has been suggested that such conflicts should be allowed, and that specifications in a lower node should be able to override specifications in a higher node. For example, the Greek os-declension and the a-declension could be subsumed under the same rule schema as shown in Figure 8.3. XVs XV XV
Xos Xo Xu
(!)Xi (!)Xus Xon
Xes Xes Xon
Xas Xa Xa
Xes Xes Xon
Figure 8.3╇ An inheritance hierarchy with a mismatch
Here there is a mismatch between the nominative and accusative plural forms /Xi/ and /Xus/ and the corresponding forms specified in the higher node (/Xes/). The exclamation mark in the notation shows that a higher specification is overridden. The forms /Xes/ in the higher rule schema are no longer fully schematic, but they are a default that applies unless it is overÂ� ridden. By using the device of default rules and overrides, the inheritance hierarchy can be simplified. Thus, in Figure 8.2 one of the rule schemas could be dispensed with if the description of Figure 8.3 were adopted. So in the end, the tree structure and inheritance are not by themselves always sufficient to capture similarities among inflection classes, but they do help us to see that inflection classes can be related to each other to greater or lesser degrees. By setting up hierarchies as in Figures 8.2 and 8.3, our formal description establishes relationships between entire inflection classes along a paradigmatic dimension. Moreover, we know that this dimension is salient for Greek speakers (and not just the creation of linguists) because of inflection class changes over time. For instance, the Modern Greek i2-declension used to have the ending -is in the nominative singular ({ [/Xis/nom.sg], [/Xi/ acc.sg], [/Xeos/gen.sg], …} ), e.g. polis/poli/poleos ‘town’. The change from /Xis/ to /Xi/ in the nominative singular was clearly a morphological, not a phonological change. The paradigm rule of the i2-declension clashed with the general schema for the other feminine microclasses in an important respect (the nominative singular in -is), and, by changing this, that schema was able to subsume the rule for the i2-declension as well. If the speakers had had only the rules for the individual declensions, this change would
172 
C H A P T E R 8 â•… I N F L E C T I O N A L P A R A D I G M S
be mysterious. Thus, diachronic change indicates that speakers establish paradigmatic relations across inflection classes, and that paradigmatic relations are therefore part of the architecture of the morphological system.
8.5╇ Stems and Priscianic formation In this section we show that not only inflectional endings, but also stems may be related along the paradigmatic dimension. In many languages, lexemes are associated with multiple inflectional stems, i.e. there is weak or strong stem suppletion. Consider the Persian verb forms in (8.25). (8.25) infinitive past tense present tense mund-œn 1sg mund-œm mi-mun-œm ‘to stay’ 2sg mund-i mi-mun-i 3sg mund mi-mun-e 1pl mund-im mi-mun-im 2pl mund-id mi-mun-id 3pl mund-œnd mi-mun-œnd (Mahootian 1997: 28, 236) All past-tense forms and the infinitive share a stem (mund-), and all presenttense forms share another (mun-). All Persian verbs behave like mundæn in this respect, as the verbs in (8.26) show. (8.26)
infinitive mund-œn xœrid-œn mord-œn šekaft-œn šetaft-œn did-œn
1st sg past mund-œm xœrid-œm mord-œm šekaft-œm šetaft-œm did-œm
1st sg present mi-mun-œm ‘stay’ mi-xœr-œm ‘buy’ mi-mir-œm ‘die’ mi-šekaf-œm ‘split’ mi-šetab-œm ‘hurry’ mi-bin-œm ‘see’ (Mahootian 1997: 231–7)
The relation between the past-tense stem and the present-tense stem is unpredictable for many verbs, but the past stem and the infinitival stem are always identical. Moreover, because the relationship between the present and the past/infinitive sometimes involves suppletion, it is natural to say that lexemes are associated in the lexicon with two stems that are restricted to occurring with particular inflectional values. Now, in the context of this chapter’s discussion, one of the more interesting facts about inflectional stems is that they can sometimes be built on other stems in the same paradigm. A well-known case in Latin is the past passive participle and the future active participle. Some representative forms are given in (8.27).
8 . 5 ╇ S T E M S A N D P R I S C I A N I C F O R M AT I O N
(8.27)
infinitive lauda¯re mone¯re du¯cere vehere mittere haere¯re premere ferre
past pass. part. lauda¯tus monitus ductus vectus missus haesus pressus la¯tus
173
future act. part. lauda¯tu¯rus ‘praise’ monitu¯rus ‘warn’ ductu¯rus ‘lead’ vectu¯rus ‘carry’ missu¯rus ‘send’ haesu¯rus ‘stick’ pressu¯rus ‘press’ la¯tu¯rus ‘bear’ (Aronoff 1994: ch. 2)
We could say here that each lexeme is associated with a set of three stems (e.g. lauda¯-, lauda¯t-, lauda¯tu¯r-). But two facts are noteworthy. First, the futureactive participle stem is the same as the past-passive participle stem with -u¯r added. Second, the relationship between the infinitive and the pastpassive participle is sometimes suppletive, but the relationship between the past-passive participle and the future-active participle is always regular. An alternative would thus be to describe the future-active participle in terms of Priscianic formation (so called because it was used by the Latin grammarian Priscian, in the sixth century ce), whereby a member of an inflectional paradigm is formed from another member of the paradigm to which it need not be closely related semantically. A Priscianic analysis of Latin would say that the form of the future-active participle is dependent upon the form of the past-passive participle. This can be represented with a word-based rule as in (8.28). (8.28) [/XY/pst.pass.part] ´ [/Xu ¯ rY/fut.act.part] This rule nicely captures the regular relationship between the past-passive participle and the future-active participle. (Note that the meaning is quite independent of the form; obviously the future-active participle cannot be based semantically on the past-passive participle.) A description in terms of Priscianic formation is equivalent to a description in terms of stem sets for most purposes, and most linguists have continued to describe examples like (8.27) in terms of stem sets. (It is, after all, the approach that is consistent with the more dominant syntagmatic perspective.) But even for these linguists, such an analysis is perhaps less attractive in cases like the following from Tümpisa Shoshoni. This language has two non-nominative case forms, an objective case and a possessive case. The formation of these cases is illustrated in (8.29). (8.29)
nominative mupin tümpi nümü
objective mupitta tümpitta nümi
possessive mupittan tümpittan nümin
‘nose’ ‘rock’ ‘person’
174 
C H A P T E R 8 â•… I N F L E C T I O N A L P A R A D I G M S
piammütsi kahni
piammütsia piammütsian kahni kahnin
‘baby’ ‘house’ (Dayley 1989: 185–6)
The objective case is related to the nominative in various ways (one of them being identity to the nominative), but the possessive is systematically formed from the objective by adding a further suffix -n. If we wanted to describe this pattern without Priscianic formation, we would have to set up a non-nominative stem that yields the possessive form by addition of -n and the objective form by addition of nothing. Of course, this kind of purely syntagmatic description is perfectly possible, but nothing seems to be gained when compared to the Priscianic solution.
8.6╇ Syncretism Not uncommonly, two word-forms in an inflectional paradigm are phonologically identical, or in other words, homonymous. For example, in the present-tense paradigm of German verbs, the third person singular and the second person plural, and the first and third person plural have the same form: (8.30)
1sg 2sg 3sg 1pl 2pl 3pl
(ich) (du) (er/sie) (wir) (ihr) (sie)
spiele spielst spielt spielen spielt spielen
‘I play’ ‘you(sg) play’ ‘he/she plays’ ‘we play’ ‘you(pl) play’ ‘they play’
When the inflectional homonymy is systematic, we speak of syncretism, and homonymous forms of a paradigm are called syncretic. Syncretism is thus a kind of ‘mismatch’ between form and inflectional function – one form for two or more (sets of) inflectional values. We will see that syncretism is perhaps the strongest piece of evidence for paradigmatic relations as part of morphological architecture. However, since the issues are somewhat complicated, we discuss them at some length in this section.
8.6.1╇ Systematic versus accidental inflectional homonymy We must first be able to distinguish between systematic and accidental homonymy. We will discuss four tests: systematicity across inflection classes, syntactic functionality, patterns of language change and whether the affected paradigm cells form a natural class (see Zwicky 1991). The first test considers the extent to which a pattern of homonymy is found in different inflection classes. The two pairs of homonymous forms
8 . 6 ╇ S Y N C R E T I S M
175
in (8.30) behave differently by this criterion. German has a small class of vowel-changing verbs that have a different stem vowel in the second and third person singular, e.g. gebe/gibst/gibt ‘give’, falle/fällst/fällt ‘fall’. In these verbs, the third person singular and the second person plural are not identical, because the vowel alternation is restricted to the third person singular (3sg gibt versus 2pl gebt, 3sg fällt versus 2pl fallt), but the first person and third person plural are still identical. In fact, the first person and the third person plural are identical in all German verb paradigms, including the suppletive paradigm of sein ‘be’ (singular: bin/bist/ist, plural: sind/seid/sind). So in this respect, the 1pl/3pl homonymy (spielen) is more systematic than the 3sg/2pl homonymy (spielt). The syntactic test concerns an interesting syntactic property of syncretic forms: they can be used in situations where two conflicting syntactic requirements must be fulfilled simultaneously. One such construction is shown in (8.31a), where the verb spielt has to agree simultaneously with both coordinands of the disjunction. Now there are situations where the two requirements are in conflict, as in (8.31b), where the verb is supposed to agree both with ich (first person singular) and with du (second person singular). Since there is no verb form that can do this, the sentence is ungrammatical. (8.31) a. Entweder Ballack oder Klose spielt gegen Bulgarien. ‘Either Ballack or Klose will play in the Bulgaria match.’ b. *Entweder ich oder du spiele/spielst gegen Bulgarien. ‘Either I or you(sg) will play in the Bulgaria match.’ c. Entweder wir oder sie spielen gegen Bulgarien. ‘Either we or they will play in the Bulgaria match.’ d. *Entweder Bierhoff oder ihr spielt gegen Bulgarien. ‘Either Bierhoff or you(pl) will play in the Bulgaria match.’ However, when the two requirements are first or third person plural, as in (8.31c), there is a way to resolve the feature-value conflict: the syncretic form Â�spielen can serve simultaneously as a first person plural and as a third Â� person plural form. In this, it contrasts with the two homonymous forms spielt ‘3rd sg’ and spielt ‘2nd pl’: as we see in (8.31d), the form spielt cannot resolve the feature-value conflict, and hence we say that, in the case of spielen, we have systematic homonymy (i.e. syncretism), whereas, in the case of spielt, we are dealing with accidental homonymy. The syntactic criterion shows that speakers treat the two syncretic forms as related. In the case of German verbs, this test gives the same results as does the first test: The identity of spielt (3sg) and spielt (2pl) is accidental, but the identity of spielen (1pl) and spielen (3pl) is systematic – i.e. in the latter case we are dealing with syncretism. The ability to resolve a feature-value conflict can be taken as a sufficient criterion for systematic homonymy, but it cannot be a necessary one
176 
C H A P T E R 8 â•… I N F L E C T I O N A L P A R A D I G M S
because sometimes there are no relevant syntactic constructions that would impose conflicting requirements. For instance, if we want to know whether the frequent homonymy of the English past tense and the past participle (as in forms like played, fed, thought) is systematic, we cannot apply the syntactic test, because there are no constructions in which a verb should simultaneously be a past tense and a past participle. It is true that, for the vast majority of verbs, these forms are homonymous, but in Old English they were distinct for all verbs, and the present-day homonymy could be explained in almost all cases by regular phonological changes. Thus, the homonymy might still be accidental for English speakers. However, here the diachronic criterion can be invoked: there are a few verbs whose past-participle form became identical with the past-tense form through morphological (analogical), not phonological change: stand/stood/stood (cf. Old English standan/sto¯d/gestanden), sit/sat/sat (cf. Old English sittan/sæt/ geseten). The morphological change is a strong indication that, at least at the time of the change, the homonymy of the two forms was perceived as systematic by the speakers. Finally, when all else fails, we can reasonably guess that forms are systematically homonymous when they form a natural class, i.e. when they can be described by a single (set of) inflectional value(s), and they are the only word-forms that express that set. Consider the Lithuanian verb paradigm in (8.32) (present tense, indicative mood of sup- ‘shake, swing’). (8.32) singular
plural
1st
supu
supame
2nd
supi
supate
3rd
supa
supa
Here the two syncretic cells are the third person singular and the third person plural and they form a natural class, which is to say they are the only word-forms expressing third person. Such syncretisms may be called natural � syncretisms, and patterns of this sort are likely to not be accidental.
8.6.2╇ Underspecification Once we have established that a case of identity is systematic and not accidental, we can return to the question of whether syncretism requires a paradigmatic perspective in order to be adequately described. This question can be rephrased in the following way: Are we really dealing with two different cells in the paradigm, and a rule stating that they must have identical form (a paradigmatic approach), or is there perhaps just a single form that simply does not distinguish the relevant values (a nonparadigmatic approach)?
8 . 6 ╇ S Y N C R E T I S M
177
Certainly, examples of natural syncretisms do not require the paradigmatic approach, and they are quite easy to describe. We can simply say that the Lithuanian third person form of the verb is supa – i.e. we do not have to mention the number feature at all. We can visualize this description by a representation in which the syncretic form occupies an enlarged cell, as in (8.33). (8.33) singular
plural
1st
supu
supame
2nd
supi
3rd
supate supa
In the more formal representation format of (8.20), we would say that the paradigm of sup- is described by the paradigm rule in (8.34), in which nothing is said about the number feature for the form supa. (8.34) { [/Xu/1.sg], [/Xi/2.sg], [/Xa/3], [/Xame/1.pl], [/Xate/2pl] } Such a mode of description is called underspecification: we simply do not specify the value of certain features in the paradigm rule. For the syntactic rule of agreement that interacts with these inflectional values, this means that it should not require feature-value identity, but only feature-value compatibility. Both a singular and a plural subject NP are compatible with a form like supa, so the agreement relation works, even though supa is not specified for number. Sometimes an underspecification analysis is possible even when the syncretic cells do not constitute a natural class. Consider another example, Standard Arabic case inflection: (8.35) singular
plural
¯haywaan-un
¯haywaan-aat-un
gen
¯haywaan-in
¯haywaan-aat-in
acc
¯haywaan-an
¯haywaan-aat-in
‘animal’
‘animals’
nom
In the plural, the genitive and the accusative have the same form, and the usual analysis is that we are dealing with syncretism here (this genitive– accusative homonymy is found in all non-singular forms, so it is unlikely to be accidental). However, this is not an example of natural syncretism: only two of the three plural cells are syncretic. One possibility is to say that Arabic has a different case system in the plural that distinguishes only a nominative and an oblique case. In other words, one might propose (8.36):
178 
C H A P T E R 8 â•… I N F L E C T I O N A L P A R A D I G M S
(8.36) singular nom
¯haywaan-un
gen
¯haywaan-in
acc
¯haywaan-an
plural nom
¯haywaan-aat-un
obl
¯haywaan-aat-in
But most linguists would not adopt this description, because it would make the rules of syntax more complicated. Instead of saying that a direct object is in the accusative case, we would have to say that it is in the accusative case in the singular and in the oblique case in the plural. Or possibly, our model could specify that oblique is somehow compatible with accusative and genitive. But a simple alternative is to posit that the nominative plural formally bears the feature values nom.pl, and the syncretic forms are underspecified for case, bearing only the value pl. This causes an apparent problem: both wordforms are then compatible with a syntactic rule that requires nominative plural, but only one of the words can appear in this context. However, we can easily solve this problem by assuming that the morphology provides to the syntax that word whose feature values are compatible in the most specific way. If the syntactic context requires a nominative plural, ¯h aywaanaatun meets these conditions in a more specific way than ¯h aywaanaatin because the former specifies both case and number values. This principle is a version of the elsewhere condition – more specific conditions apply before more general ones. The elsewhere condition is relevant to many areas of grammatical structure. Underspecification is thus a powerful tool for describing syncretism, but it is not restricted to the word-based model. The examples in this section can just as easily be described in the morpheme-based model; the Lithuanian example is given in (8.37).
{
}
 ╇╇ 
(8.37) a. /u/ b. /a/ V ___ V ___ person:1st {person:3rd} number:sg As in the word-based rule in (8.34), the morpheme lexical entry in (8.37b) is underspecified for number. This highlights that underspecification does not require a paradigmatic perspective. And inasmuch as all examples of syncretism can be described as feature underspecification, both the word-based model and the morpheme-based model offer equally viable descriptions. But of course, the key word in the preceding sentence is ‘inasmuch’. In the following section we show that many instances of syncretism cannot be described by underspecification. Such examples require a paradigmatic approach.
8 . 6 ╇ S Y N C R E T I S M
179
8.6.3╇ Rules of referral Consider the three Old Church Slavonic nominal inflection classes in (8.38). Only the case endings are given here in order to save space. (8.38) nom acc gen loc dat inst
u˘-class sg du pl -u˘ -a -i -u˘ -a -y -a -u -u˘ -u -eˇxu˘ -eˇ -u -oma -omu˘ -omı˘ -oma -y
a-class sg du pl -a -eˇ -y -o -eˇ -y -y -u -u˘ -eˇ -u -axu˘ -eˇ -ama -amu˘ -ojo. -ama -ami
˘ı -class sg du -ı˘ -i -ı˘ -i -i -ı˘ju -i -ı˘ju -i -ı˘ma -ı˘ -ı˘ma
pl -i -i -ı˘jı˘ -ı˘xu˘ -ı˘mu˘ -ı˘mı˘
Especially in the dual, we have a lot of syncretism: The nominative and accusative, the genitive and locative, and the dative and instrumental are systematically homonymous (this is true also of the other inflection classes not shown here). These syncretisms are clearly not natural syncretisms, because these three pairs of cases do not have any exclusive properties. Moreover, these examples cannot (in any obvious way) be described using the device of feature underspecification. If all three pairs of syncretic dual forms were underspecified for case, all six dual forms would have the same feature specification (i.e. they would be marked only for being dual). The elsewhere condition would not function because no one form would be associated with more specific feature values. For such cases, we need a special type of rule that says that several forms in the paradigm are identical. Such rules are called rules of referral. We can formulate the rule for the nominative–accusative dual as in (8.39).
[
]
[
(8.39) /X/n ´ /X/n ‘nom.du’ ‘acc.du’
]
This rule generalizes over all the paradigms of Old Church Slavonic. A rule of referral can thus be thought of as a kind of paradigm rule schema that relates two cells in the paradigm to each other. And as should be obvious by this point, this kind of rule encodes paradigmatic relations. That such rules of referral are real for speakers and not just thought up by linguists becomes clear when they trigger morphological changes. An example comes from Old High German (Wurzel 1987: 70–1). The paradigm of neuter nouns of the a-declension that must have existed in pre-Old High German is shown in (8.40). (8.40) nom acc gen dat
singular plural wort wort wort wort wortes worto worte wortum ‘word’
singular faz faz fazzes fazze ‘barrel’
plural fazzu fazzu fazzo fazzum
180 
C H A P T E R 8 â•… I N F L E C T I O N A L P A R A D I G M S
The original suffix of the nominative/accusative was -u, as is clear from comparative evidence. This suffix was lost by regular sound change in heavy-syllable words like wort, but it was preserved in light-syllable words like faz. Now apparently speakers formulated a rule of referral stating that the singular and the plural forms of the nominative and accusative are identical. This rule was originally based only on nouns of the wort class, but since the faz class had a much lower type frequency, it also came to be affected by this rule, and by the time of Old High German the paradigm of (8.40) had changed: the forms fazzu have been replaced by faz, in accordance with the rule of referral.
8.7╇ More form-meaning mismatches Nobody is perfect, not even inflectional paradigms. In the previous section we encountered one way in which cells in an inflectional paradigm may have a mismatch between form and meaning; they may be identical to other cells in the paradigm. In this section, we look at two more ways in which inflection fails to correspond to the principle of one-form-one-meaning, and how description of these examples benefits from a paradigmatic approach.
8.7.1╇ Defectiveness First of all, lexemes may simply lack word-forms. Lexemes with missing word-forms are called defective lexemes.7 An example of a defective lexeme is the Italian verb incombere ‘be incumbent’, which lacks a past participle and therefore cannot be used in the compound past tense. In French, the verbs frire ‘fry’, déchoir ‘fall’ and clore ‘close’ lack an imperfective past tense. In English, many speakers feel that the verb forego sounds strange in the past tense (??He foregoed/forewent treatment for cancer). In Russian, a number of verbs do not have a first person singular in the present/future tense (e.g. pobedit’ ‘win, defeat’, *pobežu), and a few nouns like mecˇta ‘dream’ lack a genitive plural form (*mecˇt). Defectiveness is surprising and interesting for a few reasons. First, it disturbs the functionality of the language. Sometimes one wants to say ‘I will win’ in Russian, but the system does not allow it. Of course, speakers are not condemned to silence in such cases – there is always a way around the defective form. A Russian speaker can resort to the expression oderžu pobedu [gain victoryN] ‘I will gain a victory’, and an English speaker can 7
Note that the term defectiveness is usually applied only to individual lexemes, not to entire inflectional values. For instance, English systematically lacks one-word forms to express the past-passive participle, but this is not normally called defectiveness.
8 . 7 ╇ M O R E F O R M - M E A N I N G M I S M AT C H E S
181
avoid foregoed by choosing a semantically similar verb or phrase such as did without or sacrificed. Second, it is surprising that speakers can learn the negative fact that a lexeme lacks certain forms. Normally there is at least one productive pattern for each inflectional value, a default pattern that is used when no other pattern is remembered. But in defective lexemes, this default pattern is not used to ‘fill in’ missing forms. This suggests that defectiveness is not simply a situation in which speakers fail to learn the missing word-form. Rather, speakers learn that no word-form is used, not even the default. Moreover, in contrast to other irregular lexemes, which usually show a high frequency of use (see Section 12.3), many defective lexemes are rather rare. Some more examples of Russian verbs that are defective in the first person singular present/future are shown in (8.41). (In Russian, the perfective future and the imperfective present are formed in the same manner. The verbs are therefore defective in either the present tense or the future, depending on aspectual value.) The summed token frequency of all present/future tense forms is also listed, based on the modern (post-1950) subpart of the Russian National Corpus, a large corpus consisting mostly of written texts of various types (newspapers, magazines, fiction, religion and philosophy, law, technical and scientific works, letters, diaries, etc.).8 (8.41) lexeme expected 1sg present/future frequency (ipm)9 pobedit´ *pobežu 30 ubedit´ *ubežu 23.2 ošcˇutit´ *ošcˇušcˇu 7.3 cˇudit´ *cˇužu 0.8 derzit´ *deržu 0.6 umiloserdit´ *umiloseržu 0
gloss ‘be victorious’ ‘persuade’ ‘feel’ ‘behave oddly’ ‘be imprudent’ ‘to take pity on’
We might posit that speakers learn that pobedit´ is defective simply by having many opportunities to observe that all of its word-forms are used – except the first person singular. But this does not seem to help explain how lexemes like derzit´ and umiloserdit´ are learned to be defective, and how defectiveness in these words can be stable for multiple generations. (A probable answer is that a kind of analogy is at work here; speakers learn that derzit´ is defective because other, similar verbs are also defective (Daland et al. 2007; Baerman 2008).) Finally, and relevant to the central discussion of this chapter, the most surprising thing about defectiveness is probably that it, too, sometimes exhibits paradigmatic dependencies. In Sections 8.2 and 8.3 we argued that the ability to predict one word-form from another is an important property
8
9
Russian National Corpus: http://ruscorpora.ru/en/. Frequency list based on modern subcorpus: http://corpus.leeds.ac.uk/serge/frqlist/ (access: July 2010). ipm = instances per million words of corpus
182 
C H A P T E R 8 â•… I N F L E C T I O N A L P A R A D I G M S
of inflection classes. Paradigm rules help explain, for example, why lexemes sometimes shift from one inflection class to another. Here we note that paradigm rules sometimes play a role in defectiveness as well – forms that cannot be accurately predicted may become defective. For instance, a handful of Spanish verbs are defective in the first person singular, e.g. abolir ‘abolish’ (1sg: *abolo, *abuelo), and sometimes other cells in the paradigm. Crucially, the defective cells are exactly those which cannot be confidently predicted based on other word-forms of the same lexeme, and are not frequent enough to have been memorized by speakers (Albright 2003).
8.7.2╇ Deponency Another phenomenon of paradigm mismatch is deponency, whereby a formal marker of an inflectional value is used in the ‘wrong’ function, i.e. to express a different value. Consider the Modern Greek active and passive forms of pléno ‘wash’ in (8.42a). This represents the inflectional pattern that most verbs follow. However, a handful of verbs like ðéxome ‘receive, accept’, erázome ‘work’, and érxome ‘come’ (8.42b), are semantically active, but nonetheless exhibit the inflection pattern that normally expresses the passive. (8.42) a. 1sg 2sg 3sg 1pl 2pl 3pl
active pléno plénis pléni plénume plénete plénun
passive plénome plénese plénete plenómaste plenósaste plénonde
b.
active érxome érxese érxete erxómaste erxósaste érxonde
Verbs like érxome, which have a paradigm from a different value but not the meaning of that value, exhibit deponency. In a purely syntagmatic approach, deponency is difficult to describe adequately. The morpheme-based model must treat -ome as meaning ‘passive first person singular’, -ese as meaning ‘passive second person singular’, and so on, because morphemes bear meaning in this approach. However, this principle runs into obvious problems with the deponent verbs. From a paradigmatic perspective this pattern is understood easily enough if we make one small assumption: paradigmatic relations akin to rules of referral can operate not only within paradigms, but also across inflection classes. The active in the deponent class can be specified as systematically taking the same inflectional endings as the passive in other inflection classes. (Proper rules of referral involve complete phonological identity, not only inflectional identity, but the same principle is at work here.) A paradigmatic approach thus offers a more intuitive description of deponency than is possible in a purely syntagmatic approach.
8 . 8 ╇ P E R I P H R A S I S
183
8.8╇ Periphrasis Not uncommonly, missing cells are not completely empty, but may be filled by multi-word phrases that express the needed concepts in a conventional way. For example, many English adjectives lack ordinary comparative forms in -er. We have warm-er, nic-er, pretti-er, but we do not have *beautifuller, *interestinger, *activer. However, morphologists do not say that the lexemes beautiful, interesting and active are defective in lacking a comparative form, because there is a well-established convention for expressing the value: more beautiful, more interesting, more active. Such comparatives are called periphrastic, and the phenomenon is periphrasis. Another example comes from Romanian, where nouns inflect for an oblique case (e.g. prietenul ‘the friend (nom)’, prietenului ‘the friend (obl)’, Ana ‘Ana (nom)’, Anei ‘Ana (obl)’). However, masculine personal names such as Petre lack an ordinary oblique case. In order to use them in a syntactic slot that requires the oblique case, a periphrasis involving the pronoun lui ‘him’ is used (lui Petre ‘Petre (obl)’). These examples represent cases of lexical periphrasis, where certain (groups of) lexemes lack word-forms for certain values. But we also find paradigmatic periphrasis – i.e. cases in which entire word-classes lack word-forms for certain combinations of inflectional values. A wellknown example of this comes from Latin verbs. The passive is sometimes expressed morphologically for verbs (e.g. when combined with present or imperfect meanings). Likewise, the perfect and the pluperfect are expressed morphologically when combined with the active. However, the combination of perfect/pluperfect and passive is periphrastic for all verbs. In (8.43) we see the third person singular forms of some tense–aspect–voice combinations of the verb scribere ‘write’. The perfect and pluperfect passive forms are expressed by the past passive participle plus the verb esse ‘be’. (8.43) present
imperfect
perfect
pluperfect
active
scribit
scribebat
scripsit
scripserat
passive
scribitur
scribebatur
scriptum est
scriptum erat
We should be careful to distinguish paradigmatic periphrasis from another kind of periphrasis that we may call categorial periphrasis. In categorial periphrasis, a given grammatical function is always expressed with a multi-word expression. For example, French is sometimes said to have a periphrastic future involving the auxiliary verb aller ‘go’, e.g. je vais faire ‘I’m going to do’, tu vas faire ‘you’re going to do’, il va faire ‘he’s going to do’, and so on. The crucial difference, as compared with (8.43), is that the French future is never expressed with a single word, whereas the Latin
184 
C H A P T E R 8 â•… I N F L E C T I O N A L P A R A D I G M S
passive sometimes is. In contrast to paradigmatic periphrasis, then, such cases of categorial periphrasis have nothing to do with morphology, and the morphologist can ignore them. Periphrasis does not appear to exhibit paradigmatic dependencies in the same way that deponency and syncretism do. But it is significant that paradigmatic (i.e. morphologically-relevant) periphrasis can only be identified in the context of other paradigmatic forms – the relevant inflectional values must somewhere in the paradigm be expressed with a single word. Moreover, some linguists have argued that paradigmatic periphrasis tends to exhibit morphological properties (e.g. noncompositional meaning), and periphrastic examples as in (8.43) therefore should be formally considered as part of a lexeme’s paradigm, but there is no consensus on this point.
Summary of Chapter 8 Inflection classes have a number of properties. They are delineated by sets of suppletive inflectional allomorphs, which are typically linked to properties such as the phonological shape of the base, the lexeme’s meaning (e.g. animacy, transitivity) and/or Â�morphological properties such as the derivational pattern. Although inflection class and gender are clearly related, their relationship is complex. Inflection classes may differ in productivity, seen in their ability to apply to novel lexemes (loanwords or productively formed neologisms) and in their ability to attract class-shifting lexemes. The balance between syntagmatic and paradigmatic description of (inflectional) structure is a major issue for the description of morphological architecture. This chapter focused on evidence that a syntagmatic perspective is insufficient by itself; formal devices for a paradigmatic description of inflection are also needed. The primary evidence comes from inflection class-shift, similarities across inflection classes (described in terms of inheritance hierarchies), Priscianic formation (one stem being built on another stem in the paradigm) and form-meaning mismatches: syncretism (one form for two sets of inflectional functions), defectiveness (lack of a form for a given function), deponency (a form with an unexpected function), and to a lesser extent, periphrasis (a function is expressed by a multiword phrase).
Further reading Book-length studies on inflection include Matthews (1972), Carstairs (1987), Wurzel (1989) and Stump (2001a). A typologically oriented overview article
COMPREHENSION EXERCISES 
185
is Bickel and Nichols (2007). For inflection classes and stems, see Aronoff (1994) and Blevins (2003), and for inheritance hierarchies, see Corbett and Fraser (1993) and Stump (2001b). For productivity, see Dressler (1997) and Bauer (2001b). Various mismatches between form and inflectional function have been studied individually. For syncretism (including discussion of rules of referral, underspecification and related approaches), see Zwicky (1985a), several of the papers in Plank (1991), Noyer (1998) and Baerman et al. (2005). The distinction between systematic and accidental homonymy is set out in Zwicky (1991). For periphrasis, particularly the debate about whether periphrastic constructions are morphological or syntactic, see Börjars et al. (1997), Embick (2000), Ackerman and Stump (2004) and Kiparsky (2005). An overview of periphrasis is Haspelmath (2000). For deponency, see the papers in Baerman et al. (2007). For defectiveness and its relationship to paradigmatic dependencies, see Hansson (1999), Albright (2003), Sims (2006), Daland et al. (2007) and the papers in Baerman et al. (2010). For the relationship between gender and inflection class, see Corbett (1982, 1991), Aronoff (1994) and Spencer (2002).
Comprehension exercises 1. Using the rules given in connection with (8.11), form the perfective form of the following Tagalog verbs: root basic form with voice affix langoy lumangoy ‘swim’ wagayway wumagayway ‘wave’ takot matakot ‘be afraid’ uhaw mauhaw ‘be thirsty’ buhat buhatin ‘raise’ punit punitin ‘rip’ punas punasan ‘wipe’ 2. Near the end of Section 8.4, we said that ‘in Figure 8.2, one of the rule schemas could be dispensed with if the description of Figure 8.3 were adopted’. Which rule schema could be dispensed with? What would the modified version of Figure 8.2 look like? 3. Take a complete list of English ‘irregular verbs’ based on past tense formation and try to group them into small inflection classes. Which classes can be established? Which verbs must be said to be truly irregular – i.e. cannot be put into a class with some other verb(s)? 4. Consider the following three inflection classes of Ancient Greek (only singular forms are given). Class (i) consists of feminines (like the Latin
186 
C H A P T E R 8 â•… I N F L E C T I O N A L P A R A D I G M S
class of insula ‘island’), class (ii) consists of masculines denoting men (like the Latin class of poeta ‘poet’) and class (iii) mostly consists of Â� masculines. The nouns of class (ii) originally inflected just like class (i). What may have motivated the change? (i) (ii) (iii) nom he¯méra¯ neanía¯s phílos acc he¯méra¯n neanía¯n phílon gen he¯méra¯s neaníou phílou dat he¯méra¯i neanía¯i phíloˉi ‘day’ ‘young man’ ‘friend’ 5. Consider the following four inflection classes of Russian nouns, and try to set up an inheritance hierarchy corresponding to Figure 8.2. (see Corbett and Fraser 1993). (Note that and stand for the same phoneme. Also, note that the spelling obscures the stem shape in (iii), and the following all stand for the phoneme /tj/: , , and plain when followed by <e> or .) (i) (ii) (iii) (iv) ‘law’ ‘room’ ‘bone’ ‘swamp’ nom.sg zakon komnata kost´ boloto acc.sg zakon komnatu kost´ boloto gen.sg zakona komnaty kosti bolota dat.sg zakonu komnate kosti bolotu inst.sg zakonom komnatoj kost´ju bolotom loc.sg zakone komnate kosti bolote nom.pl zakony komnaty kosti bolota acc.pl zakony komnaty kosti bolota gen.pl zakonov komnat kostej bolot dat.pl zakonam komnatam kostjam bolotam inst.pl zakonami komnatami kostjami bolotami loc.pl zakonax komnatax kostjax bolotax 6. English has few cases in which syncretism could be observed. However, consider the present-tense and past-tense paradigms of be: I am was you(sg) are were he/she is was we are were you(pl) are were they are were Apply the criteria of Section 8.6.1 to determine whether second person singular are and were are systematically syncretic with plural are and were.
E X P L O R AT O R Y E X E R C I S E
187
Exploratory exercise In Section 8.6 we saw a few examples of syncretism, a kind of mismatch between form and inflectional function. A question that we did not ask, however, is whether syncretism tends to affect one part of a lexeme’s paradigm more than another. In this exercise, you will explore crosslinguistic patterns of syncretism using the Surrey Person Syncretism Database (Baerman 2002). This database contains examples of person (-number-gender) syncretism in verbal inflection classes. The data come from a geographically and genetically diverse sample of 111 languages. You will use this data to develop an answer to the question above, and consider possible explanations for any observed patterns. Instructions Step 1: Develop a hypothesis and predictions. For instance, in a language in which verbs agree with controller nouns for three person values (first, second, third), and two number values (singular, plural), would you expect to find that the third person singular and third person plural are more commonly syncretic than are the third person singular and first person singular? Or the opposite pattern? Or no difference? What about the third person singular and the first person plural? Do you expect to find more examples of ‘natural’ syncretism, and if so, in which values of person/ number/gender/etc. features? In developing your predictions, you might find it helpful to inspect the examples of syncretism presented in this chapter. (Having three person values and two number values is common among verbs that agree with pronouns and nouns for person and number. But of course, verbs may express more or fewer than two number values, may agree also for gender, the first person plural and dual may be subdivided into inclusive (‘we, including the addressee’) and exclusive (‘we, not including the addressee’) forms, etc. Be sure to include other inflectional values in your predictions.) Based on what you have read in this chapter and elsewhere in the book, explain the rationale underlying your predictions. If you expect to find asymmetries in attested patterns of syncretism (more in one value than another), why do you think that such an asymmetry might exist? If you do not expect to find any asymmetrical patterns, why not? Step 2: Familiarize yourself with the data set. The Surrey Person Syncretism Database is available here (as of July 2010): www.smg.surrey. ac.uk/personsyncretism/index.aspx. Explore its content, structure, and the theoretical assumptions it makes about syncretism. Begin by reading the document ‘How to Use the Database’. Also, notice in particular that in results returned by a query, clicking the ‘Example’ button will show the entire relevant paradigm.
188 
C H A P T E R 8 â•… I N F L E C T I O N A L P A R A D I G M S
Step 3: Develop an appropriate classification/counting method. For instance, if you expect to find more syncretism among singular forms than among plural ones, or vice versa, it might seem obvious that you want to count the number of examples in the database with syncretism in the plural, and the number with syncretism in the singular. However, a single language may have more than one example of syncretism. Should each example be counted, or each language? (A question to consider here is whether multiple examples of syncretism in the same language are likely to be independent of each other.) Also, what if a language has syncretism between all plural cells, but also one singular cell; how should this kind of example be counted? There are no absolute ‘right’ answers to these questions. The important thing is that you develop precise criteria, justify them to the extent possible, and most of all, be consistent in applying them. Step 4: Collect data. Using the search interface for the Surrey Person Syncretism Database, collect data that can be used to support or reject your hypothesis. Step 5: Draw conclusions. Was your hypothesis supported? And more generally, what do the data suggest about syncretism? Are all cells in a paradigm equally likely to be syncretic, or are there asymmetries? Speculate about the reasons for any observed patterns. And as always, be sure to consider the (potential) influence of your research methodology on your results, especially the impact of the criteria that you created in Step 3. Step 6 (optional): This exercise previews discussion in Section 12.1.3. Do your results support the claims made there, or contradict them? Explain your reasoning.
Words and phrases
S
189
9
o far in this book we have pretended that it is easy to distinguish words from phrases. This has been possible because in the modern European writing system, the boundary between words is often indicated by a blank space. The segmentation of a sentence into word tokens thus seemed to be a straightforward matter – a word is surrounded by spaces. However, when we look closely, we find this procedure does not work very well. Not all writing systems indicate word boundaries. In Chinese, for instance, there are never blank spaces between characters. And even in languages that use the modern European writing system, the conventional spelling is occasionally ambiguous. Sometimes the spelling vacillates, as in English compounds (e.g. flower pot, flower-pot, flowerpot). Sometimes boundary symbols other than a blank space are used – for example, the apostrophe (as with the English genitive ’s, e.g. Joan’s book) or the hyphen (as with object pronouns in the French imperative, e.g. donne-le-moi ‘give it to me’). Sometimes the same element is spelled differently under different circumstances. In Spanish, weak object pronouns are spelled separately when they precede the verb (e.g. lo hacemos ‘we do it’), but together with the verb when they follow it (hacerlo ‘to do it’). Also, in German the infinitive marker is spelled separately in most cases (e.g. zu bringen ‘to bring’), but together with the verb when it is preceded by a prefix such as ein- ‘in’ (e.g. einzubringen ‘to bring in’). So in short, the hints from the spelling can be contradictory and misleading, and we cannot rely on the writing system of a language when trying to determine whether an expression is a word, a phrase, or an affix. The rules for orthographic word division are to some extent simply traditional in languages with a long written Â� history. And when a language is first written down, language users often disagree on where to put blank spaces between words, and when a conventional spelling is agreed on, the decisions are sometimes clearly arbitrary. To distinguish between words, phrases and affixes, we must develop other criteria.
190 
C H A P T E R 9 â•… W O R D S A N D P H R A S E S
In this chapter we begin by showing that words and phrases exhibit different properties, and that these can be used to identify word boundaries. In Section 9.1 we address a common area of difficulty – distinguishing compounds from phrases. We then go on to discuss a more complicated issue in Sections 9.2 and 9.3, namely, expressions that fall along the continuum between canonical affixes and canonical words. These are clitics. Lastly, we consider whether a distinction between words and phrases is important for a formal description of language structure. Fundamentally, this is a question of whether syntactic principles apply to word structure. The relationship between morphology and syntax arose already in Chapters 5 and 7. In Section 9.4 we look again at this issue in the context of something called the Lexical Integrity Hypothesis.
9.1╇ Compounds versus phrases A common situation in which we might ask the question whether an expression is a single word or a syntactic phrase involves (potential) compounds. For instance, are the expressions backboard, backdoor, back seat compounds or phrases? In this section we discuss some properties of compounds that allow us to distinguish them from phrases. In many cases, compounds are easy to tell apart from phrases with two content words. For instance, compounds may consist of two (or more) lexeme stems that are juxtaposed in a single word-form, and, when a language does not allow phrases consisting of two juxtaposed lexemes of those same word-classes, the combination must be a compound. For example, German Holzhaus [wood-house] must be a compound noun because two juxtaposed nouns cannot by themselves form a noun phrase in German. Also, Italian segnalibri [indicate-books] ‘bookmark’ must be a compound, because it is structurally not similar to a phrase with a similar meaning. (Italian has a phrase segna libri whose pronunciation is the same, but this is an imperative verb phrase and means ‘indicate books!’, so both syntactically and semantically it is clearly distinct from the compound segnalibri.) Occasionally compounds even have a special segmental marker. Thus, in Coast Tsimshian an -m- interfix between the two members indicates a compound, e.g. gyemg-m-dziws [light-intf-day] ‘sun’, güünks-m-hoon [dry-intf-fish] ‘dried fish’ (Dunn 1979: 55). And we saw in (7.4) that the interfixes -s- and -en- are used in German to form compounds (Liebe-s-brief ‘love letter’, Schwan-en-gesang ‘swansong’). However, there are also a great many cases in which compounds are quite similar to phrases with a similar meaning, and then we have to take a closer look in order to distinguish the two patterns. For example, Lango has an inalienable possessive construction with the order head–possessor that is expressed by simple juxtaposition (e.g. the syntactic phrases wì rwòt [head
9 . 1 ╇ C O M P O U N D S V E R S U S P H R A S E S
191
king] ‘the king’s head’, bàd dàktàl [arm doctor] ‘the doctor’s arm’). Now Lango has expressions that look like compounds at first blush, e.g. wàŋ ɔ` t [eye house] ‘window’, dɔ´ g bɔ´ŋɔ´ [mouth dress] ‘hem’ (Noonan 1992: 115, 157–8). Their most striking property is that they are idiomatic – i.e. their meaning cannot be determined from the meaning of their constituents. Idiomaticity is a typical property of compounds. However, it is neither a necessary nor a sufficient criterion for identifying a compound. On the one hand, all languages with productive compounding have some compounds with compositional meaning (English examples are piano-tuner, brake cable, spring festival). On the other hand, not all idioms are compounds. Idioms like English spill the beans, French roulette russe ‘Russian roulette’ or German goldenes Zeitalter ‘golden age’ are formally just like ordinary syntactic phrases in the language, and the general assumption is therefore that they are idiomatic phrases. Thus, one might suspect that Lango expressions like wàŋ ɔ` t ‘window’, dɔ´ g bɔ´ŋɔ´ ‘hem’ are simply phrases that happen to be semantically idiomatic. Now, in actual fact, this seems unlikely, because Lango also has clear compounds of the type N–N, e.g. ɔ` t ce` m [house-eating] ‘restaurant’, mɔ`ɔ`  ìm [oil sesame] ‘sesame oil’. These cannot be phrases, because ɔ` t and mɔ`ɔ` are not inalienable nouns, and this kind of possessive construction is possible only with inalienable nouns such as kinship terms and body part terms. Thus, wàŋ ɔ` t and dɔ´ g bɔ´ŋɔ´ are probably compounds. But the point is exactly that there can be ambiguity. So how can we distinguish a compound from a syntactic phrase when ambiguity arises? First, a semantic property of almost all compounds is that a dependent noun does not denote a particular referent but the entire class; in other words, a dependent noun in a compound is not referential but generic. For example, in the compound piano-tuner, the element piano cannot refer to a particular piano, but must refer to pianos in general. Generic meaning is also a general feature of dependent nouns in verb-headed N–V compounds (i.e. in noun incorporation), as the examples in (9.1)–(9.2) show. The (a) examples show a non-incorporated, phrasal version, and the (b) examples show an incorporated version of the sentence, with a generic interpretation of the incorporated noun. (Note in (9.1b) the absence of the second determiner ki, i.e. the one that serves to pick out particular wood in (9.1a). Likewise, the demonstrative marker -o is missing in (9.2b).) (9.1) Lakhota a. Wicˇháša ki cˇhᲠki man the wood the ‘The man is chopping the wood.’
kaksá-he. chop-cont
b. Wicˇháša ki cˇha˛-káksa-he. man the wood-chop-cont ‘The man is chopping wood.’ (Lit.: ‘The man is wood-chopping.’) (Van Valin and LaPolla 1997: 123)
192 
C H A P T E R 9 â•… W O R D S A N D P H R A S E S
(9.2) Ponapean a. I pahn kang 1sg fut eat ‘I will take that medicine.’
wini-o. medicine-dem
b. I pahn keng-wini. 1sg fut eat-medicine ‘I will take medicine.’ (Lit.: ‘I will do medicine-taking.’) (Rehg 1981: 209–14) In syntactic phrases, by contrast, a noun is more typically referential, as in (9.1a) and (9.2a). However, generic interpretation is not a sufficient criterion by itself. A dependent noun in a noun phrase need not necessarily be referential. In the German phrase Haus aus Holz ‘house from wood’, Holz ‘wood’ can be just as generic as in Holzhaus ‘wood house’, the compound that we saw above. This means that we cannot conclude that the expression is a compound just because a dependent noun is generic. But, conversely, if a dependent noun is referential (as in Lango wì rwòt ‘the king’s head’, which refers to the head of a particular king), we can be fairly certain that the expression is a phrase and not a compound. Since the typical semantic properties of compounds are not unique to compounds, we often need additional phonological, morphological and syntactic properties to identify compounds when compound and phrase patterns are otherwise formally similar. In general terms, compounds exhibit greater phonological, morphological and syntactic cohesion than phrases. A well-known phonological criterion is stress. In English, each word has one main stress, so main stress on only one member of a compoundlike expression suggests that it is a word. Thus, the expressions in (9.3a) are compounds, whereas those in (9.3b) are generally taken to be phrases. (As these examples show, word division in the spelling correlates only imperfectly with the criterion of stress.) (9.3) a. góldfìsh báckdròp Whíte Hòuse b. góld médal báckstáirs whíte kníght Stress is also one of the criteria that show that Lakhota incorporation ((9.1b) above) is a compounding pattern: the expression cˇhá˛káksahe ‘woodchopping’ is phonologically cohesive in the sense that it acts as a single unit for the purpose of stress assignment. An example of a different kind of phonological cohesion comes from
9 . 1 ╇ C O M P O U N D S V E R S U S P H R A S E S
193
Chukchi. In this language, compounding creates a single domain for vowel harmony. Within a compound, the vowels must either all belong to the set [i], [e], [u], or all belong to the set [ə], [a], [o]. Thus, when kupre-n ‘net’ occurs in a compound, it may have to be changed to kopra-n (e.g. pəlvəntəkopra-n ‘metal net’). In many better-known languages with vowel harmony (e.g. Turkish), compound nouns do not count as a single domain for vowel harmony, so an expression that does not show harmony could be either a phrase or a compound. But vowel harmony never applies across word boundaries, so when harmony does affect both lexemes, as in Chukchi, we can conclude that the expression is a compound. In some cases, morphological cohesion can give us decisive criteria for word status. In the relevant examples, a morphological pattern clearly takes the whole compound in its domain rather than just the head. Consider the English word sister-in-law, which for many speakers has the plural form sister-in-laws. The older form sisters-in-law, which has the plural suffix on the head noun, could be either a phrase or a compound noun (with the head serving as the morphosyntactic locus; see Section 7.2 for similar examples), but sister-in-laws can only be a compound. The plural suffix -s is semantically associated with the entire unit, and not only with law (it indicates multiple sisters, not multiple laws). And since the plural marker normally attaches only to words, not to phrases, sister-in-law must be a compound. Similarly, in Ponapean the aspectual suffix -(a)la attaches to verbs, as shown in (9.4a). (9.4) a. I kang-ala wini-o. 1sg eat-compl medicine-dem ‘I completed taking that medicine, i.e. I took all of that medicine.’ b. I keng-winih-la. 1sg eat-medicine-compl ‘I completed my medicine-taking.’
(Rehg 1981: 214)
The position of the completive affix in (9.4b) is evidence that keng-wini(h) is a compound verb. Where phonological and morphological criteria are not decisive, criteria of syntactic cohesion can differentiate between compounds and phrases. Most obviously, syntactic phrases and compounds differ with regard to separability: phrases are often separable, whereas compounds are inseparable. This means that other words cannot intervene between compound members. For example, Hausa has N–N compounds that resemble phrasal possessive constructions in that they show headdependent order and a relation marker (-n (masculine)/-r˜) (feminine)) on the head, e.g. gida-n-sauroo [house-rel.m-mosquito] ‘mosquito net’. There are no phonological or morphological properties that would distinguish such compounds from possessive phrases like gida-n Muusaa ‘Musa’s
194 
C H A P T E R 9 â•… W O R D S A N D P H R A S E S
house’. However, when an adjective modifies these expressions, it becomes clear that the compound is inseparable, whereas the phrase is separable. (9.5) a. gida-n-sauroo bàbba (*gidaa bàbba na sauroo) house-rel.m-mosquito big ‘big mosquito net’ b. gidaa house ‘Musa’s big house’
bàbba na big rel.m
Muusaa Musa (Newman 2000: 109)
Another clear indication of phrasal status is the expandability of the dependent element, because dependents in compounds cannot generally be expanded by modifiers such as adjectives or adverbs (e.g. English kingmaker versus *illegitimate kingmaker ‘someone who makes an illegitimate king’; crispbread versus *very crispbread ‘bread that is very crisp’). In compounds, the head noun cannot be replaced by an anaphoric pronoun.1 For instance, English allows (9.6a), but not (9.6b). Silversmith must be a compound. (9.6) a. My aunt has one gold watch and three silver ones (i.e. three silver watches). b. *My aunt knows one goldsmith and three silver ones (i.e. three silversmiths). By contrast, in Japanese, complex verbal expressions like benkyoo suru [study do] ‘study’ and rakka suru [fall do] ‘fall’ are sometimes regarded as N–V Â� compound verbs. However, the noun in these combinations can be omitted with an anaphoric interpretation. See (9.7), where the noun rakka does not occur in the response. This suggests that these expressions are phrases after all. (9.7) Sore wa rakka si-masi-ta ka? it top fall do-polite-pst int ‘Did it fall? – Yes, it did.’
– yes
Hai, si-masi-ta. do-polite-pst
(Matsumoto 1996: 41)
(The dependent noun in a compound cannot be replaced by an anaphoric pronoun either (*the king and the him-makers), but this is not very useful as a test. As we have seen already, the dependent noun is almost always generic. Anaphoric pronouns cannot be interpreted generically, so there is an independent reason for this failure.) Finally, phrases can exhibit coordination ellipsis, meaning that one of two identical elements in coordinated phrases can be optionally left out. 1
An anaphoric pronoun refers back to some noun that has already been introduced in the sentence or discourse.
9 . 1 ╇ C O M P O U N D S V E R S U S P H R A S E S
195
By contrast, a compound member generally cannot be deleted in this way; compare (9.8b) to (9.9b). (9.8) a. Large fish and small fish were mistakenly placed in the same tank. b. Large Ø and small fish were mistakenly placed in the same tank. (9.9) a. Flying fish and small fish were mistakenly placed in the same tank. b. *Flying Ø and small fish were mistakenly placed in the same tank. Flying fish must be a compound. Thus, compounds can be distinguished from phrases by semantic, phonological, morphological and syntactic criteria. These are summarized in Table 9.1. Phrases
Compounds
semantic
dependent noun may be referential head may be replaced by an anaphoric pronoun
dependent noun virtually always generic head may not be replaced 0007by an anaphoric pronoun
phonological
less cohesion
greater cohesion e.g. compound as domain 0007of stress assignment, vowel harmony
morphological
no cohesion
greater cohesion 0007e.g. compound as domain of affixation
syntactic
separable dependent noun expandable coordination ellipsis possible
inseparable dependent noun not expandable coordination ellipsis impossible
Table 9.1╇ Phrases versus compounds
While the discussion here has focused on compounds, some of these criteria (e.g. phonological and morphological cohesion, nonseparability and coordination ellipsis) apply to other types of words as well. These can thus be used as general tests for distinguishing words from phrases. Some examples that do not involve compounds will be presented below.
196 
C H A P T E R 9 â•… W O R D S A N D P H R A S E S
9.2€Free forms versus bound forms The boundary between words and phrases is not always clear – tests for wordhood may produce contradictory results. One of the more interesting complications is that an expression may be a word for the purposes of syntax, but not by itself an entire prosodic word (i.e. a word for purposes of stress assignment). These are clitics. In this section we compare clitics and affixes on the one hand to independent word-forms on the other. Clitics and affixes (collectively referred to as bound forms)2 are similar in that they exhibit prosodic dependence. This means that they cannot by themselves constitute a domain for word stress – they must ‘lean’ on a prosodic host. By contrast, canonical word-forms (often called free forms) exhibit prosodic independence. This can be seen in several ways. First, an utterance may be interrupted at a boundary between two free forms, but not at a boundary between a bound form and its host. This is true for affixes (e.g. Paul ... started to play, or Paul started ... to play, but not *Paul start ... ed to play), and also for clitics, as shown by the Croatian sentences in (9.10).3 (9.10) a. Oni = su ... pocˇeli igrati. they = aux.3pl began.m.pl play.inf ‘They began to play.’
b. *Oni ... = su pocˇeli igrati.
Croatian clitics are prosodically dependent on the preceding word. The prosodic dependence of the clitic makes it impossible to pause between the clitic and the preceding word. Also, clitics never bear their own stress. In the example above, su does not (and cannot) bear stress at all. In the French imperative joue=le! ‘play it!’, the weak object pronoun clitic le bears stress (joue=’le), but this is the stress of the whole prosodic word (which happens to be on the final syllable), not le’s own stress. Finally, in languages that use stress to express contrast, free forms can exhibit contrastive stress, whereas bound forms cannot. Thus, in English we can have paul started to play, or Paul started to play, or Paul started to play, but not *Paul started to play, or *Paul started to play, because past-tense -ed and infinitival to are prosodically dependent. Prosodic dependence can have syntactic consequences; free and bound 2
3
Note that a bound form in this sense is not the same as a bound stem (Section 2.2), although the two terms are related. A bound stem is a base that cannot stand by itself in any way – it is not a complete word for the purposes of the syntax nor is it a complete prosodic word. The term bound form, as used here, refers only to lack of prosodic independence. It is thus a broader term that encompasses bound stems, affixes and clitics. In this and other examples, we follow the convention of linking clitics to their hosts by an equal sign.
9 . 3 ╇ C L I T I C S V E R S U S A F F I X E S
197
forms differ in the kinds of syntactic constructions in which they can appear. In languages like French, where contrast is expressed by clefting, free forms can be clefted, but bound forms cannot.4 The sentence in (9.11a) can have the clefted variant (9.11b), but (9.12a) cannot have the clefted variant in (9.12b). (9.11) a. Paul commenç-ait à jou-er. Paul begin-3sg.impf to play-inf ‘Paul started to play.’ b. C’ est Paul qui commenç-ait it is Paul who begin-3sg.impf ‘It’s Paul who started to play.’
à to
jou-er. play-inf
(9.12) a. Il=commençait à jouer. ‘He started to play.’ b. *C’est il qui commençait à jouer. ‘It’s he who started to play.’ Example (9.12b) is impossible in part because cleft constructions involve prominent stress on the clefted word or phrase, and the weak subject pronoun il is a clitic (i.e. not prosodically independent). Therefore, in the clefted variant of (9.12a), French uses its independent pronoun lui ‘he’ (C’est lui qui commençait à jouer). This difference between lui (free form) and il (bound form) extends to other syntactic constructions as well. For example, the bound forms are used in normal subject + verb constructions (as in je=joue ‘I play’, tu=joues ‘you play’, il=joue ‘he plays’), but when the pronoun is topicalized, the free form is used (moi, je=joue ‘as for me, I play’, not *je, je=joue).5 And likewise in coordination, the free form is used: moi et toi jouons ‘you and I play’, not *je et tu jouons. So, the bound form cannot be used when the pronoun is separated from a viable host and/or in a position that requires sentential stress. Cross-linguistically, free forms thus exhibit more syntactic freedom of movement, and movement tests like clefting and topicalization can be useful for distinguishing free forms from bound forms.
9.3╇ Clitics versus affixes The contrast between free forms and bound forms is only half of the distinction that interests us. We also need tests that will distinguish clitics from affixes.
4
5
In cleft constructions, a word or phrase is positioned outside of its clause for the purpose of creating focus. In English, it is X that Y (e.g. It is mary that ran a marathon, (not Bill).) is a typical cleft type. Topicalization is when a constituent is moved out of its phrase, usually to the beginning of the sentence, to indicate the topic about which new information will be added.
198 
C H A P T E R 9 â•… W O R D S A N D P H R A S E S
The first thing to note is that clitics exhibit notoriously heterogeneous behaviour. (In fact, clitics are diverse enough that some linguists use the term as a kind of ‘junk’ label for anything that is not quite a word and not quite an affix.) This means that it is impossible to identify traits that all clitics will exhibit, to the exclusion of affixes. Still, even allowing that there are many different sorts of clitics, there are a number of properties that are useful for collectively distinguishing them from affixes. Perhaps the most salient property of clitics is that they have freedom of host selection – i.e. a clitic can often occur with hosts of various syntactic categories, and its host need not be syntactically related to it. Example (9.13) shows that the English clitic =’s has freedom of host selection. (9.13) a. The person you were talking about=’s walking over here. [preposition] b. Replacing the window you broke=’s going to cost a lot of money. [verb] c. That house down the street=’s going to sell quickly. [noun] Affixes do not have such freedom of host selection – they combine with stems to which they are syntactically related. Additionally, clitics may be less prosodically integrated with their hosts than are affixes. In other words, affixes are always within the domain of word stress, but clitics may or may not be. French joue=’le! ‘play it!’ is an example of word stress applying to the entire clitic group (the expression formed by one or more clitics and the host). Spanish exemplifies the opposite pattern, in which clitics are not prosodically integrated. In Spanish, stress is usually on the last or penultimate syllable of the word, and rarely on the antepenultimate (e.g. caminár ‘walk.inf’, camína ‘walk.prs.3sg’, caminábamos ‘walk.pst.1pl’), but never on the fourth syllable from the end. But this is possible with clitic groups, e.g. díga=me=lo ‘say it to me!’, suggesting that in this language, the clitics are prosodically dependent on the host, but outside of the domain for stress placement. In this respect, then, Spanish clitics behave unlike affixes. Third, morphophonological rules are less likely to operate across the boundary between a host and a clitic than across the boundary between a stem and an affix. Certainly, some languages have rules that apply equally to affixes and clitics, including the rule of vowel harmony in Finnish. Bound elements like the suffix -nsa/-nsä ‘his’ and the clitic =ko/=kö (question marker) agree in backness with the last vowel of the stem or host (koiransa ‘his dog’, ystävä-nsä ‘his friend’; koira=ko ‘dog?’, ystävä=kö ‘friend?’).6 However, many languages have morphophonological rules that operate within the domain of the word-form, but not within the clitic group. For example, in Dutch obstruents are devoiced word-finally, and no such devoicing occurs when a vowel-initial suffix follows the same morpheme 6
Orthographic <ä>, <ö>, , and correspond to International Phonetic Alphabet [æ], [ø], [y], [o] and [a], respectively. The first three are front vowels; the last two are back vowels.
9 . 3 ╇ C L I T I C S V E R S U S A F F I X E S
199
(see (9.14a)). However, when a vowel-initial clitic follows it, devoicing still occurs, as can be seen in (9.14b). Thus, the clitic is ‘invisible’ to the rule of final devoicing. (9.14) a. verband [vər´bɑnt] ‘bandage’
verband-ig ‘bandage-like’
[vər´bɑndix]
b. ik brand [ig´brɑnt] ‘I burned’
brand=ik ‘I burned’
[´brɑntik]
Similarly, in Ponapean there is a rule of vowel lengthening at the end of the word that does not apply when a suffix follows. However, when a clitic such as demonstrative et follows the noun, vowel lengthening still occurs (Ponapean spelling marks vowel length by the letter h): (9.15) sahpw ngihl pwuhs
‘land’ sapw-ei ‘voice’ ngil-ei ‘novel’ pwus-ei
‘my land’ sahpw=et ‘this land’ ‘my voice’ ngihl=et ‘this voice’ ‘my novel’ pwuhs=et ‘this novel’ (Rehg 1981: 169–70, 186)
Fourth, affixes may trigger idiosyncratic suppletive alternations in the base, whereas clitics do not. For example, consider the Finnish words in (9.16). (9.16) nainen ‘woman’
naise-llinen naise-n naise-lla naise-nsa
‘woman-like, feminine’ ‘woman’s (gen.sg)’ ‘to the woman (all.sg)’ ‘his woman’ (Kanerva 1987: 506)
In Finnish, many nouns alternate between a stem-final sequence -nen (e.g. nainen) and a sequence -se (e.g. naise-). The former occurs when the word is unsuffixed (i.e. in the nominative singular form), and the latter occurs when any kind of suffix follows, inflectional or derivational. But when a clitic follows the noun ‘woman’, the stem nainen is used (e.g. nainen=ko? ‘the woman?’), showing that the clitics behave differently from affixes in this respect, and more like word-forms. Likewise, affixes may undergo idiosyncratic suppletive alternations, whereas clitics do not. For instance, Polish has several different inflection classes of verbs, and the first person singular suffix is either -m or -e˛, depending on the class (kocha-m ‘I love’, umie-m ‘I’m able’, ucz-e˛ ‘I teach’, pij-e˛ ‘I drink’). Object pronouns, however, are clitics that attach after their hosts, and they have an invariable shape: go (kocham=go ‘I love him’, pije˛=go ‘I drink it’, ucze˛=go ‘I teach him’, etc.). Fifth, affix–base combinations may have an idiosyncratic meaning, whereas clitic–host combinations never do. Idiosyncratic meanings of affixes are widespread in derivational morphology, but occasionally they are found in inflection as well, e.g. the Dutch inflected form ouder. In its
200 
C H A P T E R 9 â•… W O R D S A N D P H R A S E S
literal meaning this word is a comparative (‘older’), but has also been extended to use as a noun, with the idiosyncratic meaning ‘parent’ (see (5.14) for more examples). Sixth, expected affix–base combinations may arbitrarily fail to exist, whereas clitic–host combinations are always possible. And finally, it is often noted that clitics may have some syntactic freedom of movement, whereas an affix must always attach to its base and cannot move independently of it. The degree to which clitics do have freedom of movement is a somewhat complicated issue, and it is useful here to divide clitics into two categories: simple and special (Zwicky 1977). A simple clitic is one that can appear in the same syntactic positions as a corresponding free form. For instance, the clitic form ’s appears largely in the same positions as the free form is, as shown in (9.17). (9.17) a. Replacing the window you broke=’s going to cost a lot of money. a’. Replacing the window you broke is going to cost a lot of money.
b. It=’s going to cost a lot of money to replace the window you broke. b’. It is going to cost a lot of money to replace the window you broke.
Of course, as we saw in the preceding section, clitics cannot be used when an accented form is required, or when there is no host to lean on (* =’s he going to replace the window?). Otherwise, however, simple clitics have the same freedom of movement as free forms. A special clitic is ‘special’ in the sense that its syntactic distribution differs from that of free forms and must be described in its own right. Special clitics typically have less freedom of movement than simple clitics, or even none. For instance, second-position clitics (also called Wackernagel clitics after the linguist who made them famous) appear after the first element of the (simple) sentence, which serves as the host. Depending on the language, the first element may be either the first stressed word, or the first syntactic constituent. The following examples come from Pitjantjatjara. (9.18) a. Tjitji-ngku =ni child-erg =acc.1sg ‘The child saw me.’
nya-ngu. see-pst
b. Tjitji nyanga child this ‘This big child saw me.’
pulka-ngku big-erg
=ni =acc.1sg
nya-ngu. see-pst
(Bowe 1990: 12)
In this language, the accusative pronominal clitic ni must occur after the first syntactic constituent, as shown in (9.18).7 It cannot occur in other positions, 7
In the Pitjantjatjara writing system, the letters l and n correspond to IPA [ı] (retroflex lateral) and [] (retroflex nasal), respectively. These contrast with [l] and [n], which are written without the underline.
9 . 3 ╇ C L I T I C S V E R S U S A F F I X E S
201
even where a free form pronoun is possible. Compare the sentence with free form of the pronoun in (9.19a) to the one with the equivalent clitic in (9.19b). (9.19) a. Trevor-lu mukuri-nganyi Mary-lu ngayunya helpamila-ntjaku. Trevor-erg want-prs.cont Mary-erg me help-purp ‘Trevor wants Mary to help me.’ b. *Trevor-lu mukuri-nganyi Mary-lu =ni helpamila-ntjaku. Trevor-erg want-prs.cont Mary-erg =acc.1sg help-purp (Bowe 1990: 72) The pronominal clitic thus has no freedom of movement. Serbian similarly has second-position clitics. An accusative clitic pronoun is shown in (9.20), and the free form pronoun in (9.21). The (c) examples are the crucial comparison: the full form can occur in third position but the clitic cannot, despite having a suitable prosodic host. (9.20) a. Marija =ga voli. Marija =him loves ‘Marija loves him.’
b. Voli =ga Marija. c. *Marija voli =ga.
(9.21) a. Marija njega voli. Marija him loves ‘Marija loves him.’
b. Voli njega Marija. c. Marija voli njega.
Notably, however, Serbian clitics have limited freedom of movement under particular circumstances. (9.22) a. Marija želi da = joj = ga Marija wants that f.dat.sg m.acc.sg ‘Marija wants to introduce him to her.’
predstavi. introduces
b. (?)Marija = joj = ga želi da predstavi. Marija f.dat.sg m.acc.sg wants that introduces ‘Marija wants to introduce him to her.’ (Franks and King 2000: 243) Here, the clitics joj ‘to her’ and ga ‘him’ are associated with the verb predstavi ‘introduce’, and in (9.22a), they appear in second position in that verb’s clause (...da joj ga predstavi). In (9.22b), however, the clitics have ‘climbed’ into the higher clause, and appear in second position within it. Some speakers consider (9.22b) to be the less preferable version of the sentence, but many speakers accept both. Thus, when Serbian second-position clitics
202 
C H A P T E R 9 â•… W O R D S A N D P H R A S E S
appear in a lower clause they may have some freedom of movement, and in this respect they are unlike affixes. The criteria for distinguishing between affixes and clitics are summarized in Table 9.3. Clitics
Affixes
freedom of host selection
no freedom of stem selection
possible freedom of movement
no freedom of movement
less prosodically integrated
more prosodically integrated
may be outside the domain of a phonological rule
within the domain of a phonological rule
do not trigger/undergo morphophonological or suppletive alternations
may trigger/undergo morphophonological or suppletive alternations
affix–base combinations... clitic–host combinations... do not have idiosyncratic meanings may have idiosyncratic meanings do not have arbitrary gaps
may have arbitrary gaps
Table 9.3╇ Clitics versus affixes
Overall, the data present us with the picture that clitics are like affixes in some respects, and like independent word-forms in others. Just as importantly, however, clitics do not themselves constitute a uniform group. All clitics are prosodically dependent on a host and have some freedom of host selection, but some clitics are prosodically or phonologically integrated with their hosts while others are not; some have special syntax, but others do not; and so on. This heterogeneous behaviour makes sense from a diachronic perspective. Inflectional morphology commonly arises from free words, and we can hypothesize that clitics represent the intermediate stages in this transition. Most likely, fast speech processes lead to reduced variants of already prosodically weak grammatical elements. These reduced variants are then susceptible to being reanalyzed by a new generation of speakers as distinct lexical expressions, rather than as straightforward instances of phonetic reduction. And over time, these clitics may acquire further affixal properties: reduced stem selection (e.g. if a clitic attaches predominantly to a single word-class, it may be reanalyzed as attaching only to that wordclass), reduced freedom of movement, morphological and phonological cohesion, etc. Cross-linguistically, then, clitics can be expected to exhibit a wide range of syntactic, morphological and phonological properties. Of course, some clitics may never become canonical affixes. The development from a free word form to an affix is not a single change with a predetermined
9 . 4 ╇ L E X I C A L I N T E G R I T Y
203
outcome; it is better thought of as a series of small changes that collectively (and to some extent coincidentally) reduce the prosodic, syntactic and morphological independence of an expression. Not surprisingly, many of the facts surrounding clitics, and how they should be accounted for in formal description, are topics of ongoing debate in morphological theory.
9.4╇ Lexical integrity We saw in the preceding sections that words differ from syntactic phrases in a number of crucial respects. We failed, however, to ask why these differences exist. We now turn to this issue. Many linguists have posited that the various differences in the behaviour of words and phrases reflect a single general principle, which can be formulated as follows: (9.23) Lexical Integrity Hypothesis: Rules of syntax can refer/apply to entire words or the properties of entire words, but not to the internal parts of words or their properties. The Lexical Integrity Hypothesis (also called the Lexical Integrity Principle) comes in many subtly different forms, but the core idea is that as far as syntactic rules are concerned, words have no internal structure. They are atomic. Moreover, (9.23) is a claim about the nature of language in general, so the generalization should apply to all languages in the world. The validity of the generalization is subject to empirical testing. So far, discussion in this chapter has focused on examples that are consistent with (9.23). It is not, however, sufficient to show that rules of syntax (e.g. movement, anaphoric replacement, coordination ellipsis) often fail to apply to the internal parts of words. The Lexical Integrity Hypothesis states that syntactic rules can never refer to the internal parts of words. We must therefore first ask whether languages ever violate the principle of lexical integrity. The Lexical Integrity Hypothesis can be evaluated in a meaningful way only if the notions ‘word’ and ‘syntactic rule’ are specified precisely. For instance, a clitic forms a prosodic word with its host, but (9.23) clearly does not hold over the prosodic word. If it did, clitics could have no independent freedom of movement – movement is a kind of syntactic rule, and a clitic is an internal part of a prosodic word. However, there is a simple resolution to this issue. We need only assume that lexical integrity holds over the morphosyntactic word, rather than the prosodic word, meaning that a clitic and its prosodic host are separate words in the relevant sense and thus not subject to (9.23) (Bresnan and Mchombo 1995). Hungarian exemplifies a more complex problem. As shown in (9.24), meg-old [pfv-solve] can serve as the input to deverbal noun derivation and deverbal adjectival formation. It is thus quite clear that constructions formed
204 
C H A P T E R 9 â•… W O R D S A N D P H R A S E S
with meg are single words in some sense. (The alternative analysis, that -ás and -hatatlan attach to old by itself seems unlikely because meg appears with nouns and adjectives only if they are deverbal.) (9.24) meg-old meg-old-ás meg-old-hatatlan
‘solve’ ‘solution’ ‘unsolvable’
Given this, it is somewhat surprising that meg can be separated from the rest of the verb, as shown in (9.25b). (9.25) a. Pál meg-old-ott-a Paul pfv-solve-pst-def.3sg ‘Paul solved the problem.’
a the
problémá-t. problem-acc
b. Pál nem old-ott-a Paul not solve-pst-def.3sg ‘Paul didn’t solve the problem.’
meg pfv
a the
problémá-t. problem-acc
Is this a violation of lexical integrity? Not necessarily. Like the forms in (9.24), by all indications megoldotta is generated by morphological rules, but some formal approaches posit that these kinds of periphrastic constructions are then inserted into syntax at more than one node (i.e. meg and oldotta occupy separate syntactic nodes). Moreover, it has been suggested that the relevant notion of a word is ‘terminal syntactic node’, rather than ‘morphologically generated object’ (Ackerman and LeSourd 1997). Thus, under these assumptions, lexical integrity is not violated. The point here is that it can be quite difficult to determine what constitutes a true violation of the Lexical Integrity Hypothesis. Nonetheless, some kinds of data are widely viewed as problematic. For instance, all syntactic models treat agreement and movement (or some equivalent of movement) as types of syntactic rules, so we should not expect to find agreement within a word, or between an internal part of a word and another element in the sentence. Likewise, we should not encounter instances of movement within a word, or syntactic placement of an element within a word. No conclusive examples of word-internal syntactic movement have been documented. (Some approaches to morphological description posit word-internal movement, but for theory-internal reasons, not empirical ones.) However, rare instances of the other three possibilities can be found. For example, in some languages verbs show agreement with incorporated nouns (i.e. with compound Â� members), as is the case in Southern Tiwa: (9.26) a. ti-khwian-mu-ban 1sg.sbj/sg.obj-dog-see-pst ‘I saw the dog.’8 8
Note that this example seems to be an exception to the generalization that incorporated nouns have generic interpretation (see (9.1) and (9.2)). The reason for this is unclear.
9 . 4 ╇ L E X I C A L I N T E G R I T Y
b. bi-khwian-mu-ban 1sg.sbj/pl.obj-dog-see-pst ‘I saw dogs.’
205
(Allen et al. 1990: 322)
Here the agreement prefixes register the number of the incorporated noun -khwian- ‘dog(s)’, so it is hard to escape the conclusion that there is agreement between the internal parts of the complex verb. Also, in Chapter 5 we encountered the following Upper Sorbian example of agreement between an internal part of a word and another word in the phrase (repeated from (5.13)). Here, the adjective mojeho agrees for gender with the root of the denominal adjective mužowa. (Muž ‘husband’ is a masculine noun.) (9.27) moj-eho muž-ow-a sotra my-m.sg.gen husband-poss-f.sg.nom sister.f.sg.nom ‘my husband’s sister’ (Corbett 1987: 303) Equally interesting are examples of clitics within words. In Udi, the clitic ne sometimes occurs between the two parts of a bimorphemic verb stem (see (9.28a)). (9.28) a. nana-n bu.γ a=ne-b-e p’a. mother-erg find=3sg-do-aorII two ‘Mother found two toys.’
acˇik’alšey toy.abs
b. nana-n te=ne bu.γ a-b-e p’a. acˇik’alšey mother-erg neg=3sg find-do-aorII two toy.abs ‘Mother did not find two toys.’ (Harris 2002: 117, 123) Bu.γ a-b-e seems to be a single word: coordination ellipsis, movement and anaphoric replacement of the internal constituents are all impossible. Yet according to properties such as host selection, ne is a clitic whose position is governed by principles of syntax, not an affix (compare (9.28a) with (9.28b)). So it seems that the syntax places the clitic internally to the verb. Moreover, to do so, the rules of syntax must know where the morpheme boundaries are, and therefore must refer to the verb’s internal morphemic structure (Harris 2000). This is problematic for the idea of lexical integrity. In the end, the Lexical Integrity Hypothesis remains controversial. The most extreme syntactic approaches to morphological structure reject it entirely, partly on empirical grounds, but also for theory-internal reasons. (A theory without a principled distinction between words and phrases entirely cannot in any meaningful way even define a condition like (9.23).) Still, some linguists consider counterexamples to be minor, and many approaches to formal description accept some version of the Lexical Integrity Hypothesis. Finally, if we believe that the Lexical Integrity Hypothesis can be upheld on
206 
C H A P T E R 9 â•… W O R D S A N D P H R A S E S
empirical grounds, we may ask about its underlying causes. One possibility is that (9.23) is an axiomatic principle of Universal Grammar – a fundamental and irreducible constraint on the nature of language. Or alternatively, lexical integrity effects might indirectly result from independent syntactic constraints on movement, binding of anaphoric pronouns, and so on. But most commonly, lexical integrity is seen as following from the architecture of grammar. If morphology and syntax are two completely separate components of Â� grammar, it can be expected that syntactic rules do not ‘look inside’ the complete words received from the morphology. In particular, lexical integrity effects have been used to motivate a language architecture in which syntactic rules apply to bundles of morphosyntactic features, and the morphological component delivers complete words only after all syntactic rules have applied. (See Section 5.5.2 for more discussion of this proposed architecture.) This would explain why compound members cannot be replaced by anaphoric pronouns, why they cannot be extracted or undergo coordination ellipsis, etc. The syntax simply has no access to word-internal structure because words are inserted post-syntactically. There are a large number of ways in which syntactic rules can be formulated, and a large number of ways in which the relevant notion of ‘word’ can be defined. It will therefore inevitably be difficult to reach a consensus about the exact relation between syntax and morphology. Clearly, morphology and syntax are different to some degree, but the nature of the relationship will probably be debated for a long time to come.
Summary of Chapter 9 There are two main difficulties that we encounter in dividing texts into word-forms: distinguishing affixed word-forms from phrases that contain a function word, and distinguishing compounds from phrases with two content words. Word-forms that are intermediate between fully independent word-forms and fully dependent affixes are called clitics, and clitics and affixes are grouped together as bound forms. Free forms differ from bound forms in that they are prosodically independent, cleftable, topicalizable and coordinatable. Clitics differ from affixes in that they have greater freedom of host selection, are phonologically less integrated, do not trigger or undergo morphophonological alternations, show no idiosyncrasies of meaning or distribution, and may have more freedom of movement. Phrases differ from compounds in that they allow referential dependent members and exhibit less phonological, morphological and syntactic cohesion. Often a ‘Lexical Integrity Hypothesis’ is postulated that forbids syntactic rules to apply to parts of words. While (apparent)
COMPREHENSION EXERCISES 
207
counterexamples to lexical integrity can be found, linguists have interpreted these exceptions differently. Among those formal approaches that accept the Lexical Integrity Hypothesis, the effects are usually seen as falling out from the architecture of the grammatical system – separate morphological and syntactic components.
Further reading The literature on clitics is strikingly large. First and foremost, much influential work regarding the affix/clitic distinction is due to Arnold Zwicky (e.g. Zwicky 1977, 1985b and Zwicky and Pullum 1983). See also Klavans (1985), Kanerva (1987), Halpern (1995), Boškovic´ (2001), Aikhenvald (2002) and Anderson (2005). Some famous examples of special clitics come from the Slavic languages; see Franks and King (2000) for descriptions. Nevis et al. (1994) is a bibliography of clitic research prior to 1991. More recent collections of papers include Beukema and den Dikken (2000) and Gerlach and Grijzenhout (2000). For compounds versus phrases, see Bauer (1998), Smirniotopoulos and Joseph (1998) and Bisetto and Scalise (1999). Lexical integrity is discussed and defended in Wasow (1977), Lapointe (1980), Di Sciullo and Williams (1987), Bresnan and Mchombo (1995), and in a modified form, in Ackerman and LeSourd (1997). Also see Rosen (1989) and Mohanan (1995) for arguments specifically related to incorporation. For counterarguments to the Lexical Integrity Principle, see Lieber (1992), Harris (2000), and Booij (2009). Syntactic approaches that do not assume lexical integrity include Baker (1988), Sadock (1991) and Halle and Marantz (1993).
Comprehension exercises 1. At the beginning of Section 9.1, we asked whether backboard, backdoor and back seat are compounds or phrases. Develop an answer to this question, and justify it using tests introduced in this chapter. 2. Provide arguments to show that English -s, the suffix of the third person singular of present-tense verbs, is an affix, not a clitic. 3. What is wrong with the following sentences? a.
Polish *Go spotka-ł-em w him meet-pst-1sg in ‘I met him in Cracow.’
Krakowie. Cracow
208 
C H A P T E R 9 â•… W O R D S A N D P H R A S E S
b. French *A: Qui joue? Robert? – B: Non, tu=joues. who plays Robert no you=play ‘A: Who is playing? Robert? – B: No, you are playing.’ c.
Serbian *Klaru cˇovek voleo=je. Klara.acc man.nom loved=aux ‘The man has loved Klara.’
d.
Ponapean *I keng-wini-o-la. 1sg eat-medicine-dem-compl ‘I completed my taking of that medicine.’
4. Sometimes the various criteria for distinguishing clitics from affixes contradict each other. For instance, in Spanish the bound pronominals undergo a morphophonological alternation when a third-person dative pronominal co-occurs with an accusative pronominal: -le is replaced by -se because another l follows: ‘tell me’ ‘tell him’ ‘tell me it’ ‘tell him it’
díga-me díga-le díga-me-lo díga-se-lo
Given what we said in this chapter about Spanish bound pronominals, where is the contradiction?
(*diga-le-lo)
5. Another case of a contradiction comes from Lithuanian, which forms reflexive verbs by means of an element s(i). (The letter e˙ stands for a long [e:].) 1sg 2sg 3 1pl 2pl
‘rock’ supu supi supa supame supate
‘rock oneself’ supuosi supiesi supasi supame˙s supate˙s
‘not rock oneself’ nesisupu nesisupi nesisupa nesisupame nesisupate
In what ways is this element like an affix, and in what way is it like a clitic?
6. Look at the example of noun incorporation in Guaraní (ex. (11.26)). Which criteria can be applied to show that (11.26b) contains a compound, not a phrase like (11.26a)?
E X P L O R AT O R Y E X E R C I S E
209
Exploratory exercise As we saw in this chapter, some linguists have posited that a word boundary operates as a kind of border below which syntactic rules cannot apply. However, it is not always clear what constitutes evidence for or against this hypothesis. For instance, in Germanic languages such as English, German and Dutch, entire phrases can act as compound members, e.g. a down in the trenches attitude, a floor of the birdcage taste, or just rolled out of bed hair. We will call these phrasal compounds. The question we want to ask is: Do phrasal compounds violate lexical integrity? Some researchers have suggested that the dependent member of the compound (i.e. the phrase) must be a set phrase or idiom – something that is stored in the lexicon as a single unit. If so, phrasal compounds are probably not true violations of lexical integrity. If the embedded phrase is stored in the lexicon similarly to a word, this may make it available to the word-formation component without the operation of syntactic rules. In this exercise you will test the claim that the dependent member of such compounds must be a set phrase or idiom, and explore implications that the results have for the Lexical Integrity Hypothesis. Many languages do not allow phrasal compounds, so the language of study should be one for which you already know that these compounds are possible to some extent. English is used here for demonstration purposes. The methodology involves observation of naturally-occurring examples. Instructions Step 1: Develop criteria for determining that a construction is (or is not) a phrasal compound. For instance, down in the trenches attitude is clearly a phrasal compound because ‘normal’ compounds cannot include prepositions or determiners. Moreover, the structure must be [[down in the trenches] attitude]. But what about gold and jewellery merchants or severe weather warning? Should the first be considered a phrasal compound (presumably with the structure [[gold and jewellery] merchants]), or conjoined phrases with ellipsis (derived from [[gold merchants] and [jewellery merchants]])? Is severe weather a phrase, or itself a compound inside the larger compound? Consider many kinds of borderline cases and develop specific criteria for including or excluding them from the category of phrasal compounds. Explain your reasoning. You might find it helpful to review the rules of compounding for your chosen language of study. Useful descriptions of English compounding patterns include Marchand (1969), Bauer (1983), Bauer and Renouf (2001) and Plag (2003). Step 2: Choose a newspaper or magazine to gather data from. Scan the text for phrasal compounds, and record any that are found. Continue until you have recorded at least 10 examples, preferably more. If working with others (e.g. in a class setting), a divide-and-conquer approach might be
210 
C H A P T E R 9 â•… W O R D S A N D P H R A S E S
useful.9 Have each person gather 10 examples from different sources and then combine the results. Step 3: Analyze the data. Remember that the starting point for this project is the claim that in phrasal compounds, the phrase must be likely to be directly stored in the lexicon rather than productively generated. Decide whether the dependent members of your phrasal compounds fit this description. Are they freely formed syntactic constructions, collocations, set phrases, idioms ...? (A collocation consists of two or more words that tend to occur together. For instance, weather tends to occur together with severe, but less often with harsh. Thus, severe weather is a collocation, but harsh weather is not. Linguists often measure the collocation strength of two words as the likelihood of two words occurring together, compared with the likelihood of each word occurring independently. Collocation strength can thus formally be expressed as a probability. However, for the purposes of this exercise, it is sufficient to use your own intuitions about the degree to which some construction constitutes a collocation or set phrase in the language.) Step 4: Consider the implications of your data for the Lexical Integrity Hypothesis. Are your data consistent with this hypothesis, or do they seem to contradict it? Explain your reasoning. Step 5 (optional): Read some of the literature about phrasal compounds and lexical integrity. Work in this area includes Botha (1981), Lieber (1992), Wiese (1996), Ackema and Neeleman (2004), Carstairs-McCarthy (2005) and Lieber and Scalise (2007). Do your observations match the description of phrasal compounds in the literature? Based on what you have read, why is it difficult to determine whether phrasal compounds represent true violations of lexical integrity? Does discussion in the literature make you look at your data in a new light? If so, explain.
9
The attentive reader has noticed that [[divide-and-conquer] approach] is a phrasal compound!
10.1 TWO TYPES OF ALTERNATIONS 
211
10
Morphophonology
I
n Chapter 2, we saw that morphemes often have different phonological shapes depending on the environment, i.e. the other morphemes and sounds with which they co-occur in a word. For example, the stem of the English lexeme leaf is pronounced [lif] in the singular, but [li:v] in the plural (leaves); the stem of pat is always pronounced [pæt] if it occurs without any suffix, but in many varieties the pronunciation is [pæɾ] if a vowel-initial suffix follows (patting [pæɾŋ]). The forms [pæt] and [pæɾ] (and [lif] and [li:v]) are phonological allomorphs – they bear the same meaning and have quite similar phonological shape (in contrast to suppletive allomorphs, which are not phonologically similar). Phonological allomorphs are interesting because they represent the point of intersection between morphological and phonological structure. In this chapter we explore issues related to the morphology–phonology interface in some detail.
10.1 Two types of alternations The formal relation between phonological allomorphs is called an alternation. Alternations come in two kinds: automatic alternations and morphophonological alternations. Like morphological patterns, alternations are often described in process terms. For example, German has a phonemic distinction between /k/ and /g/, and in positions before a vowel, both consonants are possible (e.g. Kinder [k] ‘children’, Geld [g] ‘money’, Völker [k] ‘peoples’, Tage [g] ‘days’). But in syllable-final position, both /k/ and /g/ are pronounced as [k] (e.g. Volk [volk] ‘people’, Tag [ta:k] ‘day’). This alternation of voiced and voiceless obstruents is thus called Final Devoicing (a process term) because /g/ seems to ‘lose’ its voicing feature in syllable-final position. As always when process terms are used, this terminology is probably best understood as metaphorical – speakers’ own knowledge may not include a literal process of devoicing.
212 
C H A P T E R 1 0 â•… M O R P H O P H O N O L O G Y
But the process terminology is very convenient because it gives more information than purely static terminoÂ�logy. If we simply referred to the German alternation in [ta:k]/[ta:gə] as voiceless/voiced alternation, we would not know that voiceless obstruents in syllable-final position do not always participate in this alternation – e.g. [k] in both Volk and, crucially, Völker. If, on the other hand, we call the alternation devoicing, it is immediately clear that the existence of [k] in both syllable-initial and syllable-final positions is completely expected, but [g] in syllable-final position is not. Let us look at a few representative cases of both automatic and morphophonological alternations, focusing on examples of stem allomorphy. (10.1) Some automatic alternations1 a. 0007German Final Devoicing: Voiced obstruents are voiceless when they occur in syllable-final position. Tage [ta:gə] ‘days’ Tag [ta:k] Liese [li:zə] ‘Liese (name)’ Lieschen [li:sçən] Monde [mo:ndə] ‘moons’ Mond [mo:nt]
pronounced ‘day’ ‘little Liese’ ‘moon’
b. 0007English Flapping: In many varieties of English, alveolar plosives ([d] and [t]) are pronounced as voiced flaps [ɾ] when they occur after a vowel and in front of an unstressed vowel. pat [pæt] patting [pæɾŋ] fat [fæt] fatter [fæɾər] pad [pæd] padding [pæɾŋ]
c. 0007Russian Akanie (neutralization of unstressed o and a): The vowel o is pronounced [ɑ] when it occurs in the syllable immediately before the stressed syllable, and both o and a are pronounced [ə] when they occur in an earlier syllable, or in a syllable after the stressed syllable. vol [vol] ‘ox (nom.sg)’ vol-y [vɑ’l] ‘oxen (nom.pl)’ nós-it [‘nos´it] ‘carries’ nos-í [nɑ’s´i] ‘carry!’ (impv) bórod-y [‘borəd] ‘beards’ borod-á ‘beard’ ╇ [bərɑ’dɑ] bandít [bɑn’d´it] ‘gangster’ bandit-ízm ‘gangsterism’ ╇ [bənd´i’t´izm]
d. 0007Japanese Palatalization: Alveolar obstruents ([t] and [s]) are pronounced as palatals ([t] and [], commonly written as ch and sh) when they occur before the high palatal vowel [i]. kas-e ‘lend’ (imperative) kash-i ‘lend’ (continuative) kat-e ‘win’ (imperative) kach-i ‘win’ (continuative) (Vance 1987: 177)
1
The names of alternations are capitalized here because they are often traditional names that are in general use among linguists (e.g. German Umlaut, Russian Akanie, Japanese Rendaku).
1 0 . 1 T W O T Y P E S O F A LT E R N AT I O N S
213
(10.2) Some morphophonological alternations a. 0007English Trisyllabic Shortening: The vowels or diphthongs [ei], [i:], [ai] and [ou] alternate with the short vowels [æ], [ε], [] and [ɒ] when followed by two syllables, the first of which is unstressed. [ei] nation [æ] national [i:] extreme [ε] extremity [ai] divine [] divinity [ou] globe [ɒ] globular
b. 0007German Umlaut (vowel fronting): The back vowels and diphthongs a, o, u and au alternate with front vowels ä [ε,æ], ö [ø, œ ], ü [y, y] and äu [ɔy] in certain morphological environments (plural of nouns, past subjunctive of verbs, female-noun suffix -in). Buch ‘book’ Bücher ‘books’ Vater ‘father’ Väter ‘fathers’ (cf. (3.1)) bot ‘offered’ böte ‘would offer’ Jude ‘Jewish person/man’ Jüdin ‘Jewish woman’
c. 0007Russian Zero Alternation: The vowels o/e in the last syllable of the stem sometimes alternate with zero when a vowel-initial suffix follows. zámok ‘castle (nom)’ zámk-i ‘castles (nom)’ (cf. 2.12b) ókon ‘windows (gen)’ okn-ó ‘window (nom)’ zemél’-nyj ‘relating to land’ zemlj-á ‘land’ ogrómen ‘huge (predicative)’ ogrómn-yj ‘huge (attributive)’
d. 0007Hebrew Spirantization (fricativization): The stops p, b, k alternate with the fricatives f, v, x when a vowel precedes. yi-spor ‘he will count’ sofer ‘he counts’ kotev ‘he writes’ yi-xtov ‘he will write’ pilpel ‘he peppered’ me-falpel ‘he peppers’ bakaša ‘request’ be-vakaša ‘please’
e. 0007Turkish k/g˘ alternation: The consonant k alternates with g˘ when a vowel follows. (In standard Turkish, the letter g˘ is no longer pronounced, so yatag˘-ı is [jata], but some non-standard varieties preserve a velar fricative.) inek ‘cow’ ineg˘-i ‘his cow’ kuyruk ‘tail’ kuyrug˘-u ‘its tail’ köpük ‘foam’ köpüg˘-ü ‘its foam’ yatak ‘bed’ yatag˘-ı ‘its bed’
f. 0007Japanese Rendaku (sequential voicing): Morpheme-initial voiceless obstruents alternate with voiced obstruents when a vowel precedes, mostly when they occur initially in a second
214 
C H A P T E R 1 0 â•… M O R P H O P H O N O L O G Y
compound member. (Note that [b] functions as the voiced equivalent of [h].) kami ‘paper’ (iro ‘color’) iro-gami ‘colored paper’ tooroo ‘lantern’ (ishi ‘stone’) ishi-dooroo ‘stone lantern’ shirushi ‘mark’ (hoshi ‘star’) hoshi-jirushi ‘asterisk’ hone ‘bone’ (se ‘back’) se-bone ‘backbone’ chi ‘blood’ (hana ‘nose’) hana-ji ‘nosebleed’ (Vance 1987: ch. 10)
So what makes automatic alternations different from morphophonological alternations? In other words, why do linguists recognize two distinct types? The following empirical characteristics can be used to distinguish between the two. (i) Phonological versus morphological/lexical conditioning. In automatic alternations, the conditions under which the alternations occur can always be described in purely phonological terms. In morphophonological alternations, by contrast, the conditions always have a morphological (and sometimes also lexical) component. For example, English Trisyllabic Shortening is restricted to certain suffixes (e.g. globular versus globalize) and to certain words (e.g. national exhibits Trisyllabic Shortening, but notional does not; the latter is pronounced [noυʃn l], not [nɒʃn l]). Also, Hebrew has many words where k does not undergo Spirantization although the phonological condition, a preceding vowel, is met (e.g. kocer ‘reaps’, yikcor ‘will reap’). In the extreme case, a morphophonological alternation occurs under purely morphological and lexical conditions. This can be illustrated by German Umlaut. This alternation was originally motivated by assimilation to a high front vowel in the following syllable (e.g. Jude/Jüdin), but in most contemporary words, this original front vowel has been lost completely or reduced to schwa, as shown in the examples in (10.3). (10.3) Old High German apful/epfili Apfel/Äpfel [εpfl] kalb/kelbir Kalb/Kälber [kεlbər]
Modern German ‘apple(s)’ (complete loss of final [i]) ‘calf/calves’ (reduction to schwa)
Thus, nowadays this original phonological condition is irrelevant. Umlaut occurs together with certain suffixes such as plural -er (e.g. Buch/Bücher ‘book(s)’). With other suffixes, it occurs only subject to further lexical conditions. With the plural suffix -e, the application of Umlaut has to be learned individually for each lexeme (e.g. Hund/Hunde ‘dog(s)’ versus Bund/Bünde ‘league(s)’). Automatic alternations are the synchronic consequence of phoneticallymotivated diachronic sound changes. Sound change is motivated by phonetics in the sense that it occurs because phonetic production is made
1 0 . 1 T W O T Y P E S O F A LT E R N AT I O N S
215
easier by the change. For example, pronouncing an alveolar or velar consonant before [i] is relatively more difficult than pronouncing a palatal (or palatalized) consonant, and this explains why the diachronic change of palatalization before front vowels is so common in the world’s languages (e.g. (10.1d)). Final devoicing helps pronunciation because maintaining the vibration of the vocal chords (which is made difficult by the oral obstruction of obstruents anyway) is particularly difficult in the final position (e.g. (10.1a)). Neutralization of unstressed vowels occurs for perceptual reasons: when a vowel is not stressed, it is less loud and thus differences between vowels are harder to perceive (e.g. (10.1c)). As in German Umlaut, morphophonological alternations often result when the phonetic motivation for some automatic alternation is subsequently obscured. (ii) Phonetic coherence. Often a whole range of different sounds is affected in a similar way by a sound change, so automatic alternations are more likely to be phonetically coherent in the sense that both the affected sounds and their replacements are natural classes. For example, in Old High German (c. 800–1100 CE), voiced obstruents did occur in syllable-final position, but around 1100 a sound change occurred by which all syllablefinal obstruents became voiceless. As a result, in German, the synchronic rule of Final Devoicing affects all voiced obstruents and turns them into the corresponding voiceless obstruents. Similarly, English Flapping affects all alveolar plosives. In morphophonological alternations, by contrast, the coherence of the set of affected sounds may have been lost by subsequent changes. Thus, the class of English vowels affected by Trisyllabic Shortening is not a natural class; the class of Hebrew consonants affected by Spirantization is not a natural class; and the vowels resulting from German Umlaut are not a natural class (in particular, äu [ɔy], the umlauted counterpart of au [aυ], can be described as ‘fronted’ only with great difficulty). (iii) Phonetic distance. Moreover, in automatic alternations, the alternating sounds tend to differ in one feature only, but in morphophonological alternations they may differ quite drastically. For instance, English [i:]/[ε], [oy]/[ɒ], Turkish [k]/Ø and Japanese h/b show a wide phonetic distance because the sound changes that originally created the alternations occurred a long time ago and subsequent changes have made the connections opaque. (For instance, Japanese Rendaku originally led to p/b alternations, comparable to k/g and t/d alternations, but later p became h.) (iv) Application in derived environments only. Automatic alternations result from constraints on pronunciation that are valid for all environments, and an alternation is just a special case that arises when different morphological contexts provide different phonological conditions. For instance, syllable-final obstruents are always voiceless in German, Russian [o] can never occur in an unstressed syllable and Japanese never allows [t] and [s] in front of [i] (Vance 1987: 21). Morphophonological alternations, by
216 
C H A P T E R 1 0 â•… M O R P H O P H O N O L O G Y
contrast, may be restricted to derived environments. For instance, Turkish [k] is deleted between vowels in a derived environment (see (10.2d)), but inside a morpheme there is nothing wrong with intervocalic [k] (e.g. sokak ‘street’, sitreptokok ‘streptococcus’). In Hebrew, [b] is spirantized to [v] after a vowel, but inside a morpheme there is no problem with [b] (e.g. kibuc ‘gathering; kibbutz’). English long vowels and diphthongs may get shortened when a two-syllable suffix follows (e.g. divine – divinity), but inside a morpheme there is nothing wrong with a diphthong in the antepenultimate syllable (e.g. vitamin [‘vatəmn]). (v) Application to loanwords. Automatic processes generally apply to loanwords and foreign names as they do to native words. Thus, the city Madrid is pronounced with a final [t] in German because of final devoicing; in Russian, not only Moskvá is pronounced with [a] where the spelling has o, but also Mombása and Montána (and in Mogadíšo, the pronunciation is [ə], because o is not immediately before the stressed syllable). In Japanese, loans from English have chi and shi for English [ti] and [si] (e.g. shiisoo from seesaw, shiizun from season). By contrast, the effects of morphophonological alternations need not be found in loanwords. Thus, Turkish loanwords sometimes preserve their final [k] (e.g. sitreptokok ‘streptococcus’, sitreptokoku), and Russian zero alternation is never applied in loanwords (e.g. baron/barony ‘baron’, not *barny). (vi) Speech style and obligatoriness. Automatic alternations may still be optional and sensitive to the speech style. Often, for instance, in formal, slow speech the process is less likely to occur than in informal, fast speech. For instance, English Flapping may be suppressed in formal speech. Morphophonological alternations are never sensitive to the speech style. It should be noted, however, that most automatic alternations that are described in grammars are obligatory as well. Thus, obligatoriness is not a good diagnostic of a morphophonological alternation, but sensitivity to speech style or optionality indicates an automatic process. (vii) New segments. Automatic alternations sometimes create segments that are not found under other conditions. For instance, English [ɾ] only occurs under the conditions of Flapping, and Russian [ə] occurs only under the conditions of Akanie. By contrast, morphophonological alternations tend to lead to segments that occur independently in the language. Thus, German has front vowels like ö and ü in basic words that have nothing to do with Umlaut (e.g. öde ‘bleak’, Mühle ‘mill’), and Hebrew has the fricatives f, v and x in basic words that have nothing to do with Spirantization (e.g. finjan Â� ‘coffee cup’, ʕaxšav ‘now’). (This property of morphophonological alternations is also called structure preservation.) (viii) Application across word boundaries. Automatic alternations may apply across word boundaries. Thus, Flapping occurs in English also within phrases, as in a lot of stuff [ə lɒɾ əv stf]. This is not generally possible with morphophonological alternations.
1 0 . 2 T H E P R O D U C T I V I T Y O F M O R P H O P H O N O L O G I C A L A LT E R N AT I O N S
217
These differences are summarized in Table 10.1. Automatic alternations
Morphophonological alternations
only phonologically conditioned
at least partly morphologically or lexically conditioned
phonetically coherent
not necessarily phonetically coherent
alternants are phonetically close
alternants may be phonetically distant
not contradicted by simple morphemes
may be restricted to derived environments
extend to loanwords
need not extend to loanwords
may be optional and sensitive to speech style
not sensitive to speech style
can create new segments
do not generally lead to new segments
not necessarily restricted to the word level
generally restricted to the word level
Table 10.1╇ Two types of alternations
These empirical criteria divide alternations into two types. The significance of this distinction for the formal architecture of the language system is addressed in Sections 10.4 and 10.5. But first we consider more properties of morphophonological alternations.
10.2 The productivity of morphophonological alternations When we look at morphophonological alternations in greater detail, we see that these show quite a bit of internal diversity. For the sake of convenience, we can distinguish three different classes, although they are probably just three points on a continuum: relic alternations, common alternations and productive alternations. Relic alternations are found only in a few words, and it is therefore doubtful whether a rule should be formulated for them. An example is the s/r alternation in German. This was quite regular in Old High German: in
218 
C H A P T E R 1 0 â•… M O R P H O P H O N O L O G Y
vowel-changing verbs, the past-tense plural forms and the past participle showed r, whereas the other forms showed s: (10.4)
present tense lesan ginesan kiusan friusan
past tense singular las ginas koˉs froˉs
past tense plural laˉrum ginaˉrum kurum frurum
past participle gileran gineran gikoran gifroran
‘read’ ‘be saved’ ‘choose’ ‘freeze’
In Modern German, most of these alternations have been levelled: the modern forms are lesen/las/lasen/gelesen, genesen/genas/genasen/genesen and frieren/fror/froren/gefroren. However, in the high-frequency lexeme be, the alternation was partially preserved (war/gewesen). And, when we take derived lexemes into account, we also see it in Frost/frieren ‘frost/freeze’. In these cases it really takes a historical linguist to discover anything systematic about these alternations. For contemporary speakers, the relation between war ‘was’ and gewesen ‘been’ is probably as suppletive and non-systematic as the relation between bin ‘am’ and war ‘was’. Common alternations are found in many words in a language, and often in different morphological contexts. An example is the Diphthongization alternation in Spanish, whereby ue and ie occur in stressed syllables, and o and e occur in unstressed syllables: (10.5)
ciérro cuénto buéno cuérpo
‘I close’ ‘I tell’ ‘good’ ‘body’
cerrár contár bondád corpóreo
‘to close’ ‘to tell’ ‘goodness’ ‘bodily’
Spanish has dozens of verbs such as cerrar and contar that show this alternation, and there are many derivational relationships such as bueno/ bondad where it shows up as well. So at least as linguists we want to formulate a rule rather than just say that all these cases show (weak) suppletion. And it would seem reasonable to assume that speakers, too, have some kind of rule. However, this is difficult to show, because the Diphthongization alternation is not productive. When a stem with a diphthong becomes the stem of a novel verb (e.g. a verb formed by the denominal pattern des-Nar ‘remove N’), the diphthong appears throughout the paradigm (as in deshuesár ‘remove bones’ from huéso ‘bone’, not *deshosár). When a stem with a monophthong appears in a novel verb, it shows no alternation (e.g. filosofár ‘philosophize’, which has stem-stressed forms such as filosófo ‘I Â� philosophize’, not *filosuéfo). Similarly, when a diminutive in -ito is formed from a noun with a diphthong, the diphthong is preserved (e.g. cuerpíto ‘little body’, not *corpíto, from cuérpo ‘body’). Productive alternations are not merely found in many words, but are also extended to new words such as neologisms and borrowings. German
1 0 . 2 T H E P R O D U C T I V I T Y O F M O R P H O P H O N O L O G I C A L A LT E R N AT I O N S
219
Umlaut is a famous example of such an alternation. In older German, it was productively extended to new plurals such as Mutter/Mütter ‘mother(s)’, Garten/Gärten ‘garden(s)’, which did not have Umlaut in the plural originally because their old plural suffix (now lost) did not contain an [i]. However, in modern German the Umlaut is no longer productive in plurals, and neither is it in female-noun formations of the type Jude/Jüdin (a newly formed female noun from Luchs ‘lynx’ would have to be Luchsin, not *Lüchsin). But there is one pattern in which the Umlaut is required: diminutives in -chen and -lein. For instance, one could form a diminutive Fäxchen from the new word Fax ‘fax’, and parents might refer to a medicine called Vitamnol as Vitamnölchen when talking to a small child. German Umlaut thus demonstrates clearly that a morphophonological alternation may be productive. Some other productive morphophonological alternations are: (i) Turkish k / g˘. This is extended to some loanwords, e.g. kartotek/kartoteg˘i ‘card catalog’, frikik/frikig˘i ‘free kick’, barok/barog˘u ‘baroque’. However, other loanwords preserve k (see the example sitreptokok/sitreptokoku in Section 10.1). (ii) Polish Second Palatalization. This process changes the velars k, g and ch [x] to c [ts], dz [dz] and sz [ʃ] in certain environments – e.g. in the locative singular of nouns of the a-declension (e.g. mucha ‘fly’, locative musze; sługa ‘servant’, locative słudze; matka ‘mother’, locative matce). This alternation is completely productive, and it always applies to loanwords – e.g. Braga (city in Portugal), locative Bradze; alpaka ‘alpaca’, locative alpace, and so on. (iii) Indonesian Nasal Substitution. In this alternation, the initial voiceless stop of a verb root is replaced by a nasal stop at the same place of articulation when the active-voice prefix meng- is attached to the root. In addition to t, k and p, this alternation also affects s, where the replacing nasal is ny []. (The letters ng stand for [ŋ].) (10.6)
meng meng meng meng meng
+ + + + +
urus tulis kirim pakai sewa
mengurus menulis mengirim memakai menyewa
‘take care’ ‘write’ ‘send’ ‘use’ ‘rent’
That this alternation is productive can again be seen in the behaviour of loanwords, which are also subject to Nasal Substitution: (10.7) meng meng meng
+ + +
kritik sukses + kan protes
mengritik ‘criticize’ menyukseskan ‘make successful’ memrotes ‘protest’ (Sneddon 1996: 9–13)
However, in recent borrowings the initial consonant tends to be retained, and, besides the forms in (10.7), the forms mengkritik, mensukseskan and memprotes are possible as well. This perhaps indicates that the alternation is � losing its productivity.
220 
C H A P T E R 1 0 â•… M O R P H O P H O N O L O G Y
These properties of relic alternations, common alternations and productive alternations are summarized in Table 10.2. Relic alternations
Common alternations
Productive alternations
apply to very few items
apply to many items
apply to many items
do not apply to novel words
do not apply to novel words
apply to novel words
probably not 0007recognized by speakers
probably recognized by speakers
clearly recognized by speakers
Table 10.2╇ Three types of morphophonological alternations
The different degrees of productivity exhibited by morphophonological alternations make them quite different from automatic alternations, which always apply to new words.
10.3 The diachrony of morphophonological alternations We have seen that synchronic alternations have their origin in historical sound changes, but we have not yet explained why these sometimes result in automatic alternations and sometimes in morphophonological alternations. On one level, the answer is straightforward: sound changes always yield automatic alternations initially, and automatic alternations then become morphophonological alternations in a further step of development: (10.8) 0007sound change Æ automatic alternation Æ morphophonological alternation Automatic alternations arise because phonetics and phonology are to a large extent autonomous from morphology, or, to put it in even more metaphorical terms, they act blindly, without seeing the consequences of their actions for morphology. If sound changes could, so to speak, predict the outcome of their actions and cared about morphology, they might exercise some restraint. For example, Hebrew Spirantization (see (10.2d)), which turned intervocalic [p] into [f], could have changed non-alternating words like safa ‘lip’ (from earlier sapa), and it could have spared verbs like soper/yispor, which became alternating (sofer/yispor) as a result of the sound change. The fact that this does not happen shows that phonetics and phonology mostly mind their own business and operate without consideration of morphological structure.
1 0 . 3 T H E D I A C H R O N Y O F M O R P H O P H O N O L O G I C A L A LT E R N AT I O N S
221
But this does not explain yet why automatic alternations change into morphophonological alternations. For instance, in early Old High German, the Umlaut must have been an automatic alternation, so that, for instance, Jüdin ‘Jewish woman’ (derived from Jude ‘Jew’) could not have been pronounced otherwise because a back vowel had to be assimilated to a front vowel in the next syllable. But subsequently the phonological restriction that made u-i sequences unpronounceable was lost, and already in Middle High German words like Luchsin ‘female lynx’ were no problem. But why was the alternation retained? Why did Jüdin not revert to its earlier pronunciation Judin? The reason is apparently that speakers had already begun treating the front vowel ü as a property of the word. If Old High German speakers actively derived the word Jüdin each time it was produced by applying the Umlauting rule to an underlying representation [ju:din] (producing the surface representation [jü:din]), then we might expect that loss of the rule would cause the pronunciation of Jüdin to revert to Judin. But this did not happen, and such things do not happen in general. However, perhaps speakers stored the word in the lexicon as the surface form which they heard, or maybe they stored the set of stem allomorphs ([jud-] and [jüd-]), along with a generalization about their distribution. If so, there is no reason that we should expect the effects of automatic alternations to disappear along with the phonological rule. The loss of the phonological restriction thus revealed that speakers had reanalyzed the alternation. Speakers reinterpreted the automatic alternation as signalling (or co-signalling) a particular morphological pattern. An alternation that originally was purely phonologically conditioned came to be morphologically conditioned. This reinterpretation was hidden as long as the phonological rule was in effect, and became observable only upon further change in the language. Another example of an automatic alternation being reanalyzed as a morphophonological alternation comes from Zulu Labial Palatalization in passive verb forms. In Zulu, the passive voice is marked by a suffix -w(a), as illustrated in (10.9a). In (10.9b), we see the effects of Labial Palatalization. (Note that orthographic j = [d], sh = [ʃ], ny = []; otherwise the spelling reflects the pronunciation directly.) (10.9) a. bon-a shay-a
‘see’ ‘beat’
bon-w-a shay-w-a
b.
‘hollow’ ‘take out’ ‘bite’ ‘catch’
guj-w-a ‘be hollowed’ khish-w-a ‘be taken out’ luny-w-a ‘be bitten’ banj-w-a ‘be caught’ (Ziervogel et al. 1981: 106–7, 160, 163)
gubh-a khiph-a lum-a bamb-a
‘be seen’ ‘be beaten’
The original alternation can be summarized as in (10.10). The rule applies wherever these conditions are met.
222 
C H A P T E R 1 0 â•… M O R P H O P H O N O L O G Y
  
(10.10) Labial Palatalization ╇ bh ╇╇j ╇ ph ╇╇sh ╇ m ╇ + w Æ â•‡â•‡ny ╇ mb ╇ nj mp ntsh
Interestingly, Zulu speakers evidently reinterpreted the alternation as cosignalling the passive meaning and introduced it into words that could never have developed palatals by phonological processes. Such words are shown in (10.11). (10.11) khumul-a khumbul-a bophel-a gijimis-a bophis-a
‘loosen’ ‘remember’ ‘harness’ ‘make run’ ‘make fasten’
khunyul-w-a ‘be loosened’ khunjul-w-a ‘be remembered’ boshel-w-a ‘be harnessed’ gijinyis-w-a ‘be made to run’ boshis-w-a ‘be caused to fasten’ (Ziervogel et al. 1981: 106–7, 160, 163)
In all these verbs, the root-final labial consonant is followed by some segments that would have protected it from undergoing Labial Palatalization as a sound change. The fact that it was extended to these cases shows that the alternation was reanalyzed as being morphologically conditioned.
10.4 Morphophonology as phonology With this foundation, we are now ready to address the following questions related to formal description of language architecture: Do automatic and morphophonological alternations reflect morphological structure or phonological structure? How should we describe the interface between morphology and phonology? One position, connected with the morpheme-based model, holds that both automatic and morphophonological alternations are generated by phonological structure.2 In this view, alternations are the result of phonological rules induced by affixation and in formal terms are not therefore considered markers of morphological function or meaning. To see why this approach is appealing, it is useful to look in some detail at one particular proposal of this type, known as level ordering. Level ordering is rooted in the observation that in some languages, it is 2
Linguists who adhere to this position sometimes use phonological alternation as a general term for both automatic and morphophonological alternations, implying that both belong in the same component of grammar. Since this is a controversial issue, we avoid the term in this book in favour of the more neutral (and shorter!) term alternation.
10.4 MORPHOPHONOLOGY AS PHONOLOGY 
223
useful to distinguish between two types of affixes, depending in large part on their behaviour with respect to morphophonological alternations. We call these integrated and neutral affixes here. Their typical properties are summarized in Table 10.3. We begin by briefly examining integrated and neutral affixes in two languages, Lezgian and English. Integrated affixes
Neutral affixes
are in the domain of stress assignment
are not in the domain of stress assignment
trigger and undergo morphophonological alternations
do not trigger or undergo 0007morphophonological alternations
words with integrated affixes 0007show the phonotactics of monomorphemic words
words with neutral affixes may show phonotactic peculiarities
tend to occur closer to the root
tend to occur further from the root
Table 10.3╇ Integrated and neutral affixes
In Lezgian, most inflectional suffixes are neutral, but some are integrated (all prefixes are integrated, but there are so few of them that they can be neglected here). To see the difference between the two types of suffixes, we need to consider the rule of stress assignment and two relevant morphophonological alternations (see (10.12)). (10.12) a. 0007Lezgian Stress Rule: Stress is on the second syllable in the stress domain, if there are at least two syllables in it.
b. 0007Aspirate Ejectivization: A word-final voiceless aspirate consonant (spelled Ch, where C stands for any consonant letter) alternates with an ejective (spelled C’) if the plural suffix follows: meth met’-ér ‘knee(s)’ neth net’-ér ‘louse/lice’ wakh wak’-ár ‘pig(s)’ haqh haq’-ár ‘truth(s)’
c. 0007Vowel Harmony: The stressed syllable and the pre-stress syllable agree in backness and, in the case of high front vowels, labialization – i.e. the only allowed sequences of unlike vowels are a–u, u–a, i–e, e–i, ü–e, e–ü. (Disallowed are a–e, e–u, i–ü, etc.; note that Lezgian has the five vowel phonemes a, e, i, u, ü.) The suffix vowels a/e and i/u/ü alternate:
224 
C H A P T E R 1 0 â•… M O R P H O P H O N O L O G Y
q’al q’ul q’il q’ül
q’al-ár q’ul-ár q’il-ér q’ül-ér
‘stick(s)’ ‘board(s)’ ‘head(s)’ ‘dance(s)’
gˇal cˇ’ul ric’ q’ül
gˇal-úni ‘thread’ cˇ’ul-úni ‘belt’ ric’-íni ‘bowstring’ q’ül-üni ‘dance’ (Haspelmath 1993: 56–8)
The suffixes -er/-ar (plural) and -uni/-ini/-üni (oblique stem) that are illustrated in (10.12) are examples of integrated suffixes. As the examples show, they are in the stress domain (in accordance with the Stress Rule, they receive stress when they attach to a monosyllabic base) and they trigger and undergo morphophonological alternations ((10.12b) and (10.12c), respectively). Besides these, Lezgian also has neutral plural suffixes and neutral oblique-stem suffixes, as illustrated in (10.13). (10.13) a.
Lezgian oblique-stem suffix -di (neutral) fil fíl-di ‘elephant’ tip típ-di ‘type’ nur núr-di ‘beam’ din dín-di ‘religion’
b.
Lezgian plural suffix -ar (neutral)3 tip típ-ar ‘type(s)’ kür kür-ar ‘shed(s)’ kar kár-ar ‘enclosure(s)’ li lí-jar ‘hide(s)’
(Haspelmath 1993: 68–9)
These are not in the stress domain – the stress is on the first syllable in these words, contrary to the Stress Rule – and they neither undergo any alternations (in particular, they are not subject to vowel harmony) nor do they trigger them. Integrated suffixes always follow the root immediately, whereas neutral suffixes may also come after a derivational suffix. For instance, the noun Cˇecˇen-wi ‘Chechen person’ (derived from cˇecˇen ‘Chechnya’) has the plural cˇecˇen-wi-jar and the oblique-stem suffix -di (cˇecˇen-wi-di). Lezgian words with neutral suffixes are immediately recognizable as morphologically complex: consonant sequences like pd (in tipdi) are impossible morpheme-internally. By contrast, if we disregard meaning, all words in (10.12b–c) could be monomorphemic in principle. Let us now look at English, where the distinction between integrated and neutral affixes has occupied many morphologists and phonologists. Some examples of both types of affixes are given in (10.14). (10.14) integrated affixes: -ity, in-, -ical, -ion, -ian, -al, -y1, -ous, -ive neutral affixes: -ness, un-, -ly, re-, -ize, -able, -ful, -y2, -ism 3
The neutral plural suffix -ar has the same shape as one of the alternants of the integrated plural suffix -er/-ar, but it is a distinct suffix.
10.4 MORPHOPHONOLOGY AS PHONOLOGY 
225
Integrated suffixes often lead to a stress shift, whereas neutral suffixes never do: (10.15) 0007base réal cómedy phótograph advántage
with integrated base suffix reálity nátural comédian accómpany photógraphy (-y1) ríckets advantágeous bóunty
with neutral suffix náturalness accómpaniable ríckety (-y2) bóuntiful
Integrated suffixes may also trigger Trisyllabic Shortening (cf. (10.2a)), whereas neutral suffixes never do. The integrated prefix in- shows Nasal Assimilation of the n to the first consonant of the base (elegant/inelegant, but literate/illiterate, regular/irregular), whereas the n of the neutral prefix un- is always preserved (unlimited, unrealistic, etc.). The attachment of neutral affixes may lead to the violation of morpheme-internal phonotactic constraints – e.g. cleanness and unnecessary show two consecutive instances of [n], something that never occurs within a morpheme. Likewise, the suffix -ful brings about consonant sequences such as [pf] (e.g. hopeful) and [kf] (e.g. thankful) that do not occur morpheme-internally. The integrated affixes, by contrast, only create combinations that are independently possible morpheme-internally. And finally, English shows a strong tendency for integrated suffixes to occur close to the root, whereas neutral suffixes occur further away from the root. Integrated affixes do not, as a rule, attach to words derived by a Â� neutral affix (*[hope-ful]-ity, *in-[friend-ly], *[kind-ness]-ical), whereas the opposite order is unproblematic ([natur-al]-ness, un-[product-ive], [Rastafari-an]-ism). The contrast between integrated and neutral affixes in English gave rise to the idea that the innate architecture of the grammar provides the possibility of several levels of affixes that are linked to particular (morpho)phonological rules. English derivation would have two levels, commonly called level I and level II (see Table 10.4). The basic idea is that the levels are ordered relative to each other, and within each level rules introducing affixes are ordered relative to phonological rules: sets of derivational affixes are paired with sets of phonological rules that apply after the affix has been introduced. Level
Affixes
(Morpho)phonological rules
level I: (≈ integrated affixes)
-ity, in-, -ical, -ion, -ian, -al, -y1, -ous, -ive
Trisyllabic Shortening Stress Assignment Nasal Assimilation
level II: (≈ neutral affixes)
-ness, un-, -ly, re-, -ize, -able, -ful, -y2, -ism
Flapping
Table 10.4╇ The two levels of English morpho(phono)logy
226 
C H A P T E R 1 0 â•… M O R P H O P H O N O L O G Y
This architecture requires level I affixes to be attached before level II affixes. In this way, level ordering explains the restriction that prohibits integrated affixes from attaching to words with neutral affixes. Sample derivations are given in (10.16).4 (10.16) nation acid debate Root ‘nejʃn ‘æsd də’bejt Level I Morphology ‘nejʃn-l ‘æsd-ti Level I Phonology Trisyllabic Shortening ‘næʃn-l Stress Assignment æ’sdti Level II Morphology də’bejt-əbl Level II Phonology Flapping æ’sɾɾi də’bejɾəbll Surface [‘næʃnl] [æ’sɾɾi] [də’bejɾəbl] Representation national acidity debatable
create kri’eit kri’eit-v
kri’eitv-nεs kri’eiɾvnεs [kri’eiɾvnεs] creativeness
In addition, it explains why neutral affixes are not affected by morphophonological rules. Note that the word-forms national and chastity exhibit Trisyllabic Shortening. But in debatable and creativeness the rule does not apply, even though both have surface forms that meet the conditions for the rule. In a level ordering account, this is because Trisyllabic Shortening applies at level I, whereas for debatable and creativeness, the conditions for the morphophonological rule are met only at level II. The level ordering approach thus has appealing powers of explanation. Moreover, it has been argued that level ordering has an added advantage in being a restrictive hypothesis about morphological architecture: like other models that posit only concatenative rules, it disallows many types of morphological patterns that are not observed in the world’s languages.5 Not surprisingly, level ordering has been influential among generative morphologists and phonologists. At the same time, when the details are pinned down, application of level ordering to English encounters some serious problems. Some counterexamples to the ordering restriction are obvious and widely recognized: the level I suffix -ity can attach to the level II suffix -able as in readability, and -ation (a variant of -ion) can attach to -ize (e.g. realization). There are also problems with the pairing of affixes and rules. For example, the rule of Velar Softening (which changes underlying [k] into [s], and [g] 4
5
Note that there are multiple senses of the term derive. In Chapter 3 and elsewhere, we have used this term mostly to describe a static relationship between complex word-forms and their bases. Here, however, the term is intended in the sense of ‘derivational phonology’ – constructing surface representations from underlying representations by applying a series of rules. At least, it is restrictive in this way if zero affixes are excluded. See Chapter 3 for a discussion of restrictiveness as a goal in morphological analysis, and problems with zero affixes.
10.4 MORPHOPHONOLOGY AS PHONOLOGY 
227
into [d] before certain suffixes – e.g. electric/electricity) is a clear example of a morphophonological rule that should go with level I affixes. And, indeed, many level I affixes do trigger this rule (e.g. analogous/analogy, music/ musician, opaque/opacity), but there are also two level II suffixes that trigger it, -ize and -ism (e.g. public/publicize, fanatic/fanaticism). Also with respect to stress, integrated and neutral affixes may behave alike: words prefixed with in- and un- both share the same stress pattern, with secondary stress on the prefix (ùnnátural, ùnafráid, ìmprecíse). This stress pattern contrasts with that of monomorphemic words like ínnocent, ímpudent, ínfidel. Thus, in this respect in- behaves as we would expect from a level II prefix (Raffelsiefen 1999b). Now, a few counterexamples do not in general invalidate a generalization, but if the generalization is supposed to be a direct consequence of the architecture of the grammar, counterexamples do become a big problem, because there is no way in which they could arise if the system of Table 10.4 is assumed. Even more damaging to the level ordering hypothesis is the fact that there appears to be an alternative explanation for the observed ordering restrictions. Most integrated affixes in English are quite unproductive anyway, so it seems unnecessary to invoke a level ordering architecture in order to explain why they do not attach to words derived with neutral affixes. The integrated affixes were borrowed along with complex words from French or Latin, and most of them never became truly productive in English. Even the most common suffix, -ity, cannot in general be used with new bases (cf. *chivalrosity, *naturality, ?*effectivity), only in the special case of adjectives derived by -able (readability, bagelizability, etc.). True, within strict limits it is sometimes possible to form new words with the integrated affixes; for instance, use of telescopy, grammophonic and credentious can be found with a Google search (telescopy is by far the most frequent of these three). Still, limited productivity probably goes a long way towards explaining affix ordering restrictions. Factors that affect the saliency of the morpheme boundary may also play a role. Remember that a word with an integrated affix has the same phonotactic structure as a monomorphemic word. The internal morpheme boundary is thus less salient in some sense. According to one proposal, this (among other factors) tends to make it more efficient for an English speaker to store a stem and integrated affix together as a single unit in the lexicon. (See Chapter 4 and Section 6.4.1 for details about lexical storage and its interaction with morphological structure.) And lexical storage, in turn, has consequences for affix ordering: since a stem and integrated affix constitute a unit, integrated affixes must occur closer to the root than neutral affixes (Hay and Plag 2004). Furthermore, in this complexity-based ordering hypothesis, integration is a matter of degree, which helps explain ordering restrictions within the two categories of affixes. More integrated affixes must occur closer to the root than less integrated affixes, and so on. Level
228 
C H A P T E R 1 0 â•… M O R P H O P H O N O L O G Y
ordering incorrectly predicts that there should be no ordering restrictions within a given level. In the following section it is shown that the problems not only concern the hypothesis of level ordering, but in fact extend to the more general claim that morphophonological alternations fall into the domain of phonology.
10.5 Morphophonology as morphology The alternative hypothesis – that morphological alternations properly belong to the domain of morphology – begins from the observation that these alternations behave in ways that are typical of morphological structure more generally. For instance, we have already seen that morphophonological alternations can vary in productivity. This is a typical property of affixes, but not of phonological rules. Alternations can also serve as the basis for back-formation. An example of this comes from Polish. A widespread (and productive) alternation in Polish is the First Palatalization, whose effects are shown in (10.17). (Note that this is somewhat different from the Second Palatalization, which we saw in Section 10.2, and which occurs in different environments.)
 
 
 
(10.17) [k] [tʃ] k cz . [g] Æ [] (spelling: g Æ z [x] [ʃ] ch sz
The First Palatalization occurs, for instance, with the verb-deriving suffix -yc´, with the adjective-deriving suffix -ny and with the diminutive suffixes -ek and -ka: (10.18)
kaleka dynamika pon´czocha kra g
‘cripple’ ‘dynamics’ ‘stocking’ ‘circle’
kaleczyc´ dynamiczny pon´czoszka kra z˙ek
‘mutilate’ ‘dynamic’ ‘little stocking’ ‘little circle’
Polish has a productive pattern of back-forming words from non-diminutive words ending in -ek or -ka. These derivatives get an augmentative interpretation, as in (10.19). (10.19) ogórek szpilka
‘cucumber’ ogór ‘pin’ szpila
‘big cucumber’ ‘big pin’
Now when this rule of subtractive augmentative formation is applied to words ending in -szka or -czka, the result is a new word ending in -cha or -ka:
10.5 MORPHOPHONOLOGY AS MORPHOLOGY 
(10.20)
broszka flaszka gruszka Agnieszka beczka taczka
‘brooch’ ‘bottle’ ‘pear’ (name) ‘barrel’ ‘wheelbarrow’
brocha flacha grucha Agniecha beka taka
229
‘big brooch’ ‘big bottle’ ‘big pear’ ‘big Agnieszka’ ‘big barrel’ ‘big wheelbarrow’
The words in the left-hand column in (10.20) all have [ʃ] (sz) and [tʃ] (cz) originally. For example, broszka was borrowed from French broche [bɔʃ], flaszka was borrowed from German Flasche and gruszka was derived from grusza ‘pear tree’. The ch/k in the back-formed augmentatives is clearly new, and it shows that morphophonological alternations can operate in the reverse direction under certain circumstances. In this respect, these alternations are just like morphological rules. Based on this kind of data, many linguists would say that only automatic alternations are truly phonological, whereas morphophonological alternations are really morphological in nature. This means that morphophonological alternations can signal (or co-signal) morphological meaning. From the perspective of language change, an automatic alternation that becomes closely associated with a morphological pattern is susceptible to being reanalyzed as part of the morphology, and it can then be expected to behave like other ingredients of morphological patterns. For instance, it is possible for a morphophonological alternation to become the sole formal marker of a pattern – e.g. when the original marker disappears for phonological reasons. This has happened in Modern Irish, where the past tense of verbs is marked by Lenition of the initial consonant. Lenition involves fricativization and some other changes and originally it occurred only in intervocalic position (like Hebrew Spirantization (10.2d)). (10.21) Modern Irish Lenition { k, g, t, d, p, b, s, f}╇ Æ╇ { x, , h, , f, w, h, Ø} (spelling: c, g, t, d, p, b, s, f╇ Æ╇ ch, gh, th, dh, ph, bh, sh, fh) (10.22)
present tense molaim brisim sábhálaim díbrím
past tense mhol mé bhris mé shábháil mé dhíbir mé
‘I praise(d)’ ‘I break/broke’ ‘I save(d)’ ‘I banish(ed)’
The past tense was originally formed with a prefix do-, but this was lost, and nowadays only the Lenition is a signal of the past tense (but there are also different person-number markers, -(a)im and mé for first person singular). The result in Modern Irish was thus a base modification pattern, and we saw similar cases earlier in the book: German plurals signalled solely by the Umlaut (e.g. Mutter, Mütter ‘mother(s); see (3.1)), and Albanian plurals signalled solely by Palatalization (e.g. [fik], [fic] ‘fig(s); see (3.3)). The loss of
230 
C H A P T E R 1 0 â•… M O R P H O P H O N O L O G Y
the original affixes makes sense here if the morphophonological alternations co-signalled the relevant morphological meaning, i.e. if they were part of the morphological system of expression. These data seem to further support a morphological approach to morphophonological alternations. Furthermore, as pointed out already in Chapter 3, the existence of base modification without (overt) affixation has led many linguists to conclude that we cannot reasonably reduce the operation of morphology to concatenation – even if morphophonological alternations are factored out of the equation. An adequate description of base modification patterns still requires either non-concatenative operations or zero affixes. And if non-concatenative operations must be posited anyway, nothing is really sacrificed by positing that the same kinds of principles govern morphophonological alternations as well. And since zero affixes are virtually unrestricted theoretical devices, shuttling morphophonological alternations off to the phonology does not clearly lead to a simpler, more restrictive morphology (despite frequent assumptions to the contrary). Theoretical concerns thus also seem to come down on the side of morphology. Now, a word of caution is in order here, lest we think that the data unques- tionably favour the morphological approach. The need for nonconcatenative operations in the morphological system does not entail that all morphophonological alternations are morphology. Adescription of morphology that includes only concatenative rules forces morphophonology to be treated as phonology, but a model that allows non-concatenative morphological operations is, in principle, compatible with both a morphological and a phonological approach to these alternations (although perhaps more naturally aligned with the former). The key question thus becomes: Is phonology ever sensitive to morphological structure? If not, we can safely conclude that morphophonological alternations necessarily belong to the domain of morphology. But if not, we must be more careful in our conclusions. It turns out that there is at least one good candidate for morphologysensitive phonology: phonotactic constraints at morpheme boundaries. Affixation sometimes creates combinations of sounds that are not allowed morpheme-internally (e.g. pd in Lezgian tip-di ‘type’, the consonant cluster [ksts] in English text-s). More violations of morpheme structure conditions were presented in Section 4.2, including a pattern in Standard Northern Italian whereby an alternation between [z] and [s] ([z] between vowels and [s] elsewhere) is violated morpheme initially, e.g. a[s]immetrico ‘asymmetric’. So it seems that phonotactic constraints are sometimes sensitive to morphological structure. And given that phonotactics are considered a core aspect of phonological structure, it may turn out that little is sacrificed by allowing morphologically-conditioned phonological rules, i.e. (some) morphophonological alternations as affix-induced phonology. The proper place of morphophonology has not been conclusively settled. As should be obvious from the discussion in this chapter, analyses of the
FURTHER READING 
231
morphology–phonology boundary depend crucially on a host of other assumptions about how morphology and phonology function. As long as debates in those areas continue, morphophonological alternations will probably continue to be a contentious issue.
Summary of Chapter 10 Two types of alternations can be distinguished: automatic alternations and morphophonological alternations. They differ in a variety of ways: automatic alternations show clear signs of phonetic motivation, may be optional and may apply across word boundaries, whereas morphophonological alternations have lost their connection to phonetics, have morphological conditioning, are obligatory and apply within words. Morphophonological alternations vary in productivity. Diachronically, automatic alternations turn into morphophonological alternations. Some languages have two types of affixes (here called neutral and integrated), depending on the type of alternation that they trigger. It is widely agreed that automatic alternations are phonological in nature, but the status of morphophonological alternations is controversial. A strict morpheme-based model requires morphophonological rules to belong to the phonological component of the grammar. For instance, in a level ordering account, morphophonological and automatic alternations (and neutral and integrated affixes) correspond to different rule blocks that are ordered relative to each other. However, this approach encounters a number of serious theoretical and empirical problems. The alternative hypothesis – that morphophonological alternations belong to the morphological component – is appealing because it can account for commonalities that morphophonology has with morphology.
Further reading Alternations and derivational phonology are discussed in every phonology textbook (e.g. Gussenhoven and Jacobs 2005: ch. 6). The most influential work in derivational phonology is Chomsky and Halle (1968). The view that only automatic alternations are phonological and morphophonological alternations really belong to the � morphology is highlighted in Hooper (1976) and Bochner (1993), among many others. The opposing view is defended in Mohanan (1995) and Kiparsky (1996). A
232 
C H A P T E R 1 0 â•… M O R P H O P H O N O L O G Y
variety of approaches to morphophonology are discussed in the papers in Singh (1996). For diachronic change from phonological to morphophonological rules, see Wurzel (1980) and Joseph and Janda (1988). For the theory of level ordering (also called ‘Lexical Phonology’), see Kiparsky (1982, 1985) and Kaisse and Shaw (1985). Optimality Theory is a constraint-based framework that has been developed to explore some of the same interactions between morphology and phonology that motivated level ordering. Major references include Prince and Smolensky (1993), McCarthy and Prince (1993a, b), and Kager (1999). (Classical OT rejects the principle of level ordering, but it is worth mentioning Stratal Optimality Theory (Kiparsky 2000), which has adopted some of its fundamental insights.) Kiparsky (1982), Fabb (1988), Aronoff and Fuhrhop (2002), Hay and Plag (2004) and Plag and Baayen (2009) investigate affix order restrictions (in English) and their relation to productivity and other factors. The most comprehensive book on morphophonology is Dressler (1985). Important insights on morphophonology are found in Bybee (1985, 2001).
Comprehension exercises 1.
Is the voicing alternation of English fricatives in leaf/leaves knife/knives house/houses, etc.
an automatic or a morphophonological alternation?
2. English has a morphophonological alternation of [ŋ] and [ŋg] – e.g. young [jŋ], younger [jgr]. Is this a relic alternation, a common alternation or a productive alternation? 3. Decide whether the following alternations are automatic or morphophonological, on the basis of the (necessarily incomplete) Â� information given here. a. 0007In Hausa, the alveolars t, d, s, z palatalize to c [tʃ], j [d], sh [ʃ], j [d] when they occur before an original front vowel (Newman 2000: 414–15): kàazaa ‘hen’ kàajii ‘hens’ cìizaa ‘bite’ cìiji ‘bite’ (imperative) Hausa ‘Hausa’ Bàhaushèe ‘Hausa person’ gwadàa ‘measure’ gwajìi ‘experiment’ (deverbal noun)
COMPREHENSION EXERCISES 
233
Recent sound changes have created new cases of ee and i: original form current form ai > ee tàibà > tèebà ‘cooked cassava flour’ oosai > oosee ‘fried beancake’ u > i tukaatukii > tikàatikii ‘calf, shin’
Some English loanwords: laasiisìi ‘licence’ teebur ‘table’ gàzêt ‘gazette’
b. 0007In Spanish, the voiced stops b, d, g alternate with the fricatives [β, ð, ] if a vowel or fricative precedes them: el dedo [el deðo] ‘the finger’ los dedos [loz ðeðos] ‘the fingers’ Damiano viene [damjano βjene] ‘Damiano is coming’ viene Damiano [bjene ðamjano] ‘Damiano is coming’
c. 0007In Modern Greek, the velar phonemes [k], [g], [x], [] alternate with the palatal phonemes [c], [j¯], [ç], [j] whenever they precede a front vowel ([e] or [i]), e.g. 1sg steko exo 2sg stecis eçis 3sg steci eçi 1pl stekume exume 2pl stecete eçete 3pl stekun exun ‘stand’ ‘have’
0007Some loanwords: [cinino] ‘chinine’, [j¯emi] ‘reins’ (from Turkish gem). d. 0007In Polish, the vowel [o] alternates with [u] (spelled ó) in certain morphological forms when the morpheme-final consonant does not start a new syllable, e.g. głowa ‘head.nom.sg’ głów ‘head.gen.pl’ głodu ‘hunger.gen.sg’ głód ‘hunger.nom.sg’ woda ‘water’ wódka ‘vodka’
0007However, there are numerous exceptions to this rule, not just loanwords: spora ‘spore.nom.sg’ spor ‘spore.gen.pl’ kodu ‘code.gen.sg’ kod ‘code.nom.sg’ wódeczka ‘little vodka’ wódka ‘vodka’ 4. 0007We saw that Zulu Labial Palatalization is a morphophonological alternation (and not an automatic alternation), because it is tied to particular morphological contexts. What other criteria can be invoked to support that conclusion?
11.1
Morphology and valence
11
S
o far we have focused our attention primarily on form-related aspects of Â� morphology. But this chapter will be entirely devoted to one type of function of morphological patterns. We will examine various ways in which morphology can affect valence – i.e. the expression of arguments in verbs and deverbal formations. We will first look at valence-changing operations such as passives and causatives (Section 11.1), then move on to the way in which valence is affected by compounding (Section 11.2), and finally discuss what happens to verbal arguments in transpositional derivation (i.e. derivational patterns that change the base’s word-class) (Section 11.3) and transpositional inflection (Section 11.4).
11.1╇ Valence-changing operations 11.1.1╇ Semantic valence and syntactic valence (argument structure and function structure) Most verbs are associated with one, two or three arguments as part of their lexical entries (verbs with zero or more than three arguments are very rare, and many languages lack them completely). When we know a verb’s meaning, we also know the semantic roles (alternatively, thematic relations) of the participants of the verbal event. For example, a verb that means ‘eat’ will have an agent participant (the entity doing the eating) and a patient participant (the thing being eaten) in all languages, a verb meaning ‘please’ will have an Â�experiencer and a stimulus participant, and a verb that means ‘steal, rob’ will have an agent, a theme (the thing that is taken away) and a source participant. But this knowledge is not sufficient if we want to use these verbs, because the syntactic functions (such as subject, object, oblique) by which these participants are expressed differ
1 1 . 1 ╇ VA L E N C E - C H A N G I N G O P E R AT I O N S
235
from language to language and from verb to verb. As a concrete example, the semantic-role structures and the syntactic-function structures of five English verbs are given in (11.1). (11.1) a. eat: sbj –– obj agent patient (Robert ate a mango.) b. like: sbj –– obj experiencer stimulus (I like this song.) c. please: sbj –– obj stimulus experiencer (This song pleases me.) d. steal: sbj –– obj –– agent theme (Baba stole my bike from me.)
oblfrom source
e. rob: sbj –– obj –– agent source (Baba robbed me of my bike.)
oblof theme
The verbs please and like, and the verbs steal and rob, are roughly synonymous, so there is no way to predict their different behaviour from their meanings. Hence speakers must store not only the meaning of every verb, but also the syntactic functions that are associated with the semantic roles. Thus, the lexical entries of the verbs please and rob would look as in (11.2).
 
 
(11.2) a. ╇ /pli:z/v b. ╇╇ /rɒb/v ╇ sbj –– obj sbj –– obj –– oblof ╇ ╇ stimulus experiencer agent source theme ‘please’╇╇ ‘rob’

The information that these entries contain in addition to the Â� pronunciation, the word-class and the meaning is called the valence of the verb. The valence has two parts: the syntactic-function structure (‘syntactic valence’,
236 
C H A P T E R 1 1 â•… M O R P H O L O G Y A N D VA L E N C E
also called simply function structure),1 and the semantic-role structure (‘semantic valence’, also called argument structure).2 The argument structure can in principle be derived from the meaning (or conceptual structure, or event structure) of a verb. For example, a Â� formal decomposition of the meaning of steal or rob looks as in (11.3) (see Jackendoff 1990).
[
]
(11.3) cause ([A], [goposs ([B], [from ([C])])]) {by-force} ╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇
(11.3) can be paraphrased as ‘A causes B to go from C’s possession by force’ – i.e. A robs C of B. The participant A must be an agent because it is the first role of the semantic element cause; the participant B must be a theme because it is the first role of the semantic element go; and C must be a source because it is the participant of from. Thus, it would in principle be possible to formulate the linking rules as direct links between the conceptual structure and the syntactic-function structure. The Â� lexical entry of the verb steal would then be as in (11.4), where there is no separate argument structure.

(11.4) /sti:l/v sbj obj oblfrom ‘cause ([A], [goposs ([B], [from ([C])])])’ {by-force} ╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇╇

Although it is actually quite likely that the format of (11.4) is closer to the truth than the format of (11.2), in the present context a practical problem is that there is much less agreement about the right form of the conceptual decomposition of verb meanings than about semantic roles. Thus, we will mostly continue to use the simplified format of (11.2), bearing in mind that this is just an abbreviation and that the complete picture requires a more elaborate specification of verb meaning along the lines of (11.3). Now morphological operations may change the valence of a verb in two different ways. On the one hand, they may change the linking of semantic roles to syntactic functions. Such operations are called function-changing operations (or voice). On the other hand, they may change the conceptual 1
2
The most important syntactic functions are subject (sbj), (direct) object (obj) and oblique (obl – i.e. adpositional phrases and phrases in oblique cases). Two further functions that are needed less commonly are indirect object (iobj) and adverbial (adv). A syntactic function that is needed for noun phrase structure is possessor (poss). The links between argument-structure positions and function-structure positions (indicated here by lines) are governed by a set of rules that have been extensively discussed by syntacticians and that we cannot go into here. The crucial point is that the rules cannot be entirely independent of knowledge of particular verbs. At least for some verbs such as like and please the function structure must be part of the lexical entry because it is idiosyncratic.
1 1 . 1 ╇ VA L E N C E - C H A N G I N G O P E R AT I O N S
237
structure (or event structure) of the verb in such a way that the argument structure is affected. We will refer to such operations as event-changing operations. Examples of both subtypes of valence-changing operations will be seen in the following sections.
11.1.2╇ Agent-backgrounding operations The best-known function-changing operation is the passive, where the agent is backgrounded in that it is no longer the subject. Instead, the patient usually becomes the subject. English has a passive construction, of course: compare the active sentence Mark wrote a letter with the corresponding passive one The letter was written (by Mark). The passive in English has a complicated form, however, (involving both an auxiliary and a participle), so we will look here at the Chichewa passive, which is more clearly morphological and therefore serves our Â� purposes better (the Chichewa type is far more common in the world’s languages anyway). Examples of an active and a passive sentence from Chichewa are given in (11.5). (11.5) a. Naphiri a-na-lemba Naphiri 3sg-pst-write ‘Naphiri wrote a letter.’
kalata. letter
b. Kalata i-na-lemb-edwa (ndi Naphiri). letter 3sg-pst-write-pass by Naphiri ‘The letter was written (by Naphiri).’ (Dubinsky and Simango 1996: 751–2) In Chichewa, the passive is marked by the suffix -idw/-edw, which is attached directly to the verb stem (the ending -a is a stem extension that need not concern us here). Its syntactic effect is that the patient is linked to the subject function and the agent is linked to the oblndi function. As in English, the oblique agent is optional, as is indicated by the parentheses. Thus, we can formulate the rule for passivization as in (11.6).

 
(11.6) /Xa/v /Xidwa/v sbj –– obj (oblndi) –– ´ agent patient agent ‘dox’ ‘be donex’
sbj patient

Here, all that changes is the phonological form of the verb and the function structure of the verb (as well as the linking to the semantic roles). The argument structure is unaffected – it still specifies both an agent participant and a patient participant. Even when the oblique agent is omitted, it is still present implicitly: the sentence kalata inalembedwa means that some unspecified agent wrote the letter (not just that some agentless letter writing
238 
C H A P T E R 1 1 â•… M O R P H O L O G Y A N D VA L E N C E
took place), as is clear from a sentence like (11.7), where the adverb mwadala ‘deliberately’ presupposes such an agent. (11.7) Chitseko chi-na-tsek-edwa mwadala. door 3sg-pst-close-pass deliberately ‘The door was closed deliberately.’ (Dubinsky and Simango 1996: 751) The passive is thus a prototypical example of a function-changing operation, or voice. A clear example of an event-changing operation is the anticausative, where the agent-backgrounding is much more radical than in the passive. Here, the agent is completely removed from the argument structure. An example comes from Russian, where the anticausative is expressed by the suffix -sja/-s’. (11.8) a. Vera zakryla Vera.nom closed ‘Vera closed the door.’
dver’. door.acc
b. Dver’ zakryla-s’. door.nom closed-antic ‘The door closed.’ (11.9)

 
/X/v /Xsja/v sbj — obj sbj ´ agent patient patient ‘cause ([A], [become ([statex ([B])])])’ ‘become ([statex ([B])])’

In (11.9) we see that not only is the agent removed from the argument structure, but also the cause element is eliminated from the conceptual structure (hence the term ‘anticausative’). It is in this sense that the anticausative is event changing and not merely function changing. The function change (patient becoming subject) is an almost trivial consequence of the main function of the anticausative. That the agent is not present in the argument structure (and in the verb meaning) can also be seen from the fact that it cannot appear as an oblique argument (*Dver’ zakrylas’ Veroj ‘The door closed by Vera’), and no agent-oriented adverbials may occur in the sentence (*Dver’ zakrylas’ namerenno ‘The door closed deliberately’; this sentence is possible only in an unlikely world in which doors have intentions). An even more radical change in the event structure of the verb is effected by the resultative (or stative) operation, which removes not only the ‘cause’ part of the event structure together with the agent, but also the ‘become’ part. An example of a resultative (marked by the suffix -ik/-ek) from Chichewa,
1 1 . 1 ╇ VA L E N C E - C H A N G I N G O P E R AT I O N S
239
which contrasts with the passive in (11.7), is given in (11.10a). The active and resultative event structures are given in (11.10b). (11.10) a. Chitseko chi-na-tsek-eka. door 3sg-pst-close-result ‘The door was closed (= in a closed state).’
b. ‘cause ([A], [become ([closed ([B])])])’ ´ ‘closed ([B])’
As in the Russian anticausative, neither an oblique agent nor an agentoriented adverb is permitted (*Chitseko chinatsekeka ndi Naphiri ‘The door was in a closed state by Naphiri’; *Chitseko chinatsekeka mwadala ‘The door was in a closed state deliberately’) (Dubinsky and Simango 1996: 751). Note that an interesting feature of the anticausative and resultative operations is that they are semantically subtractive – i.e. the derived form removes part of the conceptual structure of the base. Finally, another example of a valence-changing operation is the reflexive, where the agent and the patient are coreferential and hence can be thought of as occupying a single syntactic function. Examples of an active and a reflexive verb in Eastern Armenian are given in (11.11a) and (11.11b), respectively, and the rule is given in (11.12).3 (11.11) a. Mayr-ə lvan-um mother-art wash-prs ‘Mother is washing Seda.’
e aux
b. Seda-n lva-cv-um Seda(nom)-art wash-refl-prs ‘Seda is washing (herself).’
e. aux

 
Seda-yi-n. Seda-dat-art
(Kozinceva 1981: 83)

(11.12) /Xnum/v /Xcvum/v ╇ sbj –– ╇ obj sbj ╇╇ ╇╇ ´ agent patient ╇ agenti patienti ‘A actsx on B’╇╇╇╇ ‘A actsx on self’╇╇╇╇╇
In the reflexive voice, the meaning of the verb remains the same, but it is specified that the agent and the patient are coreferential (indicated in the right-hand word-schema in (11.12)). Thus, although the reflexive is not really event-changing, its effect is not strictly limited to function changing either. It is thus a borderline case between the two subtypes of valencechanging operations.
3
Coreferentiality is conventionally indicated by subscript letters. So, the subscript i in both agenti and patienti in (11.12) indicates that the agent and patient refer to the same entity. Different subscript letters are used when two participants are not coreferential.
240 
C H A P T E R 1 1 â•… M O R P H O L O G Y A N D VA L E N C E
11.1.3╇ Patient-backgrounding operations Antipassive is the term for a morphological operation whose effect is to background the patient in much the same way as the agent is backgrounded in the passive. An example of an active and an antipassive construction from West Greenlandic is shown in (11.13a–b). Note that the oblique patient is marked by the instrumental case in West Greenlandic. The relevant part from the antipassive rule is given in (11.14). (11.13) a. Qimmi-p inu-it dog-erg.sg person-abs.pl ‘The dog killed the people.’
tuqup-pai. kill-3sg.sbj/3pl.obj.ind
b. Qimmiq (inun-nik) dog(abs) person-inst.pl ‘The dog killed (people).’
tuqut-si-vuq. kill-antip-3sg.ind


(11.14) sbj — obj ´ agent ╇ patient

(Fortescue 1984: 86, 206)
sbj — agent

(oblinst) patient
Now we might ask whether there is also a patient-backgrounding operation that completely removes the patient from the argument structure (parallel to how the agent is removed from the argument structure in the anticausative). And, indeed, some languages have a valence-changing affix whose effect is that the patient cannot be expressed at all. We may call this operation deobjective. An example comes from Tzutujil. (11.15) a. x-Ø-uu-ch’ey pst-3sg.obj-3sg.sbj-hit ‘he hit him’ b. x-Ø-ch’ey-oon-i pst-3sg.sbj-hit-deobj-pst ‘he was hitting’
(Dayley 1985: 89, 116)
(11.15b) is an intransitive verb in all respects: it has the suffix -i in addition to the prefix x- in the past tense (cf. x-eel-i ‘he went out’, contrasting with x-uuch’ey in (11.15a) where there is no -i), it has only a single person–number prefix for the subject, and it does not allow a patient to be expressed. However, it is unlikely that (11.15b) has a different event Â� structure from (11.15a), because it is difficult to conceive of a hitting event without a patient participant – for hitting to occur, there must be something that is being hit. In anticausatives, agents can be eliminated from the event structure because the ‘cause’ element is eliminated: we can think of opening, breaking and similar events as occurring either through an external agent or spontaneously, but we
1 1 . 1 ╇ VA L E N C E - C H A N G I N G O P E R AT I O N S
241
cannot easily think of such events as occurring without a patient. Thus, the most likely valence-changing effect of the deobjective is that shown in (11.16). The crossed linking line above ‘patient’ means that this semantic role cannot be linked to any syntactic function.

 
(11.16) sbj — obj ´ agent patient

sbj ¥ agent patient
Thus, patient-backgrounding operations seem to be exclusively function changing.
11.1.4╇ Agent-adding operations: causatives When a new participant is added to a verb, the event structure must be enriched as well, so the causative is clearly an event-changing operation. Two examples of causative constructions from Japanese are given in (11.17b)–(11.18b), and (11.17c)–(11.18c) show the valence-changing rules. (11.17) a. Taroo ga Taro nom ‘Taro goes.’
ik-u. go-prs
b. Hanako ga Taroo Hanako nom Taro ‘Hanako made Taro go.’ c.


sbj ´ agent

o acc
ik-ase-ta. go-caus-pst (Shibatani 1990: 308–10)
sbj — obj causer agent

(11.18) a. Taroo ga hon o yom-u. Taro nom book acc read-prs ‘Taro reads a book.’ b. Hanako ga Taroo ni hon Hanako nom Taro dat book ‘Hanako made Taro read a book.’ c.


sbj — obj ´ agent patient

o acc
yom-ase-ta. read-caus-pst
(Shibatani 1990: 310)
sbj — iobj — obj causer agent patient

The semantic change in the event structure is obvious: it consists in adding the element ‘cause’ and with it a causer role (e.g. for ‘go’: [go ([A])] ´
242 
C H A P T E R 1 1 â•… M O R P H O L O G Y A N D VA L E N C E
[cause ([D], go ([A]))]). The linking of semantic roles to syntactic functions in causatives is complicated because languages cannot simply create a new syntactic function for the new role. Instead, causative verbs are made to fit into the existing function structures. The agent of an intransitive verb becomes an object as in (11.17b–c), but the agent of a transitive verb often becomes an indirect object (as in 11.18b–c), especially in languages that do not allow two equal objects. Causatives are probably the most common type of morphological valence-changing operation in the world’s languages, but since they happen to be rare in Europe, linguists have often paid more attention to the agentbackgrounding constructions that are so common and varied in Europe.
11.1.5╇ Object-creating operations: applicatives The applicative operation creates a completely new object in the function structure of the verb or shifts a non-object to the object function. An example of the latter kind comes from German, where the productive verbal prefix be- can have the effect of turning an indirect object into a direct object. The Â� original direct object can be omitted or expressed as an oblique phrase. Example (11.19b) is the applicative construction corresponding to (11.19a). (11.19) a. IKEA liefert dem Nachbar-n die IKEA delivers the neighbour-dat the ‘IKEA delivers furniture to the neighbour.’
Möbel. furniture.acc
b. IKEA be-liefert den Nachbar-n (mit Möbeln). IKEA appl-delivers the neighbour-acc with furniture ‘IKEA delivers furniture to the neighbour.’

 
c. sbj –– obj –– iobj sbj (oblmit) obj ´ agent patient recipient agent patient recipient

This construction is called a recipient applicative because it is the recipient that becomes a direct object, and therefore receives accusative case. Crosslinguistically, almost all roles apart from the agent can become direct objects when an applicative marker is added to the verb. An example of a locative applicative from Ainu is shown in (11.20b). (11.20) a. A-kor kotan ta 1sg-poss village to ‘I arrived at my village.’
sirepa-an. arrive-1sg.intr
b. A-kor kotan a-e-sirepa. 1sg-poss village 1sg.tr-appl-arrive ‘I arrived at my village.’ (Shibatani 1990: 65)
1 1 . 1 ╇ VA L E N C E - C H A N G I N G O P E R AT I O N S

 
c. sbj –– adv sbj –– obj ´ agent location agent location
243

Ainu has no case marking, but the subject-agreement marker a-, which is restricted to transitive verbs, clearly shows that the applicative prefix e- Â� creates a direct-object function in the derived verb’s function structure. However, an applicative may also add an object argument that was not in the function structure of the verb before. For example, Chamorro has a benefactive applicative, illustrated in (11.21b). (11.21) a. Ha hatsa i acho’. he.erg lift abs stone ‘He lifted the stone.’ b. Ha hatsa-yi si Pedro ni acho’. he-erg lift-appl abs Pedro obl stone ‘He lifted the stone for Pedro.’ (Topping 1973: 253) Thus, here the applicative adds a new participant (a beneficiary) to the argument structure: (11.21) c.


sbj — obj ´ agent patient

sbj — obl — obj agent patient beneficiary

In the non-applicative construction, the patient is associated with the syntactic function of the object (indicated by the absolutive case in (11.21a)), but in the applicative construction the new argument of the verb – the beneficiary – is linked to the object function, and so it receives absolutive case. Correspondingly, the patient is linked to the oblique. This means that applicatives can be either function changing (as in (11.19) and (11.20)) or event changing (as in (11.21)). One might propose that these two subtypes of applicatives should be given different names, but it is in fact not so easy to keep them apart. One might argue, for instance, that the ‘location argument’ of the Ainu verb sirepa is not in fact an argument but an adjunct; this would make it event changing as well. Moreover, some languages use the same affix for both benefactive and recipient applicatives, suggesting that this is indeed the same kind of operation. Thus, a sharp distinction between event-changing and function-changing operations can be problematic.
11.1.6╇ General properties of valence-changing operations As we have seen, valence-changing operations primarily affect agents/ subjects and patients/direct objects. Other participants can be promoted
244 
C H A P T E R 1 1 â•… M O R P H O L O G Y A N D VA L E N C E
to object (or occasionally to subject) status, but there are no operations that change an oblique to an indirect object, for example. Explaining such possible restrictions on valence changing is a matter for syntactic and semantic theories of verbal event structure and argument linking, and thus largely beyond the scope of this book. However, here it still needs to be pointed out that the semantic/syntactic contrast between event-changing and function-changing operations shows a clear correlation with derivational and inflectional status of the valencechanging affixes. Passives and antipassives are primarily inflectional, whereas anticausatives, resultatives and causatives are primarily derivational. Reflexives and applicatives tend to show mixed behaviour, again correlating with their intermediate status with respect to the eventchanging/function-changing contrast. Since inflectional operations may apply after derivational ones, but the reverse is not typically true (see Chapter 5), an important consequence of this contrast is the prediction that it should be possible to apply a functionchanging operation to an event-changing operation, but not vice versa. This seems generally to be true. For example, in Chichewa the passive suffix -idw (a function-changing operation; see (11.5) above) can be attached to a benefactive applicative verb in -ir. (11.22) a. Chibwe a-na-phik-ir-idwa nyemba. Chibwe 3sg-pst-cook-appl-pass beans ‘Chibwe was cooked beans for.’ (Dubinsky and Simango 1996: 752) b. active:

benefactive applicative:
passive:
sbj — obj sbj — obj — oblØ (oblndi) sbj oblØ ´ ´ agent patient agent beneficiary patient agent beneficiary patient

 

The reverse ordering is not possible in Chichewa, although it would make sense semantically (cf. 11.23a). However, the applicative suffix can follow the resultative suffix, as in (11.23b), because the applicative and the resultative are both event-changing operations. (11.23) a. *Chitseko chi-na-tsekul-idw-ira Chibwe. door 3sg-pst-open-pass-appl Chibwe ‘The door was opened for Chibwe.’ b. Chitseko chi-na-tseku-k-ira Chibwe. door 3sg-pst-open-result-appl Chibwe ‘The door was opened (= in an opened state) for Chibwe.’ (Dubinsky and Simango 1996: 757)
1 1 . 2 ╇ VA L E N C E I N C O M P O U N D I N G
245
Valence-changing operations are in many ways syntactic phenomena, and in languages where they are signalled by specific morphological patterns, they also clearly belong to morphology. However, most of the operations that we have seen in this section also occur with no specific formal coding. For instance, English has alternations such as (11.24)–(11.25). (11.24) a. b. (11.25) a. b.
I opened the door. The door opened. I baked a cake for her. I baked her a cake.
The alternation in (11.24) clearly resembles the anticausative and the causative operation, and (11.25) is very much like a benefactive applicative. The English alternations are not usually discussed under the heading of morphology, but there is really no deep reason why they should not. Morphological operations need not be associated with a particular change in the pronunciation, as we saw earlier (e.g. Section 3.1.4). When they are not, morphologists speak of conversion, and, while this term is mostly applied to word-class-changing operations, it could easily be transferred to valencechanging operations. Note also that such valence-changing operations may vary in productivity, from sporadic to extremely productive, much like other morphological processes.
11.2╇ Valence in compounding When one of the members of a compound takes arguments, this may be affected by the compound structure and the result may be a kind of valence change. We will look at valence change in three different compound types: noun incorporation, V–V compound verbs and synthetic compounds.
11.2.1╇ Noun incorporation Noun incorporation is the traditional term for N–V compounds with a verbal head; we have encountered examples already in Sections 7.1 and 9.1. Since verbs typically require arguments, it is natural for incorporated nouns to occupy an argument position of the verb. Consider (11.26) from Guaraní. (11.26) a. A-jogua-ta petei mba’e. 1act-buy-fut one thing ‘I will buy something.’ b.
A-mba’e-jogua-ta. 1act-thing-buy-fut ‘I’ll go shopping.’ Lit: ‘I’ll thing-buy.’
(Velázquez-Castillo 1996: 107)
246 
C H A P T E R 1 1 â•… M O R P H O L O G Y A N D VA L E N C E
In incorporated structures like (11.26b), the dependent noun is clearly part of the compound verb, and not a separate word. For instance, as shown in (11.27), the incorporated noun cannot be modified by an adjective. (The different placement of the future marker in (11.26) and (11.27) is not directly relevant here.) This indicates that the dependent noun is not an independent word, but rather a member of the compound. (See Section 9.1 for discussion of compounding tests.) (11.27) *A-ha-ta a-mba’e-hepy-jogua 1act-go-fut 1act-thing-expensive-buy ‘I’ll go shopping for expensive items.’ Lit: ‘I will go expensive-thing-buying.’ (Velázquez-Castillo 1996: 108) Now, the interesting thing here is that in (11.26b) the noun is also the patient of the verbal action. This raises the question: In noun incorporation, is specification of valence necessarily part of the compounding rule? One possibility would be to simply say that the semantic relation between the head verb and the dependent noun is vague, as in English N–N compounds (e.g. lipstick). The patient interpretation (‘buy things’) would then be a natural implicature, but not strictly speaking part of the compound verb’s meaning. If this is so, we would expect incorporated nouns to be able to fulfil other semantic roles besides the patient, and this is indeed possible in quite a few languages with noun incorporation. Example (11.28) is from Huauhtla Nahuatl. (11.28) Ya’ ki-koccˇillo-tete’ki panci. he 3sg.obj-knife-cut bread ‘He cut the bread with the knife.’
(Merlan 1976: 185)
Here, the incorporated noun -koccˇillo- ‘knife’ is the instrument of the action. Thus, it may be that the noun incorporation rule in these languages does not affect the syntax of the verb at all, and that the valence change is only apparent. Still, languages seem to differ in this respect. In many languages there is clear morphosyntactic evidence for a valence-changing effect of noun incorporation. This is the case, for example, in Ainu, which has different subject-agreement affixes in transitive and intransitive sentences, as we saw earlier in (11.20) (e.g. -an for first person singular intransitive, a- for first person singular transitive). (11.29) a. Inaw a-ke. inaw 1sg.tr-make ‘I make an inaw (a wooden prayer symbol).’
1 1 . 2 ╇ VA L E N C E I N C O M P O U N D I N G
b. Inaw-ke-an. inaw-make-1sg.intr ‘I make an inaw.’
247
(Shibatani 1990: 11, 28)
In contrast to the transitive simple verb ke ‘make’, the compound verb inaw-ke ‘make an inaw’ is intransitive, as is clearly seen in the choice of the subject affix. Thus, it is not sufficient to say that the patient interpretation in (11.29b) arises as a pragmatic implicature – here, it must be part of the compounding rule, which can be formulated as in (11.30).
 

(11.30) /X/n /Y/v /XY/v sbj — obj sbj & ´ agenti patientj agenti ‘x’ ‘A Aii acts ‘Ai acts on x’ acts on Bj’

As this rule shows, the patient variable of the semantic structure of the simple verb is filled by the meaning of the incorporated noun, so that the semantic structure of the compound verb contains only a single variable and hence only a single argument. As in the case of the reflexive voice (Section 11.1.2), we have here a borderline case between event changing and function changing.
11.2.2╇ V–V compound verbs A compound type that is rarely found in European languages, but that is very interesting from the point of view of valence, is V–V compounding. Two well-known languages in which such compounds are common are Chinese and Japanese. The simplest and least problematic case involves two verbs with the same argument structure – e.g. Japanese ukare-sawagu [make.merry-be.noisy] ‘go on a spree’, Mandarin Chinese tang-huai [iron-break] ‘ruin by ironing’. Example (11.31) shows how the Chinese verb is used. (11.31) Meimei tang-huai le nei jian xin yi. sister iron-break prf that clf new clothes ‘Sister ruined those new clothes by ironing them.’ (C. H. Chang 1998: 82) The rule for Chinese tang-huai could be formulated as in (11.32).4
4
Here we use a simplified representation because we are only interested in the way that elements of argument structure combine in compounding. Naturally, in a full representation, function structure and links between it and argument structure would also be specified.
248 
C H A P T E R 1 1 â•… M O R P H O L O G Y A N D VA L E N C E

 
 

(11.32) /X/v /Y/v /XY/v agenti patientj & agenti patientj ´ agenti patientj ‘Ai actsx on Bj’ ‘Ai actsy on Bj’ ‘Ai actsx and actsy on Bj’ ╇╇╇╇╇╇╇╇╇╇
However, both Chinese and Japanese allow verbs with different argument structures to be compounded as well. In Japanese, where compounds are usually right-headed, it is mostly the second verb that determines the argument structure of the compound. An example is given in (11.33), and the correspondence is shown in (11.34). (11.33) Sono booru wa sora takaku (*Jon ni yotte) uchi-agat-ta. the ball top sky high (*John by hit-go.up-pst ‘The ball was hit high up in the air (by John).’ (Matsumoto 1996: 204)

 
 
(11.34) /uchi/v /agaru/v /uchiagaru/v agenti patientj & themei directionk ´ themej directionk ‘Ai hits Bj’ ‘Aj goes up to Bk’ ‘Aj is hit upwards to Bk’ ╇╇╇╇╇╇╇╇╇╇

In uchi-agaru [hit-go.up] ‘be hit high up in the air’, the first verb is transitive and the second is intransitive (with an additional direction argument), and the theme of the head verb is identified with the patient of the dependent verb. The agent of the dependent verb completely disappears from the argument structure, as is shown by the fact that it cannot be expressed as a kind of passive agent. The head verb agaru contributes its arguments to the compound. The association of intransitive theme and transitive patient is very natural (both of these semantic roles are affected by the processes in which they are involved), but an intransitive theme may also be identified with an intransitive agent: (11.35) a. Japanese hataraki-tsukareru [work-get.tired] ‘get tired from working’ tatakai-yabureru [battle-lose] ‘lose as a result of fighting’ (Matsumoto 1996: 204) b. Chinese zou-lei xiao-jiang
[walk-get.tired] ‘get tired from walking’ [laugh-stiff] ‘be stiff from laughing’ (C. H. Chang 1998: 83)
1 1 . 2 ╇ VA L E N C E I N C O M P O U N D I N G

 
 
249

(11.36) /hataraki/v /tsukareru/v /hataraki-tsukareru/v agenti ´ themei & themei ‘Ai works’╇╇ ‘Ai gets tired’╇ ‘Ai gets tired from Ai working’
Perhaps the most interesting type of V–V compound is the argumentmixing type, in which the compound verb’s argument structure has arguments from both constituent verbs. An example is Japanese mochi-kaeru [have-return] ‘bring back’ – see (11.37) and the correspondence in (11.38). (11.37) Jon wa kamera o ie ni mochi-kaet-ta. John top camera acc house to have-return-pst ‘John brought the camera back home.’ (Matsumoto 1996: 208) (11.38) /mochi/v /kaeru/v /mochikaeru/v possessori themej & agenti directionk ´ agenti themej directionk ‘Ai has Bj’╇╇╇╇╇╇╇ ‘Ai returns to Ck’╇╇ ‘Ai brings Bj back to Ck’╇

 
 

In this compound verb, all arguments of the constituent verbs end up as arguments of the compound verb. In this way, mochikaeru contrasts with uchiagaru. Clearly, Japanese V–V compounding consists of different Â� subrules in which the argument linking is crucially different. The final case to be mentioned here is the ambiguous type represented by Chinese qi-lei [ride-tired]. This can mean two different things: (11.39)
Zhangsan qi-lei le ma. Zhangsan ride-tired pfv horse a. ‘Zhangsan was tired from riding horses.’ b. 0007‘The horse was tired from Zhangsan’s riding/Zhangsan rode the horse tired.’ (C. H. Chang 1998: 82)
Thus, here the theme argument of lei ‘tired’ can be identified either with the agent or with the patient of qi ‘ride’.
11.2.3╇ Synthetic nominal compounds A compound whose head is a deverbal noun and whose dependent member fills an argument position in the head’s valence is often called a synthetic compound. Examples from English include N–N compounds of the following type:
250 
C H A P T E R 1 1 â•… M O R P H O L O G Y A N D VA L E N C E
(11.40) a.
truck driver b. pipe smoker air-cleaner fire-fighter
whale hunting rat poisoning duck-shooting meter-feeding
The deverbal head noun inherits the verb’s valence requirements. Thus, the noun driver can be analyzed as taking a patient argument, like its base verb drive, and this argument is filled by truck, the dependent member of the compound. Likewise, the noun hunting can be analyzed as taking a patient argument, which is filled by whale. (Argument inheritance is discussed in Section 11.3 below.) There are at least three different ways in which the semantic relationship between the head and the dependent member in a synthetic compound could be described. The first approach derives argument interpretation from a constituent structure that is different from that of ordinary N–N compounds. Consider the compounds in (11.41). (11.41) a.
race driver b. dog hunting chain smoker food poisoning vacuum cleaner gang shooting freedom fighter breast feeding (some examples from Oshita 1995: 183, 189)
While formally similar to the examples in (11.40), these are not synthetic compounds because the dependent member does not fill an argument of the head. A chain smoker does not smoke chains (cf. pipe-smoker), a freedom fighter does not fight freedom (cf. fire-fighter), and food poisoning does not involve poisoning food (cf. rat poisoning). In this approach, the structure of pipe-smoker would be as in (11.42a), contrasting with that of chain smoker in (11.42b). (11.42) a. b. N
V
Nsuff
N
N N
N V V Nsuff
pipe smoke -er chain smoke -er
An argument in favour of having different structures for synthetic and non-synthetic compounds is the fact that, in more complex compounds, a dependent noun that is interpreted as an argument must be closest to
1 1 . 2 ╇ VA L E N C E I N C O M P O U N D I N G
251
the deverbal head: chain pipe smoker, but *pipe chain smoker. The constituent structure analysis provides a natural account of the ungrammaticality of the second example. However, the approach also makes some incorrect predictions. If synthetic compounds have the constituent structure [[N V]V -er]N, it is difficult to explain the systematic lack of N–V compounds in English: pipe-smoker, but *to pipe-smoke; hat-seller, but *to hat-sell; churchgoer, but *to churchgo; cabdriving, but *to cab-drive. And the few instances of N–V compounding that do exist seem to have been created by back-formation from the corresponding synthetic compounds, not directly as N–V compounds. For instance, to skydive is a backformation from skydiving and skydiver. The second approach to synthetic compounds involves a special rule of argument linking, analogous to the incorporation rule above in Section 11.2.1 (ex (11.30)). Let us assume that the noun hunting has the argument structure [agent patient], just like the verb hunt, and the function structure [possessorof — obliqueby] (e.g. hunting of whales by traditional fishermen). Then the compound whale hunting eliminates the patient/possessorof argument and the resulting compound is ‘intransitive’ – i.e. it takes only a single obliqueby argument (e.g. whale hunting by traditional fishermen). The complete rule is shown in (11.43).



(11.43) /X/n /Y/n /XY/n poss ╇ — oblby oblby & ╇╇╇ ´ ╇╇ patientj agenti agenti ‘x’╇╇ ‘event of Ai actingy ‘event of Ai actingy on Bj’ ╇╇╇╇╇╇╇╇╇╇╇ on x’ ╇╇╇╇╇╇╇╇╇╇╇╇
This approach is less radical than the first approach in that it does not assume a completely different compounding structure for synthetic compounds. (11.43) is an instantiation of the general English compounding rule (3.30), being merely more specific in that it specifies what happens to the arguments and the syntactic functions. This seems necessary, at least for action nouns like hunting, because the possibilities of associating semantic roles and syntactic functions are severely restricted (for instance, we cannot have *fisherman hunting of whales). Finally, the simplest approach is to deny that any special rule is needed at all. In this view, compounds like truck-driver and whale hunting are described as ordinary N1–N2 compounds that do not mean more than ‘N2 that has some relation to N1’. This is the approach that we took in Section 7.1 for other types of compounds (consider the semantic relationship of the dependent to the head in lipstick vs. sea bird vs. swansong). In truck-driver, this meaning (‘driver who has some relation to a truck’) is then naturally interpreted as ‘driver who drives a truck’ by a pragmatic implicature. Similarly, whale hunting really means only ‘hunting that has some relation to
252 
C H A P T E R 1 1 â•… M O R P H O L O G Y A N D VA L E N C E
whales’, but a natural pragmatic implicature gives rise to the interpretation ‘hunting in which whales are hunted’. This analysis does not imply that individual compounds cannot acquire the argument interpretation. But it does mean that unlike in the second approach, there would be no general rule to account for argument interpretation in synthetic compounds. This approach has the advantage of being able to capture similarities to other types of compounds. We can think of synthetic compounds as a special type of what are called affix compounds here. Affix compounds are patterns that consist of more than one stem plus an affix. In addition to the examples in (11.41) above (chain smoker, food poisoning), English also has affix compounds of the following type: (11.44) green-eyed dark-haired red-roofed
‘having green eyes’ ‘having dark hair’ ‘having a red roof’
Given that in affix compounds, the affix often attaches to a base that is not itself a compound (*a green-eye, *to pipe-smoke, *to race-drive), the descriptions in (11.45) seem preferable to any kind of description that relies on hierarchical constituent structure for semantic interpretation. (Note that it is possible to say that someone has a green eye, but this is a phrase, not a compound – there is primary stress on both lexemes.)
    

(11.45) a. /X/a & /Y/n ´ /XYd/a ‘x’ ‘y’ ‘having (a) y(s) with the property x’╇╇╇╇
    

b. /X/n & /Y/v /XYer/n ‘x’ ‘y’ ´ ‘A person who does y, having to do with x’
This approach utilizes the general rule of English compounding, but unlike the second approach, it leaves interpretation of the relationship between the head and the dependent to pragmatic implicature. This allows for a unified description of the formal similarities between synthetic and other types of affix compounds. Of course, not directly specifying the semantic relationship between the head and the dependent in synthetic compounds also creates the potential for overprediction. This approach has difficulty accounting for semantic restrictions of the kind *fisherman hunting of whales. Thus, there are arguments for all three approaches, and it is possible that different approaches are appropriate for different compounds or for different languages (synthetic compounding has been most intensively discussed for English). Other things being equal, it would of course be desirable to have just a single type of rule, but it remains to be seen whether
1 1 . 3 ╇ T R A N S P O S I T I O N A L D E R I VAT I O N
253
other things are in fact equal. Thus far, no consensus has formed about how to formally describe synthetic compounds.
11.3╇ Transpositional derivation 11.3.1╇ Transposition and argument inheritance A derivational process is called transpositional when it changes the word-class of the base lexeme. Some typical examples of transpositional derivation are shown in (11.46). (11.46)
a. b. c. d. e. f.
N V A V A N
Æ Æ Æ Æ Æ Æ
V N V A N A
English Russian Basque Italian Japanese Indonesian
computer Æ computerize napolnit’ ‘fill’ Æ napolnenie ‘filling’ luze ‘large’ Æ luza-tu ‘lengthen’ mangiare ‘eat’ Æ mangiabile ‘edible’ atarashii ‘new’ Æ atarashisa ‘newness’ tahun ‘year’ Æ tahunan ‘annual’
Valence may be affected by transposition when a verb or an adjective is transposed into another word-class (non-derived nouns normally cannot be said to have a valence potential, so transpositional derivations of nouns are hardly relevant here). When a verb such as examine is transposed into an action noun such as examination, its basic meaning (referring to an event with an agent and patient) is still intact, but the arguments cannot be expressed in the same way as with the base verb. We can say The vet� examined the pet, but not *[The vet examination the pet]np (took one hour). This is because deverbal nouns behave much like ordinary nouns in that they do not take subject and object arguments, but only possessor and oblique � arguments. Thus, we have The examination of the pet by the vet (took one hour). The patient argument becomes an of-possessor, and the agent argument becomes a by-oblique. The resulting noun phrase is similar to noun phrases with non-derived noun heads such as the portrait of Charles V by Titian. The relation between the valences of examine and examination can be described with our usual notation as in (11.47).

 
(11.47) ╇ /examine/v ╇ /examination/n ╇ sbj — obj ╇ (oblby) — possof ´ agent patient agent patient

In transpositional derivation, when a derived word has a valence that corresponds to the valence of the base, we say that the derivative inherits the base’s valence.
254 
C H A P T E R 1 1 â•… M O R P H O L O G Y A N D VA L E N C E
In the following subsections, we will take a closer look at various kinds of transpositional derivation.
11.3.2╇ Action nouns (V Æ N) Perhaps the most interesting type of transpositional derivation is the action noun (or event noun) – one that refers to the event or action itself, not to a participant of the event – because action nouns show the greatest variety of argument structure phenomena both within a language and across languages. In English and other European languages, two different types of event noun can be distinguished, the simple event noun (e.g. (11.48a)) and the complex event noun (e.g. (11.48b)). (11.48) a. I have an examination tomorrow. b. The vet’s careful examination of Fido’s eyes took a long time. c. The examination is on your desk. The basic difference between them is that complex event nouns preserve more verbal properties than simple event nouns. Sometimes a third type of event noun is distinguished, called concrete noun, and illustrated in (11.48c). However, this is not really an event noun, because it does not refer to an event. But it is necessary to mention this type in the present context because this is a widespread phenomenon: in many languages, the derivational patterns used for action nouns can also have concrete meanings. However, the kinds of concrete meanings are unpredictable: the product of an action (building, painting, judgement, composition), a group of people (management, government) or a manner (conjugation). Concrete nouns seem to arise by illunderstood and unsystematic (though frequent) processes of metonymic meaning shift, not by a word-formation rule, so we need not discuss them further. Returning to simple and complex event nouns, we note that when the verbal arguments are expressed with a complex event noun, it must be definite (see (11.49b)) and cannot be pluralized (see (11.49c)). (11.49) a. the examination of Fido’s eyes by the vet b. *an examination of Fido’s eyes by the vet c. *three examinations of Fido’s eyes by the vet Simple event nouns are more like ordinary nouns in that they can be indefinite or definite (Tomorrow I have an/the examination), and they can be pluralized (Tomorrow I have three examinations). Moreover, complex event nouns can be modified by duration adverbs like frequent and constant, whereas simple event nouns cannot (cf. the frequent examination of Fido’s eyes/*a frequent examination). But, in the present context, the most important
1 1 . 3 ╇ T R A N S P O S I T I O N A L D E R I VAT I O N
255
difference between complex event nouns and simple event nouns is that only the former inherit the verb’s argument structure. Thus, for complex event nouns, the function-changing transposition rule in (11.47) is Â� appropriate, whereas, in simple event nouns, the argument structure is not preserved. As a result, simple event nouns may occur on their own, with no arguments expressed, as in (11.50). (11.50) a. The examinations took a long time. b. We are witnessing a new development. c. The destruction was awful to see. By contrast, complex event nouns derived from transitive verbs require the overt expression of the patient, while the agent may be optionally present, as seen in (11.51).5 (11.51) a. 0007The frequent examination *(of the evidence) (by the scientists) isÂ� necessary. b. 0007The constant development *(of new inexpensive housing) (by the city) was applauded. c. 0007The continuing destruction *(of rainforests) (by humans) will speed up desertification. In some languages, complex event nouns have an argument structure that is even more verb-like in that the patient is coded as an accusative NP. An example comes from Modern Hebrew. (11.52) ha-hafcaca ha-tedira šel ha-cava the-bombing the-frequent of the-army ‘the army’s frequent bombing of the city’
et acc
ha-ʕir the-city
(Siloni 1997: 170)
In English, only oblique arguments coded by a PP and clausal arguments may be retained in an action noun construction (e.g. they rely on her Æ their reliance on her; they elected Maria as president Æ their election of Maria asÂ� president; I predict that it will rain Æ my prediction that it will rain).
11.3.3╇ Agent nouns (V Æ N) and deverbal adjectives (V Æ A) An agent noun is one that refers to the agent of the action, rather than the action itself. In contrast to (complex) event nouns, agent nouns in English and in many other languages do not seem to inherit the verb’s argument
5
An asterisk before an expression in parentheses means that the expression cannot be left out.
256 
C H A P T E R 1 1 â•… M O R P H O L O G Y A N D VA L E N C E
structure. Expressions such as *voter for Mitterrand, *thinker about deep problems or *claimer that Armageddon is near are systematically impossible. However, it is, of course, possible to have a possessive phrase that serves the same purpose as a verbal argument: explorer of Antarctica, founder of Lund University, Mitterrand’s voters, and so on. One could see this as evidence that to some extent the verbal argument structure may be inherited after all, but a Â� simpler account is available: possessive phrases have a very general meaning, and often the precise interpretation is left to pragmatic inferences from the context, as in the case of compounds (see Section 11.2). Given the meaning of an agent carrying out some action, the interpretation of a Â� possessive phrase as a patient of that action is readily available, so we do not need to say that it arises as a result of argument inheritance. This view is also confirmed by the fact that agent nouns, unlike complex event nouns, do not admit an agent-oriented adverbial such as a purpose clause: (11.53) a. *an explorer of America in order to discover El Dorado b. the exploration of America in order to discover El Dorado Thus, the rule for deriving an agent noun of a transitive verb would be as in (11.54), where the derived noun lacks an argument structure.

 
(11.54) /X/v ╇ /Xer/n sbj — obj ╇ ´ ╇ agenti ╇ patientj ‘Ai actsx on Bj’╇╇╇ ‘person who actsx’

(Note that if this is the right analysis, we have to revise what we said about synthetic compounds above. The idea that agent nouns do not have an argument structure is incompatible with the second approach to describing synthetic compounds that was presented in Section 11.2.3 (see (11.43)). Perhaps we ought to say that the third approach to synthetic compounds outlined in that section, which rests on pragmatic inference, is appropriate for agent nouns and other deverbal formations that lack an argument structure, whereas the second approach is appropriate for complex event nouns.) English deverbal adjectives in -able seem to be similar to agent nouns in that they do not generally inherit oblique or clausal arguments from the base verb (*convincible of the eventual success, *emptiable of water, *persuadable that I’m right, but cf. deductible from income tax).
11.3.4╇ Deadjectival transposition (A Æ N, A Æ V) Adjectives are much less often associated with their own argument and function structure, but many languages have at least a few argument-
1 1 . 4 ╇ T R A N S P O S I T I O N A L I N F L E C T I O N
257
taking adjectives (such as English proud of, full of, similar to, obedient to, different from, responsible for, ready to do something). In English, most of these oblique arguments are preserved in deadjectival quality nouns (similarity to, obedience to, responsibility for, readiness to do something, ??difference from), though in some cases we have idiosyncratic changes (pride in, not *pride of). In deadjectival verbs, the oblique argument may also be preserved. The examples in (11.55)–(11.56) show that in Russian the adjectival argument structure is inherited. The adjective gordyj ‘proud’ takes an instrumental oblique argument, and so does the deadjectival verb gordit´sja ‘to pride oneself (on)’. The adjective gotovyj ‘ready’ takes an infinitival argument, and so does the deadjectival verb gotovit´sja ‘get ready’. (11.55) a. On gord svoimi dostiženijami. he proud self’s.inst achievements.inst ‘He is proud of his achievements.’ b. On gorditsja svoimi dostiženijami. he proud.3sg.refl self’s.inst achievements.inst ‘He prides himself on his achievements.’ (11.56) a. On gotov vyexat´ iz he ready leave.inf from ‘He is ready to leave the country.’
strany. country
b. On gotovitsja vyexat´ iz strany. he get.ready.3sg.refl leave.inf from country ‘He is getting ready to leave the country.’ A counterexample would be English fill, which does not behave like full (cf. full of, fill with). A difficulty in determining whether the adjectival argument structure is inherited is the fact that the choice of the preposition or oblique case that marks the adjectival argument is rarely completely arbitrary. In many cases, it could be argued that the choice of the preposition or case is determined semantically and is independent of the base adjective.
11.4╇ Transpositional inflection A particular challenge for morphologists and syntacticians is the description of transpositional (word-class-changing) inflection. In transpositional inflection, not just some, but all of the argument structure of the base is preserved, plus its other combinatory possibilities. An inflectional V Æ A transposition is called a participle in many languages (see (11.57)
258 
C H A P T E R 1 1 â•… M O R P H O L O G Y A N D VA L E N C E
from German), and an inflectional V Æ N transposition is called a masdar in some languages (cf. example (11.58) from Lezgian). (11.57) der im Wald laut pfeif-end-e Wanderer the in.the forest loud whistle-ptcp-m.sg hiker ‘the hiker who is whistling loud in the forest’ (11.58) Wun fad qaragˇ-un-i cˇun tažub iji-zwa. you.abs early get.up-masd-erg we.abs surprise do-impf ‘That you are getting up early surprises us.’ (Haspelmath 1993: 153) A less well-known example of word-class-changing inflection is the Hungarian proprietive (‘having’, N Æ A): (11.59) rendkívül nagy hatalm-ú uralkodó extremely great power-propr monarch ‘monarch with extremely great power’ (Kenesei 1995–96: 164) The participle is similar to the deverbal adjective (Section 11.3.3), but note that it also inherits the possibility to combine with a locative modifier (im Wald ‘in the forest’) and a manner modifier (laut ‘loud’). The masdar is similar to the action noun, but it preserves the verbal valence completely: in (11.58), the agent argument is in the absolutive case, and in this respect it is very different from a noun’s modifier or argument. Moreover, (11.58) also shows that the masdar is like a verb, not like an action noun, in that it can combine with an adverb (cf. the behaviour of English action nouns: *My perusal carefully of the article versus my careful perusal of the article). The Hungarian proprietive is similar to denominal adjectives like powerful, but, unlike such adjectives in English, Hungarian proprietives can take prenominal modifiers that only nouns can take. This suggests that, if we want to describe the syntactic behaviour of participles, masdars and proprietives (and other inflectional transpositions not mentioned here), instead of invoking a mechanism of inheritance from the base lexeme, we should say that we do not have a new lexeme here at all but an inflected word-form of the same lexeme. Participles and masdars are verbs, and Hungarian proprietives are nouns. Combined with their dependents (i.e. their arguments and modifiers), they yield verb phrases and noun phrases:
1 1 . 4 ╇ T R A N S P O S I T I O N A L I N F L E C T I O N
(11.60) German
VP
PP P
AdvP
im Walde laut in.the forest loud ‘whistling loud in the forest’
NP
AdvP
wun fad you.abs early ‘you rising early’
Adv
pfeif-end whistle-ptcp
VP
(11.62) Hungarian
V
NP
(11.61) Lezgian
259
AP
V
qaragˇun-i get.up-masd
NP N
A
rendívül nagy hatalm-ú extremely great power-propr ‘having extremely great power’ If we want to account for the behaviour with respect to their dependents, this description of these constructions is unexceptionable, but now we face a paradox: we have just said that participles, masdars and proprietives do not change the word-class of their base, although at the beginning of this section we said that they were examples of word-class-changing inflection.
260 
C H A P T E R 1 1 â•… M O R P H O L O G Y A N D VA L E N C E
And, of course, there are good reasons for saying that a participle is an adjective. For instance, in German it shows exactly the same agreement inflection as adjectives, and it precedes the noun in an NP. There are also good reasons for saying that the Lezgian masdar is a noun: it shows nominal case inflection and occurs in the same syntactic environment as non-derived nouns. The Hungarian proprietive, too, is adjective-like with respect to its position and its pluralization. A possible solution to this paradox is the following. Participles, masdars and proprietives show dual behaviour – they act like verbs, verbs and nouns with respect to their dependents, but like adjectives, nouns and adjectives with respect to the other elements in the sentence. We conclude from this dual behaviour that they have a dual nature: a lexeme word- class and a word-form word-class (Haspelmath 1996). As a lexeme, a participle is a verb, just like the other verb forms. But, as a word-form, a participle is an adjective. The internal syntax of a word is determined by its lexeme word-class, and the external syntax of a word is determined by its word-form word-class. Let us now see how we could describe the external syntax of the phrases in (11.60)–(11.62). One possibility would be to assume a structure as in (11.63) for the German phrase in (11.57). (11.63)
NP
D
AP
N
VP
der the
PP
Adv
V
im Wald in.the forest
laut loud
pfeif- whistle-
A
-end-e Wanderer -ptcp-m.sg hiker
This representation has two disadvantages. First it makes the claim that the participle pfeifende belongs to two different syntactic constituents, although usually one assumes that a unitary word-form must also be a unitary syntactic constituent. Second, it works only for transpositional formations that are characterized by affixes. Participles such as Hebrew šorek ‘whistling’ behave just like German pfeifend, but they cannot be represented as in (11.63) because they have no participial affix – the participle is Â� signalled by the vowel pattern o-e (cf. the past tense šarak of this verb).
1 1 . 4 ╇ T R A N S P O S I T I O N A L I N F L E C T I O N
261
An alternative proposal that does not have these disadvantages is to indicate the dual word-class membership in the syntactic trees. A participle can be represented as a word-syntactic tree as in (11.64a), contrasting with a derivational transpositional form such as an agent noun, given in (11.64b). (11.64) a. pfeifend ‘whistling’
b. Pfeifer ‘whistler’
〈〈V〉A〉
V pfeif
A end
N
V Pfeif
N er
In (11.64a), the lexeme word-class is given in the inner angled brackets, and the word-form word-class is given in the outer angled brackets. Thus, in inflectional transposition, properties of the word-class of both constituents are preserved. By contrast, in derivational transposition, the derivative has primarily the head’s word-class properties. If such dual-word-class representations are admitted in the syntax, we get (11.65), where the phrasal node dominating pfeifend also has dual category membership. The notation ‘〈〈V〉A〉P’ can be read as ‘VP with respect to internal syntax, AP with respect to external syntax’. (11.65)
NP
〈〈V〉A〉P
D
der the
PP
Adv
im Wald in.the forest
laut loud
N
〈〈V〉A〉
pfeifende Wanderer whistle.ptcp hiker
The difference between transpositional inflection and transpositional derivation is interestingly similar to the difference between event-changing and function-changing operations that we saw in Section 11.1. Eventchanging operations are generally derivational and involve a change in the argument structure of the base, like most transpositional derivation. Function-changing operations are generally inflectional and involve no change in the argument structure of the base, like transpositional inflection.
262 
C H A P T E R 1 1 â•… M O R P H O L O G Y A N D VA L E N C E
The main difference is that function-changing operations of course change syntactic functions, whereas in prototypical transpositional inflection no functions are changed. Moreover, it should be recalled that the difference between eventchanging and function-changing operations is not always clear-cut, and we often find intermediate cases. Transpositional operations are no different. Some inflectional forms do require some limited function changing – e.g. English masdar-like expressions of the type Maria’s criticizing Robert, the guest’s arriving late, where the verb’s subject is coded not as a subject but as a prenominal possessor. On the other hand, derivational formations in some languages allow the expression of adverbials. Examples (11.66a–b) are from Spanish (adverbials: hoy ‘today’ and todavía más ‘even more’), and example (11.67) is from Modern Greek (adverbial: prosektiká ‘carefully’). (11.66) a. la inauguración hoy en Barcelona del Congreso the inauguration today in Barcelona of.the Congress ‘the inauguration today in Barcelona of Congress’ b. la caída de los precios todavía más the falling of the prices still more ‘the falling of the prices even more’
(Rainer 1993: 214)
(11.67) i katastrofí ton engráfon prosektiká the destruction the.genpl documents.genpl carefully ‘the destruction of the documents carefully’ (Alexiadou 1999: 19) However, this blurring of the boundaries between word-class-changing inflection and derivation is not surprising if we remember what we said in Chapter 5 – that there is reason to view the boundary between inflection and derivation generally as a continuum, rather than a dichotomy.
Summary of Chapter 11 The most interesting inflectional values and derivational meanings are those that affect the valence of the base: valence-changing operations, some types of compounding and transpositional derivation (in transpositional inflection, the base’s valence remains unaffected). Valence-changing operations may be event changing (i.e. the event structure of the base and therefore its argument structure are modified) or function changing (i.e. only the function structure of the base is modified). The most important valence-changing operations
COMPREHENSION EXERCISES 
263
are passive, reflexive, anticausative, resultative, antipassive, causative and applicative. In compounds, valence is potentially affected if at least one of the bases is a verb (as in incorporation and V–V Â� compounding) or a deverbal derivative (as in synthetic nominal compounds). Transpositional derivatives such as action nouns and agent nouns inherit the base’s valence to a greater or lesser extent. In transpositional inflection, the base’s valence is completely preserved, but, in order to arrive at a coherent description, one needs to differentiate between a word’s lexeme word-class and word-form word-class.
Further reading For syntactic theories that are deeply concerned with semantic valence (argument structure) and syntactic valence (function structure), see Dik (1997), Van Valin and LaPolla (1997) and Bresnan (2001). Passive morphology is discussed in Haspelmath (1990). For antipassives, see Cooreman (1994), for resultatives, Nedjalkov (1988), and for causatives, Dixon (2000). An overview of noun incorporation is given in Mithun (1984), and see Mithun and Corbett (1999) for noun incorporation and valence. Japanese compound verbs are discussed in Matsumoto (1996); for Chinese compounds, see Packard (2000). Synthetic compounds are discussed lucidly in Oshita (1995). For action nouns, see Koptjevskaja-Tamm (1993) and Grimshaw (1990). Transpositional inflection is discussed in Haspelmath (1996).
Comprehension exercises 1. Formulate the morphological rule for adjectives of the type supportive of (derived from the verb support) (cf. Sections 11.3.3–11.3.4), analogous to the rule in (11.47). 2. English has one kind of verbal valence-changing prefix that can be regarded as an applicative marker, the prefix out-, as in run outrun play outplay shine outshine
Formulate the rule for out-, stating how the function structure, the argument structure and the meaning are affected.
264 
C H A P T E R 1 1 â•… M O R P H O L O G Y A N D VA L E N C E
3. The phrase ruler over a large empire is accepted by many speakers of English. Which generalization of this chapter is the phrase a counterexample to?
12.1 ASYMMETRIES IN INFLECTIONAL VALUES 
265
12
Frequency effects in morphology
T
he ways in which speakers use language have a profound influence on language structure, and frequency is one of the most important sources for system-external explanation of language structure. In fact, we have already seen examples in this book of frequency affecting the content of the lexicon (Section 4.3), productivity (Sections 6.4.1–6.4.2) and word-class shift (Chapter 8). In this chapter we explore how frequency matters for language structure, and why. Frequency influences word structure in many ways, but one of the most striking effects is found in inflection, where frequency asymmetries result in asymmetrical structural behaviour of various kinds. We look at some examples of this interaction.
12.1 Asymmetries in inflectional values In inflectional systems, we often observe asymmetries in the behaviour of inflectional values that belong to the same inflectional feature, including number (singular versus plural), case (nominative versus accusative), voice (active versus passive) and polarity (affirmative versus negative).
12.1.1 Frequent and rare values Frequency differences among some common inflectional values are summarized in Table 12.1, where ‘>’ means ‘is more frequent than’. It should be noted that not every word in a language will exhibit these frequency asymmetries. These generalizations should instead be taken as describing the overall pattern of a language. Also, not all languages have inflection for all these features, but the claim is that, when a language has inflection for one of these features and values, it will conform to the generalization expressed in the table.
266 
C H A P T E R 1 2 â•… F R E Q U E N C Y E F F E C T S I N M O R P H O L O G Y
Feature
Values, ordered by frequency
number case person degree voice mood polarity tense
singular > plural > dual nominative > accusative > dative 3rd > non-3rd (1st/2nd) positive > comparative > superlative active > passive indicative > subjunctive affirmative > negative present > future
Table 12.1╇ Frequent and rare values
The generalizations in Table 12.1 can be illustrated by examining the results produced by counting usage of number values in six languages: (12.1) French Latin Russian Sanskrit Slovene Upper Sorbian
Singular 74.3% 85.2% 77.7% 70.3% 72.5% 64%
Plural Dual Number of nouns 25.7% ╇ 1,000 14.8% ╇ 8,342 22.3% ╇ 8,194 25.1% 4.6% 93,277 26.9% 0.6% 11,711 30% 6% unknown
(Corbett 2000: 281; data partly from Greenberg 1966: 31–2)
The differences between languages that we see here could be due to slight differences in the meanings of the number values, or they might simply be due to the genre or style of the text chosen. Ideally, the token frequency of inflectional values should be counted in a text that is representative of the everyday spoken language in the community, and finding such representative texts is not straightforward. But, fortunately, the asymmetries in the usage of number values are so robust that the same result is generally obtained, no matter which texts we look at. So we can safely say that the singular has a higher frequency than the plural. Why should such frequency asymmetries exist? To start with, the nominative can be expected to be more frequent than the accusative, at least in languages that do not allow unexpressed arguments, because all verbs require a nominative argument (i.e. a subject), but only transitive verbs also have an accusative. Similarly, the subjunctive must be rarer than the indicative because subjunctives are used primarily in subordinate clauses, and a subordinate clause presupposes a main clause with, at least typically, an indicative verb. However, the ultimate reason for the different frequencies
1 2 . 1 A S Y M M E T R I E S I N I N F L E C T I O N A L VA L U E S
267
of different inflectional values is outside language. Some expressions are more frequent simply because humans find them more relevant: we all talk more about singular entities than about plural entities, more about third persons and things than about speech act participants (first/second person), more about present events than about future events, and so on. The linguist has no privileged skills for explaining these preferences, so we will not discuss them further. Instead, we will focus on structural properties that correlate with frequency.
12.1.2 The correlation between frequency and shortness Quite generally, frequent expressions tend to be short in human languages. Frequent words are shorter than rare words. For example, in French the 10 most frequent words are de, le, la, et, les, des, est, un, une, du, and long words like éléphant or questionnaire are used rarely. Even more strikingly, frequently used inflectional values may not be expressed overtly at all but are left to be inferred from the context – i.e. they sometimes show zero expression. This is just one more manifestation of the correlation between frequency and shortness. As an example, consider the partial inflectional paradigm of regular nouns in Udmurt, given in (12.2). (12.2) nominative accusative ablative abessive
singular val valez valle´s valtek
plural valjos valjosty valjosle´s valjostek
‘horse(s)’ ‘horse(s) (dir. obj.)’ ‘from the horse(s)’ ‘without the horse(s)’ (Perevošcˇikov 1962: 86–7)
In this paradigm, the more rarely used cases, ablative and abessive, have a longer form than the more frequently used accusative. The nominative and the singular are the shortest: they are both expressed by zero. This Udmurt paradigm is quite typical of inflectional systems. Zero expression is found in frequent values, and when two contrasting values are both overtly coded, typically the more frequently used value has the shorter expression. Two more examples from verbal inflection are given in (12.3) and (12.4). (12.3)
Tzutujil 1sg 2sg 3sg 1pl 2pl 3pl
completive x-in-wari x-at-wari x-wari x-oq-wari x-ix-wari x-ee-wari
incompletive n-in-wari n-at-wari n-wari n-oq-wari n-ix-wari n-ee-wari
potential xk-in-wari xk-at-wari xti-wari xq-oo-wari xk-ix-wari xk-ee-wari (Dayley 1985: 87–8)
268 
C H A P T E R 1 2 â•… F R E Q U E N C Y E F F E C T S I N M O R P H O L O G Y
(12.4)
Kobon 1sg 2sg 3sg 1du 2/3du 1pl 2pl 3pl
present ar-ab-in ar-ab-ön ar-ab ar-ab-ul ar-ab-il ar-ab-un ar-ab-im ar-ab-öl
future ar-nab-in ar-nab-ön ar-nab ar-nab-ul ar-nab-il ar-nab-un ar-nab-im ar-nab-öl
conditional ar-bnep ar-bnap ar-böp ar-blop ar-blep ar-bnop ar-bep ar-blap (Davies 1981: 166, 181)
Both these paradigms show zero expression in the third person singular. The Tzutujil paradigm shows that the non-indicative form (called ‘potential’) has a longer marker than the indicative forms, and the Kobon paradigm shows a longer marker for future tense than for present tense. The conditional mood in Kobon is marked by the two consonants b and p, so it is longer than the present indicative form, which has just a single consonant (this assumes that consonants are more important in counting length than vowels).
12.1.3 The correlation between frequency and differentiation In three different senses, frequently used values tend to be more differentiated than rarely used values.1 First, frequent values show less syncretism than rare values. Consider the partial paradigm of the Old English verb bindan ‘bind’ in (12.5). (12.5) 1 sg 2 sg 3 sg 1–3 pl
present ind binde bintst bint bindaþ
present sbjv binde binde binde binden
past ind band bunde band bundon
past sbjv bunde bunde bunde bunden
This paradigm shows that there is more syncretism in the plural than in the singular (in fact, all plural forms of all verbs are syncretized in Old English), more syncretism in the subjunctive than in the indicative, and more syncretism in the past indicative than in the present indicative. The same tendency is found in Khanty possessive suffixes:
1
Generalizations about a correlation between frequency of use and linguistic structure are based on strong tendencies, but should not be treated as inviolable principles. For every generalization there are counterexamples.
1 2 . 1 A S Y M M E T R I E S I N I N F L E C T I O N A L VA L U E S
(12.6) 1st 2nd 3rd
singular -ēm -ēn -l
plural -ēw -lən -ēl
dual -ēmən -lən -lən
269
(Nikolaeva 1999: 14)
Here, syncretism is found in the rarest of the three number values, the dual, and in one of the rarer person values, second person. (More syncretism in the dual can also be seen in Kobon (see (12.4) above)). More syncretism in the passive than in the active voice can be exemplified from Gothic (niman ‘take’). (12.7) active passive singular plural singular plural 1st nima nimam nimada nimanda 2nd nimis nimiþ nimaza nimanda 3rd nimiþ nimand nimada nimanda The active has five different shapes, and the passive has only three. The second sense in which frequent values are more differentiated is that inflection classes differ primarily with respect to the frequent values, less so with respect to rare values. In other words, the classes have fewer shared exponents in the frequently used values. This can be seen in Russian noun inflection. The endings of the four most important Russian inflection classes are shown in (12.8) (the inflection classes are labelled I–IV, as is traditional. See Exercise 5 of Chapter 8 for full word-forms belonging to each of these classes). (12.8) singular IV nom acc
I
-o
III
II -a
Ø
gen
-a
dat
-u
loc
-e
instr
plural
-om
-u -i
IV
-e
-ju
-oj
III
-a Ø
-i
I
II
-i -ov
-ej
Ø
-am -ax -ami
The contrast between singular and plural is clear: in the singular, there are at least twelve distinct endings, while in the plural there are at most eight. And, at least in the plural, the rarer cases (dative, locative, instrumental) show fewer allomorphs than the more frequent cases. Likewise in Standard
270 
C H A P T E R 1 2 â•… F R E Q U E N C Y E F F E C T S I N M O R P H O L O G Y
Arabic, transitive verbs belong to one of four inflection classes, characterized by different vowels before the final stem consonant. However, in the rarer passive voice the inflection is uniform and the difference between the inflection classes disappears (see (12.9)). (12.9) a-u: a-i: i-a: a-a:
active passive perfect imperfect perfect imperfect qatala yaqtulu i-a: qutila yuqtalu d.araba yad.ribu i-a: d.uriba yud.rabu ˉhafiz. a yah ˉ faz. u i-a: ˉh ufiz. a yuh ˉ faz. u jamaʕa yajmaʕu i-a: jumiʕa yujmaʕu
‘kill’ ‘hit’ ‘protect’ ‘gather’
The third sense in which frequently used values are more differentiated is that they tend to show more cross-cutting values. For example, as we saw in Section 5.1, the Latin future tense lacks a subjunctive mood (or one could alternatively say that the subjunctive mood lacks a future tense). In (12.10), we again see the third person singular of the verb laudare ‘praise’. (12.10) indicative subjunctive
present tense laudat laudet
past tense laudabat laudaret
future tense laudabit —
Lack of cross-cutting values is similar, but not identical to syncretism. In Latin, the distinction between indicative and subjunctive is not neutralized in the future tense. The form laudabit (‘she will praise’) expresses only the indicative, and future tense cannot be expressed directly in the subjunctive.
12.1.4 Local frequency reversals Table 12.1 shows the frequency asymmetries that hold in general in languages. However, in particular lexemes, the frequency relations may be reversed. For instance, while most nouns (such as ‘table’, ‘head’ or ‘doctor’) occur more often in the singular than in the plural, a small group of nouns tend to occur more often in the plural in many, if not all, languages. These are nouns referring to some paired or multiple body parts (‘eyes’, ‘lips’, ‘hair(s)’), small animals (‘ants’, ‘fish’, ‘mice’), small parts of plants (‘beans’, ‘strawberries’, ‘leaves’), and some others (‘sand grains’, ‘splinters’). In the case feature, nouns that denote a place occur in the locative case more often than in the nominative, in contrast to other nouns. And, while the greater relative frequency of the nominative case is clearly true of animate nouns that may occur as subjects of transitive clauses, it is not so clear that inanimate nouns, which are typically patients, are also used more frequently in the nominative than in the accusative case. Local frequency reversals may also be found in particular cross-cutting
1 2 . 1 A S Y M M E T R I E S I N I N F L E C T I O N A L VA L U E S
271
values. While in general the third person is more frequent than the second person, in the imperative mood this relation is reversed: commands are more often addressed to the person who is supposed to carry them out, and indirect imperatives (with the subject in the third person) are rare in all languages. Structural effects of these frequency reversals can be observed in many languages. In Welsh, plurals are normally marked by suffixes as in other Indo-European languages (cath/cathod ‘cat/cats’, draenog/draenogod ‘hedgehog/hedgehogs’; see (8.10) for more examples). However, in certain nouns that are used frequently in the plural, it is the singular that is marked by a special suffix: (12.11)
dail pysgod ffa cacwn mefus tywys
‘leaves’ ‘fish (pl)’ ‘beans’ ‘wasps’ ‘strawberries’ ‘corn’
deilen pysgodyn ffäen cacynen mefusen tywysen
‘leaf’ ‘fish (sg)’ ‘bean’ ‘wasp’ ‘strawberry’ ‘ear of corn’ (King 1993: 67–9)
There are also languages in which case marking is found only in inanimate nouns (or non-personal pronouns). Godoberi is such a language, and here it is the (transitive) subject case that is overtly marked, whereas the directobject case is zero (12.12).2 Presumably, this is connected to the fact that inanimate nouns are more likely to be used as objects than as subjects. (12.12) (transitive) subject case direct-object case
den-Ø den-Ø
‘I’ ‘me’
hanqu-di ‘house’ hanqu-Ø ‘house’ (Kibrik 1996: 119, 36)
And in the imperative, the second person form is often zero while the third person form is overtly marked (e.g. Latin second person imperative lauda ‘praise!’, third person imperative laudato ‘let him/her praise!’). Local frequency reversals occur in derivational morphology as well. In many languages female person nouns are derived by a special affix from the corresponding male or general person noun – e.g. Dutch handelaar ‘(male) merchant’, handelaarster ‘female merchant’, Hausa àbookii ‘(male) friend’, àbookìyaa ‘female friend’. From the point of view of the semantics, it would be equally possible to have a special affix that denotes male persons, but such affixes seem to be extremely rare. One reason for this asymmetry is probably that, in most societies, men tended to have more specialized
2
Instead of the familiar terms ‘nominative/accusative’, the terms subject case/object case are used here, because overtly marked subject cases are usually called ‘ergative’ rather than ‘nominative’.
272 
C H A P T E R 1 2 â•… F R E Q U E N C Y E F F E C T S I N M O R P H O L O G Y
roles, so that at least person nouns that denote professions and occupations are more frequently applied to men. Thus, the direction of derivation (from male/general to female) is related to frequency of use. However, the frequency relations tend to be reversed with nouns like nurse (because more women are nurses than men) and widow (probably not because husbands die more often than wives, but because marital status has traditionally been considered more relevant for women than for men). As a result of the unusual frequency relations, we get unusual male forms with overt marking (widow-er, male nurse).
12.1.5 Explaining the correlations The correlation between frequency and shortness is clearly motivated by language users’ preference for economical structures. Speakers can afford shorter expressions (or even zero expressions) when these are frequent, because frequent expressions are more predictable and are therefore those that are expected by default. The basic principle here is the same as in many other areas of human communication. For instance, in many countries local phone calls do not require an area code because phone calls to the local area are more common than phone calls to other areas. In language, such economical structures may arise when a new distinction is introduced that is coded only in one of the two contrasting values. For instance, Spanish has a new subject/object distinction, which is marked by the preposition a with animate NPs (e.g. Veo a mi hermano [see.1sg to my brother] ‘I see my brother’). This does not have morphological status yet, but if it becomes grammaticalized as an accusative case prefix, we will have a case system that conforms to the pattern in (12.12), in which the less frequently used case form gets the overt marking. The nominative was never marked overtly from the beginning of this change. Another way in which an economical case-marking system may arise is by selectively preserving older markers. For example, in the Old High German n-declension, animate and inanimate nouns alike had a distinction between nominative and accusative (see (12.13)). (12.13) nom.sg acc.sg
Old High German affo knoto affon knoton ‘ape’ ‘knot’
Modern German Affe Knoten Affen Knoten ‘ape’ ‘knot’
Then the nominative/accusative distinction was lost in inanimate nouns, and in Modern German only animates preserve the zero marking in the nominative. Again, the resulting pattern conforms to (12.12), but it has come into existence via a different diachronic route. The correlation between frequency and differentiation is due to the
12.2 THE DIRECTION OF ANALOGICAL LEVELLING 
273
greater memory strength of frequent values. When a value occurs rarely, it is more difficult to remember all the details of that value, so syncretism is more common in rare values, and various suppletive allomorphs are more easily kept apart in the frequent values.
12.2 The direction of analogical levelling Analogical levelling is a common type of morphological change. Levelling eliminates morphophonological alternations by extending one stem alternant to other word-forms in the paradigm. For instance, many speakers of English have eliminated the alternation in house/houses, which in the traditional pronunciation has a voiced final stem consonant in the plural: [haʊs]/[haʊzəz]. Now crucially, it is the form of the singular stem that is extended by the innovating speakers ([haʊs]/[haʊsəz]), not the plural stem. There are no English speakers that pronounce the singular noun house as [haʊz]. This change is typical of analogical levelling in general: the form of the stem that is extended within the paradigm is usually the value with higher frequency. That frequency is the crucial factor is particularly clear from cases of local frequency reversals. A particularly striking case of this comes from West Frisian, where in the traditional language many nouns show a vowel alternation in singular–plural pairs. In innovative varieties of the language, this alternation is eliminated and the singular and plural stems are identical again, (see (12.14)). (12.14) conservative a. hoer/hworren koal/kwallen miel/mjillen poel/pwollen
innovative hoer/hoeren koal/koalen miel/mielen poel/poelen
b. earm/jermen kies/kjizzen hoarn/hwarnen trien/trjinnen
jerm/jermen kjizze/kjizzen hwarne/hwarnen trjin/trjinnen
‘whore(s)’ ‘coal’ ‘meal, milking’ ‘pool(s)’ ‘arm(s)’ ‘tooth/teeth’ ‘horn(s)’ ‘tear(s)’ (Tiersma 1982: 834)
In (12.14a), the singular stem form is extended in analogical levelling, but, in (12.14b), the plural stem form is extended. The choice of the form that is extended is by no means arbitrary: when the noun denotes a thing that tends to occur in groups and hence is more frequent in the plural, the plural stem wins out. An example from case inflection is Latin oleum ‘olive tree’, which goes back to an earlier form oleivum (cf. oleiva, later olıˉva, ‘fruit of the olive tree,
274 
C H A P T E R 1 2 â•… F R E Q U E N C Y E F F E C T S I N M O R P H O L O G Y
olive’). Then three sound changes occurred: (i) the diphthong ei turned into ˉe and later into ˉı , (ii) the semivowel v [w] was dropped before u and (iii) long vowels were shortened before another vowel. As a result, the nominative/ accusative form oleivum successively became oleˉvum, oleˉum and oleum, whereas the genitive and dative forms oleivıˉ/oleivoˉ became olıˉvıˉ/olıˉvoˉ. Then, analogical levelling extended the nominative/accusative stem to the other case forms (oleiva became olıˉva and retained the stem olıˉv-, because the v never dropped from its paradigm): (12.15) oldest form later form Classical Latin nom/acc.sg oleivum oleum oleum gen.sg oleivıˉ olıˉvıˉ oleıˉ dat.sg oleivoˉ olıˉvoˉ oleoˉ The greater stability of frequent stem forms can be explained again by memory strength and speed of lexical access. The genitive singular olıˉvıˉ is replaced by oleıˉ because the stem ole- has higher memory strength and may thus be used when a speaker (temporarily) forgets the old form olıˉv-, or because ole- can be retrieved more quickly from the lexicon and combined with the suffix -ıˉ than the form olıˉvıˉ, with its rarer stem form olıˉv-.
12.3 Frequency and irregularity In language after language, if there are irregularities in inflection, these primarily affect the most frequent lexemes. Our first example comes from Koromfe, which has scores of regular verbs like those in (12.16a), and a few irregular verbs like those in (12.16b). (12.16) a. habitual past b. habitual past kam kamε ‘squeeze’ bε bεn-ε ‘come’ tarı tarε ‘plaster’ bo bol-e ‘say’ leli lele ‘sing’ tε tεr-ε ‘arrive’ (Rennison 1997: 271–5) In Welsh, there are four irregular verbs whose past tense is totally unlike the past tense of a regular verb such as gwel- in (12.17a). Three of them are shown in (12.17b). (12.17) a. gwel-d ‘see’ b. myn-d ‘go’ 1sg gwel-es i es i 2sg gwel-est ti est ti 3sg gwel-odd e aeth e
gwneu-d ‘do’ nes i nest ti naeth e
do-d ‘come’ des i dest ti daeth e (King 1993: 183)
12.3 FREQUENCY AND IRREGULARITY 
275
In Old English, grammars list just four verbs that are totally irregular and cannot be fitted into any of the inflectional classes. These are shown in (12.18b), and a regular verb is shown in (12.18a). (12.18) a. ‘bind’ b. ‘be’ 1sg.prs binde eom 2sg.prs bintst eart 3sg.prs bint is 1–3pl.prs bindaþ sint 1sg.pst band wæs participle gebunden —
‘do’ doˉ deˉst deˉþ doˉþ dyde gedoˉn
‘go’ gaˉ gœ¯st gœ¯þ gaˉþ eˉode gegaˉn
‘want’ wille wilt wille willaþ wolde —
Thus, the verbs that tend to show irregularities are those that mean ‘be’, ‘do’, ‘go’, ‘come’, ‘say’, and so on – i.e. precisely those verbs that are used the most frequently across languages. In nouns, the situation is more or less the same. For example, in Lango regular plural suffixes are -ê, -nì and -í. Some regular and most of the irregular nouns are listed in (12.19). (12.19) a. rɛ́c réc-ê ‘fish(es)’ b. dákô pʊ́nʊˆ pùn-nì ‘pig(s)’ ɲákô lɛ̀ ley-í ‘axe(s)’ ìcɔ̀ dánô dyàŋ gìn
món ‘woman/women’ àɲìrà ‘girl(s)’ cɔ̀ ‘man/men’ jɔ̀ ‘person/people’ dòk ‘cattle’ gìgù ‘thing(s)’ (Noonan 1992: 83–5)
Irregular noun plurals in Bulgarian include oko/ocˇ i ‘eye(s)’, uxo/uši ‘ear(s)’, dete/deca ‘child(ren)’, and Italian has the three irregular nouns uomo/uomini ‘man/men’, dio/dei ‘god(s)’, bue/buoi ‘ox(en)’. The appearance of words for ‘cattle’ and ‘ox’ on several of these lists may at first seem surprising – these are certainly not among the most frequent nouns in modern Italian and modern English. But in modern Lango they may well be (cattle herding is one of the main economic activities of Lango speakers), and in older Italian and older English the situation may have been similar. There are two rather different ways in which frequency may cause irregularity in morphology. On the one hand, frequency leads to phonological reduction, because frequent expressions are relatively predictable, so that speakers can afford to articulate less clearly. This factor must be invoked to explain the irregularities in Koromfe verbs in (12.16). Examples from English are the verbs have, say and make, which were completely regular in earlier English (haved, sayed, maked), but became irregular because they were subjected to greater phonological reduction than comparable rarer verbs (e.g. said versus played, had versus behaved, made versus faked).
276 
C H A P T E R 1 2 â•… F R E Q U E N C Y E F F E C T S I N M O R P H O L O G Y
On the other hand, frequency leads to memory strength and fast lexical access, so that frequent items are less susceptible to analogical levelling and other regularizations. So, while frequency causes faster phonological change, with respect to morphology it has a conserving, decelerating function. For example, the irregular Italian noun uomo/uomini ‘man/men’ preserves an old declension type inherited from Latin (homo/homines) that was otherwise eliminated by regularizing changes (cf. Latin virgo/virgines ‘virgin(s)’, Italian vergine/vergini). This conserving effect of frequency is also the cause of the Bulgarian irregular plurals ocˇi ‘eyes’ and uši ‘ears’. These were originally dual forms, and, because eyes and ears typically occur in pairs, these word-forms were probably the most frequent forms in the paradigm. Since eyes and ears are among the most frequently used paired body parts, it is not surprising that these forms survive. From a diachronic point of view, the least well-understood type of irregularity is stem suppletion, as seen in Welsh myn-/es-/aeth, Old English is/wæs, gæˉþ/e¯ ode, and Lango dákô/món. It is difficult to understand why speakers would begin to associate roots that originally came from two different lexemes and integrate them as word-forms of the same lexeme. But, granted that speakers sometimes do that, the conserving effect of frequency will maintain the suppletion in the most frequent lexemes. It is also worth pointing out that inflection class differentiation (which we discussed in Section 12.1.3) works in exactly the same way: different markers for the same meaning/inflectional values can be maintained if the items affected are sufficiently frequent, whether owing to the frequency of the inflectional value or to lexeme frequency.
Summary of Chapter 12 Token frequency is relevant to morphology because frequent words occur more predictably in context, are more easily remembered and are retrieved faster than rare words. Because speakers favour economical structures, the greater predictability of frequent values typically results in zero expression (or otherwise short expression). Frequent values are also more differentiated – they show less syncretism, fewer shared exponents and more cross-cutting values. Because frequent words and values are more easily remembered, they are less subject to analogical levelling, and this is also one of the reasons why irregularities exist mostly in frequent words. Another reason is that frequent words are subject to greater phonological reduction, again because of predictability. Over time, frequency effects thus shape (inflectional) morphological structure in a number of ways.
COMPREHENSION EXERCISES 
277
Further reading Frequency differences between inflectional values of the same feature are discussed (under the name of ‘markedness’) by Greenberg (1966) and Croft (1990: ch. 4). Haspelmath (2006) argues that an abstract notion of markedness is superfluous once the role of frequency is appreciated. The insight that frequency is the explanation for shortness was already emphasized by Zipf (1935). For local frequency reversals, see Tiersma (1982). For the relation between frequency and irregularity, see Man´czak (1980a, b), Werner (1989), Bybee (1995), Nübling (2001), Corbett et al. (2001) and Brown et al. (2007). For the view that grammatical structure (including morphology) cannot be adequately understood without considering frequency effects, see Bybee (2006) and the papers in Bybee and Hopper (2001).
Comprehension exercises 1. The general correlation between frequency and shortness leads to certain expectations about inflectional paradigms. Consider the following (partial) paradigms and determine where these expectations are fulfilled, and where we should be surprised.
a. Udmurt conjugation: past tense of uc´k- ‘look’ 1sg uc´ki 1pl uc´kimy 2sg uc´kid 2pl uc´kidy 3sg uc´kiz 3pl uc´kizy (Perevošcˇikov 1962: 203)
b. Even declension: ǰuu ‘house’ sg pl nom ǰuu ǰuul acc ǰuuw ǰuulbu dat ǰuudu ǰuuldu com ǰuuñun ǰuulñun abl ǰuuduk ǰuulduk
(Malchukov 1995: 9)
c. Pipil possessive inflection: nu-chi:l ‘my chilli pepper’, etc. 1sg nu-chi:l 1pl tu-chi:l 2sg mu-chi:l 2pl amu-chi:l 3sg i-chi:l 3pl in-chi:l (Campbell 1985: 43)
278 
C H A P T E R 1 2 â•… F R E Q U E N C Y E F F E C T S I N M O R P H O L O G Y

d. Tauya possessive inflection: ya-potiyafo ‘my hand’, etc. 1sg ya-potiyafo 1pl sono-potiyafo 2sg na-potiyafo 2pl tono-potiyafo 3sg potiyafo 3pl nono-potiyafo (MacDonald 1990: 129–30)
2. The Modern French verb trouver ‘find’ used to have two different forms of the stem in older French, trouv- and treuv-. The former occurred in word-forms that were stressed on the suffix, and the latter occurred in word-forms that were stressed on the stem. (A dot below the syllable indicates the position of the stress.) This stem alternation no longer exists in modern French: all forms of the verb trouver have the same stem vowel. Why is this change surprising after what we learned in this chapter? older French modern French ‘I find’ je tre.uve je tro.uve ‘you find’ tu tre.uves tu tro.uves ‘he finds’ il tre.uve il tro.uve ‘we find’ nous trouvo.ns nous trouvo.ns ‘you(pl) find’ vous trouve.z vous trouve.z ‘they find’ ils tre.uvent ils tro.uvent 3. Go back to Chapter 10, where morphophonological alternations were discussed. Where did we make reference to frequency in that chapter? How did what we said there fit with the claims of this chapter?
Exploratory exercise In this chapter we saw that frequency asymmetries and structural asymmetries are correlated. For instance, a noun lexeme’s singular forms tend to be more frequently used than its plural forms, and correspondingly, case forms tend to be more differentiated in the singular and singular exponents tend to be shorter than their plural counterparts. This pattern can be seen even at the level of individual lexemes. Where the plural is more frequently used (a frequency reversal), that lexeme is likely to exhibit more differentiation in the plural, and shorter plural forms. We also claimed that frequent lexemes tend to be irregular. The reader may have noticed, however, that this discussion was based on a different type of frequency comparison. Rather than looking at the relative frequency of different paradigm cells, we compared the frequency of use of different lexemes. We did not consider whether irregularity correlates with frequency asymmetries within a lexeme. But given the importance of frequency at this level, we might ask whether such a correlation exists. In other words,
E X P L O R AT O R Y E X E R C I S E
279
are more frequently used cells in a paradigm more (or less) likely to be irregular? You will develop an answer to this question. This exercise is based on (but simplifies) Corbett et al.’s (2001) study of Russian, and we use Russian for demonstration purposes below. You need not choose this language to investigate, but a good frequency dictionary or frequency list, with information for individual word-forms, must be available. (Hint: to make the process of finding word-form frequencies more efficient, it is helpful if the frequency dictionary/list is available for electronic searching.) Instructions Step 1: Choose a morphological pattern that exhibits (strong) stem suppletion in at least a handful of lexemes. A few examples of Russian singular–plural noun pairs are given below. As can be seen, the examples in (12.20a) have the same stem for the singular and plural, but the ones in (12.20b) exhibit suppletion. (12.20) singular a. zavod student
plural zavody studenty
gloss ‘factory’ ‘student’
b. syn rebënok cˇelovek
synov’ja deti ljudi
‘son’ ‘child’ ‘person’
Since suppletion in the Russian examples is according to number, we want to ask how frequently the plural is used, compared to the singular, and whether this differs depending on whether the noun is regular or suppletive. An appropriate measure is thus the ratio of plural frequency to singular frequency, e.g. token frequency of ljudi divided by token frequency of cˇelovek. (For demonstration purposes we are ignoring the fact that Russian has nominal cases.) Step 2: Develop a hypothesis. Based on what you have read in this chapter and elsewhere in the book (e.g. discussion of frequency in Section 4.3), make a guess about what the answer to the research question will be. For instance, would you expect Russian lexemes with suppletive stems in the plural to be more/less frequently used in the plural than in the singular? What would you expect for lexemes with the same stem throughout? Explain your reasoning. Step 3: Build a list of lexemes with stem suppletion. Identify as many relevant words as possible. Step 4: Using a frequency dictionary or frequency list, gather token frequency counts. For each of the suppletive lexemes from Step 3, find the token frequency of each word-form in its paradigm. Then, do the same for at least 10 lexemes that have the same stem throughout the paradigm (the
280 
C H A P T E R 1 2 â•… F R E Q U E N C Y E F F E C T S I N M O R P H O L O G Y
(12.20a) type), and which are of similar overall frequency to the suppletive lexemes (or as close as possible). Step 5: For each lexeme, calculate the relevant frequency ratio. For instance, according to one count, the lexeme cˇelovek occurs in the singular 1678 times per million words of text, and in the plural 1267 times per million words of text (Sharoff 2002). Its ratio is thus 0.755 (= 1267/1678). Step 6: Evaluate the data and draw conclusions. Compare the relative frequency ratios for the two groups. Are there any notable differences in the frequency ratios? Do the results match your predictions? Consider the implications for understanding the relationship (or lack thereof) between the frequency of paradigm cells and irregularity. Alternative: Irregularity can be treated as a scale ranging from strong suppletion to full regularity (see Chapter 2). Rank different kinds of stem irregularity, and look for a correlation between degree of irregularity and frequency. (This is much harder!)
281
Key to comprehension exercises
Chapter 1 1.
complex words: nights (night-nights, cat-cats, trick-tricks, lap-laps,…) playing (play-playing, think-thinking, run-running, hop-hopping,…) affordable (afford-affordable, accept-acceptable, form-formable,…) indecent (decent-indecent, accurate-inaccurate, adequate-inadequate,…) searched (search-searched, cough-coughed, pass-passed, laugh-laughed,…) hopeful (hope-hopeful, mercy-merciful, colour-colourful, deceit-deceitful,…) redo (do-redo, think-rethink, absorb-reabsorb, read-reread,…)
not complex: owl, religion, indolent, bubble, during
2.
chàng chuán deˉ ng dıˇng fáng huaˉ jiaˇo lì shaˉn shù weˇi yóu yuán
‘music’ ‘ship’ ‘lamp, light’ ‘top’ ‘house’ ‘flower’ ‘leg’ ‘power’ ‘mountain’ ‘number’ ‘tail, part’ ‘oil’ ‘garden’
cheˉ cí diàn dòngwù feˉi jıˉ kè qì shì shuıˇ xué yú zhıˇ
‘vehicle’ ‘word’ ‘electric’ ‘animal’ ‘fly’ ‘machine’ ‘customer, visitor’ ‘steam, gas’ ‘sight, see’ ‘water’ ‘study of’ ‘fish’ ‘paper’
3.
wari in- ix- oq- x-
‘sleep’ ‘I’ ‘you(pl)’ ‘we’ ‘past tense’
eeli e7- at- n-
‘leave’ ‘they’ ‘you(sg)’ ‘present tense’
282 
KEY TO COMPREHENSION EXERCISES
Note that the third person singular (‘he or she’) does not correspond to a morphological constituent, but the word neeli ‘he or she leaves’ can (and must) be interpreted as having the meaning ‘he or she’ because of the absence of any other person marker. 4. In Hebrew, the lexical meaning is represented by the consonants, so the root dbr corresponds to speak, which shows up in the word-forms diber ‘he spoke’, dibra ‘she spoke’, dibur ‘speech’. The vowels correspond to grammatical information and word-class. There are four identifiable sets of word pairs in this data that exhibit a morphological relationship. Set 1: diber kimet milmel
‘he spoke’ ‘he wrinkled’ ‘he muttered’
dibra kimta milmla
‘she spoke’ ‘she wrinkled’ ‘she muttered’
In Set 1, the masculine (‘he’) past tense has the structure CiC(C)eC, where C indicates a consonant, and anything in parentheses is optional. The feminine (‘she’) past tense has the structure CiCC(C)a. The difference between masculine and feminine (past tense) verbs thus has to do with both the vowel quality, and the position of the vowel in the word. Set 2:
ħašav kalat sagar
‘he thought’ ‘he received’ ‘he shut’
ħašva kalta sagra
‘she thought’ ‘she received’ ‘she shut’
Set 2 is similar to Set 1, except that the masculine and the feminine have a for both vowels. The difference between masculine and feminine (past tense) verbs is thus represented only by the position of the vowel in the word. Set 3:
dibur kimut
‘speech’ ‘wrinkling’
Set 3 contains abstract nouns derived from verbs. These nouns have the form CiCuC. Set 4: ma-klet ma-sger ma-ħšev
‘radio receiver’ ‘lock’ ‘computer’
Set 4 contains concrete nouns. In this set, there is a prefix ma-, and a stem with the form CCeC. Chapter 2 1. Allomorph 1: -ayaal (e.g. awowe, awowayaal ‘grandfather(s)’). Conditions: Used when the stem ends in an [e], which is removed in the plural form. Allomorph 2: -oyin (e.g. baabaco, baabacooyin ‘palm(s)’). Conditions: Used when the stem ends in an [o].
KEY TO COMPREHENSION EXERCISES 
283
Allomorph 3: -aC, where C stands for a consonant that is the same as the final consonant in the stem (e.g. beed, beedad ‘egg(s)’). Conditions: Used when the stem has only one syllable (and ends in a consonant). Allomorph 4: -Co, where C stands for a consonant that is the same as the final consonant in the stem (e.g. cashar, casharro ‘lesson(s)’). Conditions: Used when the stem has (at least) two syllables and ends in a consonant. Plurals: tuulooyin, togag, albaabbo, bustayaal
2. Phonologically conditioned. The past tense allomorph [d] appears in verbs whose stems end in a vowel or any voiced consonant except [d]. The allomorph [t] appears in verbs whose stems end in any voiceless consonant except [t]. The allomorph [əd] appears in verbs whose stems end in [t] or [d]. 3.
Venezi-a Trent-o Milan-o Savon-a Volterr-a Piacenz-a Ancon-a Palerm-o Trevis-o Gubbi-o Bressanon-e Palestrin-a Domodossol-a Ivre-a Napoli Bologn-a
Venezi-ano Trent-ino Milan-ese Savon-ese Volaterr-ano Piacent-ino Anconet-ano Palermit-ano Trevigi-ano Eu-gub-ino Brissin-ese Prenest-ino Dom-ese Eporedi-ese Partenopeo Petroni-ano
(0 segments difference) (0) (0) (0) (1) (1) (2) (2) (2) (3) (4) (6) (7) (7+) (full suppletion) (full suppletion)
Chapter 3 1.
a. b. c. d. e. f.
tonal change suffixation vowel change (fronting) reduplication infixation circumfixation, vowel change
[ ] [ ] [ ] [ ] [
2. /hiɹ/ /ɪŋ/ /ɹi/ /gʊd/ /z/ n v __ __ v adj v __ ‘hear’ ‘prog’ ‘again’ ‘good’ ‘3sg pres’
[
]
[
3. /X/adj ↔ /Xɹˌ/adj ‘x’ ‘more x’
]
]
284 
KEY TO COMPREHENSION EXERCISES
[ [ [ [ [
]
[ [ [
/X/adj ↔ /ənX/adj ‘x’ ‘not x’
]
]
/X/v ↔ /ɹiX/v ‘x’ ‘dox again’
]
]
]
/Xi/adj ↔ /Xili/adv ‘x’ ‘in a x manner’ ´ 4. /CV CVC/n ↔ /CVCV´ C/adj ‘x’ ‘having the quality x’
]
]
[
[
]
5. /Xi/n ↔ /Xɹˌ/n ‘field of study x’ ‘person who studies x’
]
6. Subtraction is an attractive analysis because the final consonant of the feminine is different in each of the given words ([t,d,g,l,z]). A morphological relationship exists if a group of words shows identical partial resemblances in both form and meaning. These words show partial resemblance in meaning (they are masculine/feminine pairs), but since the final consonant is different in each of the feminine forms, the only generalization that can be made about form of all of the words is that the masculine form is one phonological segment shorter than the feminine form. If we tried to derive the feminine from the masculine, each word would require a different rule. If we delete a segment of the feminine to form the masculine, only one rule is needed. Chapter 4 1. actual: abundance, libertarian, replay possible: fraternitarian, itinerance, rebagelize impossible: happytarian: The morpheme -(t)arian can attach only to bases that end in t or ty, e.g. liberty-libertarian, humanity-humanitarian, document-documentarian. (It is possible to have a base that is truncated at /t/, e.g. vegetable-vegetarian, but this does not apply to happy.) â•… penchance: While penchant-penchance looks formally similar to abundant-abundance, in the latter pair, the noun abundance is derived from the adjective abundant. In the first pair, penchant is a noun, so penchance cannot be derived from it. â•… reknow: re- must attach to a verb that expresses a repeatable action. 2. compositional meaning: ability, legalize, modernize, morality, popularity, vaporize non-compositional meaning: authority, community, materialize, specialize 3. The (a) forms all have allomorphy (stress placement is different in the base form and in the derived form, e.g. pópular, populárity); the (b) forms have no allomorphy (e.g. módern, módernize).
KEY TO COMPREHENSION EXERCISES 
285
4. Finnish: cumulative expression for person and number (e.g. me indicates both 1st person and plural number), and zero expression in the nominative (unlike in the other cases, there is no form overtly marking the nominative case). Ndebele: empty morph. When the root contains only a consonant or consonant cluster, yi- is prefixed to the root in the imperative form. (The underlying generalization is that the word-form must have at least two syllables, and yi- is added when this condition would not otherwise be met.) However, these forms bear an imperative marker (-a, e.g. yidl-a ‘eat!’). Since yi- does not express any distinct inflectional or lexical meaning, it is an empty morph. Serbian: the answer depends on the segmentation that is assumed. One possibility is: singular 1st person govor-i-m 2nd person govor-i-š 3rd person govor-i
plural singular govor-i-mo 1st person tres-e-m govor-i-te 2nd person tres-e-š govor-e 3rd person tres-e
plural tres-e-mo tres-e-te tres-u
Under this analysis, the Serbian data exhibit all three phenomena. The morphemes -m, -mo, -š, -te, and -e/-u express person and number cumulatively because it is not possible to subdivide them into morphemes meaning ‘singular’, ‘plural’, ‘1st person’, etc. The forms -i and -e, which occurs in five of the six word-forms, are empty morphs because they do not directly correspond to any aspect of meaning. The third person singular has zero expression because there is no morpheme directly corresponding to this grammatical meaning. Another possible segmentation is: singular 1st person govor-im 2nd person govor-iš 3rd person govor-i
plural singular govor-imo 1st person tres-em govor-ite 2nd person tres-eš govor-e 3rd person tres-e
plural tres-emo tres-ete tres-u
This analysis has a disadvantage, in that it does not capture that the suffixes that attach to govor- and very similar to the ones that attach to tres-. However, under this segmentation, the Serbian data still has cumulative expression, but no empty morphs or zero expression. Chapter 5
[
]
[
]
1. caminabas tense: past person: 2nd number: singular insulam case: accusative number: singular
286 
KEY TO COMPREHENSION EXERCISES
[
]
[ [
]
cantabit mood: indicative aspect: infectum tense: future person: 3rd number: singular
[number: plural]
books
ovci case: dative number: singular incal person: 3rd number: plural
]
[degree: comparative]
bigger
2. affirmative polarity imperfective
perfect
habitual
kat-zawa
kat-nawa
kat-da
kat-zawa-j
kat-nawa-j
kat-da-j
imperfective
perfect
habitual
kat-zawa-cˇ
kat-nawa-cˇ
kat-da-cˇ
kat-zawa-cˇ -ir
kat-nawa-cˇ -ir
kat-da-cˇ -ir
present past
negative polarity
present past 3.
Denominal verbs act like N: cannibalize put into N: categorize cover with N: butter use N: phone, skate, ski (new category) create N: unionize, terrorize, peel (new category)
Deadjectival verbs factitive: flatten, legalize, blacken, modernize 4. vi
-ly (i) not relevant to the syntax (= derivational) (ii) not obligatory expression (= derivational) (iii) limited applicability (= derivational) (e.g. *longly, *brownly, etc.) (iv) (difficult to apply) (v) 0007relatively abstract meaning (= more typical of inflection than derivation)
KEY TO COMPREHENSION EXERCISES 
287
(vi) 0007compositional meaning (= typical of inflection, but also possible for derivation) (vii) 0007expression close to the base (= derivational) (there are examples of derivation followed by -ly, e.g. abus-ive-ly, content-ed-ly, but this is consistent with either derivation or inflection. The decisive examples are -ly followed by derivation, e.g. clean-li-ness, man-liness, like-li-hood). (viii) 0007no base allomorphy (= typical of inflection, but also possible for derivation) (ix) changes word-class (= derivational) (x) no cumulative expression (= derivational) (xi) 0007not iterable (= typical of inflection, but also possible for derivation) (e.g. *nicelyly)
Chapter 6 1. The primary restriction prohibiting *bountifulity, *sonorantity, *aimlessity, *darkishity and *fearsomity has to do with borrowed vocabulary strata. The suffix -ity attaches only to Latinate bases. The suffixes -ic, -able, -ive, -ous and -al are all Latinate, so they create the Latinate bases electric, probable, captive, curious and abnormal. However, -ful, -ant, -less, -ish and -some are all Germanic, so bases ending in these suffixes are also Germanic. (Also, since Germanic bases routinely form abstract nouns with -ness, even if it were possible to add -ity, some of the impossible words would be subject to synonymy blocking (e.g. aimlessness would likely block *aimlessity)). 2. *reknow is subject to semantic restrictions. The prefix must attach to a verb that expresses a repeatable action. *happytarian is subject to phonological restrictions. The morpheme -(t)arian can attach only to bases that end in t or ty, e.g. liberty-libertarian, humanity-humanitarian, document-documentarian. (It is possible to have a base that is truncated at /t/, e.g. vegetable-vegetarian, but this does not apply to happy.) 3. The suffix -simo is used when the stem is monosyllabic. The suffix -ma is used when the stem has (at least) two syllables.
dˉjávas-ma mángo-ma skónda-ma tínaγ-ma
‘reading’ ‘squeezing’ ‘stumbling’ ‘shaking’
kóp-simo lú-simo pjá-simo trék-simo
‘cutting’ ‘bathing’ ‘seizing’ ‘running’
4. *musting is not an example of blocking; must does not have a form musting because it is a modal verb (and have to and must are interchangeable otherwise, e.g. I have to go/I must go.) â•… *foots is blocked by feet. Irregular (inflectional) forms, including
288 
KEY TO COMPREHENSION EXERCISES
feet, must be stored in the lexicon and lexical items block otherwise productive rule application. â•… *cooker is blocked by cook. â•… *bishopdom is blocked by bishopric. â•… *teacheress is blocked by teacher. â•… *slickize is not an example of blocking; -ize does not attach monosyllabic words, or words that end in a stressed syllable. *certainness is blocked by certainty. *sisterlily is not an example of blocking; the impossibility of this word probably represents a phonological constraint against adjacent identical syllables (-li-ly).
5. The suffix -(er)ati developed from analogical extensions. Literati is an established term and was the model for the original analogical extension. Glitter-ati (< glitter) was the first term coined by analogy, and not coincidentally the two stems contain both the sequence lit(t)er, which was the basis for the analogy. By analogy the pattern then spread to other stems ending in er (liber-ati (cf. liber-al), chatter-ati, soccer-ati), and was even further extended to stems that do not end in er (digit-erati, cf. digit-al). The suffix -scape is similarly the result of analogy, with landscape as the original model, and seascape as the first term coined by analogy. This was followed by cloudscape, skyscape, waterscape, winterscape, etc. This example is exactly parallel to -gate (Watergate, Irangate, coingate). Chapter 7 1.
N N N family
N
N
planning adviser
KEY TO COMPREHENSION EXERCISES 
N N N N
N
N
undersea
cable
N
repair
team
N
N
N
ADJ
N
fixed-
line
N
N
phone
system
N N N ADJ mad
N cow
N
N
disease
hysteria
289
290 
KEY TO COMPREHENSION EXERCISES
N N N N N
N
World
Trade
N
N
N
Center
rescue
worker
N N
N N
N N N
N
N
N
N credit
N card
N agreement
N form
credit
Ncard
agreement
form
N N ADJ
N
major
league
N baseball
N game
KEY TO COMPREHENSION EXERCISES 
2.
291
endocentric exocentric exocentric endocentric endocentric exocentric endocentric appositional appositional coordinative
as´vakovida- bahuvrıˉhi- divyaruˉpa- gr̥hapati- mahaˉ raˉ ja- mahaˉ tman- priyasakhıˉ - raˉ jars.i- s´uklakr̥s.n.a- sukhaduh.kha-
3. ADJ
ADJ
V
ADJ
V
V
Pref
ADJ
Vsuf
in-
móvil
-iz
ADJsuf -able
‘immobilizable’
[
]
[
Pref
ADJ
Vsuf
ADJsuf
in-
móvil
-iz
-able
‘unmobilizable’
]
[
]
4. /Xyj/ADJ & /Ya/N ↔ /XoYyj/ADJ ‘x’ ‘y’ ‘having y with the property x’ Chapter 8 1.
lumangoy wumagayway natakot nauhaw binuhat pinunit pinunasan
2. If the description in Figure 8.3 were adopted, the following rule schema from Figure 8.2 could be eliminated: { [ /XVs/nom.sg ] , [ /XV/acc.sg ] , [ /XV/gen.sg ] , [ /XVZ/nom.pl ] , [ /XVs/acc.pl ] , [ /Xon/gen.pl ] }
292 
KEY TO COMPREHENSION EXERCISES
The modified version of Figure 8.2 would look like: XVZ XVs XV XVs XVZ Xon
Xos Xo Xu
!Xi !Xus Xon
XVs XV XV
Xes Xes Xon
Xas Xa Xa
Xes Xes Xon
XV XV XVs
Xus Xu Xu
Xa Xa Xas
XVs XVs Xon
Xuðes Xuðes Xuðon
XV XV XVs
Xes Xes Xon
Xi Xis Xi Xis Xeos Xeon
Xes Xes Xon
Xi Xi Xis
Xes Xes Xon
Xu Xu Xus
Xuðes Xuðes Xuðon
Figure 13.1
3. An answer to this question depends on how we choose to define the classes. On a narrow definition, it is possible to identify at least 46 ‘irregular’ past tense patterns (shown below). Of these, 22 contain only one word (catch/caught), or have only words built on the same stem (stand/stood, understand/understood, withstand/withstood). One answer, then, is that the words belonging to these 22 classes are truly irregular, whereas words belonging to the remaining 24 groups at least exhibit some kind of shared pattern and therefore belong to an inflection class. Looking more broadly, however, some of these small and singleton groups have traits in common. For instance, there is only one word with the pattern [æC] Æ [ɔt] (catch/caught), but there are five other words with past tense [ɔt]: seek/sought, teach/taught, bring/brought, think/ thought and buy/bought. Also, words that have both vowel change and a suffix ([d] or [t]) can plausibly be grouped with words that have only one or the other pattern. And so on. The only verb that does not form a past tense in any way similarly to other words is be/was (even highly suppletive go/went has a final [t]!). Thus, for almost all ‘irregular’ verbs, inflection classes can be established at a broad level.
KEY TO COMPREHENSION EXERCISES 
293
[d] or [t] patterns [k] Æ [d]: make/made [d] Æ [t]: bend/bent, build/built, lend/lent, send/sent, spend/spent Ø Æ [t]: burn/burnt, learn/learnt, spill/spilt Vowel change patterns [i] Æ [e]: eat/ate [i] Æ [ɛ]: 0007bleed/bled, breed/bred, feed/fed, lead/led, meet/met, plead/pled, read/ read, speed/sped [i] Æ [ɔ]: see/saw [i] Æ [o]: freeze/froze, shear/shore, speak/spoke, steal/stole, weave/wove [ɪ] Æ [e]: bid/bade, forbid/forbade, forgive/forgave, give/gave [ɪ] Æ [æ]: 0007begin/began, drink/drank, ring/rang, shrink/shrank, sing/sang, sink/ sank, sit/sat, spit/spat, spring/sprang, stink/stank, swim/swam [ɪ] Æ [ʌ]: 0007cling/clung, dig/dug, fling/flung, sling/slung, spin/spun, spring/ sprung, stick/stuck, sting/stung, string/strung, swing/swung, win/ won, wring/wrung [e] Æ [o]: awake/awoke, break/broke, wake/woke [e] Æ [ʊ]: 0007forsake/forsook, mistake/mistook, overtake/overtook, shake/shook, take/took [ɛ] Æ [ɔ]: forget/forgot, get/got, tread/trod [ɛ] Æ [o]: bear/bore, swear/swore, tear/tore, wear/wore [a] Æ [ɛ]: fall/fell [æ] Æ [ʌ]: hang/hung [aj] Æ [ɪ]: hide/hid, slide/slid [aj] Æ [e]: lie/lay [aj] Æ [æ]: bind/bound, find/found, grind/ground, wind/wound [aj] Æ [o]: dive/dove, drive/drove, ride/rode, rise/rose, shine/shone, strive/strove [aj] Æ [u]: fly/flew [aj] Æ [ʌ]: strike/struck [ɔ] Æ [u]: draw/drew [o] Æ [ɛ]: hold/held, uphold/upheld, withhold/withheld [o] Æ [u]: 0007blow/blew, grow/grew, know/knew, overthrow/overthrew, throw/ threw [u] Æ [ɔ]: shoot/shot [u] Æ [o]: choose/chose [ʌj] Æ [ɪ]: bite/bit, light/lit [ʌ] Æ [e]: become/became, come/came, overcome/overcame [ʌ] Æ [æ]: run/ran [ʌj] Æ [ɔ]: fight/fought [aj]/[ʌj] Æ [o]: smite/smote, stride/strode, write/wrote Past tense vowel change + [d] or [t] patterns [i] Æ [ɛd]: flee/fled
294 
KEY TO COMPREHENSION EXERCISES
[iC] Æ [ɛCt]: 0007creep/crept, deal/dealt, dream/dreamt, feel/felt, keep/kept, kneel/ knelt, leap/leapt, leave/left, mean/meant, sleep/slept, sweep/ swept, weep/wept [e] Æ [ɛd]: say/said [ɛl] Æ [old]: sell/sold, tell/told [u] Æ [ıd]: do/did, overdo/overdid [uC] Æ [ɔCt]: lose/lost Weak suppletive patterns with past tense [ɔt] [iC] Æ [ɔt]: seek/sought, teach/taught [ıŋ(C)] Æ [ɔt]: bring/brought, think/thought [æC] Æ [ɔt]: catch/caught [aj] Æ [ɔt]: buy/bought Other suppletive patterns be/was,were forego/forewent, go/went stand/stood, understand/understood, withstand/withstood Zero expression beat/beat, beset/beset, bet/bet, bid/bid, broadcast/broadcast, burst/burst, cast/cast, cost/cost, cut/cut, fit/fit, forbid/forbid, hit/hit, hurt/hurt, knit/knit, let/let, put/put, quit/quit, rid/rid, set/set, shed/shed, shut/shut, slit/slit, spit/spit, split/split, spread/ spread, thrust/thrust, upset/upset, wed/wed 4. Based on the given data, there are two factors that may have motivated the change. The first issue is grammatical gender. Both class (ii) and class (iii) contain masculine nouns, whereas class (i) contains feminine nouns. If the inflectional pattern for class (i) was considered by speakers to be the canonical feminine pattern, and class (iii) was considered to be the canonical masculine pattern, this could have motivated inflectional change in class (ii) towards the more canonical masculine pattern. The second potential factor is overlapping inflectional patterns. Class (i/ii) and class (iii) have similar patterns in the accusative (final [n]) and the dative (final [i]). The similarity of class (ii) to the class (iii) pattern in these paradigm cells could have served as the basis for shift in the other cells (nom and gen) to be more similar to class (iii). 5. The following is an inheritance hierarchy for Russian nominal inflection classes. Inflectional information that is shared by most or all of the classes is introduced at the highest node in the hierarchy (e.g. dative, instrumental and locative plural forms), and more specific information is introduced at lower nodes (e.g. the genitive plural form). In a few instances information at lower nodes overrides the inherited information (e.g. nominative plural form in (iv)).
KEY TO COMPREHENSION EXERCISES 
NOM ACC GEN DAT INST LOC
SG XZ XZ XV XV XZ Xe
295
PL Xy Xy XZ Xam Xami Xax
ACC = NOM
NOM ACC GEN DAT INST LOC
SG XZ XZ Xa Xu Xom Xe
PL Xy Xy XZ Xam Xami Xax
ACC = NOM
NOM ACC GEN DAT INST LOC
SG X X Xa Xu Xom Xe
ACC = NOM
(i)
PL Xy Xy Xov Xam Xami Xax
NOM ACC GEN DAT INST LOC
SG Xo Xo Xa Xu Xom Xe
ACC = NOM
(iv)
PL !Xa !Xa X Xam Xami Xax
NOM ACC GEN DAT INST LOC
SG Xa Xu Xy Xe Xoj Xe
PL Xy Xy X Xam Xami Xax
!ACCSG = NOMSG
(ii)
NOM ACC GEN DAT INST LOC
SG X' X' X' i X' i X' ju !X' i
PL X' i X' i X'ej X'am X'ami X'ax
ACC = NOM
(iii)
Figure 13.2
Intuitively, an important fact about this data is that the accusative is almost always the same as the nominative; the singular forms in (ii) (komnata, komnatu) are the exception. This generalization is captured at the highest node with the notation that ACC = NOM, and it then is inherited by lower nodes, and overridden in the class represented by komnata. (Note that all of the nouns in this exercise are inanimate. Animate nouns belonging to class (i), however, exhibit the pattern in ACC = GEN in the singular, as do all animates in the plural.)
6. The quantitative criterion does not provide clear evidence on the issue. In other inflection classes the second person singular is syncretic with the plural, but so is the first person singular (unlike in the verb be). The two inflection classes thus do not exhibit identical patterns of syncretism. Moreover, the relevance of this criterion is particularly unclear in the past tense – only be has more than one form.
296 
KEY TO COMPREHENSION EXERCISES
I you(sg) he/she we you(pl) they
According to the syntactic criterion, 2sg are/were is systematically syncretic with plural are/were. When the verb has to agree simultaneously with the second person singular and a plural subject, the same verb can agree with both: Either we or you are/were supposed to play in the Bulgaria match. The diachronic evidence also suggests systematic syncretism. In the history of English, the 2sg had a form distinct from the plural form: art in the present tense, wast in the past tense (wast had earlier replaced etymological were under influence of other verbs that had the form Xt in the 2sg past tense). Over time, the pronoun you displaced thou, and around the same time, are/were displaced art/wast. The ultimate result was that the 2sg was reformed on the basis of the 2pl.
I thou(sg) he/she we you(pl) they
am are is are are are
prs am art is are are are
walk walk walks walk walk walk
pst was wast > you was were were were
was were was were were were
are
walked walked walked walked walked walked
were
Chapter 9 1. Referentiality and morphological cohesion are not relevant to these examples. Moreover, English does not generally use special segmental markers (e.g. interfixes) to indicate compounds, so this criterion is uninformative. However, four tests can be applied: phonological cohesion, anaphoric replacement, expandability and coordination ellipsis. backboard: compound a. 0007Phonological cohesion: one main word stress on the first syllable (báckboard) b. 0007Anaphoric replacement impossible: *Every basketball court has one scoreboard and two back ones. (i.e. two backboards) c. 0007Not expandable: *The ball hit the very backboard. (i.e. the board in the very back) d. 0007Coordination ellipsis impossible: *The back and scoreboards were damaged by vandals.
KEY TO COMPREHENSION EXERCISES 
297
backdoor: phrase a. Lacks phonological cohesion: two main word stresses (báckdóor) b. 0007Anaphoric replacement possible: My house has one front door and two back ones. (i.e. two backdoors) c. 0007Expandable: When you get to the house, go through the far backdoor (i.e. the door in the far back of the house). d. 0007Coordination ellipsis possible: The back and side doors were mistakenly left open. back seat: phrase a. Lacks phonological cohesion: two main word stresses (báck séat) b. 0007Anaphoric replacement possible: My car has one front seat and two back ones. (i.e two back seats) c. 0007Expandable: There is a lever in the trunk to release the far back seat (i.e. the seat in the far back of the car). d. 0007Coordination ellipsis possible: The back and front seats we had upholstered in red velvet, but the middle row we upholstered in purple leather. 2. Third singular present tense -s has no freedom of stem selection; it must attach to verbs. It also has no freedom of movement. For instance, it cannot be clefted separately from its verb. (*It is -s that Mary walk.) Finally, it undergoes morphophonological alternation: [-s], [-əz] or [-z], depending on the final consonant of the verb stem. 3. Polish: clitics attach after their hosts (they are enclitic), so the clitic go cannot occur in sentence-initial position. French: the pronoun occurs in a position that requires emphatic (technically, contrastive) stress, so use of the clitic tu is not grammatical. Free form toi should be used. Serbian: the clitic je is a second-position clitic, but in this sentence it occurs in fourth position. Ponapean: the position of the completive suffix -la indicates that kengwini- is a compound verb with an incorporated noun (wini- ‘medicine’ is the dependent member). However, dependent members generally cannot be referential (including in Ponapean), so the demonstrative -o (which is associated with wini-) makes the word ungrammatical. 4. As discussed in this chapter, the pronominals are outside of the domain of word stress. This is evidence that they are clitics. However, clitics do not generally undergo morphophonological alternations; in this respect the pronominals behave like affixes. 5. In Lithuanian, s(i) is like a clitic in that it appears either after the verb root (when not negated), or before it (when negated). It thus has freedom of host selection. It is like an affix in that it undergoes a morphophonological alternation.
298 
KEY TO COMPREHENSION EXERCISES
6. Two criteria from Table 9.1 seem to be relevant. First, the first person active marker a- is prefixed to the verb root jogua ‘buy’ in (11.26a), but immediately precedes the noun mba’e ‘thing’ in (11.26b). If we assume that a- attaches only to verbs, and not to nouns, then the entire incorporated structure apparently serves as the domain of affixation. This is parallel to example (9.4). Second, as (11.27) shows, the dependent noun mba’e is not expandable – it cannot be modified by the adjective hepy ‘expensive’. Both of these criteria suggest that (11.26b) is a compound. Chapter 10 1. Morphophonological alternation. The most important issue is that stem voicing is both morphologically and lexically conditioned. It occurs in the plural, but not under the same phonological conditions in the possessive (e.g. [fs] in leaf’s), and does not occur in all relevant words (e.g., briefs, roofs, cuffs). The affected sounds form a natural class (voiceless obstruents), and produce a natural class (voiced obstruents). The ‘input’ and ‘output’ are also phonetically close. While phonetic coherence and phonetic closeness are typical traits of automatic alternations, morphophonological alternations may or may not be phonetically coherent. These properties thus provide no solid evidence either way. Finally, the alternation does not apply to loanwords, its application is not sensitive to speech style, and it does not create a new segment (voiced obstruents exist in English independently of the alternation). It does not apply across word boundaries (a lea[f] sat on the porch; *a lea[v] sat on the porch). 2. Relic alternation. The alternation applies to very few words, all of which are comparative adjectives: long/longer, strong/stronger, young/younger. (The rare example wrong/wronger seems to vary; some speakers have the alternation, and some have only [ŋ] in both word-forms.) Nouns do not exhibit the alternation (e.g. sing/singer). And it is not productive even in comparatives (e.g. winningest, blinger than bling). 3. a. 0007Morphophonological alternation: the palatalization rule does not apply to loanwords (laasiisìi ‘licence’) or to new instances of the conditioning environment (tèebà ‘cooked cassava flour’), so it must be at least in part lexically conditioned. b. 0007Automatic alternation: the rule applies across word boundaries and in a phonetically motivated environment. c. 0007Automatic alternation: the palatalization rule applies in a phonetically motivated environment, and whenever that environment arises. This rule is similar to the Hausa example in (a),
KEY TO COMPREHENSION EXERCISES 
299
but the crucial difference is that in Modern Greek, the palatalization rule applies to loanwords (Greek [cinino] ‘chinine’). d. 0007Morphophonological alternation: the fact that the alternation occurs only in ‘certain morphological forms’ indicates that the rule is morphologically conditioned. It is also lexically conditioned, as evidenced by the fact that it fails to apply to some (native) words.
4. Phonetic distance: the sounds that are affected sometimes differ from their replacements by more than one phonological feature (e.g. ph and sh differ in both place (labial vs. alveolar) and manner (stop vs. fricative) of articulation). No new segments: based on the given data, at least some of the output sounds are independently attested: j in gijimisa ‘make run’; sh in shaya ‘beat’. It is unclear from this data whether the remaining output sounds (esp. ny, nj) are new segments, or also independently exist in the language. Phonological motivation: in (10.11), the affected sounds are not immediately adjacent to the conditioning environment (-w). Given that automatic alternations are phonologically motivated, and phonological rules generally apply to adjacent sounds, we can take this as further evidence that the alternation in (10.11) is morphophonological in nature. Chapter 11
[
] [
1. /support/v /supportive/a sbj — obj ↔ oblof agent patient patient
[ ] [
]
]
2. /X/v /outX/v ↔ sbj — obj sbj agenti agenti patientj ‘ Ai actsx’ ‘Ai actsx better than Bj
3. Section 11.3.3: ‘In contrast to (complex) event nouns, agent nouns in English … do not seem to inherit the verb’s argument structure.’ In the example ruler over a large empire, ruler is an agent noun derived from the verb rule, but it seems to have inherited the argument structure of the verb (The king rules over a large empire). Ruler over a large empire has a possessive phrase, similar to explorer of Antarctica. Chapter 12 1. a. 0007The plural is usually less frequent than the singular, so we expect that in Udmurt, plural inflected forms will be longer than singular
300 
KEY TO COMPREHENSION EXERCISES
inflected forms. This expectation is fulfilled. Since the third person is usually more frequent than other values of the feature person, we likewise expect that the third person will be shorter. This expectation is not fulfilled; the first person singular form is shorter than the third person singular form. b. 0007The plural forms are longer than the singular forms in Even; this fits expectations for the same reasons as in (a). The nominative is also shorter than the other case forms; this is expected given that the nominative is usually the more frequently used form. c. 0007In the Pipil example, the plural forms are longer than the corresponding singular form in two instances; they are the same length in one (1sg nuchi:l ‘my chilli pepper’; 1pl tuchi:l ‘your chilli pepper’). This mostly matches expectations. Also, the third person singular form is the shortest – this also matches expectations. d. 0007The Tauya data entirely matches expectations. The plural forms are longer than the singular forms, and the third person singular form exhibits zero expression and is therefore the shortest of all forms.
2. In this chapter it was noted that within a paradigm, the forms with higher relative frequency usually serve as the model for analogical levelling, and forms with lower relative frequency usually undergo analogical levelling. Since singular forms are usually more frequent than plural forms, we would expect the singular to have served as the model for analogical levelling in the French example. The change is therefore surprising because the singular forms have undergone stem levelling based on the (presumably less common) plural forms. 3. Frequency was discussed in Section 10.2, where it was noted that in German, the lexeme be, which has high absolute frequency, preserves an alternation between r and s (war/ge-wes-en ‘is/been’) that has been levelled in other words (las/ge-les-en ‘read(prs)/read(part))'. This fact fits with the discussion in this chapter in that lexemes with high absolute frequency resist analogical change, whereas lexemes with low absolute frequency are likely to undergo it.
301
References
Abraham, R.C. 1940. The principles of Tiv. London: Crown Agents for the Colonies, on behalf of the Government of Nigeria. Ackema, Peter and Neeleman, Ad. 2004. Beyond morphology: interface conditions on word formation. Oxford: Oxford University Press. Ackerman, Farrell, and LeSourd, Philip S. 1997. Toward a lexical representation of phrasal predicates. In Alsina, Alex, Bresnan, Joan, and Sells, Peter (eds), Complex predicates. Stanford, CA: CSLI, 67–106. Ackerman, Farrell, and Stump, Gregory T. 2004. Paradigms and periphrastic expression: a study in realization-based lexicalism. In Sadler, Louisa and Spencer, Andrew (eds), Projecting morphology. Stanford, CA: CSLI, 111– 158. Aikhenvald, Alexandra Y. 2002. Typological parameters of clitics, with special reference to Tariana. In Dixon, R.M.W. and Aikhenvakd, Alexandra Y. (eds), Word: a cross-linguistic typology. Cambridge: Cambridge University Press, 1–41. Albright, Adam. 2003. A quantitative study of Spanish paradigm gaps. In Garding, Gina and Tsujimura, Mimu (eds), West Coast Conference on Formal Linguistics 22 proceedings. Somerville, MA: Cascadilla Press, 1–14. —— and Hayes, Bruce. 2002. Modeling English past tense intuitions with minimal generalization. In Maxwell, Michael (ed.), Proceedings of the Sixth Meeting of the Association for Computational Linguistics Special Interest Group in Computational Phonology in Philadelphia, July 2002. Cambridge, MA: Association for Computational Linguistics, 58–69. Aldrich, R.I. 1966. The development of -scape. American Speech 41: 155–7. Alegre, Maria and Gordon, Peter. 1999. Rule-based versus associate processes in derivational morphology. Brain and Language 68.2: 347–54. Alexiadou, Artemis. 1999. On the syntax of nominalization and possession: remarks on patterns of ergativity. Unpublished. Habilschrift, Universität Potsdam.
302 
REFERENCES
Algeo, John (ed.) 1991. Fifty years among the new words: a dictionary of neologisms 1941–1991. Cambridge: Cambridge University Press. Allen, Barbara J., Frantz, Donald G., Gardiner, Donna B. and Perlmutter, David M. 1990. Verb agreement, possessor ascension, and multistratal representation in Southern Tiwa. In Postal, Paul M. and Joseph, Brian D. (eds), Studies in relational grammar 3. Chicago: University of Chicago Press, 321–83. Anderson, Stephen R. 1988. Morphological change. In Newmeyer, Frederick (ed.), Linguistics: the Cambridge survey. Vol. I. Cambridge: Cambridge University Press, 324–62. —— 1992. A-morphous morphology. Cambridge: Cambridge University Press. —— 2005. Aspects of the theory of clitics. Oxford: Oxford University Press. Annamalai, E. and Steever, Sanford B. 1998. Modern Tamil. In Steever, Sanford B. (ed.), The Dravidian languages. London: Routledge, 121–53. Anshen, F. and Aronoff, Mark. 1988. Producing morphologically complex words. Linguistics 26: 641–55. Arensen, Jon. 1982. Murle grammar. Juba: SIL Sudan. Aronoff, Mark. 1976. Word formation in generative grammar. Cambridge, MA: MIT Press. —— 1980. The relevance of productivity in a synchronic description of word formation. In Fisiak, J. (ed.), Historical morphology. The Hague: Mouton, 71–82. —— (ed.) 1992. Morphology now. Albany, NY: SUNY Press. —— 1994. Morphology by itself: stems and inflectional classes. Cambridge, MA: MIT Press. —— 2007. In the beginning was the word. Language 83(4), 803–30. —— and Fudeman, Kirsten. 2005. What is morphology? Malden, MA: Blackwell. —— and Fuhrhop, Nanna. 2002. Restricting suffix combinations in German and English: closing suffixes and the monosuffix constraint. Natural Language and Linguistic Theory 20.3: 451–90. Ashton, E. O. 1947. Swahili grammar. London: Longman. Baayen, R. Harald. 1993. On frequency, transparency and productivity. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 1992. Dordrecht: Kluwer, 181–208. —— and Lieber, Rochelle. 1991. Productivity and English derivation: a corpus-based study. Linguistics 29: 801–43. ——, Piepenbrock, R. and Gulikers, L. 1995. The CELEX Lexical Database (Release 2). Philadelphia: Linguistic Data Consortium. —— and Schreuder, Robert (eds) 2003. Morphological structure in language processing. Berlin: Mouton de Gruyter. Baerman, Matthew. 2002. The Surrey person syncretism database. http:// www.smg.surrey.ac.uk/personsyncretism/index.aspx.
REFERENCES 
303
—— 2008. Historical observations on defectiveness: the first singular nonpast. Russian Linguistics 32(1): 81–97. ——, Brown, Dunstan, and Corbett, Greville. 2005. The syntax-morphology interface: a study of syncretism. Cambridge: Cambridge University Press. ——, Corbett, Greville, and Brown, Dunstan (eds) 2010. Defective paradigms: missing forms and what they tell us. London: British Academy Press in coordination with Oxford University Press. ——, Corbett, Greville G., Brown, Dunstan, and Hippisley, Andrew (eds) 2007. Deponency and morphological mismatches. London: British Academy Press in coordination with Oxford University Press. Baker, Mark. 1985. The Mirror Principle and morphosyntactic explanation. Linguistic Inquiry 16.3: 373–415. Baker, Mark C. 1988. Incorporation. Chicago: University of Chicago Press. —— 1998. Comments on the paper by Sadock. In Lapointe, Steven G. et al. (eds), 188–212. Barker, Chris. 1998. Episodic -ee in English: a thematic role constraint on new word formation. Language 74.4: 695–727. Baroni, Marco. 2001. The representation of prefixed forms in the Italian lexicon: evidence from the distribution of intervocalic [s] and [z] in Northern Italian. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 1999. Dordrecht: Kluwer, 121–52. Bauer, Laurie. 1983. English word-formation. Cambridge: Cambridge University Press. —— 1990. Be-heading the word. Journal of Linguistics 26: 1–31. —— 1992. Scalar productivity and -lily adverbs. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 1991. Dordrecht: Kluwer, 185–91. —— 1998. When is a sequence of two nouns a compound in English? English Language and Linguistics 2.1: 65–86. —— 2001a. Compounding. In Haspelmath, Martin et al. (eds) Language typology and language universals. Vol. I. Berlin: Walter de Gruyter, 695–707. —— 2001b. Morphological productivity. Cambridge: Cambridge University Press. —— 2002. What you can do with derivational morphology. In Bendjaballah, Sabrina, Dressler, Wolfgang U., Pfeiffer, Oskar and Voeikova, Maria (eds), Morphology 2000: selected papers from the 9th Morphology Meeting, Vienna, 24–28 February 2000. Amsterdam: Benjamins, 37–48. —— 2003. Introducing linguistic morphology. 2nd edn. Washington, DC: Georgetown University Press. —— 2004. A glossary of morphology. Washington, DC: Georgetown University Press. —— 2005. Productivity: theories. In Štekauer, Pavol and Lieber, Rochelle (eds), Handbook of word-formation. Dordrecht: Springer, 315–34. —— and Renouf, Antoinette. 2001. A corpus-based study of compounding in English. Journal of English Linguistics 29.2: 101–23.
304 
REFERENCES
Beard, Robert and Szymanek, Bogdan. 1988. Bibliography of morphology 1960–85. Amsterdam: Benjamins. Becker, Thomas. 1990. Analogie und morphologische Theorie. Munich: Fink. —— 1993a. ‘Back-formation, cross-formation, and ‘bracketing paradoxes’ in paradigmatic morphology. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 1993. Dordrecht: Kluwer, 1–25. —— 1993b. Morphologische Ersetzungsbildungen im Deutschen. Zeitschrift für Sprachwissenschaft 12.2: 185–217. Berchem, Jörg. 1991. Referenzgrammatik des Somali. Cologne: Omimee. Bertram, Raymond, Schreuder, Robert and Baayen, R. Harald. 2000. The balance of storage and computation in morphological processing: the role of word formation type, affixal homonymy and productivity. Journal of Experimental Psychology: Learning, Memory and Cognition 26.2: 489–511. Beukema, Frits and den Dikken, Marcel (eds) 2000. Clitic phenomena in European languages. Philadelphia: J. Benjamins. Beyer, Stephan B. 1992. The Classical Tibetan language. Albany, NY: SUNY Press. Bickel, Balthasar and Nichols, Johanna. 2007. Inflectional morphology. In Shopen, Timothy (ed.), Language typology and syntactic description, vol. 3: grammatical categories and the lexicon. 2nd edn. Cambridge: Cambridge University Press, 169–240. Bisetto, Antonietta and Scalise, Sergio. 1999. Compounding: morphology and/or syntax? In Mereu (ed.), 31–48. Blevins, James P. 2003. Stems and paradigms. Language 79.4: 737–67. —— and Blevins, Juliette (eds) 2009. Analogy in grammar: form and acquisition. Oxford: Oxford University Press. Bloomfield, Leonard. 1933. Language. New York: Holt. Bochner, Harry. 1993. Simplicity in generative morphology. Berlin: de Gruyter. Booij, Geert. 1993. Against split morphology. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 1993. Dordrecht: Kluwer, 27–49. —— 1996. Inherent versus contextual inflection and the split morphology hypothesis. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 1995. Dordrecht: Kluwer, 1–16. —— 2007. The grammar of words: an introduction to linguistic morphology. Oxford: Oxford University Press. —— 2009. Lexical integrity as a formal universal: a constructionist view. In Scalise et al. (eds), 83–100. —— Lehmann, Christian and Mugdan, Joachim (eds) 2000–2004. Morphologie/Morphology: ein internationales Handbuch zur Flexion und Wortbildung/An international handbook on inflection and word-formation. Vol. 1: 2000, Vol. 2: 2004. Berlin: de Gruyter. Börjars, Kersti, Vincent, Nigel and Chapman, Carol. 1997. Paradigms, periphrases and pronominal inflection: a feature-based account. In Booij,
REFERENCES 
305
Geert and van Marle, Jaap (eds), Yearbook of morphology 1996. Dordrecht: Kluwer, 155–80. Boškovic´, Željko. 2001. On the nature of the syntax-phonology interface: cliticization and related phenomena. Amsterdam: Elsevier. Botha, Rudolf. 1981. A base rule theory of Afrikaans synthetic compounding. In Moortgat, M., van der Hulst, H. and Hoekstra, T. (eds), The scope of lexical rules. Dordrecht: Foris, 1–77. Bowe, Heather J. 1990. Categories, constituents, and constituent order in Pitjantjatjara: an aboriginal language of Australia. New York: Routledge. Bresnan, Joan. 2001. Lexical functional syntax. Malden, MA: Blackwell. —— and Mchombo, Sam M. 1995. The lexical integrity principle: evidence from Bantu. Natural Language and Linguistic Theory 13.2: 181– 254. Brown, Dunstan, Tiberius, Carol and Corbett, Greville. 2007. The alignment of form and function: corpus-based evidence from Russian. International Journal of Corpus Linguistics 12.4: 511–34. Bubenik, Vit. 1999. An introduction to the study of morphology. (Lincom coursebooks in linguistics, 7.) Munich: Lincom Europa. Buchholz, Oda and Fiedler, Wilfried. 1987. Albanische Grammatik. Leipzig: Enzyklopädie. Bugenhagen, Robert D. 1995. A grammar of Mangap-Mbula: an Austronesian language of Papua New Guinea. (Pacific linguistics, C-101.) Canberra: Australian National University. Butterworth, Brian. 1983. Lexical representation. In Butterworth, Brian (ed.), Language production, vol. 2: development, writing and other language processes. London: Academic Press, 257–94. Bybee, Joan L. 1985. Morphology: a study of the relation between meaning and form. Amsterdam: Benjamins. —— 1988. Morphology as lexical organization. In Hammond and Noonan (eds), 119–41. —— 1995. Regular morphology and the lexicon. Language and Cognitive Processes 10.5: 425–55. —— 2001. Phonology and language use. Cambridge: Cambridge University Press. —— 2006. From usage to grammar: the mind’s response to repetition. Language 82.4: 711–33. —— and Hopper, Paul (eds) 2001. Frequency and the emergence of linguistic structure. Amsterdam: Benjamins. —— and Moder, Carol. 1983. Morphological classes as natural categories. Language 59.2: 251–70. —— and Newman, Jean E. 1995. Are stem changes as natural as affixes? Linguistics 33: 633–54. —— and Slobin, Dan. 1982. Rules and schemas in the development and use of the English past tense. Language 58.2: 265–89.
306 
REFERENCES
Campbell, Lyle. 1985. The Pipil language of El Salvador. Berlin: Mouton de Gruyter. Caramazza, Alfonso, Laudanna, Alessandro and Romani, Christina. 1988. Lexical access and inflectional morphology. Cognition 28.3: 297–332. Carstairs, Andrew. 1987. Allomorphy in inflexion. Beckenham: Croom Helm. Carstairs-McCarthy, Andrew. 1991. Current morphology. London: Routledge. —— 1993. Morphology without word-internal constituents: a review of Stephen R. Anderson’s A-Morphous morphology. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 1992. Dordrecht: Kluwer, 209–33. —— 2005. Phrases inside compounds: a puzzle for lexicon-free morphology. SKASE Journal of Theoretical Linguistics 2.3: 34–42. Chang, Claire Hsun-huei. 1998. V–V compounds in Mandarin Chinese: argument structure and semantics. In Packard, Jerome (ed.), New approaches to Chinese word-formation. Berlin: Mouton de Gruyter, 77–101. Chang, Suk-Jin. 1996. Korean. Amsterdam: Benjamins. Chomsky, Noam A. and Halle, Morris. 1968. The sound pattern of English. New York: Harper & Row. Clahsen, Harald, Eisenbeiss, Sonja and Sonnenstuhl-Henning, Ingrid. 1997. Morphological structure and the processing of inflected words. Theoretical linguistics 23.3: 201–49. Coates, Richard. 1999. Word structure. (Language workbooks.) London: Routledge. Comrie, Bernard. 1981. The languages of the Soviet Union. Cambridge: Cambridge University Press. Cooreman, Ann. 1994. A functional typology of antipassives. In Fox, Barbara and Hopper, Paul (eds), Voice: form and function. Amsterdam: Benjamins, 49–88. Corbett, Greville. 1982. Gender in Russian: an account of gender specification and its relationship to declension. Russian Linguistics 6: 197–232. —— 1987. The morphology/syntax interface: evidence from possessive adjectives in Slavonic. Language 63.2: 299–345. —— 1991. Gender. Cambridge: Cambridge University Press. —— 2000. Number. Cambridge: Cambridge University Press. —— and Fraser, Norman M. 1993. Network Morphology: a DATR account of Russian nominal inflection. Journal of Linguistics 29: 113–42. —— and Hippisley, Andrew, Brown, Dunstan and Marriott, Paul. 2001. Frequency, regularity and the paradigm: a perspective from Russian on a complex relation. In Bybee and Hopper (eds), 201–26. Crocco-Galèas, Grazia. 1991. Gli etnici italiani: studio di morfologia naturale. Padova: Unipress. Croft, William. 1990. Typology and universals. Cambridge: Cambridge University Press. Daland, Robert, Sims, Andrea D. and Pierrehumbert, Janet. 2007. Much ado about nothing: a social network model of Russian paradigmatic
REFERENCES 
307
gaps. In Zaenen, Annie and van den Bosch, Antal (eds), Proceedings of the 45th Annual Meeting of the Association for Computational Linguistics in Prague, Czech Republic, June 24th–29th, 2007. Prague: Association for Computational Linguistics, 936–43. Dasgupta, Probal, Ford, Alan and Singh, Rajendra. 2000. After etymology: towards a substantivist linguistics. Munich: Lincom Europa. Daugherty, Kim G. and Seidenberg, Mark S. 1994. Beyond rules and exceptions: a connectionist approach to inflectional morphology. In Lima et al. (eds), 353–88. Davies, John. 1981. Kobon. (Lingua Descriptive Studies, 3.) Amsterdam: North-Holland. Dayley, Jon P. 1985. Tzutujil grammar. (University of California Publications in Linguistics, 107.) Berkeley: University of California Press. —— 1989. Tümpisa (Panamint) Shoshone grammar. (University of California Publications in Linguistics, 115.) Berkeley: University of California Press. Derwing, Bruce L. 1990. Morphology and the mental lexicon: psycholinguistic evidence. In Dressler, Wolfgang U., Luschutszky, Hans C., Pfeiffer, Oskar E. and Rennison, John R. (eds), Contemporary morphology. Berlin: Mouton de Gruyter, 249–65. —— and Skousen, Royal. 1994. Productivity and the English past tense: testing Skousen’s analogy model. In Lima et al. (eds), 193–218. Dik, Simon. 1997. The theory of Functional Grammar. Parts 1–2. Berlin: Mouton de Gruyter. Di Sciullo, Anne-Marie and Williams, Edwin. 1987. On the definition of word. Cambridge, MA: MIT Press. Dixon, R.M.W. 2000. A typology of causatives: form, syntax and meaning. In Dixon, R.M.W. and Aikhenvald, Alexandra Y. (eds), Changing valency. Cambridge: Cambridge University Press, 30–83. Dressler, Wolfgang U. 1981. General principles of poetic license in wordformation. In Logos semantikos: studia linguistica in honorem Eugenio Coseriu. Vol. II. Tübingen: Narr, 423–31. —— 1985. Morphonology. Ann Arbor: Karoma. —— 1989. Prototypical differences between inflection and derivation. Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 42: 3–10. —— 1997. On productivity and potentiality in inflectional morphology. (CLASNET Working Papers, 7.) Montreal: University of Montreal. —— and Ladányi, Mária. 2000. Productivity in word-formation: a morphological approach. Acta Linguistica Hungarica 47: 103–44. —— and Mayerthaler, Willi and Panagl, Oswald and Wurzel, Wolfgang U. 1987. Leitmotifs in natural morphology. Amsterdam: Benjamins. Dryer, Matthew S. 2008a. Order of subject, object and verb. In Haspelmath et al. (eds), Chapter 81.
308 
REFERENCES
—— 2008b. Order of adposition and noun phrase. In Haspelmath et al. (eds), Chapter 85. —— 2008c. Prefixing vs. suffixing in inflectional morphology. In Haspelmath et al. (eds), Chapter 26. —— 2008d. Order of object and verb. In Haspelmath et al. (eds), Chapter 83. Dubinsky, Stanley and Simango, Silvester Ron. 1996. Passive and stative in Chichewa: evidence for modular distinctions in grammar. Language 72.4: 749–81. Dunn, John Asher. 1979. A reference grammar for the Coast Tsimshian language. (Canadian Ethnology Service Paper, 55.) Ottawa: National Museum of Canada. Embick, David. 2000. Features, syntax, and categories in the Latin perfect. Linguistic Inquiry 31.2: 185–230. Embick, David and Noyer, Rolf. 2007. Distributed Morphology and the syntax/morphology interface. In Ramchand, Gillian and Reiss, Charles (eds), Oxford handbook of linguistic interfaces. Oxford: Oxford University Press, 289–324. Fabb, Nigel. 1988. English suffixation is constrained only by selectional restrictions. Natural Language and Linguistic Theory 6.4: 527–39. Feldman, Laurie Beth (ed.) 1995. Morphological aspects of language processing. Hillsdale, NJ: Lawrence Erlbaum Associates. Foley, William A. 1991. The Yimas language of New Guinea. Stanford, CA: Stanford University Press. Fortescue, Michael. 1984. West Greenlandic. (Croom Helm descriptive grammars.) London: Croom Helm. Fox, G.J. 1979. Big Nambas grammar. Canberra: Australian National University. Franks, Steven and King, Tracy Holloway. 2000. A handbook of Slavic clitics. Oxford: Oxford University Press. Gerlach, Birgit, and Grijzenhout, Janet (eds) 2000. Clitics in phonology, morphology and syntax. Amsterdam: J. Benjamins. Greenberg, Joseph. 1959. A quantitative approach to the morphological typology of language. International Journal of American Linguistics 26: 178–94. —— 1966. Language universals, with special reference to feature hierarchies. (Janua Linguarum, Series Minor, 59.) The Hague: Mouton. Grimshaw, Jane. 1990. Argument structure. Cambridge, MA: MIT Press. Guevara, Emiliano and Scalise, Sergio. 2009. Searching for universals in compounding. In Scalise, et al. (eds), 101–28. Gussenhoven, Carlos and Jacobs, Haike. 2005. Understanding phonology, 2nd edition. London: Arnold. Hall, Christopher J. 1992. Morphology and mind: a unified approach to explanation in linguistics. London: Routledge.
REFERENCES 
309
Halle, Morris and Marantz, Alex. 1993. Distributed morphology and the pieces of inflection. In Hale, Kenneth and Keyser, Samuel Jay (eds), The view from Building 20. Cambridge, MA: MIT Press, 111–76. —— 1994. Some key features of Distributed Morphology. MIT Working Papers in Linguistics 21: 275–88. Halpern, Aaron 1995. On the placement and morphology of clitics. Stanford, CA: CSLI. Hammond, Michael and Noonan, Michael (eds) 1988. Theoretical morphology: approaches in modern linguistics. Orlando: Academic Press. Hankamer, Jorge. 1989. Morphological parsing and the lexicon. In MarslenWilson, William (ed.), Lexical representation and process. Cambridge, MA: MIT Press, 392–408. Hansson, Gunnar Olafur. 1999. ‘When in doubt…’: intraparadigmatic dependencies and gaps in Icelandic. In Tamanji, Pius, Hirotani, Masako and Hall, Nancy (eds), NELS 29: Proceedings of the 29th Meeting of the North Eastern Linguistic Society. Amherst, MA: GLSA Publications, The University of Massachusetts, 105–19. Harley, Heidi and Noyer, Rolf. 1998. Licensing the non-lexicalist lexicon: nominalizations, vocabulary items and the encyclopedia. MIT Working Papers in Linguistics 32: 119–37. Harris, Alice C. 2000. Where in the word is the Udi clitic? Language 76.3: 593–616. —— 2002. Endoclitics and the origins of Udi morphosyntax. Oxford: Oxford University Press. Haspelmath, Martin. 1990. The grammaticization of passive morphology. Studies in Language 14.1: 25–71. —— 1992. Grammaticization theory and heads in morphology. In Aronoff, Mark (ed.), Morphology now. Albany, NY: SUNY Press, 69–82; 194–8. —— 1993. A grammar of Lezgian. Berlin: Mouton de Gruyter. —— 1995. The growth of affixes in morphological reanalysis. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 1994. Dordrecht: Kluwer, 1–29. —— 1996. Word-class-changing inflection and morphological theory. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 1995. Dordrecht: Kluwer, 43–66. —— 2000. Periphrasis. In Booij, Geert, Lehmann, Christian and Mugdan, Joachim (eds), Morphology: a handbook on inflection and word formation. vol. 1. (Handbücher zur Sprach- und KommunikationsÂ�wissenschaft.) Berlin: de Gruyter, 654–64. —— 2006. Against markedness (and what to replace it with). Journal of Linguistics 42.1: 25–70. ——, Dryer, Matthew S., Gil, David and Comrie, Bernard (eds) 2008. The world atlas of language structures online. Munich: Max Planck Digital Library. Available online at: http://wals.info/feature.
310 
REFERENCES
Hay, Jennifer. 2001. Lexical frequency in morphology: is everything relative? Linguistics 39.6: 1041–70. —— 2003. Causes and consequences of word structure. New York: Routledge. —— and Baayen, R. Harald. 2002. Parsing and productivity. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 2001. Dordrecht: Kluwer, 203–35. —— and Plag, Ingo. 2004. What constrains possible suffix combinations? On the interaction of grammatical and processing restrictions in derivational morphology. Natural Language and Linguistic Theory 22.3: 565–96. Hooper, Joan B. 1976. An introduction to Natural Generative Phonology. New York: Academic Press. Inkelas, Sharon and Zoll, Cheryl. 2000. Reduplication as morphological doubling. Rutgers Optimality Archive: http://roa.rutgers.edu/view. php3?id=425. Access date: July 2010. Isaev, M.I. 1966. Osetinskij jazyk. In Vinogradov, V.V (ed.), Jazyki narodov SSSR. Tom I: Indoevropejskie jazyki. Moscow: Nauka, 237–56. Jackendoff, Ray. 1990. Semantic structures. Cambridge, MA: MIT Press. Jacobsen, Thorkild. 1974. Very ancient linguistics: Babylonian grammatical texts. In Hymes, Dell (ed.), Studies in the history of linguistics: traditions and paradigms. Bloomington: Indiana University Press, 41–62. Janda, Richard D. 1982. On limiting the form of morphological rules: German umlaut, diacritics, and ‘cluster constraint’. NELS 12: 140–52. —— 1984. Why morphological metathesis rules are rare: on the possibility of historical explanation in linguistics. Proceedings of the annual meeting of the Berkeley Linguistics Society 10: 87–103. Jarema, Gonia and Libben, Gary (eds) 2007. The mental lexicon: core perspectives. Amsterdam: Elsevier. Jarvikivi, Juhani, Bertram, Raymond and Niemi, Jussi. 2006. Affixal salience and the processing of derivational morphology: the role of suffixal allomorphy. Language and cognitive processes 21.4: 394–431. Joseph, Brian D. 1998. Diachronic morphology. In Spencer and Zwicky (eds), 351–73. —— 2003. Morphologization from syntax. In Joseph, Brian D. and Janda, Richard D. (eds), The handbook of historical linguistics. Oxford: Blackwell Publishing, 472–92. —— and Janda, Richard. 1988. The how and why of diachronic morphologization and demorphologization. In Hammond and Noonan (eds), 193–210. Kager, René. 1999. Optimality theory. Cambridge: Cambridge University Press. Kageyama, Taro. 1982. Word formation in Japanese. Lingua 57: 215–58. Kaisse, E. and Shaw, Patricia A. 1985. On the theory of Lexical Phonology. Phonology yearbook 2: 1–30.
REFERENCES 
311
Kanerva, Jonni M. 1987. Morphological integrity and syntax: the evidence from Finnish possessive suffixes. Language 63.3: 498–521. Kastovsky, Dieter. 1986. The problem of productivity in word-formation. Linguistics 24: 585–600. Katamba, Francis and Stonham, John. 2006. Morphology. New York: Palgrave Macmillan. Keegan, John M. 1997. A reference grammar of Mbay. Munich: Lincom Europa. Keenan, Edward L. and Polinsky, Maria. 1998. Malagasy. In Spencer and Zwicky (eds), 563–623. Kemmer, Suzanne. 2003. Schemas and lexical blends. In Cuyckens, Hubert, Berg, Thomas, Dirven, Rene and Panther, Klaus-Uwe (eds), Motivation in language: studies in honor of Günter Radden. Amsterdam: Benjamins, 69–98. Kenesei, István. 1995–6. Bracketing paradoxes in Hungarian. Acta Linguistica Hungarica 43: 153–73. Kibrik, Aleksandr E. (ed.) 1996. Godoberi. Munich: Lincom Europa. King, Gareth. 1993. Modern Welsh: a comprehensive grammar. London: Routledge. Kiparsky, Paul. 1982. Lexical morphology and phonology. In Linguistic Society of Korea (ed.), Linguistics in the morning calm: selected papers from SICOL-1981. Seoul: Hanshin, 3–91. —— 1985. Some consequences of Lexical Phonology. Phonology yearbook 2: 85–138. —— 1996. Allomorphy or morphophonology? In Singh (ed.), 12–42. —— 2005. Blocking and periphrasis in inflectional paradigms. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 2004. Dordrecht: Springer, 113–35. Klavans, Judith L. 1985. The independence of syntax and phonology in cliticization. Language 61.1: 95–120. Koptjevskaja-Tamm, Maria. 1993. Nominalizations. London: Routledge. —— and Muravyova, Irina A. 1993. Alutor causatives, noun incorporation, and the Mirror Principle. In Comrie, Bernard and Polinsky, Maria (eds), Causatives and transitivity. Amsterdam: Benjamins, 287–313. Kornfilt, Jaklin. 1997. Turkish. London: Routledge. Kozinceva, Natalija A. 1981. Refleksivnye glagoly v armjanskom jazyke. In Xrakovskij, V[iktor] S. (ed.), Zalogovye konstrukcii v raznostrukturnyx jazykax. Leningrad: Nauka, 81–98. Lapointe, Steven G. 1980. A theory of grammatical agreement. Ph.D. dissertation, University of Massachusetts, Amherst. ——, Brentari, Diane K. and Farrell, Patrick M. (eds) 1998. Morphology and its relation to phonology and syntax. Stanford: CSLI. Layton, Bentley. 2000. A Coptic grammar with chrestomathy and glossary: Sahidic dialect. Wiesbaden: Harrassowitz Verlag. Libben, Gary and Jarema, Gonia (eds) 2006. The representation and processing of compound words. Oxford: Oxford University Press.
312 
REFERENCES
Lieber, Rochelle. 1992. Deconstructing morphology. Chicago: University of Chicago Press. —— and Scalise, Sergio. 2007. The Lexical Integrity Hypothesis in a new theoretical universe. In Booij, Geert, Ducceschi, Luca, Fradin, Bernard, Guevara, Emiliano, Ralli, Angela, and Scalise, Sergio (eds), On-line proceedings of the Fifth Mediterranean Morphology Meeting (MMM5), Fréjus 15-18 September 2005. http://mmm.lingue.unibo.it/mmm-proc/ MMM5/001-024-Lieber-Scalise.pdf. Access date: July 2010. Lima, Susan D., Corrigan, Roberta L. and Iverson, Gregory (eds) 1994. The reality of linguistic rules. Amsterdam: John Benjamins. Macaulay, Monica. 1996. A grammar of Chalcatongo Mixtec. (University of California Publications in Linguistics, 127.) Berkeley: University of California Press. MacDonald, Lorna. 1990. A grammar of Tauya. Berlin: Mouton de Gruyter. MacWhinney, Brian and Leinbach, Jared. 1991. Implementations are not conceptualizations: Revising the verb learning model. Cognition 40.1–2: 121–57. Mahootian, Shahrzad. 1997. Persian. London: Routledge. Malchukov, Andrei L. 1995. Even. Munich: Lincom Europa. Man´czak, Witold. 1980a. Frequenz und Sprachwandel. In Lüdtke, Helmut (ed.), Kommunikationstheoretische Grundlagen des Sprachwandels. Berlin/ New York: de Gruyter, 37–79. —— 1980b. Laws of analogy. In Fisiak, Jacek (ed.), Historical morphology. The Hague: Mouton, 283–8. Marchand, Hans. 1969. The categories and types of present-day English wordformation: a synchronic-diachronic approach. München: Beck. Matsumoto, Yo. 1996. Complex predicates in Japanese: a syntactic and semantic study of the notion ‘word’. Stanford, CA: CSLI Publications. Matthews, Peter. 1972. Inflectional morphology: a theoretical study based on aspects of Latin verb conjugation. Cambridge: Cambridge University Press. —— 1991. Morphology. 2nd edn. Cambridge: Cambridge University Press. McCarthy, John J. and Prince, Alan. 1993a. Generalized alignment. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 1993. Dordrecht: Kluwer, 79–153. —— 1993b. Prosodic morphology: constraint interaction and satisfaction. Rutgers Optimality Archive: http://roa.rutgers.edu/view.php3?id=590. Accessed July 2010. Mel’čuk, Igor A. 1993–2000. Cours de morphologie générale. Vols. 1–5. Montreal: Presses universitaires de Montréal. Mereu, Lunella (ed.) 1999. Boundaries of morphology and syntax. Amsterdam: Benjamins. Merlan, Francesca. 1976. Noun incorporation and discourse reference in modern Nahuatl. International Journal of American Linguistics 42: 177–91.
REFERENCES 
313
Mithun, Marianne. 1984. The evolution of noun incorporation. Language 60: 847–94. —— 1998. The sequencing of grammaticization effects: a twist from North America. In Schmid, Monika S., Austin, Jennifer R. and Stein, Dieter (eds), Historical linguistics 1997. Amsterdam: Benjamins, 291–314. —— and Corbett, Greville G. 1999. The effect of noun incorporation on argument structure. In Mereu (ed.), 49–71. Mohanan, Tara. 1995. Wordhood and lexicality: noun incorporation in Hindi. Natural Language and Linguistic Theory 13: 75–134. Nedjalkov, Vladimir P. (ed.) 1988. Typology of resultative constructions. (Typological Studies in Language, 12.) Amsterdam: Benjamins. Nevis, Joel A., Joseph, Brian D., Wanner, Dieter, and Zwicky, Arnold M. (eds) 1994. Clitics: a comprehensive bibliography, 1892–1991. Amsterdam: J. Benjamins. Newman, Paul. 2000. The Hausa language: an encyclopedic reference grammar. New Haven: Yale University Press. Nguyen, Dinh-Hoa. 1997. Vietnamese. (London Oriental and African Language Library, 9.) Amsterdam: Benjamins. Nikolaeva, Irina. 1999. Ostyak. Munich: Lincom Europa. Noonan, Michael. 1992. A grammar of Lango. (Mouton Grammar Library, 8.) Berlin: Mouton de Gruyter. Noyer, Rolf. 1998. Impoverishment theory and morphosyntactic markedness. In Lapointe et al. (eds), 264–85. Nübling, Damaris. 2001. The development of ‘junk’: Irregularization strategies of have and say in the Germanic languages. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 1999. Dordrecht: Kluwer, 53–74. Olsen, Susan. 2001. Copulative compounds: a closer look at the interface between syntax and morphology. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 2000. Dordrecht: Kluwer, 279–320. Oshita, Hiroyuki. 1995. Compounds: a view from suffixation and a-structure alteration. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 1994. Dordrecht: Kluwer, 179–205. Packard, Jerome L. 2000. The morphology of Chinese. Cambridge: Cambridge University Press. Payne, Doris L. 1990. Morphological characteristics of Lowland South American Languages. In Payne, Doris L. (ed.), Amazonian linguistics. Austin, TX: University of Texas Press, 213–41. Perevoščikov, P.N. (ed.) 1962. Grammatika sovremennogo udmurtskogo jazyka: Fonetika i morfologija. Iževsk: Udmurtskoe knižnoe izdatel´stvo. Perlmutter, David. 1988. The split-morphology hypothesis: evidence from Yiddish. In Hammond and Noonan (eds), 79–100. Pesetsky, David. 1985. Morphology and logical form. Linguistic Inquiry 16.2: 193–246.
314 
REFERENCES
Pinker, Steven. 1991. Rules of language. Science 253.5019: 530–35. —— and Prince, Alan. 1988. On language and connectionism: analysis of a parallel distributed processing model of language acquisition. Cognition 28.1: 73–193. Plag, Ingo. 1999. Morphological productivity: structural constraints in English derivation. Berlin: Mouton de Gruyter. —— 2003. Word-formation in English. Cambridge: Cambridge University Press. —— and Baayen, R. Harald. 2009. Suffix ordering and morphological processing. Language 85.1: 109–52. Plank, Frans. 1981. Morphologische (Ir-)Regularitäten. Tübingen: Narr. —— (ed.) 1991. Paradigms: the economy of inflection. Berlin: Mouton de Gruyter. —— 1994. Inflection and derivation. In R.E. Asher (ed.), The encyclopedia of language and linguistics, vol. 3. Oxford: Pergamon Press, 1671–8. Plungian, Vladimir A. 2000. Obščaja morfologija: vvedenie v problematiku. Moscow: Èditorial URSS. Prince, Alan and Smolensky, Paul. 1993. Optimality Theory: constraint interaction in generative grammar. Rutgers Optimality Archive: http:// roa.rutgers.edu/view.php3?id=845. Accessed July 2010. Raffelsiefen, Renate. 1996. Gaps in word-formation. In Kleinhenz, Ursula (ed.), Interfaces in phonology. Berlin: Akademie Verlag, 194–209. —— 1999a. Phonological constraints on English word-formation. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 1998. Dordrecht: Kluwer, 225–87. —— 1999b. Diagnostics for prosodic words revisited: the case of historically prefixed words in English. In Hall, T. Alan and Kleinhenz, Ursula (eds.) Studies on the phonological word. Amsterdam: Benjamins, 133–201. Rainer, Franz. 1988. Towards a theory of blocking: Italian and German quality nouns. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 1988. Dordrecht: Foris, 155–85. —— 1993. Spanische Wortbildungslehre. Tübingen: Niemeyer. Reh, Mechthild. 1985. Die Krongo-Sprache (nìino mó-dì). Beschreibung, Texte, Wörterverzeichnis. (Kölner Beiträge zur Afrikanistik, 12.) Berlin: Reimer. Rehg, Kenneth L. 1981. Ponapean reference grammar. Honolulu: University of Hawaii Press. Reis, Marga. 1983. Gegen die Kompositionstheorie der Affigierung. Zeitschrift für Sprachwissenschaft 2: 110–29. Rennison, John R. 1997. Koromfe. (Descriptive grammars.) London: Routledge. Rosen, Sara Thomas. 1989. Two types of noun incorporation: a lexical analysis. Language 65.2: 294–317. Rowlands, E.C. 1969. Yoruba. (Teach Yourself Books.) Sevenoaks: Hodder & Stoughton.
REFERENCES 
315
Rumelhart, David E. and McClelland, James L. 1986. On learning the past tenses of English verbs. In McClelland, James L. and Rumelhard, David E. (eds), Parallel distributed processing: explorations in the microstructure of cognition, vol. 2: foundations. Cambridge, MA: MIT Press, 216–71. Sadock, Jerrold M. 1991. Autolexical syntax: a theory of parallel grammatical representations. Chicago: The University of Chicago Press. —— 1998. On the autonomy of compounding morphology. In Lapointe, Steven G. et al. (eds), 161–87. Scalise, Sergio. 1988a. Inflection and derivation. Linguistics 26: 561–81. —— 1988b. The notion of ‘head’ in morphology. In Booij, Geert and van Marle, Jaap (eds), Yearbook of morphology 1988. Dordrecht: Foris, 229–45. —— 1994. Morfologia. Bologna: Il Mulino. —— Magni, Elisabetta and Bisetto, Antonietta (eds) 2009. Universals of language today. Amsterdam: Springer. Schachter, Paul and Otanes, Fe T. 1972. Tagalog reference grammar. Berkeley: University of California Press. Schleicher, August. 1859. Zur Morphologie der Sprache. Ser. VII. Vol. I, No. 7. St Petersburg: Eggers. Schreuder, Robert and R. Harald Baayen, 1999. War and peace: morphemes and full forms in a noninteractive activation parallel dual-route model. Brain and Language 68: 27–32. Schultink, Henk. 1961. Produktiviteit als morfologisch fenomeen. Forum der Letteren 2: 110–25. Selkirk, Elizabeth. 1982. The syntax of words. Cambridge, MA: MIT Press. Sharoff, Serge. 2002. Cˇastotnyj spisok, vtoraja versija. http://www.artint.ru/ projects/frqlist.asp. Accessed: July 2010. Shibatani, Masayoshi. 1990. The languages of Japan. Cambridge: Cambridge University Press. Siloni, Tal. 1997. Event nominals and the construct state. In Haegeman, Liliana (ed.), The new comparative syntax. London: Longman, 165–88. Sims, Andrea D. 2006. Minding the gaps: inflectional defectiveness in paradigmatic morphology. Ph.D. dissertation, The Ohio State University. Singh, Rajendra (ed.) 1996. Trubetzkoy’s orphan. Amsterdam: Benjamins. —— and Starosta, Stanley (eds) 2003. Explorations in seamless morphology. New Delhi: Sage. Skousen, Royal. 1989. Analogical modeling of language. Dordrecht: Kluwer. —— 1992. Analogy and structure. Dordrecht: Kluwer. Smirniotopoulus, Jane C. and Joseph, Brian D. 1998. Syntax versus the lexicon: incorporation and compounding in Modern Greek. Journal of Linguistics 34.2: 447–88. Sneddon, James Neil. 1996. Indonesian: a comprehensive grammar. London: Routledge. Sohn, Ho-min. 1994. Korean. (Descriptive grammars.) London: Routledge.
316 
REFERENCES
Spencer, Andrew. 1991. Morphological theory: an introduction to word structure in generative grammar. Oxford: Blackwell. —— 1997. Inflectional morphology and functional heads. In Dressler, Wolfgang U. Prinzhorn, Martin and Rennison, John R. (eds), Advances in morphology. Berlin: Mouton de Gruyter, 31–50. —— 2002. Gender as an inflectional category. Journal of Linguistics 38.2: 279–312. —— and Zwicky, Arnold M. (eds) 1998. The handbook of morphology. Oxford: Blackwell. Sridhar, S.N. 1990. Kannada. (Descriptive grammars.) London: Routledge. Štekauer, Pavol. 2000. Beheading the word? Please, stop the execution. Folia Linguistica 34: 333–55. Stephany, Ursula. 1982. Inflectional and lexical morphology: a linguistic continuum. Glossologia 1: 27–55. Stump, Gregory T. 2001a. Inflectional morphology: a theory of paradigm structure. Cambridge: Cambridge University Press. —— 2001b. Default inheritance hierarchies and the evolution of inflectional classes. In Brinton, Laurel (ed.), Historical linguistics 1999. Amsterdam: Benjamins, 293–307. Sullivan, Thelma D. 1988. Compendium of Nahuatl grammar. Salt Lake City: University of Utah Press. Sutton-Spence, Rachel and Woll, Bencie. 1999. The linguistics of British sign language: an introduction. Cambridge: Cambridge University Press. Švedova, N. Ju. (ed.) 1980. Russkaja grammatika. Tom I. Moscow: Nauka. Taft, Marcus. 1994. Interactive-activation as a framework for understanding morphological processing. Language and cognitive processes 9.3: 271–94. —— and Forster, Kenneth. 1975. Lexical storage and retrieval of prefixed words. Journal of Verbal Learning and Verbal Behavior 14.6: 638–47. Thompson, Laurence and Thompson, M. Terry. 1969. Metathesis as a grammatical device. International Journal of American Linguistics 35: 213– 19. Tiersma, Peter. 1982. Local and general markedness. Language 58.4: 832–49. Toman, Jindrich. 1998. Word syntax. In Spencer and Zwicky (eds), 306–21. Topping, Donald M. 1973. Chamorro reference grammar. Honolulu: University Press of Hawaii. Vance, Timothy J. 1987. An introduction to Japanese phonology. Albany, NY: SUNY Press. van Marle, Jaap. 1985. On the paradigmatic dimension of morphological creativity. Dordrecht: Foris. Van Valin, Robert D. and LaPolla, Randy. 1997. Syntax: structure, meaning and function. Cambridge: Cambridge University Press. Velázquez-Castillo, Maura. 1996. The grammar of possession: inalienability, incorporation and possessor ascension in Guaraní. Amsterdam: Benjamins.
REFERENCES 
317
Wasow, Thomas. 1977. Transformations and the lexicon. In Culicover, Peter, Wasow, Thomas and Akmajian, Adrian (eds), Formal syntax. New York: Academic Press, 327–60. Weber, David John. 1989. A grammar of Huallaga (Huánuco) Quechua. Berkeley: University of California Press. Werner, Otmar. 1989. Sprachökonomie und Natürlichkeit im Bereich der Morphologie. Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 42: 34–47. Wiese, Richard. 1996. Phrasal compounds and the theory of word syntax. Linguistic Inquiry 27.1: 183–93. Wilkins, David. 1989. Mparntwe Arrernte (Aranda): studies in the structure and semantics of grammar. Ph.D. dissertation, Australian National University. Williams, Edwin. 1981a. On the notions ‘lexically related’ and ‘head of a word’. Linguistic Inquiry 12: 245–74. —— 1981b. Argument structure and morphology. Linguistic Review 1: 81– 114. Wurzel, Wolfgang Ulrich. 1980. Ways of morphologizing phonological rules. In Fisak, Jacek (ed.), Historical morphology. The Hague: Mouton, 43–62. —— 1987. System-dependent morphological naturalness in inflection. In Dressler et al., 59–96. —— 1989. Inflectional morphology and naturalness. Dordrecht: Kluwer. —— 1996. On similarities and differences between inflectional and derivational morphology. Sprachtypologie und Universalienforschung 49: 267–79. Ziervogel, D., Louw, J. A. and Taljaard, P. C. 1981. A handbook of the Zulu language. Pretoria: J. L. van Schaik. Zipf, George Kingsley. 1935. The psycho-biology of language: an introduction to dynamic philology. Boston: Houghton Mifflin. Zwicky, Arnold. 1977. On clitics. Bloomington: Indiana University Linguistics Club. ——1985a. How to describe inflection. In Niepokuj, Mary, VanClay, Mary, Nikiforidou, Vassili (eds), Proceedings of the Eleventh Annual Meeting of the Berkeley Linguistics Society. Berkeley, CA: Department of Linguistics, University of California Berkeley, 371–86. —— 1985b. Clitics and particles. Language 61: 283–305. —— 1991. Systematic vs. accidental phonological identity. In Plank (ed.), 113–31. —— and Pullum, Geoffrey. 1983. Cliticization vs. inflection: English n’t. Language 59.3: 502–13.
Glossary of technical terms
abessive: an inflectional value of the feature case: ‘without, lacking’ (e.g. Udmurt val-tek [horse-abe.sg] ‘without the horse’) (Section 12.1.2). ablative: an inflectional value of the feature case: ‘(away) from’ (e.g. Huallaga Quechua mayu-pita[river-abl.sg] ‘(away) from the river’) (Section 5.1). absolutive: an inflectional value of the feature case that is used to mark both the subject of an intransitive verb and the object of a transitive verb (Section 11.1.5). acceptability judgement: a native speaker’s assessment of whether a word or sentence is a possible word/sentence of the language (Section 6.1). accusative: an inflectional value of the feature case that is used to mark the direct object (e.g. Latin Marcus rosa-m [rose-acc] vidit ‘Marcus saw a rose’) (Section 5.1). acronym: an abbreviation consisting of initial letters that are read like an ordinary word, e.g. NATO [neitoʊ] (as opposed to alphabetism) (Section 3.1.5). action noun = event noun. active: an inflectional value of the feature voice (‘semantic agent is the syntactic subject’) (Section 5.1). actual word (= usual word): a lexeme that is familiar to most speakers (Section 4.2) (cf. neologism, possible word, occasionalism). adjunct: a participant in an event that is optionally expressed (Section 11.1.5) (cf. argument). adposition: a term that refers collectively to prepositions and postpositions (Chapter 7 Exploratory Exercise). affirmative: an inflectional value of the feature mood that indicates the veracity of the argument.
G L O S S A RY
319
affix: a morpheme that must attach to a base and cannot occur by itself. Usually a short morpheme with an abstract meaning (Section 2.2) (subtypes: circumfix, infix, prefix, suffix). affixation: the operation of stringing together a base and an affix (Section 3.1.1). affix compound: a morphological pattern that involves at least two stems and one affix (Section 11.2.3). agent: a semantic role; the instigator of an action (Chapter 11). agentive adjective: a deverbal adjective denoting an action performed by the modified noun, e.g. English typing, as in a typing monkey sat on the floor (Section 5.2). agent noun: a deverbal noun that refers to the agent participant of the action, e.g. English drinker (Sections 5.2, 11.3.3). agglutination = concatenation. agglutinative language: a language in which almost all words are formed by concatenation of morphemes (cf. fusional language, isolating language) (Section 4.2). agreement: a syntactic relation that requires related constituents to show identical marking for certain inflectional values (e.g. verbs often agree for number with subject or object nouns) (Section 5.3). allative: an inflectional value of the feature case: ‘motion towards, onto’. allomorph (= morpheme alternant): two roots or morphological patterns are allomorphs (of the same abstract morpheme) if they express the same meaning and occur in complementary distribution (Section 2.3) (subtypes: phonological allomorph, suppletive allomorph). alphabetism: an abbreviation consisting of initial letters that are read with the letters’ alphabet values, e.g. CD [si:di:] (Section 3.1.5). alternation: the differences in pronunciation between two (or more) phonological allomorphs (Section 2.3, Chapter 10) (subtypes: automatic alternation, morphophonological alternation). analogical extension: a kind of analogical change in which an existing morphological pattern is applied to a different or new lexeme (Section 6.4.3). analogical levelling (= levelling): a kind of analogical change in which a word-form is changed based on another word-form belonging to the same lexeme (Sections 10.2, 12.2). analogy (= analogical change): the use of similar existing words as models in the modification and creation of words (Section 6.4.3).
320 
G L O S S A RY
analytic language: a language that uses little morphology (Section 1.2) (cf. polysynthetic language, synthetic language). animacy: a semantic property of nouns that has to do with whether a noun denotes a living (or sentient) thing (Section 8.2.1). anticausative: an event-changing operation signalling that there is no ‘cause’ element and no agent role in the derived event structure (e.g. Russian anticausative Tarelka slomalas’. ‘The plate broke.’ vs. Ona slomala tarelku ‘She broke the plate’) (Section 11.1.2). antipassive: a function-changing operation that backgrounds the patient (Section 11.1.3). aorist: an inflectional value for verbs of the feature tense that indicates the occurrence of an action in the past, without indicating whether the action is completed. applicative: a valence-changing operation that creates a new object argument (e.g. German beladen ‘load onto’ requires a direct object, whereas laden ‘load’ does not) (Section 11.1.5) (subtypes: recipient applicative, locative applicative, benefactive applicative). appositional compound: a compound denoting an entity that fulfils several descriptions simultaneously, e.g. English student worker ‘worker who is also a student’ (Section 7.1). argument: a semantic role that is assigned to a noun by the verb (Section 11.1.1) (cf. adjunct). argument inheritance: a deverbal derived word is said to exhibit argument inheritance when its argument structure (and function structure) match and are dependent upon the argument/function structure of the verbal base from which it is derived (Section 11.3.1). argument structure (= semantic valence): a verb’s argument structure is the set of semantic roles that it assigns (Chapter 11). aspect: an inflectional feature of verbs that has to do with the internal temporal constituency of an event (values: perfective, imperfective, habitual, etc.) (Section 5.1). attenuative adjective: a deadjectival adjective that denotes a reduced degree of the base (e.g. English bluish from blue) (Section 5.2) (cf. intensive adjective). augmentative noun: a denominal noun denoting a larger (or otherwise pragmatically special) version of the base noun, e.g. Russian borodišča ‘huge beard’ is an augmentative of boroda ‘beard’ (Section 5.2). automatic alternation: a sound alternation that is purely phonologically conditioned (Section 10.1).
G L O S S A RY
321
auxiliary: a verb that co-occurs with a main verb in a phrase to indicate values of verbal features such as tense or mood. back-formation: the formation of a shorter, simpler word from a longer word that is perceived as morphologically complex (Section 3.2.2). bahuvrihi compound = exocentric compound. base: the base of a morphologically complex word is the element to which a morphological operation applies (Sections 2.2, 3.1.2). base modification: a collective term for morphological operations that change the pronunciation of part of the base, usually without adding segmentable material (Sections 3.1.2, 10.5). benefactive applicative: a valence-changing operation that creates a new direct object argument for the participant who is the beneficiary of the action (Section 11.1.5). blend: a lexeme whose stem was created by combining parts of two other lexeme stems, e.g. smog from smoke and fog (Section 3.1.5). blocking (= synonymy blocking): when the application of a productive rule is pre-empted by an existing word with the same meaning (Section 6.4.2). bound form: an element (word-form or affix) that is prosodically dependent on its host and cannot stand on its own in a variety of ways (Section 9.3). bound stem: a base that is not by itself a word-form and which therefore occurs only in combination with another morpheme (Section 2.2). case: an inflectional feature of nouns that serves to code the noun phrase’s semantic role (values: nominative, accusative, genitive, dative, locative, ablative, instrumental, etc.) (Section 5.1). categorial periphrasis: when a given inflectional value is always expressed by a periphrastic (multi-word) expression (e.g. the French future always involves a multi-word expression) (Section 8.8). category-conditioned degree of productivity: a measure of productivity; the ratio of the number of hapax legomena with a given morphological pattern to the total sum token frequency of all word-forms with that morphological pattern (Section 6.5). causative: an event-changing operation referring to an event that is a caused version of the base event (Section 11.1.4). causative verb: a deverbal verb denoting an action that has caused the action represented by the base verb to occur, e.g. Korean cwuk-i- ‘kill’ is the causative of cwuk- ‘die’ (Section 5.2).
322 
G L O S S A RY
cell: a position in a paradigm defined by the possible combinations of inflectional values (Chapter 2 Exploratory Exercise, Appendix to Chapter 5). circumfix: a discontinuous affix that occurs on both sides of the base (Section 2.2). citation form: a word-form that is used by convention to refer to a lexeme – e.g. when listing a lexeme in a dictionary. classifier: a morpheme, usually a lexeme stem, used to classify a noun as belonging to a semantically-based group. clipping: (a method of forming) a shortened word that does not differ semantically from the longer version (Section 3.1.5). clitic: a bound word-form – i.e. a word-form that is prosodically dependent on a host (Sections 9.2–9.3). clitic group: an expression formed by one or more clitics and the host (Section 9.3). coalescence: a diachronic change whereby two formerly free syntactic elements turn into a single word-form (Section 3.2.2) (cf. grammaticalization). combinatory potential (= subcategorization frame): the information in a lexical entry about the surrounding elements with which a word or morpheme can or must combine (Section 3.1.1). combining form: a bound stem that occurs only in a compound. comparative: an inflectional value of the feature degree (‘having a higher degree, more’) (Section 5.1). competence: the speaker’s knowledge of the linguistic system (Section 6.1) (cf. performance). complementary distribution: when two morphs occur in non-overlapping environments, they are in complementary distribution; a partial criterion for identifying allomorphs of the same morpheme (Section 2.3). complex event noun: a deverbal noun that refers to the event or action itself, and which inherits the base verb’s argument structure (Section 11.3.2). complex word: a word that is one of a group of words that show systematic covariation in their form and meaning – i.e. morphological structure (Section 1.1). complexity-based ordering: the idea that restrictions on affix order are emergent from the structure of the lexicon: affixes that are more likely to be stored in the lexicon together with their stems must occur closer to the root
G L O S S A RY
323
than affixes that are likely to be decomposed during lexical access (Section 10.4) (cf. level ordering). compositional meaning: when the meaning of a complex word is equal to the sum of the meanings of its component morphemes (Sections 4.1, 5.3). compound (= compound lexeme): a complex lexeme that is made up of more than one lexeme stem (Sections 2.1, 7.1) (subtypes include affix compound, appositional compound, coordinative compound, endocentric compound, exocentric compound, phrasal compound, synthetic compound). compounding: the formation of compounds (Sections 2.1, 3.1.1). concatenative operation: an operation that consists of stringing morphemes together – i.e. affixation or compounding (as opposed to non-concatenative operations such as base modification or reduplication) (Section 3.1.1). conceptual structure (= event structure): the formal semantic decomposition of a verb’s meaning (Section 11.1.1). concrete noun: a deverbal noun that is similar to an event noun, but which does not refer to the event or action itself; instead, it refers to the product of the action (e.g. building), a group of people associated with the action (e.g. management), etc. (Section 11.3.2). conditional: an inflectional value of the feature mood that indicates a hypothetical, unrealized action. conditioning: the environments in which different allomorphs of the same morpheme occur (Section 2.3) (subtypes: phonological conditioning, morphological conditioning, lexical conditioning). conjugation: (i) an inflection class of a verb; (ii) verb inflection in general (Section 8.2). constituent: a subgrouping within the structure of a word or sentence (Sections 1.1, 3.2.1). contextual inflection: a part of inflectional morphology consisting of features that are assigned to a word because of the syntactic context in which it appears, i.e. as a result of agreement or government; an inflectional feature may be inherent for one word-class and contextual for another (Section 5.4.1) (cf. inherent inflection). continuative: an inflectional value of the feature aspect that indicates an ongoing action. continuum approach: a hypothesis according to which morphological patterns are understood as lying on a continuum ranging from canonical inflection to canonical derivation, without sharp boundaries between types (Section 5.3) (cf. dichotomy approach).
324 
G L O S S A RY
controller (of agreement): in syntax, the constituent whose properties determine the properties of the agreeing constituent, e.g., when a noun determines the gender property of an adjective that agrees with it, the noun is the controller (Section 5.3) (cf. target). converb: a verb-form that is used for adverbial subordination (Section 5.1). conversion: a morphological pattern in which the pronunciation of the base does not change (Section 3.1.4). coordination ellipsis: a test for word status; one of two identical elements in a coordinated phrase can usually be deleted, but a compound member cannot be deleted in this way (Section 9.1). coordinative compound (= dvandva compound): a compound that refers to multiple referents corresponding to the compound members: e.g. Korean elun-ai ‘adult and child’ (elun ‘adult’, ai ‘child’) (Section 7.1). coreferential: two nouns are coreferential if they refer to the same entity (Section 11.1.2). count noun: a noun that can refer to individual entities, and can have both singular and plural forms (e.g. English table) (Section 8.2) (cf. mass noun). creativity: the creation of neologisms by unproductive patterns (Section 6.2). cross-formation: the formation of a complex word from a base that is itself complex, by removing part of the base (Section 3.2.2). cumulative expression (= fusion): the expression of multiple morphological meanings simultaneously by a single un-analyzable element (Section 4.1). dative: an inflectional value of the feature case that indicates the indirect object of a verb (Section 5.1). deadjectival: a formation whose base is an adjective is called deadjectival (Section 5.2). declarative: an inflectional value of the feature mood that indicates the proposition expressed is an unqualified statement of fact. declension: (i) an inflection class of a noun; (ii) noun inflection in general (Section 8.2). decomposition route: a means of lexical access in which a complex word is broken up into component morphemes, and information about the word (e.g. meaning) is retrieved from the individual morphemes’ lexical entries (Section 4.3) (cf. direct route). default rule: a default rule is one that applies in the general case, when no more specific rule applies (Sections 8.4, 8.6.2).
G L O S S A RY
325
defective: a lexeme is defective if some cells of its inflectional paradigm are not filled – i.e. if there are some inflectional meanings that it cannot express (Section 8.7.1). definite: an inflectional value of the feature definiteness. degree: an inflectional feature of adjectives having to do with comparison of gradable properties (values: comparative, superlative). degree of exhaustion: a measure of productivity; the ratio of words exhibiting a given morphological pattern to all words within the domain of that pattern (Section 6.5). degree of generalization (= profitability): a measure of productivity; the type frequency of the relevant morphological pattern (Section 6.5). denominal: a formation whose base is a noun is called denominal (Section 5.2). deobjective: a valence-changing operation in which the patient is removed from argument structure, or not linked to any element in function structure (Section 11.1.3). dependent: an element in a compound or syntactic phrase that modifies the head (Sections 7.1–7.3). dependent verb: a verb that is confined to a subordinate clause – i.e. to a clause that cannot stand alone as a sentence. Dependent verbs are nonfinite (Section 5.1). deponency: a type of mismatch between form and meaning. Imagine that a language has inflectional feature A with value {a}, usually expressed by form a, and value {b}, usually expressed by form b. A lexeme is deponent if it has a paradigm with the form a, but this form expresses the meaning {b}. In other words, a deponent verb has inflectional values that are ‘opposite’ to what is expected, given its formal marking (Section 8.7.2). derivation1 (= derivational morphology): the relationship between lexemes of a word family; a part of morphology that is characterized by relatively concrete morphological meanings, potential semantic irregularity, restrictions on applicability, etc. (Section 2.1 and Chapter 5) (Note: derivation1 is closely related neither to derive1 nor to derive2!). derivation2: the process of deriving1 or deriving2. derivational morphology = derivation1. derivative: a lexeme that is related to another lexeme by virtue of having been derived1 from it (Section 2.1). derive1 (A from B): build or form a complex word A on the basis of a base B (Chapter 3).
326 
G L O S S A RY
derive2 (A from B): construct a (phonological) surface representation A by applying a series of modifying rules to an underlying representation B (Section 10.4). desiderative: a deverbal derivational meaning: ‘want to do’, e.g. Inuktitut sinikkuma- ‘want to sleep’ is the desiderative of sini- ‘sleep’ (Section 5.2). deverbal: a deverbal lexeme is one whose base is a verb (Section 5.2). diachronic: having to do with language change over time (Section 6.1) (cf. synchronic). diachronic productivity: a measure of productivity; the number of neologisms with a given morphological pattern attested over a period of time (Section 6.5). dichotomy approach: a hypothesis according to which inflectional and derivational patterns have sufficiently different traits as to suggest that they represent distinct subsystems in morphological architecture (Section 5.3) (cf. continuum approach). diminutive noun: a denominal noun denoting a smaller (or otherwise pragmatically special) version of the base noun, e.g. Spanish gatito ‘little cat’ is a diminutive of gato ‘cat’ (diminutive adjectives, adverbs and verbs are also possible) (Section 5.2). direct route: a means of lexical access in which information about a complex word (e.g. meaning) is retrieved directly from the lexical entry for the complex word (Section 4.3) (cf. decomposition route). domain: the domain of a rule is the set of bases to which a rule could apply in principle (Sections 6.1, 6.3). dual: an inflectional value of the feature number (‘two’) (Section 5.1). dual-processing model: a psycholinguistic model of inflection that assumes two completely separate modes of mental processing and storage of inflected forms (Section 4.3). duplifix: an element attached to the base that consists of both copied segments and fixed segments (= a mixture of affix and reduplicant) (Section 3.1.3). durative: an inflectional value of the feature aspect that indicates an ongoing action or state. dvandva compound = coordinative compound. economy: a common way of measure the elegance of a description, particularly in the context of the lexicon (smaller lexicon = more economical description = more elegance) (Section 4.1).
G L O S S A RY
327
elative: an inflectional value of the feature case (‘motion away from’) (Chapter 4 Comprehension Exercises). elsewhere condition: the principle that more specific conditions apply before more general conditions. For instance, if two word-forms are compatible with syntactic requirements, the word-form that expresses the more specific set of features will be inserted into syntactic structure (Section 8.6.2). empty morph: a morph (generally an affix) that has no meaning but that must be posited for the sake of descriptive elegance (Section 4.1). enclitic: a clitic that follows its host. endocentric compound: a compound that consists of a head and a dependent (or several dependents); the meaning of the semantic head is a hyponym of the meaning of the entire compound (Section 7.1). ergative: an inflectional value of the feature case that indicates the agent of a transitive verb. ergative-absolutive language: a language that uses the same grammatical markers to indicate the argument of an intransitive verb and the object of a transitive verb, and a different grammatical marker to indicate the agent of a transitive verb (cf. nominative-accusative language). essive: an inflectional value of the feature case that indicates a state of being (Chapter 4 Comprehension Exercises). event-changing operation: A morphological operation that changes the argument structure of a verb (Section 11.1) (cf. function-changing operation). event noun (= action noun): a deverbal noun that refers to the event or action itself, e.g. English replacement ‘act of replacing’ (Section 11.3.2) (subtypes: simple event noun, complex event noun). event structure = conceptual structure. exclusive first person: refers to a group including the speaker but not including the addressee (Chapter 8 Exploratory Exercise) (cf. inclusive first person). exocentric compound (= bahuvrihi compound): a compound pattern that does not contain a (semantic) head and a dependent (Section 7.1). expandability: a test for compound status; dependents in phrases can be expanded by modifiers (adjectives and adverbs), but dependent members in compounds generally cannot (Section 9.1). experiencer: a semantic role: the participant that experiences an experiential situation (Section 11.1).
328 
G L O S S A RY
exponent: when a morphological pattern (e.g. -ed) expresses an inflectional feature value (e.g. past tense), it is the exponent of that feature value (Chapter 12). facilitative adjective: a deverbal derivational meaning: ‘able to undergo an action’ (e.g. Basque jangarri ‘edible’ is the facilitative adjective corresponding to jan ‘eat’) (Section 5.2). factitive verb: a deadjectival verb with the derivational meaning ‘cause something to be Adj’ (e.g. Russian černit’ ‘to blacken’ is the factitive of černyj ‘black’) (Sections 5.2–5.3). feature = inflectional feature. feature-value compatibility: the principle that the morphology must provide to the syntax a word-form whose morphosyntactic values do not conflict with those required by the syntax (Section 8.6.2) (cf. feature-value identity). feature-value identity: the principle that the morphology must provide to the syntax a word-form whose morphosyntactic values exactly match those required by the syntax (Section 8.6.2) (cf. feature-value compatibility). feature-value notation: a convention for representing the inflectional values associated with a word-form, e.g. [case: genitive]. female noun: a derivational meaning of nouns (‘female’) – e.g. English poetess Â� (derived from poet) (Section 5.2). feminine: an inflectional value of the feature gender. first (person): an inflectional value of the feature person (‘refers to the speaker’) (Section 5.1). focus: indicates prominent or new information in a discourse. formal head: in (endocentric) compounds, the lexeme that serves as the morphosyntactic locus and determines word-class, gender and inflection class (Sections 7.2–7.4). formalist orientation = generative orientation. free form: a word-form that is not bound (Section 9.2). freedom of host selection: a property of clitics such that they can take various syntactic categories as a host (Section 9.3). freedom of movement: a test for word status; constituents of a phrase may be clefted, topicalized, etc., but constituents of a word may not (Section 9.2). fronting: a type of base modification that involves changing the place of articulation of a sound (usually a vowel) so that it is pronounced closer to the front of the mouth (Section 3.1.2).
G L O S S A RY
329
function-changing operation: A morphological operation that changes the way semantic roles are linked to syntactic functions; function-changing operations are often encoded by the inflectional feature of voice (Section 11.1) (cf. event-changing operation). function structure (= syntactic valence): the set of syntactic functions of a verb’s arguments (Section 11.1.1). functionalist orientation: an approach to (morphological) research that emphasizes system-external explanation (Section 1.3) (cf. generative orientation). fusion = cumulative expression. fusional language: A language that makes a fair amount of use of morphology, but in which there is not usually a one-to-one correspondence between morphs and meanings (cf. agglutinative language, isolating language). future: an inflectional value of the feature tense (‘occurring later than the moment of speech’) (Section 5.1). gemination: a type of base modification that involves making a sound (usually a consonant) longer (Section 3.1.2). gender: an inherent property of nouns in some languages that is reflected in agreement (by adjectives, verbs and other agreement targets) and that serves to group the nouns into classes. Gender is an inflectional feature for agreement targets; typical values are masculine, feminine, but sometimes simply gender 1, gender 2, etc. generality: a common way of measuring the elegance of a description, according to which fewer descriptions capturing a larger portion of the data each is deemed more general (and thus more elegant) than more descriptions capturing a smaller portion of the data each (Section 1.3). generative orientation (= formalist orientation): an approach to (morphological) research that seeks to discover the principles of Universal Grammar, and emphasizes this mode of explanation (Section 1.3) (cf. functionalist orientation). generic: in a phrase or sentence, an expression is generic if it refers to a whole class, rather than a particular item (Section 9.1) (cf. referential). genitive: an inflectional value of the feature case (‘Â�possessor’, e.g. English ‘s in student’s book) (Section 5.1). Germanic suffixes: A class of suffixes in English that historically were mostly inherited from Germanic (Section 6.3.5). government: a syntactic relation in which one word requires another word or phrase to have a particular inflectional value (e.g. assigning a case value to a noun) (Section 5.3) (cf. agreement).
330 
G L O S S A RY
gradable adjective: an adjective that is semantically compatible with the feature degree (e.g. English happy is gradable because the comparative form happier can be formed from it) (Section 5.3). grammatical theory: an architecture for the description of grammatical structure (Section 1.3). grid: a notation convention for representing paradigms (Chapter 2 Exploratory Exercise, Appendix to Chapter 5). habitual: an inflectional value of the feature aspect (‘an event that is repeated regularly’) (Section 5.1). hapax (legomenon): a word that occurs exactly once in some corpus (Section 6.5). hapax-conditioned degree of productivity: a measure of productivity; the ratio of the number of hapax legomena with a given morphological pattern to the number of hapax legomena with all morphological patterns (Section 6.5). head: see semantic head and formal head. head-final language: a language in which syntactic heads are located at the end of their phrases (e.g. adpositions come after nouns) (Chapter 7 Exploratory Exercise). head-initial language: a language in which syntactic heads are located at the beginning of their phrases (e.g. adpositions come before nouns) (Chapter 7 Exploratory Exercise). hierarchical structure: the constituent structure of and dominance relations between elements in a word (morphemes, bases) or elements in a sentence (words, phrases) (Section 3.2.1, Chapter 7). homonymy: two word-forms are homonymous if their pronunciation is identical (Section 8.6). host: a clitic’s host is the element that a clitic combines with; a clitic’s prosodic host and syntactic host may differ (Section 9.2). hypothetical: an inflectional value of the feature mood (‘counterfactual but possible event’). idiomaticity = non-compositional meaning. imperative: an inflectional value of the feature mood (‘speaker issues command to hearer’) (Section 5.1). imperfective: an inflectional value of the feature aspect (‘an event seen from within or as not completed’) (Section 5.1).
G L O S S A RY
331
implicational relationship: a predictive relationship such that if A is true, B must also true, but not necessarily the reverse; in language typology, implicational relationships are usually probabilistic (if A is true, B has a high probability of being true), rather than absolute (Chapter 7 Exploratory Exercise). inalienable possession: indicates a relationship such that the possessor of an object cannot be changed; body parts are classic examples of inalienably possessed objects (Section 9.1). inchoative verb: a deadjectival verb expressing the derivational meaning ‘begin to be Adj., become Adj.’ (e.g. Spanish verdear ‘become green’ is the inchoative of verde ‘green’) (Sections 5.2–5.3). inclusive first person: refers to a group including both the speaker and the addressee (Chapter 8 Exploratory Exercise) (cf. exclusive first person). indeclinable: a lexeme that does not have different inflected forms, and differs from other lexemes in this respect (Section 8.2.3). indicative: an inflectional value of the feature mood (‘an event thought of as occurring in reality’) (Section 5.1). inessive: an inflectional value of the feature case (‘inside of’) (Chapter 4 Comprehension Exercises). infinitive: an inflectional meaning of verbs: a nonfinite form used for clausal complements when the complement subject is identical to the matrix subject; often has nominal properties (Section 5.1). infix: an affix that occurs inside the base (Section 2.2). inflect: When we say that a word inflects (for some value) we mean that it has (inflectional) word-forms for that value, e.g. ‘Russian verbs inflect for gender’, i.e. Russian verbs distinguish different word-forms for different genders (of the subject argument). inflection (= inflectional morphology): the relationship between wordforms of a lexeme; a part of morphology that is characterized by relatively abstract morphological meanings, semantic regularity, almost unlimited applicability, etc. (Section 2.1 and Chapter 5). inflection class: a class of lexemes that inflect in the same way – i.e. that show the same suppletive allomorphy in all word-forms of their paradigm (Sections 8.2, Chapter 12). (inflection) class shift: a diachronic change in which a lexeme belonging to one inflection class becomes associated with a different class (Sections 8.3, 6.4.3). inflectional exponent = exponent.
332 
G L O S S A RY
(inflectional) feature (= inflectional category): a set of inflectional values that share a semantic property and are mutually exclusive, e.g. tense, case and voice (Section 5.1). (inflectional) value (= inflectional property, inflectional feature value): a specified value within an inflectional feature – e.g. future (from the feature tense), accusative (from the feature case), passive (from the feature voice) (Section 5.1). inherent case: a value of the feature case that meets the criteria for inherent inflection, e.g. locative, ablative, and instrumental for nouns (Section 5.4.1) (cf. structural case). inherent inflection: a part of inflectional morphology consisting of features that are relevant to the syntax but convey a certain amount of independent information; inherent inflection is distinguished from contextual inflection by virtue of sharing properties with derivation; an inflectional feature may be inherent for one word-class and contextual for another (Section 5.4.1). inheritance1 hierarchy: a descriptive device in which a tree structure is used to represent similarities among inflection classes (i.e. similarities among rule schemas). Shared information is specified on higher nodes, and information that is specific to individual classes on lower nodes. A lower node inherits information from its mother node in the default situation (Section 8.4). inheritance2 = argument inheritance. instrumental: an inflectional value of the feature case that indicates that the marked noun is the means by which an action is accomplished (e.g. Russian rezat’ [cut.inf] nož-om [knife-ins] ‘to cut with a knife’) (Section 5.4.1). integrated affix: an affix that triggers or undergoes a morphophonological alternation, exhibits the phonotactics of a monomorphemic word, and tends to occur close to the root (Section 10.4) (cf. neutral affix). intensive adjective: a deadjectival adjective signalling an increased degree of the base, e.g. Turkish yepyeni ‘brand new’ is derived from yeni ‘new’ (Section 5.2) (cf. attenuative adjective). interfix: a semantically empty affix that occurs between the two members of a N + N compound (especially in German and some other European languages) (Section 7.1). interlinear (morpheme-by-morpheme) gloss: a notation convention used by linguists to help readers understand the structure of a morphological example (Appendix to Chapter 2). intransitive: a verb that does not take a direct object is called intransitive (cf. transitive).
G L O S S A RY
333
isolating language: a language that makes only minimal use of morphology (Section 1.2) (cf. agglutinative language, fusional language). Latinate suffixes: A class of suffixes in English that historically derive mostly from Latin or Greek via large-scale borrowing of vocabulary; Latinate suffixes combine (almost) exclusively with Latinate bases (i.e. bases of Latin or Greek origin) (Section 6.3.5). lengthening: a type of base modification that involves making a sound (usually a vowel) longer (Section 3.1.2). lenition = weakening. level ordering: a hypothesis about grammar architecture, according to which affixation rules are separated into two or more levels. These levels are ordered relative to each other in derivation2, and within each level, affixation rules operate prior to sets of phonological rules that are associated with that level (Section 10.4). levelling = analogical levelling. lexeme: a word in an abstract sense; an abstract concept representing the core meaning shared by a set of closely-related word-forms (e.g. lives, live, lived) that form a paradigm (Section 2.1). lexeme formation = word-formation. lexical access: the mental process of looking up a word in the lexicon (Section 4.3). lexical alternation = morphophonological alternation. lexical category = word-class. lexical conditioning: when the choice of allomorph is determined by the lexeme it attaches to (Section 2.3). lexical entry: a listing in the lexicon (Chapter 3). (lexical) gang: a group of words that shares an (inflectional) morphological pattern and that is also highly similar in phonological form, e.g. sling, wring, swing (Section 6.4.3). Lexical Integrity Hypothesis (= Lexical Integrity Principle): a hypothesis about the universal nature of language, stating that rules of syntax can refer and/or apply to entire words or the properties of entire words, but not to the internal parts of words or the properties of the internal parts of words (Section 9.4). lexical item: any item that is listed in the lexicon, ranging from morphemes to words to phrases (e.g. idioms) (Chapter 4).
334 
G L O S S A RY
lexical neighbourhood: a set of words that are minimally (phonologically) different from some other word. lexical periphrasis: when a given set of inflectional values is expressed by a periphrastic (multi-word) expression for some lexemes, but the same set is expressed as a single word for other lexemes (e.g. happier versus periphrastic more beautiful – both are comparatives) (Section 8.8). lexicon: the list of elements that speakers have to know in addition to the rules of grammar (Chapter 4, Section 6.4). loanword (= borrowing): a word taken from another language (Section 8.2.3). locative: an inflectional value of the feature case that indicates a location: ‘in, on, at, by,…’ (e.g. Turkish ev-de [house-loc] ‘in the house’) (Section 5.4.1). locative applicative: a valence-changing operation that creates a new direct object argument for the locational participant (Section 11.1.5). masculine: an inflectional value of the feature gender. masdar: an inflectional action noun (Sections 5.1, 11.4). mass noun: a noun that refers to a group of objects as a collective entity, rather than as a group of individual member entities; often has a form for only one number value (e.g. English information, *informations, or furniture, *furnitures) (Sections 5.3, 8.2) (cf. count noun). memory strength (= resting activation level): the relative strength of the representation of a word or morphological pattern in the lexicon (e.g. as a result of token frequency) (Sections 4.3, Chapter 12). metathesis: a type of base modification that involves switching the order of two contiguous sounds or groups of sounds (e.g. syllables) (Section 3.1.2). moderate word-form lexicon: the hypothesis that a speaker’s lexicon includes simple and complex words, morphemes (or morphological patterns), and derived stems. Whether any particular complex word is listed in the lexicon depends on a variety of factors (Sections 4.3, 6.4.1) (cf. morpheme lexicon, strict word-form lexicon). monomorphemic: containing one morpheme. monosyllabic: containing one syllable. mood: an inflectional feature of verbs that has to do with the speaker’s level of commitment to the actuality of the event, or its desirability or conditionality (values: imperative, subjunctive, indicative, conditional, optative, etc.) (Section 5.1). morph: a concrete primitive element of morphological analysis.
G L O S S A RY
335
morpheme: the smallest meaningful part of a linguistic expression that can be identified by segmentation; a frequently occurring subtype of morphological pattern (Section 1.1). morpheme alternant = allomorph. morpheme-based model: a collective term for approaches to morphological analysis in which morphological rules are thought of as combining morphemes in much the same way as syntactic rules combine words (Section 3.2.1) (cf. word-based model). morpheme lexicon: the hypothesis that a speaker’s lexicon lists (to the extent possible) only simple monomorphemic elements, i.e. roots and affixes, rather than complex words (Section 4.1) (cf. moderate word-form lexicon, strict word-form lexicon). morpheme structure condition: a restriction on the co-occurrence of sounds within a morpheme (Sections 4.2, 10.5). morphological conditioning: when the morphological context (usually, grammatical function) determines the choice of allomorph (Section 2.3). (morphological) correspondence: a convention used to represent the association between morphologically related sets of words; a morphological rule in the word-based model (Section 3.2.2). morphological pattern: a pattern across words in which a recurrent aspect of meaning corresponds to a recurrent aspect of form (Section 3.1). (morphological) rule: a formal description of a morphological pattern (Chapter 3). morphology: (the study of) systematic covariation in the form and meaning of words (Section 1.1). morphophonological alternation (= lexical alternation): an alternation that is at least partly morphologically or lexically conditioned (Chapter 10). morphophonological rule: a formal description of an alternation; a rule that derives2 a surface representation from an underlying representation by changing the shape of the word when certain (morpho)phonological conditions are met (Section 2.3). morphophonology: the study of morphophonological alternations. morphosyntactic feature: a(n inflectional) feature that is relevant to both morphology and syntax, such as case. morphosyntactic locus: the stem in a compound where morphosyntactic features are expressed (Sections 7.2, 7.4). morphosyntactic representation: a set of (inflectional) values that can be combined with a lexeme and that are relevant to both morphology and
336 
G L O S S A RY
syntax; in some theories, the bottom node in syntactic trees consists of morphosyntactic representations (i.e. inflectional values + word-class, but without specification of a particular lexeme). natural class: a group of inflected forms constitute a natural class if they are all and the only forms that express some set of inflectional values (Section 8.6.2). This term has a parallel use in phonology – a group of sounds constitute a natural class if they share some phonological feature or set of features (e.g. labial) (Section 10.1). natural syncretism: a pattern of syncretism in which the syncretic forms constitute a natural class (Section 8.6.2). negative adjective: a deadjectival adjective signalling a reversal of the (positive) quality of the base (e.g. English unhappy, derived from happy). neologism: a new lexeme that is attested, but had not previously been observed in the language. (Section 4.2, Chapter 6) (cf. actual word, occasionalism, possible word). neuter: an inflectional value of the feature gender. neutral affix: an affix that does not trigger or undergo a morphophonological alternation, may show phonotactic peculiarities, and tends to occur further from the root (Section 10.4) (cf. integrated affix). nominative: an inflectional value of the feature case (‘the case of the subject, the case-form that is used as citation form’) (Section 5.1). nominative-accusative language: a language that uses the same grammatical markers to indicate the argument of an intransitive verb and the agent of a transitive verb, and a different grammatical marker to indicate the patient of a transitive verb (cf. ergative-absolutive language). nonce formation = occasionalism. non-compositional meaning (= idiomaticity): see compositional meaning (Sections 3.2.2, 5.3, 9.1). non-concatenative operation: a morphological operation that cannot be straightforwardly described as stringing together of two morphemes (Sections 3.1.2–3.1.4). non-gradable adjective: see gradable adjective (Section 5.3). non-word: a sequence of sounds that adheres to the phonological rules of a language, and therefore sounds like a word, but has no meaning (Chapter 3 Exploratory Exercise). noun incorporation: N + V compounding in which the verb is the head; found especially in polysynthetic languages (Sections 7.1, 9.1, 11.2.1).
G L O S S A RY
337
NP = noun phrase. number: an inflectional feature of nouns, having to do with the number of items a noun refers to (values: singular, plural, dual, paucal, etc.) (Section 5.1). objective case: an inflectional value of the feature case that indicates an object role (Section 8.5). oblique1: oblique cases are all morphological cases apart from the most basic ones (e.g. all but the nominative and accusative). oblique2: syntactic functions other than the subject and direct object; the term is often used even when these functions are expressed by syntactic means, rather than morphological case (Section 11.1.1). occasionalism (= nonce formation): a neologism that has not caught on and is restricted to occasional occurrences (Sections 4.2, 6.1) (cf. actual word, neologism, possible word). optative: an inflectional value of the feature mood that indicates a desire or wish for some event to occur. palatalization: a type of base modification that involves changing the place of articulation of a sound (usually a consonant) so that it is pronounced at the palate (Section 3.1.2). paradigm: the structured set of word-forms of a lexeme (Section 2.1). (Often subsets that belong together (e.g. all past-tense forms of a verb) are also referred to as paradigms.) paradigm rule: a word-based rule consisting of multiple correspondences between word-forms in an inflectional paradigm (Section 8.3). paradigmatic gap: the ‘missing’ word-form in a defective lexeme. paradigmatic periphrasis: when a given set of inflectional values is always expressed by a periphrastic (multi-word) expression, but individual values in the set are expressed by a single-word expression elsewhere in the paradigm (e.g. in Latin, the combination of perfect and passive is always periphrastic, but the combination of perfect and active is not periphrastic) (Section 8.8). paradigmatic relations: relations between units that could (potentially) occur in the same slot (Section 8.1). parsing ratio: the percentage of words with a given morphological pattern that are decomposed (parsed) in lexical access, according to a parallel dual route model of lexical access (Section 6.4.1). participle: a deverbal adjective that may retain some verbal properties (Sections 5.1, 11.4).
338 
G L O S S A RY
partitive: an inflectional value of the feature case that denotes a subpart of a collective entity (Chapter 4 Comprehension Exercises). passive: an inflectional value of the feature voice that signals that the patient is the grammatical subject; a function-changing operation in which the agent is backgrounded (Sections 5.1, 11.1.2). past: an inflectional value of the feature tense (‘occurring earlier than the moment of speech’) (Section 5.1). patient: a semantic role; the participant that undergoes an action (Chapter 11). patient noun: a deverbal noun that refers to the verb’s patient, e.g. English invit-ee (derived from the verb invite, and indicating the person who is the recipient of the action) (Section 5.2). paucal: an inflectional value of the feature number (‘a few’) (Section 5.1). perfect: an inflectional value of the feature aspect (‘an event that took place in the past but has current relevance’). perfective: an inflectional value of the feature aspect (‘an event seen from the outside or as completed’) (Section 5.1). performance: use of language (Section 6.1) (cf. competence). periphrastic construction: a multi-word phrase that cumulatively expresses some set of inflectional values. A periphrastic construction fills a cell of an inflectional paradigm (Sections 8.8, 9.4) (subtypes: lexical periphrasis, paradigmatic periphrasis, categorial periphrasis). person: an inflectional agreement feature of verbs (person of subject or object) and nouns (person of possessor) (values: 1st, 2nd, 3rd) (Section 5.1). phonetic alternation = automatic alternation. phonological allomorph: two allomorphs are phonological if they have quite similar phonological shape (Section 2.3, Chapter 10). phonological conditioning: when the phonological context determines the choice of allomorph (Section 2.3). phrasal compound: a compound in which an entire phrase is the dependent member, e.g. a [[floor of the birdcage] taste] (Chapter 9 Exploratory Exercise). phrasal node: a non-terminal node in a syntactic tree diagram, representing a phrasal constituent (Section 7.4). phrase structure rule: in syntax, a rule stating how words may be combined to form a phrase (Section 3.2.1). plural: an inflectional value of the feature number (‘more than one’) (Section 5.1).
G L O S S A RY
339
poetic licence: the creation of neologisms by unproductive rules in a way that violates ordinary language norms (Section 6.2). polarity: an inflectional feature of verbs that indicates the positive or negative status of the event (e.g. Japanese kir-u [cut-prs] ‘cuts’, kir-ana-i [cut-neg-prs] ‘doesn’t cut’) (Section 5.1). polysyllabic: containing multiple syllables. polysynthetic language: a language that makes very extensive use of morphology (Section 1.2) (cf. analytic language, synthetic language). portmanteau morph: an affix or stem that cumulatively expresses two meanings that would be expected to be expressed separately (Section 4.1). possible word (= potential word): a lexeme that could be formed according to word-formation rules but is novel and perhaps never used before (Sections 4.2) (cf. actual word, neologism, occasionalism). postposition: similar to a preposition, except that postpositions are syntactically positioned after noun phrases rather than before them (Chapter 7 Exploratory Exercise). potential word = possible word. prefix: an affix that precedes the base (Section 2.2). present: an inflectional value of the feature tense (‘occurring simultaneously with the moment of speech’) (Section 5.1). Priscianic formation: the formation of an inflected form on the basis of another inflected form that is not closely related semantically (Section 8.5). privative adjective: a denominal adjective signalling lack of possession of the base noun (N-priv ‘lacking N’, e.g. Russian bezgolosyj ‘voiceless’, derived from the noun golos ‘voice’) (Section 5.2). proclitic: a clitic that precedes its host. productivity: a morphological pattern or rule is productive if it can be applied to new bases to create new words (Section 4.2, Chapter 6). profitability = degree of generalization. progressive: an inflectional value of the feature of aspect (‘an event that is in progress’). proportional equation: a way of representing the relationship between the model and the target in analogical change (Section 6.4.3). proprietive adjective: a denominal adjective signalling possession of the base noun (‘having N’, e.g. Hungarian nagy hatalm-ú uralkodó [great powerpropr monarch] ‘monarch having great power’) (Sections 5.2, 11.4).
340 
G L O S S A RY
prosodic dependence: a property of clitics and affixes; a prosodically dependent element is smaller than a prosodic word and must ‘lean on’ a host (Section 9.2). prosodic word: a unit that acts as the domain for word stress (Section 9.2). prototypical member: in a lexical gang, one or more word-forms that exhibit the most specific set of phonological properties relevant to the gang; the ‘centre’ of the gang (Section 6.4.3). pure stem: a stem that is not identical to any member of the inflectional paradigm (Section 7.1). purposive: a derivational meaning expressing purpose or goal-oriented action. quality noun: a derivational meaning of deadjectival nouns (e.g. English goodness from good) (Section 5.2). recipient applicative: a valence-changing operation that creates a new object argument for the recipient participant (Section 11.1.5). reduplicant: the copied element in a reduplication (Section 3.1.3). reduplication: a formal operation whereby (part of) the base is copied and attached to the base (Section 3.1.3). referential: in a phrase or sentence, an expression is referential if it refers to some specific, existing entity (Section 9.1) (cf. generic). reflexive: a primarily function-changing operation signalling that agent and patient are coreferential (Section 11.1.2). relational adjective: a denominal adjective signalling some kind of relation to the base noun, e.g. Russian korolevskij ‘royal’ is the relational form derived from korol’ ‘king’ (Section 5.2). relative clause: a subordinate clause that modifies a noun (e.g. ‘the man who ate everything’). relic alternation: an instance of allomorphy that occurs in very few words and is not productive. Typically, relic alternations were productive at an earlier stage of the language, but subsequently levelled in all but a few (high frequency) words (Section 10.2). repetitive: a deverbal derivational meaning of verbs: ‘again’ (e.g. English rewrite, derived from write) (Section 5.2). restrictions = selectional restrictions. resultative: an event-changing operation signalling that there is no ‘cause’ and ‘become’ element in the event structure (Section 11.1.2).
G L O S S A RY
341
reversive: a deverbal derivational meaning of verbs: ‘reverse or undo the effect of the base verb’ (e.g. English unfasten, derived from fasten) (Section 5.2). root: a base that cannot be analyzed further – i.e. a base that consists of a single morpheme (Section 2.2). rule = morphological rule. rule of referral: an inflectional rule stating that two cells in a paradigm have the same phonological form (Section 8.6.3). rule schema: a schema that generalizes over several different morphological rules that exhibit similarities (Section 8.4). schema = word-schema. second (person): an inflectional value of the feature person (‘refers to the addressee’) (Section 5.1). second-position clitic (= Wackernagel clitic): a well-known type of special clitic; second-position clitics appear after the first element (word or syntactic constituent) of a simple sentence (Section 9.3). segment: to break-up complex words into individually meaningful parts (Chapter 2). selectional restrictions (= restrictions): conditions that define the domain of a rule, e.g. some affixes ‘select’ only bases that have a particular phonological shape, particular semantic meaning, etc. (Sections 3.1.1, 6.1, 6.3). semantic head: the semantic head of a compound or a syntactic phrase is the hyponym of the whole expression (Sections 7.1–7.2). semantic role (= thematic relation): the semantic role of a noun phrase is the manner in which the noun participates in the action or state expressed by the verb, e.g. as agent, patient, theme, etc. (Chapter 11). semantic scope: when an affix C combines with a complex unit AB, C has semantic scope over AB; the meaning of ABC should be equal to the meaning of AB + the meaning of C (which crucially may be different than the meaning of A + the meaning of BC) (Section 7.3). semantic valence = argument structure. separability: a syntactic test of word status; phrasal constituents can be separated by other words, but words cannot intervene between parts of a word (incl. compound members) (Section 9.1). shortening: a type of base modification that involves making a sound shorter (Section 3.1.2).
342 
G L O S S A RY
simple clitic: a clitic that can appear in the same syntactic positions as a free form of the same word-class (Section 9.3) (cf. special clitic). simple event noun: a deverbal noun that refers to the event or action itself, but which does not preserve the base verb’s argument structure (Section 11.3.2). single-component hypothesis: a hypothesis about the architecture of the linguistic system according to which morphological rules are collected together into a single grammatical component that applies before, or in parallel to, syntax (Section 5.4.2) (cf. split-morphology hypothesis). singular: an inflectional value of the feature number (‘one’) (Section 5.1). source: a semantic role: the participant that is the initiating location or state for the action (Section 11.1.1). special clitic: a clitic whose syntactic distribution differs from that of free forms of the same word-class, and must be described in its own right (Section 9.3) (cf. simple clitic). split-morphology hypothesis: a hypothesis about the architecture of the linguistic system according to which morphology is divided between two grammatical components: word-formation rules apply before syntactic rules, whereas inflectional rules apply after syntactic rules (Section 5.5.1) (cf. single-component hypothesis). stative verb: a verb with the semantic property of referring to a state of existence, rather than a physical action (e.g. be is a stative verb in English) (Section 5.3). stem: the base of an inflected word-form (Section 2.2). stimulus1: a semantic role: the participant that represents the content of the experiencer’s experience (Section 11.1.1). stimulus2: in psycholinguistics, a test item presented to a participant during the course of an experiment. stress shift: a type of base modification that involves changing the syllable in a word with which primary stress is associated (Section 3.1.2). strict word-form lexicon: the hypothesis that a speaker’s mental dictionary lists both simple and complex words, regardless of whether the complex words have predictable meaning or form, but does not list affixes (Section 4.2) (cf. moderate word-form lexicon, morpheme lexicon). strong suppletion: a kind of allomorphy in which allomorphs of the same morpheme are phonologically radically different (Section 2.3).
G L O S S A RY
343
structural case: a value of the feature case that meets the criteria for contextual inflection, e.g. nominative, accusative, and genitive for nouns (Section 5.4.1) (cf. inherent case). structure preservation: the property of morphophonological alternations of not introducing new segments (Section 10.1). subcategorization frame = combinatory potential. subjunctive: an inflectional value of the feature mood (‘a non-realized event in a subordinate clause’) (Section 5.1). subtraction: a type of base modification that consists in deleting a segment (or more than a segment) from the base (Section 3.1.2). suffix: an affix that follows the base (Section 2.2). superlative: an inflectional value of the feature degree (‘highest degree, most’) (Section 5.1). suppletion: a kind of allomorphy in which two allomorphs of the same morpheme are not similar in pronunciation (Sections 2.3, 8.2) (subtypes: strong suppletion, weak suppletion). suppletive allomorph: two allomorphs are suppletive (= show suppletion) if they are not similar in pronunciation (i.e. cannot be related to each other by (morpho)phonological rules) (Sections 2.3, 8.2). surface representation: a word-form as it is actually pronounced by speakers; a form derived2 from an underlying representation by morphophonological rules (rules of derivational phonology) (Sections 2.3, 10.3). synchronic: having to do with language at a given point in time (Section 6.1) (cf. diachronic). syncretism: systematic homonymy of inflected words in a paradigm (Sections 8.6, 12.1.3). synonymy blocking = blocking. syntactic agreement = agreement. syntactic function: the syntactic function of a noun phrase is the way in which the noun phrase’s semantic role is encoded in syntactic structure, e.g. as subject, object, oblique (Chapter 11). syntactic government = government. syntactic valence = function structure. syntagmatic relations: relations between units that (potentially) follow each other in speech (Section 8.1).
344 
G L O S S A RY
synthetic compound: an N–N compound with a deverbal nominal head and with a dependent noun that fills an argument position in the head’s valence (Section 11.2.3). synthetic language: a language that uses a fair amount of morphology (Section 1.2) (cf. analytic language, polysynthetic language). target (for agreement): in syntax, the constituent whose properties are determined by the properties of another constituent, e.g. when a noun determines the gender property of an adjective that agrees with it, the adjective is the target (Section 5.3) (cf. controller). tense: an inflectional feature of verbs that has to do with the temporal location of the verbal event, especially with respect to the speech event (values: present, future, past, etc.) (Section 5.1). thematic relation = semantic role. theme1: a semantic role: the participant that undergoes a movement or other change of state (Section 11.1.1). theme2: an older term for ‘stem’. third (person): an inflectional value of the feature person (‘refers to a third party’) (Section 5.1). token frequency: a count of how frequently some structure (word-form, morpheme, lexeme, etc.) is used in some sample of language (Section 4.3, Chapter 12) (cf. type frequency). tonal change: a type of base modification that involves changing the tone pattern of a word (Section 3.1.2). transitive: a verb that takes a direct object is called transitive. Transitivity is the property of being either transitive or intransitive (Section 8.2.1). transposition: change of word-class by a morphological operation (Sections 11.3–11.4). tree diagram: in syntax and morphology, a convention for representing hierarchical constituent structure (Sections 3.2.1, 7.2–7.4). type frequency (of a morphological pattern): the number of lexemes that exhibit a given morphological pattern is that pattern’s type frequency (Chapter 3 Exploratory Exercise; Section 6.5) (cf. token frequency). underlying representation: an abstract representation that is not actually used by speakers, but that linguists postulate to simplify the rule system; morphophonological rules (rules of derivational phonology) operate on underlying representations to produce actually pronounced surface representations (Sections 2.3, 10.3).
G L O S S A RY
345
underspecification (of inflectional value): a descriptive device in which a word-form or morpheme is not specified for a given inflectional value; used to formally describe some types of syncretism, esp. natural syncretisms (Section 8.6.2). Universal Grammar: the innate part of speakers’ grammatical knowledge (Section 1.3). unproductive pattern: see productivity. usual word = actual word. valence: information about the semantic roles and syntactic functions of a verb (or sometimes another word-class) (Chapter 11). value = inflectional value. voice: an inflectional (and sometimes derivational) feature of verbs that indicates a function-changing operation (values: active, passive, reflexive, antipassive) (Section 5.1). voicing: a type of base modification that involves changing an unvoiced sound to its voiced counterpart (Section 3.1.2). vowel harmony: a phonological assimilation process that constrains the cooccurrence of vowels in the same word-form or lexeme, e.g. in Chukchi, the vowels in a word-form must be either all [i,e,u], or all [ə,a,o] (Section 9.1). VP = verb phrase. Wackernagel clitic = second-position clitic. weakening (= lenition): a type of base modification that involves changing the manner of articulation of a sound so that it is produced with less constriction (e.g. stops become fricatives, or fricatives become sonorants) (Section 3.1.2). weak suppletion: a kind of allomorphy in which allomorphs of the same morpheme are not radically different phonologically, but neither are they similar enough to be related by (morpho)phonological rules (Section 2.3). word: a word-form or a lexeme, or less commonly, a word token (Section 2.1, Chapter 9). word-based model: a collective term for approaches to morphological analysis in which the fundamental significance of the word is emphasized and the relationship between complex words is captured by formulating word-schemas that represent common features (Section 3.2.2) (cf. morphemebased model). word-class: a category of words such as ‘noun’, ‘verb’, ‘adjective’, ‘adverb’, etc. (Section 3.1.1).
346 
G L O S S A RY
word family: a set of morphologically related lexemes (Section 2.1). word-form: a word in a concrete sense; a sequence of sounds that express the combination of a lexeme and a set of grammatical meanings; a wordform can be isolated from surrounding elements in speech because it is either prosodically independent (= a free form) or a clitic and not an affix (Section 2.1, Chapter 9). word-formation (= lexeme formation): derivation and compounding (Section 2.1). word-schema: a representation of a set of morphologically related words (Section 3.2.2). word-structure rule: in the morpheme-based model, a rule stating how morphemes may combine to form a word (Section 3.2.1). word token: an instance of use of a word in some text or in speech (Section 2.1, Chapter 9 intro). zero affix (= zero expression): an inflectional value is said to be expressed by zero if there is nothing in the pronunciation that corresponds to the value, so that the presence of the value’s meaning must be inferred from this absence of form. (In derivational morphology, morphologists usually talk about conversion rather than zero expression, though there is really no deep difference); a device used within the morpheme-based model to preserve the principle of having only concatenative rules (Sections 3.2.1, 4.1).
LANGUAGE INDEX 
347
Language index
Language Index
1
Language
Family
Geographical area
ISO 639-3 Page code1 numbers
Afrikaans Indo-European, southern Africa afr 98 West Germanic Ainu Isolate northern Japan ain 242–243, (Hokkaido) 246 Akkadian Semitic ancient Mesopotamia akk 1 Albanian Indo-European Albania sqi 28, 35, 36, 45, 154, 229 Alutor Chukotko- Russian Far East alr 138 Kamchatkan (Kamchatka Peninsula) Arabic Afro-Asiatic, Middle East, arb 9, 20, 36, 87, Standard Semitic northern Africa 000795–96, 101, 149, 160, 162, 177– 178, 270 Armenian Indo-European Armenia hye 160, 239 (Eastern) Arrernte Pama-Nyungan central Australia aer 86, 154 (Mparntwe) Isolate Spain and France eus 000789, 253, 328 Basque Bella Coola Salishan central British blc 21 Columbia coast
1
ISO 639-3 is an international standard for identifying the world’s languages. Every language is assigned a unique three-letter code. The reader is encouraged to use these codes to find more information about the languages discussed in the book; the online language encyclopaedia Ethnologue (www.ethnologue.com) is helpful in this regard.
348 
LANGUAGE INDEX
Language
Family
Geographical area
ISO 639-3 Page code1 numbers
Big Nambas Austronesian, Vanuatu nmb 86 Oceanic Bulgarian Indo-European, Bulgaria bul 275–276 South Slavic Capanahua Panoan Peru kaq 145 Chamorro Austronesian, Guam cha 243 Malayo- Polynesian Chichewa Niger-Congo, Malawi nya 237–239, Benue-Congo 244 Bantu Chinese Sino-Tibetan China cmn 12, 121– (Mandarin) 0007122, 189, 247–249, 263 Chukotko- Russian Far East ckt 193, 345 Chukchi Kamchatkan (Kamchatka Peninsula) Clallam Salishan NW United States clm 37–38 Coptic Afro-Asiatic Egypt (extinct, still cop 55, 64 used as liturgical language of the Coptic church) Croatian Indo-European, Croatia, Bosnia and hrv 57–59, 196 South Slavic Herzegovina Dutch Indo-European, Netherlands, nld 69, 98, 101, West Germanic Belgium 0007198–200, 209, 271 English Indo-European, England, USA, etc. eng 2–12, 14–26, West Germanic 000730–31, 34– 35, 37–51, 53, 56, 60– 64, 67, 69, 71–73, 76– 77, 81–89, 91, 93–96, 98–100, 104, 107, 109–112, 115–119, 121–125, 127–131, 133–135, 137–149, 151–152,
LANGUAGE INDEX 
Language
Family
Geographical area
349
ISO 639-3 Page code1 numbers
English (contd) 0007156–157, 163–164, 166, 176, 180–181, 183, 185–186, 189–198, 200, 207, 209–216, 224–227, 230, 232–233, 235–237, 245–246, 249–258, 262–264, 272, 275, 319–321, 324, 327–330, 333–334, 336, 338, 340–342 Even Altaic, Tungusic Siberia eve 277 Evenki Altaic, Tungusic China, Mongolia, evn 154 Asian Russia Finnish Uralic, Finno Finland fin 64, 66, 76, -Ugric, 0007153, 198– Balto-Finnic 199 French Indo-European, France, Quebec, etc. fra 56, 86, 94, Romance 0007127, 165, 180, 183, 189, 191, 196, 198, 208, 227, 229, 266– 267, 278, 321 Frisian (West) Indo-European, northern Netherlands fry 273 West Germanic Gaelic Indo-European, Scotland gla 36 (Scottish) Celtic
350 
LANGUAGE INDEX
Language German
Family
Geographical area
ISO 639-3 Page code1 numbers
Indo-European, Germany etc. deu 6, 20, 23–24, West Germanic 000734–36, 55, 63, 66–67, 69, 87–89, 95-96, 98, 101, 104, 117, 119– 120, 124, 126, 130, 139, 142, 144, 160, 174–175, 189–192, 209, 211– 219, 221, 229–230, 242, 258– 261, 272, 320, 332 Godoberi North Caucasian, Daghestan (Russia) gdo 271 Andic Gothic Indo-European, Balkan peninsula got 269 East Germanic Greek Indo-European Greece grc 2, 4, 140, (Ancient, 185–186, Classical) 333 Greek Indo-European Greece, Cyprus ell 16, 30–31, (Modern) 000762, 133, 167–172, 182, 233, 262 Greenlandic Eskimo-Aleut Greenland kal 5–6, 88, 240 (West) Guaraní Tupi-Guarani Paraguay grn 208, 245 Hausa Afro-Asiatic, Nigeria, Niger hau 55, 193, West Chadic 0007232–233, 271 Hebrew Afro-Asiatic, Israel heb 7, 13, 120, (Modern) Semitic 0007213–216, 220, 229, 255, 260 Hindi Indo-European, India, Pakistan, hin 36-37, 48, 86 Indic, Nepal, etc. Western Hindi
LANGUAGE INDEX 
Language
Family
Hungarian Igbo Indonesian Irish Italian
Uralic, Finno- Hungary, Romania, hun 7, 203–204, Ugric Moldova 0007258–260, 339 Niger-Congo, Nigeria ibo 82 Benue-Congo Austronesian, Indonesia ind 219, 253 Malayo- Polynesian, Malay Indo-European, Ireland gle 158, 229 Celtic Indo-European, Italy, Switzerland ita 21, 30, 69, Romance 000792, 96, 125– 126, 140– 141, 146, 162–165, 180, 190, 230, 253, 275–276 Isolate Japan jpn 000727, 29, 86– 87, 121–122, 152, 194, 212–216, 241, 247– 249, 253, 263, 339 Dravidian, Karnataka kan 122 Southern (southern India) Uralic, Finno- western Siberia kca 268–269 Ugric, Ugric Niger-Congo, Kenya kik 154 Benue-Congo, Bantu Trans-New Papua New Guinea kpw 268–269 Guinea, Madang Isolate Korea kor 000722–23, 26, 86, 88, 141, 159, 321, 324 Niger-Congo, northern Burkina Faso kfz 274–275 North, Gur Nilo-Saharan, northern Sudan kgo 63–64 Kadugli- Krongo
Japanese
Kannada Khanty Kikuyu Kobon Korean
Koromfe Krongo
Geographical area
351
ISO 639-3 Page code1 numbers
352 
LANGUAGE INDEX
Language
Family
Geographical area
ISO 639-3 Page code1 numbers
Lakhota Siouan North and South lkt 191–192 Dakota Lango Nilo-Saharan, southern Sudan lno 190–192, Eastern 275–276 Sudanic, Nilotic Latin Indo-European, Italy lat 2, 16–19, Italic 000728–29, 83–84, 92, 94, 96, 98, 103–105, 110, 122, 159–160, 165–167, 172–173, 183–186, 227, 266, 270–271, 273–274, 276, 318, 333, 337 Lezgian East Caucasian, southern Daghestan lez 5–6, 64-66, Lezgic, (Russia) and 110, 160, Nakh- northern 223-224, Daghestanian Azerbaijan (eastern 230, 258– Caucasus) 260 Lithuanian Indo-European, Lithuania lit 176–178, Baltic 208 Malagasy Austronesian, Madagascar mlg 38 Malayo- Polynesian, Barito Malay Austronesian, Malaysia, Indonesia zlm 121 Malayo- Polynesian, Malayic Mangap-Mbula Austronesian, Papua New Guinea mna 38-39 Malayo- Polynesian, Oceanic Mbay Nilo-Saharan, Chad myb 55 Central Sudanic Mixtec Oto-Manguean, Oaxaca (Mexico) mig 37, 154 (Chalcatongo) Mixtecan
LANGUAGE INDEX 
Language
Family
Geographical area
353
ISO 639-3 Page code1 numbers
Murle Nilo-Saharan, southern Sudan mur 37 0007 0007Eastern Sudanic, Surmic Nahuatl Uto-Aztecan central Mexico nci 19–20, 91, (Classical) Aztecan 110 Nahuatl Uto-Aztecan, Mexico nhe 246 (Huauhtla) Aztecan Ndebele Niger-Congo, Zimbabwe, Botswana, nde 76 Benue-Congo, Zambia Bantu Old Church Indo-European, liturgical language chu 179 Slavonic Slavic 0007of various Eastern European churches Old English Indo-European, England ang 6, 53, 158, West Germanic 0007176, 268, 275–276 Old High Indo-European, Germany goh 179–180, German West Germanic 0007214–215, 217–218, 221, 272 Ossetic Indo-European, northern Caucasus oss 161 Iranian (Russia and Georgia) Paiute Uto-Aztecan, Western United States pao 154 (Northern) Numic Persian Indo-European, Iran, Afghanistan fas 26, 172 Iranian Pipil Uto-Aztecan, El Salvador ppl 277 Aztecan Pitjantjatjara Pama-Nyungan Southern Australia pjt 200–201 Polish Indo-European, Poland pol 90-91, 127, West Slavic 0007146, 199, 207, 219, 228–229, 233 Ponapean Austronesian, Micronesia pon 38-39, 89, Malayo- 192–193, Polynesian, 193, 199, Oceanic 208 Quechua Quechuan, Peru qub 36–37, 98, (Huallaga) Quechua I 0007100, 159, 318 Romanian Romance Romania ron 183
354 
LANGUAGE INDEX
Language Russian
Family
Geographical area
ISO 639-3 Page code1 numbers
Indo-European, Russia etc. rus 18–20, East Slavic 000723–25, 70, 87–89, 94, 97, 100, 117, 119–120, 134–136, 146, 148, 151–152, 163–165, 180–181, 186, 212– 213, 215– 216, 238– 239, 253, 257, 266, 269, 279, 320, 328, 331–332, 339–340 Sanskrit Indo-European, India san 6, 69, 109, Indic 0007121–122, 138, 151, 266 Serbian Indo-European, Serbia, etc. srp 63, 66, South Slavic 000776–77, 96, 110, 201– 202, 208 Somali Afro-Asiatic, Somalia som 29–30, 39, Cushitic 48 Sorbian (Upper) Indo-European, Germany hsb 97, 205, 266 West Slavic Spanish Indo-European, Spain, Latin America spa 25–26, 52, Romance 000787–89, 92–93, 99, 106–110, 118, 139, 141–142, 144, 146, 151, 182, 189, 198, 208, 218, 233, 262, 272, 326, 331 Isolate ancient Mesopotamia sux 1, 4–5 Sumerian
LANGUAGE INDEX 
Language Swahili Swedish
Family
Geographical area
355
ISO 639-3 Page code1 numbers
Niger-Congo, Tanzania etc. swh 5–6, 84–85, Benue-Congo, 88, 92 Bantu Indo-European, Sweden swe 7, 85, 88 0007North Germanic Tagalog Austronesian, northern Philippines tgl 20, 55–56, Malayo- 88, 161, 185 Polynesian, Phillipine Tamil Dravidian, India, Sri Lanka tam 159–160 Southern Tauya Trans-New Papua New Guinea tya 278 0007Guinea, Madang Sino-Tibetan, Tibet xct 141–142 Tibetan (Classical) 0007Tibeto- Burman, Tibetic Niger-Congo Nigeria tiv 65 Tiv Benue-Congo, Bantu Tiwa (Southern) Kiowa-Tanoan New Mexico tix 204–205 Tsimshian Penutian northern coast of tsi 190 (Coast) British Columbia Tümpisa Uto-Aztecan, California, Nevada par 173–174 Shoshoni Numic Turkish Altaic, Turkic, Turkey tur 22, 28, Southern 000768–69, 89, 95, 100, 106, 159, 193, 213, 215–216, 219, 233, 332, 334 Tzutujil Mayan Guatemala, Mexico tzj 00077, 13, 28, 39, 89, 240, 267–268 Udi East Caucasian, Azerbaijan udi 205 Lezgic Udmurt Uralic, Finno- central Russia udm 267, 277, Ugric, 318 Permic
356 
LANGUAGE INDEX
Language
Family
Urdu Vietnamese Welsh
Indo-European, Pakistan, India, urd 36-37, 48, 86 Indic, Nepal, etc. Western Hindi Austro-Asiatic, Vietnam vie 4–6, 82 Mon-Khmer Indo-European, Wales cym 160–161, Celtic 0007164–165, 271, 274, 276 Sepik-Ramu Papua New Guinea yee 55, 91 Niger-Congo, Nigeria yor 4–7 Benue-Congo, Defoid Niger-Congo, southern Africa zul 162, 221– Benue-Congo, 222, 233 Bantu
Yimas Yoruba Zulu
Geographical area
ISO 639-3 Page code1 numbers
SUBJECT INDEX 
357
Subject index
Page numbers in bold are pages where the term in question is in bold within the text.
Subject Index acceptability judgements, 129, 134, 136, 318 acquisition, 122 acronyms, 40, 318 action nouns, 67, 87, 99–100, 120, 133, 318 valence inheritance, 251–255, 258, 263 See also event nouns actual words, 71, 114, 129–130, 134, 318 adjectives derivational types, 89; attenuative, 320; facilitative, 86, 328; intensive, 332; privative, 339; proprietive, 258–260, 339; relational 89, 340 gradable, 93, 330 inflectional features and agreement, 82, 85, 91–93, 101 participles, relation to, 257–261 valence inheritance, 255–257 adjuncts, 243, 318 adpositions, 82, 152, 318 adverbs, 238, 262 affix compounds, 252, 319 affix ordering complexity-based ordering hypothesis, 227, 322 level ordering hypothesis, 222–223, 226–228, 231–232, 333
affixes, 19–22, 39, 54, 230, 319 combinatory potential, 34–35, 43, 47, 322 diachronic sources of, 52, 202–203 integrated and neutral, 223–227, 231, 332, 336 types of, 20; circumfixes, 322; duplifixes, 39, 326; infixes, 331; interfixes, 139, 332; prefixes, 34– 35, 39, 227, 339; suffixes, 20–21, 34, 343 agents, 85, 234–243, 247–249, 319 agent-adding operations, 241–242 agent-backgrounding operations, 237–239, 248, 253 agentive adjectives, 89, 319 agent nouns, 86–87, 255–256, 263, 319 agglutination. See concatenation agglutinative languages, 68, 319 agreement, 90–92, 99, 103, 149, 162–163, 319 interaction with syncretism, 177 word-internal, 97, 149, 150, 204–205 allomorphy, 22–27, 69, 319 complementary distribution of allomorphs, 23, 27, 158–159, 211–214, 322 conditioning of, 25–27, 214–215, 231, 323; lexical, 333; morphological, 335; phonological, 338
358 
SUBJECT INDEX
comparison of inflection and derivation, 90, 96, 101, 111–112 influence on lexical access, 73–75 triggered by zero affixes, 45, 54 See also alternations; suppletion alphabetisms, 40, 319 alternations, 23–24, 25, 211–233, 319 in the architecture of the grammar, 222–231 levelling of, 218, 273–274 productivity of, 217–220; relic alternations, 340 properties of; optionality, 216, 217, 231; phonetic coherence, 215; phonetic distance, 215 reanalysis of, 220–222, 229–230, 232 sensitivity to word/morpheme boundaries, 69, 199, 202, 206 types of; automatic, 211–212, 214–217, 231, 320; morphophonological, 211–217, 231, 232, 335 See also allomorphy analogy, 127, 132, 171–172, 181, 319 extension, 127–129, 319 levelling, 218, 273–274, 276, 319 proportional equations, 127–129, 339 analytic languages, 4–5, 320 animacy, 160, 184, 270–271, 320 anticausative verbs, 88, 238–240, 244– 245, 263, 320 antipassive verbs, 240, 244, 263, 320 applicative verbs, 88, 242–245, 263, 320, 321, 340 appositional compounds, 141, 150, 320 architecture of the grammar, 8–9, 40–41, 51–52, 68 See also morphology-phonology interface; syntax-morphology interface argument structure, 234–235, 236–262, 320 coreferentiality, 239 inheritance, 253–257, 320 linkage to function structure, 236– 237, 241–242, 244, 251 argument-mixing compounds, 249 aspect, 85, 92, 106, 320 inflectional values, 82, 84; continuative, 323; durative, 326; habitual, 330; imperfective, 330; perfective, 85 338; progressive, 339 attenuative adjectives, 89, 320
augmentative nouns, 87, 320 auxiliary verbs, 321 back-formation, 49, 51, 54, 228, 251, 321 bahuvrihi compounds. See exocentric compounds bases, 20–22, 36, 137, 200, 321 lexical access of, 73–75, 123 in the morpheme-based model, 43 semantic relation to derived words, 90, 93–94, 139–145 in the word-based model, 47–50, 51 See also affixes; allomorphy; alternations; compounds; stems; transposition base modification patterns 35–38, 54, 149, 230, 321 diachronic sources of, 53, 229 formal approaches to, 43, 45–46, 48, 63, 65–66, 75 influence on lexical access, 73 beneficiary participants, 243–244, 321 blending, 40, 321 blocking, 125–127, 132, 321 borrowing. See loanwords bound forms, 196–197, 206, 321 bound roots/stems, 21–22, 196, 321 case, 16, 82, 83–84, 321 diachronic sources of, 272, 274 frequency of use, 265–267, 269–272, 274 government, 90, 103 inflectional traits of, 90, 92, 100–101, 106 inflectional values 83; abessive, 318; ablative, 83, 318; absolutive, 243, 318, 327; accusative, 82, 83, 318; allative, 319; dative, 83, 324; elative, 327; ergative, 271, 327; essive, 327; genitive, 83, 100, 329; inessive, 331; instrumental, 100, 332; locative, 100, 270, 334; nominative, 82, 83, 100, 265, 266, 336; objective, 337; oblique, 234, 236, 237, 243, 244, 257, 337; partitive, 338; possessive, 256 categorial periphrasis, 183–184, 321 category-conditioned degree of productivity, 124, 130–131, 321 causative verbs, 88, 241–242, 321 circumfixes, 20, 322 citation forms, 83, 322 classifiers, 322 clausal arguments, 255
SUBJECT INDEX 
clefting, 197 clipping, 40, 322 clitic groups, 198, 322 clitics, 196–203, 207, 322, 327, 339 hosts for, 196–197, 198–202, 330 and lexical integrity, 203, 205 simple, 200, 342 special, 200–203, 207, 342; secondposition, 200–202, 341 coalescence, 52, 322 cognitive realism, 6–7, 11, 40–41, 64, 68, 74 combinatory potential, 34–35, 38, 43, 47, 322 See also selectional restrictions competence, 114–115, 129, 131–132, 322 complementary distribution (of allomorphs), 23, 27, 158–159, 211–214, 322 complex words, 2, 18, 33, 34, 137, 322 borrowing of, 121–122 storage in the lexicon, 60–63, 67, 70– 75, 77–80, 227; implications for affix order, 227–228; implications for blocking; 125–126; implications for productivity 123–124 complexity-based affix ordering hypothesis, 227–228, 322 compounds, 8, 18–19, 34–35, 48–49, 54, 323 agreement with dependent member, 204–205 diachronic sources of, 49, 52; borrowing of, 121 distinguishing from affixed words, 21–22 distinguishing from phrases, 190– 195, 206 heads, 139–144, 147, 149, 150 inflection of dependent member, 95, 101, 104 productivity of, 117, 121, 130, 138 types of, 137–142; affixal, 48, 252, 319; appositional, 141, 150, 320; coordinative, 141, 144, 150, 324; endocentric, 139–141, 143–144, 147, 150, 327; exocentric, 140–141, 144, 150, 327; phrasal, 209, 338; synthetic, 245, 249–253, 256, 263, 344 stress, 192, 195 valence, 245–253, 263
359
concatenation, 34–35, 40–41, 46–47, 323 diachronic sources of, 52–53 in the morpheme-based model, 41–46, 53, 54, 226 in the word-based model, 46–47, 50–51, 53, 54, 230 reduplication as, 38–39 conceptual structure, 236, 239, 323 concrete nouns, 254, 323 conditioning environments, 25–27, 214–215, 231, 323 conjugation, 159, 323 See also aspect; mood; number; person; tense constituent structure, morphological 3, 14, 44, 323 semantic interpretation of, 44–45, 145, 250–251 sensitivity to, 69, 205 and word-class membership, 261 See also hierarchical structure contextual inflection, 100–102, 104–105, 107, 109, 323 controllers, agreement, 91, 97, 103, 162, 187, 324 converbs, 86, 324 conversion, 39–40, 43, 45–46, 54, 245, 324 coordination ellipsis, 194, 195, 203, 205, 206, 324 coordinative compounds, 141, 144, 150, 324 coreferentiality, 239 correspondences. See morphological correspondences creativity, 116–117, 132, 325 cross-formation, 50–51, 54, 324 cumulative expression, 63–64, 66, 75, 324 comparison of inflection and derivation, 90, 98, 99 in the architecture of the grammar, 105, 107 deadjectival transposition, 87, 88, 89, 256–257 declension, 159, 168–170, 324 See also case, gender, number, person default rules/patterns, 171, 181, 324 defectiveness, 180–182, 184–185, 337 arbitrary derivational gaps, 93 definiteness, 325
360 
SUBJECT INDEX
degree, 85, 325 frequency of use, 266 inflectional values; comparative, 85, 93, 183, 322; positive, 266; superlative, 85, 266, 343 degree of exhaustion of a rule, 130, 325 degree of generalization of a rule, 129, 325 demonstratives, 82 denominal transposition, 87, 88, 89, 253, 258, 325 deobjective verbs, 240–241, 325 dependent members of compounds, 139–140, 150, 209–210, 248, 325 expandability, 194 inflection of, 95, 101, 104 order relative to heads, 154–155, 193 semantic interpretation of, 147, 149, 191–192, 194–195, 246, 250–252, 329 See also noun incorporation dependent verb forms, 86, 325 deponency, 182, 184, 185, 325 derivation, 18, 27, 81, 149, 225, 325 meanings, range of, 87–89 relationship to inflection continuum approach, 81, 90, 99, 102, 105–107, 109, 323 dichotomy approach, 81, 89, 98– 106, 107, 326 typical properties of, 89–98, 111–112 See also transposition derivational phonology, 226, 231 derivatives, 18, 125, 253, 261, 325 See also derivation; transposition desiderative verbs, 88, 326 deverbal transposition, 88, 89, 90, 253– 258, 260–261 diachronic changes. See analogy; back-formation; coalescence; inflection classes, diachronic shift; metonymic meaning shift; phonological reduction; reanalysis; sound change diachronic productivity, 130, 326 diminutive nouns, 87, 99, 326 Distributed Morphology, 54, 109–110 domains of rule application, 115, 326 restrictions on, 118–126 as tests for wordhood, 193, 196, 198, 202, 223 duplifixes, 39, 326
dvandva compounds. See coordinative compounds economy of expression, 272, 276 economical description, 61–62, 66, 74 elegant description, 6, 7, 11, 40 elsewhere condition, 178–179, 327 empty morphs, 63, 64–66, 75, 285, 327 endocentric compounds, 139–141, 143– 144, 147, 150, 327 ergative-absolutive languages, 327 event nouns, 254, 327 complex event nouns, 254–256, 322 simple event nouns, 254–255, 342 See also action nouns events conceptual structure, 236, 239, 323 event-changing operations, 238, 241, 244, 327 exocentric compounds, 140–141, 144, 150, 327 expandability (test for wordhood), 194, 327 experiencer participants, 234–235, 327 exponence, 269, 276, 328 external syntax, 260–261 extraction (test for wordhood), 197 factitive verbs, 88, 95, 286, 328 features. See inflectional features and values female nouns, 87, 93, 95, 271–272, 328 final devoicing, 198–199, 211–212, 215, 216 form-meaning mismatches, 180–182, 184, 185 formalisms, 41–50, 54, 70, 157, 166, 169–171 formalist orientation, 9, 328 free forms, 196–197, 200–201, 206, 328 freedom of host selection, 198, 202, 206, 328 freedom of movement, 197, 200–203, 206, 328 frequency of use of concatenative patterns, 41, 53 and defectiveness, 181 local reversals, 270–272, 273, 277 memory strength, 334; and lexical access, 73–74, 75; and productivity, 123–127, 132; and analogical levelling, 273, 274, 276 and structural asymmetries in inflection, 265–273, 276
SUBJECT INDEX 
type frequency and analogy, 128–129 and word length, 267–268, 272, 277 fricativization, 213 fronting, 36, 213, 328 function structure, 234–236, 251, 256, 262–263, 329 function-changing operations, 329; agent-backgrounding 236–238; reflexives, 239; patientbackgrounding, 240; objectcreating, 242–243; relationship to inflection, 244, 261–262 linkage to argument structure, 236– 237, 241–242, 244, 251 functionalist orientation, 9, 12, 329 fusion. See cumulative expression fusional languages, 329 gemination, 36, 54, 329 gender, 82, 106, 329 agreement, 91–92 as a head property, 144 inflectional values, 82; feminine, 328; masculine, 334; neuter, 336 relationship to inflection classes, 162–163, 184, 185 generality as a goal of description, 6, 329 generative orientation, 9, 11–12, 226, 329 generic reference, 191–192, 194–195, 329 Germanic bases/suffixes, 122, 329 glossing. See notation conventions government, syntactic, 90–91, 100, 148–149, 150, 329 gradable and non-gradable adjectives, 93, 330 grammatical functions, 16–17, 19, 26, 183 diachronic sources of, 52 syntactic relevance, 90 See also meaning grammatical theory, 8, 330 grammaticality acceptability judgements, 129, 134, 136, 318 impossible words, 118, 120 grammaticalization, 52 grid notation, 107–108, 330 hapax legomena / hapaxes, 130–131, 131–132, 135–136, 330 hapax-conditioned degree of productivity, 131, 135
361
heads, morphological, 139, 193–195, 330 formal, 143–144, 146, 147, 149, 150, 328; as morphosyntactic locus, 143, 148–149, 335 semantic, 139–142, 143, 150, 341 head-final languages, 153, 330 head-initial languages, 153, 330 hierarchical structure, 44–45, 137, 142– 150, 250, 252, 330 See also constituent structure, morphological homonymy, 174–176, 330 See also syncretism idiomaticity, 191, 330 See also meaning, compositionality of implicational relationships (typological), 153, 331 implicature, 246, 247, 251–252 impossible words, 118, 120 inalienable possession, 190, 331 inchoative verbs, 88, 93, 331 incorporation. See noun incorporation indeclinable nouns, 165, 331 infinitives, 86, 331 infixes, 20, 331 inflection, 18, 19, 27, 331 contextual and inherent, 100–102, 104–105, 107, 109, 323 diachronic sources of, 202–203 implications for a word-form lexicon, 68, 75 relationship to derivation continuum approach, 81, 90, 99, 102, 105–107, 109, 323 dichotomy approach, 81, 89, 98– 106, 107, 326 relationship to function-changing operations, 244 segmentability of, 63–65 typical properties of, 89–98 word-class changing (transpositional), 257–263 See also agreement; conjugation; declension; government, syntactic; inflection classes; inflectional features and values inflection classes, 144, 146, 179, 184, 269–270, 331 assignment of words to, 160–161 diachronic shift, 163, 165–167, 176, 184, 331 macroclasses and microclasses, 170–171
362 
SUBJECT INDEX
productivity of, 163–165, 184, 185 relationship to gender, 162–163 See also inheritance hierarchies inflectional features and values, 81–86, 332 cross-cutting, 270, 276 expression by multi-word constructions (periphrasis), 183–184 frequency of use, 265–277 notation conventions, 107–109, 328, 330 underspecification of, 176–178 See also agreement; aspect; case; gender; government; inflection; mood; number; person; tense inhabitant nouns, 30, 87 inherent case, 100–102, 104–105, 107, 109, 332 inherent inflection, 100–102, 104–105, 107, 109, 332 inheritance of arguments. See argument structure, inheritance inheritance hierarchies, 169–171 cross-classification, 170 innateness, 9, 44, 225 See also Universal Grammar instrument nouns, 87, 140–141, 246 integrated affixes, 223–227, 231, 332 intensive adjectives, 89, 332 interfixes, 139, 332 internal syntax, 260–261 intransitive verbs, 242, 246–251, 332 irregularity and frequency of use, 274–276, 277 and lexical access, 75 morphological, 159, 165 phonological, 124 semantic, 94–95, 100, 124, 132 isolating languages, 5–6, 333 iteration, 98 Latinate bases/suffixes, 122 learning, 122 lengthening, 36, 38, 39, 54, 333 lenition. See weakening level ordering hypothesis, 222–223, 226–228, 231–232, 333 levelling. See analogy lexemes, 15–18, 27, 333 word-class membership, 260–263 See also derivation; compounds; transposition
lexical access, 33, 72–75, 333 memory strength, 334; and productivity, 123–127, 132; and analogical levelling, 273, 274, 276 via decomposition, 72–75, 106, 324 via whole-word / direct-route, 72–75, 326 See also parallel dual-route processing models lexical class. See word-class lexical conditioning, 26, 27, 214, 333 Lexical Integrity Hypothesis, 203–206, 206–207, 333 lexical periphrasis, 183, 334 Lexical Phonology, 232 See also level ordering hypothesis lexicon, 33, 60–75, 103–106, 122–129, 132, 227, 334 entries (items) in, 33, 43, 60, 236, 333 morphemic, 61–66, 68, 75, 335 word-based, 61, 66–69, 70–74, 75, 123, 324 lexical gangs, 128, 132, 333, 340 lexical neighbourhoods, 334 redundancy of storage, 70–71, 169 See also lexical access loanwords, 163–165, 184, 216–219, 334 borrowed vocabulary strata, 121–122, 132 locative applicative operations, 242, 334 markedness, 277 masdars, 86, 258–260, 262, 334 mass-count distinction, 160, 324 meaning, 2, 11, 14–16 abstractness, 19, 21, 27, 90, 94 compositionality of, 50, 62, 75, 173, 323; comparison of inflection and derivation, 90, 94–95, 97, 99, 101, 107; compounds, 191; clitic-host combinations, 199–200, 202. conceptual structure, 236 derived from constituent structure, 44–45, 144–145, 250 predictability of, 17–18, 60–61, 66, 67 restrictions on word-formation, 119 role of pragmatic implicature in, 139–140, 247, 251–252, 256 semantically empty morphs, 64–65, 139, 327 See also blocking; form-meaning mismatches; valence memory. See lexical access, lexicon
SUBJECT INDEX 
mental processing. See lexical access metathesis, 37, 53, 54 metonymic meaning shift, 254 moderate word-form lexicon, 61, 70–74, 123, 324 monomorphemic words, 15, 121, 223, 224, 227, 334 mood, 82, 84–85, 92, 106, 334 inflectional values 82; conditional, 84, 323; hypothetical, 85, 330; imperative, 82, 84, 85, 271, 330; indicative, 82, 84–85, 266, 268, 331; optative, 337; subjunctive, 82, 84, 266, 343 morpheme structure conditions, 69, 223, 225, 227, 230, 335 morpheme-based model, 335 approach to morphophonological alternations, 222, 231 lexicon, 61–66, 75, 335 rules, 41–46, 53, 54, 226 syntagmatic approach to description, 157, 167, 178, 182 morphemes, 3, 11, 14–15, 19–27, 34, 54, 335 boundaries between, 69, 205, 227– 228, 230 See also affixes morphological conditioning, 26, 231, 335 morphological correspondences, 47–50, 54–55, 335 morphological patterns, types of, 34–40, 335 morphology-phonology interface, 8–9, 61, 103–105, 220, 222–231 morphophonological alternations. See alternations morphophonological rules, 23–25, 198, 227, 231, 232, 335 morphs, 334 movement tests, 197, 200–203, 206, 328 nasal assimilation, 225 nasal substitution, 219 natural classes, 174, 176, 176–177, 215, 336 neologisms, 71, 115–117, 132, 218–219, 336 blocking of, 125–126 measuring productivity, 130–131, 163–164, 184 neutral affixes, 223–227, 336
363
nominative-accusative languages, 336 nonce formations. See occasionalisms non-words, 336 notation conventions, 4, 27–29 feature-value notation, 109, 328 grid notation, 107–108, 330 interlinear morpheme-by-morpheme glossing, 27–29, 332 tree diagrams, 44–45, 142–143, 344 nouns, 7–8, 82, 83, 85, 148 derivational types, 87–88; action, 67, 87, 99–100, 120, 133, 251–255, 258, 263; agent, 86, 255–256, 263, 319; augmentative, 320; diminutive, 99, 326; event, 254–256, 322, 327, 342; female, 93, 95, 271–272, 328; inhabitant, 30; instrument, 140–141, 246; person, 271; quality, 86, 126, 340; status, 87 concrete, 254, 323 count, 160, 324 derived, 87, 256 noun incorporation, 138, 150, 191, 245– 247, 263, 336 noun phrases, 41, 42, 91, 148, 337 number, 16, 82, 83–85, 337 inflectional values, 83; dual, 83, 179, 266, 326; paucal, 83, 338; plural, 7, 82, 83, 338; singular, 16, 82, 83, 265–267, 342 object arguments, 83, 84, 234–247 object-creating operations, 242–243 obligatoriness of inflection, 92–93, 107 oblique arguments, 238, 257 occasionalisms, 71, 130–131, 136, 337 Optimality Theory, 232 palatalization, 36 alternation, 212, 215, 219, 221–222, 228 morphological pattern by itself, 36, 54, 337 paradigmatic gaps. See defectiveness paradigmatic periphrasis, 183–184, 337 paradigms, 16–17, 19, 27, 156–188, 337 in the architecture of the grammar, 156–158, 165–185 cells, number of, 93, 176–177, 322 paradigm rules, 166–167, 168, 171, 182, 337; underspecification of, 177 rule schemas, 168–171, 179, 341
364 
SUBJECT INDEX
parallel dual-route processing models, 72, 75, 326 parsing ratios, 123–124, 337 participles, 86, 172–173, 257–261, 337 patients, 85, 87, 234–235, 255, 338 incorporated, 245–252 passive constructions, 237–239 patient-backgrounding operations, 240–241, 243, 247 performance, 114–115, 131, 132, 338 periphrasis, 183–184, 321, 334, 338 person (inflectional feature), 82, 84, 91, 106, 338 inflectional values, 84; first, 84, 181– 182, 187, 327, 328, 331; second, 84; third, 84 person nouns, 87, 271 phonetic motivation for alternations, 214–215 for restrictions on morphological productivity, 119 phonological allomorphy, 23–25, 73 phonological conditioning, 25, 26, 338 phonological reduction, 52, 275–276 phonological rules, 196–197, 202, 221 phonology, 2, 3, 220–221, 222–230, 231–232 sensitivity of, to morphological structure, 69, 230–231 See also morphology-phonology interface phrasal compounds, 209, 338 phrase structure rules, 41–43, 103 poetic licence, 117, 339 polarity, 86, 339 inflectional values 86; affirmative, 265, 266; negative, 89, 265–266 polysynthetic languages, 5, 138, 339 portmanteau morphs, 64, 339 possessors, 83, 251, 253, 262 possible words, 34, 71, 114, 120, 130, 339 postpositions, 152, 339 potential words. See possible words prefixes, 20, 34–35, 69, 227, 339 Priscianic formation, 172–174, 184, 339 privative adjectives, 89, 339 processing. See lexical access productivity, 67, 114, 181, 232, 339 of alternations, 217–220, 231 compared to creativity, 116–117, 132, 325 of compound patterns, 138, 140 gradience of, 116–117
of inflection classes, 163–165, 181, 184, 185 measuring, 118, 124, 129–131, 132 restrictions on, 117–129 speakers’ knowledge of, 114–116 unproductive patterns, 49, 67, 124, 126, 132, 345; relationship to restrictedness, 130 of valence-changing operations, 245 profitability of a rule, 129, 325 pronouns, 82, 91, 199, 206 anaphoric, 194–195, 206 See also clitics proprietive adjectives, 89, 258–260, 339 prosodic dependence of bound forms, 196–197, 340 prosodic words, 196, 203, 340 pure stem, 138, 340 purposive, 340 quality nouns, 86, 87, 126, 340 reanalysis, 55, 202, 221–222, 229 reduplication, 38–39, 43, 48, 54, 340 referentiality, 100, 141, 191–192, 194– 195, 206, 340 reflexivity, 239, 244, 247, 263, 340 regularity of inflection, 75, 124, 132, 274–276, 277 relevance of inflection to the syntax, 90–92, 101–106, 257–263 relic alternations, 217–218, 220, 340 repetitive verbs, 88, 340 resultative verbs, 238–239, 244, 263, 340 retrieval from memory. See lexical access reversive verbs, 88, 341 roots, 21–22, 23, 27, 34, 341 rules of referral, 179–180, 341 rule schemas, 168–171, 179, 341 rules, 60, 75, 335 domains of, 115 in the morpheme-based model, 41–46, 53, 54, 226 in the word-based model, 46–53, 54–55, 141, 157, 165, 335. See also paradigms, paradigm rules; rule schemas; morphological correspondences schemas. See rule schemas; wordschemas second-position clitics, 200–202, 341 segmentability, 64–65, 189
SUBJECT INDEX 
selectional restrictions, 34, 115, 341 See also combinatory potential semantic roles, 85, 234–253, 341 See also agents; beneficiary participants; experiencer participants; patients; source participants; stimulus participants; theme semantic scope, 145, 341 semantic valence. See argument structure semantic-role structure. See argument structure separability (test for wordhood), 193– 195, 341 shortening alternation, 213–215, 225–226 morphological pattern by itself, 37–38, 54, 341 sign languages, 2, 3, 12 simple clitics, 200, 342 single-component hypothesis, 106, 107, 342 sound change, 214–215, 220 source participants, 234–236 special clitics, 200–203, 207, 342 speech style, 216, 217 Split Morphology Hypothesis, 102–105, 107, 109, 342 stative verbs, 93, 238, 342 status nouns, 87 stems, 20–22, 23, 172–174, 342 alternations, 36–38 suppletion, 25–26, 27, 172, 276 stimulus participants, 234–235, 342 storage. See lexicon, lexical access stress, 223–227 clitic groups, 198 compound, 192, 195 contrastive, 196–197 stress shift, 37, 54, 342 strict word-form lexicon, 61, 66–69, 70, 75, 342 structural cases, 100, 343 structure preservation, 216, 343 subcategorization frames. See combinatory potential subject arguments, 82, 83, 84, 234–235 subtraction, 37, 54, 343 suffixes, 20–21, 34, 343 suppletion, 24–26, 27, 172, 218, 276, 343 affixal, 25, 26; basis for inflection
365
classes, 158–165, 184 and clitics, 199, 202 surface representations, 23–25, 221, 343 syncretism, 184, 185, 343 compared to accidental inflectional homonymy, 174–176 formal description of rules of referral, 179–180, 341 underspecification, 176–178, 185, 345 and frequency of use, 268–273, 276 natural, 176,177, 336 synonymy blocking. See blocking syntactic functions, 82, 85, 234–244, 251, 262, 343 syntactic valence. See function structure syntax, 41–44, 67, 137, 145, 178, 260–262 agreement and government, 90–92, 100, 163, as a diachronic source of morphology, 52–53 heads in, 147–149, 150, 153–155 phrases compared to words, 190–206 productivity of, 114, 123 See also external syntax; internal syntax; syntax-morphology interface syntax-morphology interface, 8–9, 61–62, 109–110 comparison of morphology and syntax, 44–45, 147–149, 150 feature-value compatibility, 177, 328 feature-value identity, 177, 328 Lexical Integrity Hypothesis, 203– 206, 206–207, 333 morphosyntactic representations, 103–105, 105–106, 335 morphosyntactic features, 143, 206, 335 single-component hypothesis, 105–106 Split Morphology Hypothesis, 102–105 See also inflection synthetic compounds, 245, 249–253, 256, 263, 344 synthetic languages, 4–5, 344 system-external explanation, 6–9, 11, 41, 53, 106, 265 targets, agreement, 344
366 
SUBJECT INDEX
tense, 82–85, 92, 344 inflectional values, 84; aorist, 320; future, 82, 329; past, 82, 83, 84, 85, 338; perfect, 183, 338; pluperfect, 183; present, 81–85, 181, 266, 339 tests for wordhood, 191–202, 206 thematic relations. See semantic roles theme, 234–236, 248–249, 344 tonal change, 37, 38, 54, 344 topicalization, 197 transitive verbs, 248, 255–256, 266, 344 transposition, 87–89, 90, 96–98, 253–262, 344 tree diagrams, 44–45, 142–143, 148, 344 trisyllabic shortening, 213, 214–215, 225–226 underlying representations, 23–25, 221, 344 underspecification, 176–178, 185, 345 uninflectedness, 92, 104, 149 Universal Grammar, 9, 206, 345 universal patterns, 7 usual words. See actual words valence, 234–263, 345 valence-changing operations, 234– 245, 262 velar softening, 226 verbs, 82, 83, 84–87, 88 auxiliaries, 321 dependent verb forms, 86 derivational types 88; anticausative, 88, 238–240, 244–245, 263, 320; antipassive, 240, 244, 263, 320; applicative, 88, 242–245, 263; causative, 88, 241–242, 321; deobjective, 240–241, 325; desiderative, 88, 326; factitive, 88, 95, 286, 328; inchoative, 88, 93, 331; repetitive, 88, 340; resultative, 238–239, 244, 263, 340; reversive, 88, 341 transitivity, 242, 246–251, 255–256, 266, 332, 344 statives, 93, 238, 342 voice, 85, 161, 266, 345 inflectional values 85; active, 85, 219, 269; passive, 85, 183, 237–239, 244, 263, 265, 338 vowel harmony, 193, 195, 198, 345
vowel lengthening. See lengthening Wackernagel clitics. See second-position clitics weakening, 36, 38, 54, 229, 345 word boundaries, 189–190, 193, 196, 216–217, 231 word families, 17, 18, 62, 346 word formation, 18, 19, 346 in the architecture of the grammar, 103–105 restrictions on, 117–122 See also compounding; derivation Word and Paradigm morphology, 54 Word Syntax, 54, 109–110 word tokens, 15, 16, 27, 131, 189, 346 word-based model, 41, 46–53, 54, 61, 67–69, 70–74, 75, 128–129, 165–167, 177–178, 345 lexicon, 61, 66–69, 70–74, 75, 123, 324 paradigmatic approach to description, 156–158, 165–185 rules, 46–53, 54–55, 141, 157, 165, 335. See also paradigm rules; rule schemas; morphological correspondences word-class 35, 43, 183, 345 comparison of inflection and derivation, 90, 96–98, 101 determined by compound head, 144 determined by derivational affixes, 35, 87, 146 word-class-changing (transpositional) derivation, 87–89, 253–257 word-class-changing (transpositional) inflection, 97–98, 257–262, 263 word-forms, 15–19, 21, 27, 346 words, 1–3, 15–19, 345 distinguishing from phrases, 189–210 word-schemas, 46–53, 54, 128, 166, 346 zero affix (zero expression), 45, 346 lack of overt inflectional exponence, 92, 158; and frequency of use, 267–268, 272, 276 problems for the morpheme-based model, 45–46, 54, 63, 64–66, 75, 230