Roots and Patterns: Hebrew Morpho-syntax

ROOTS AND PATTERNS Studies in Natural Language and Linguistic Theory VOLUME 63 V Managing Editors Marcel den Dikken, ...

Author: Maya Arad

328 downloads 2156 Views 9MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

ROOTS AND PATTERNS

Studies in Natural Language and Linguistic Theory VOLUME 63 V Managing Editors Marcel den Dikken, City University of New York Liliane Haegeman, University of Lille Joan Maling, Brandeis University Editorial Board Guglielmo Cinque, University of Venice Carol Georgopoulos, University of Utah Jane Grimshaw, Rutgers University Michael Kenstowicz, Massachusetts Institute of Technology Hilda Koopman, University of California, Los Angeles Howard Lasnik, University of Connecticut at Storrs Alec Marantz, Massachusetts Institute of Technology John J. McCarthy, University of Massachusetts, Amherst Ian Roberts, University of Cambridge

The titles published in this series are listed at the end of this volume.

ROOTS AND PATTERNS Hebrew Morpho-syntax

by

MAYA ARAD Stanford University, Stanford, CA, U.S.A.

A C.I.P. Catalogue record for this book is available from the Library of Congress.

ISBN-10 ISBN-13 ISBN-10 ISBN-13 ISBN-10 ISBN-13

1-4020-3245-5 (PB) 978-1-4020-3245-5 (PB) 1-4020-3243-9 (HB) 978-1-4020-3243-1 (HB) 1-4020-3244-7 (e-book) 978-1-4020-3244-8 (e-book)

Published by Springer, P.O. Box 17, 3300 AA Dordrecht, The Netherlands. www.springeronline.com

Printed on acid-free paper

All rights reserved. C 2005 Springer No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, microfilming, recording or otherwise, without written permission from the Publisher, with the exception of any material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work. Printed in the Netherlands.

CONTENTS A CKNOWLEDGEMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii C HAPTER 1. R OOTS : W HERE S YNTAX , M ORPHOLOGY , AND THE L EXICON M EET . . . . . . . . . . . . . . . . . . . . . . . . .

1

1.1. W HY R OOTS ? T HE D ECOMPOSITION D EBATE . . . . . . . . . . . . . . . . 2 1.2. D ISTRIBUTED M ORPHOLOGY AND THE S YNTAX M ORPHOLOGY I NTERFACE . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.3. H EBREW M ORPHOLOGY : R OOTS AND P ATTERNS . . . . . . . . . . . . . 12 1.4. T HE A RGUMENT FOR THE R OOT : S TRUCTURE AND S COPE OF THE B OOK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 C HAPTER 2. T HE N OUN V ERB A SYMMETRY IN H EBREW: W HEN A RE P ATTERNS O BLIGATORY ? . . . . . . . . . . . . . 2.1. 2.2. 2.3. 2.4.

I NTRODUCTION : R OOTS AND F EATURES . . . . . . . . . . . . . . . . . . . . . . H EBREW R OOTS AND P ATTERNS : T HE V ERBAL S YSTEM . . . . . . T HE N OUN V ERB A SYMMETRY IN H EBREW . . . . . . . . . . . . . . . . . A CCOUNTING FOR THE A SYMMETRY : T HE O BLIGATORINESS OF I NFLECTION ? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5. A CCOUNTING FOR THE A SYMMETRY : T HE R EALIZATION OF G RAMMATICAL F EATURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.6. T HE S TUFF R OOTS ARE M ADE OF : C ONSTRAINTS ON H EBREW V ERB -F ORMATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.7. S UMMARY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

24 25 27 31 36 39 44 52

C HAPTER 3. T HE C ONTENTS OF THE R OOT : M ULTIPLE C ONTEXTUALIZED M EANING IN H EBREW . . . . . . . . . . 53 3.1. I NTRODUCTION : R OOTS AND THEIR I NTERPRETATIONS . . . . . . . . 55 3.2. M ULTIPLE C ONTEXTUALIZED M EANING IN H EBREW . . . . . . . . . . 71 3.3. M ULTIPLE C ONTEXTUALIZED M EANING AND THE R OOT H YPOTHESIS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

v

vi

C ONTENTS

C HAPTER 4. T HE M ORPHOLOGICAL C ONSEQUENCES OF MCM: A N I NTERMEDIATE S UMMARY . . . . . . . . . . 106 4.1. B INYANIM AND THEIR P ROPERTIES . . . . . . . . . . . . . . . . . . . . . . . . . . 107 4.2. R OOTS ACROSS P ATTERNS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 4.3. R EGULARITY AND I RREGULARITY P REDICTED AND E XPLAINED 115 C HAPTER 5. R OOTS ACROSS P ATTERNS IN H EBREW . . . . . . . . . . . . 118 5.1. I NTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2. R OOTS ACROSS P ATTERNS : T YPES AND T OKENS . . . . . . . . . . . . . 5.3. T HE R ELATION BETWEEN V ERB A LTERNATIONS AND M ORPHOLOGICAL F ORM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4. A M ORPHOLOGICAL A CCOUNT OF THE H EBREW V ERBAL S YSTEM: B INYANIM AS I NFLECTIONAL C LASSES (A RONOFF 1994) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5. A S YNTACTIC A CCOUNT OF THE H EBREW V ERBAL S YSTEM: B INYANIM AS A S PELL -O UT OF S YNTACTIC H EADS (D ORON 1999, 2003) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.6. B INYANIM AND THE T YPOLOGY OF V ERB A LTERNATIONS . . . .

118 121 127

150

157 173

C HAPTER 6. A T HEORY OF H EBREW V ERBAL M ORPHO - SYNTAX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182 6.1. P RELIMINARIES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 6.2. A T HEORY OF H EBREW V ERBAL M ORPHO - SYNTAX . . . . . . . . . . . 189 6.3. S UMMARY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235 C HAPTER 7. R OOTS IN W ORD -F ORMATION : T HE R OOT H YPOTHESIS R EVISITED . . . . . . . . . . . . . . . . . . . . . . . . . . . 240 7.1. R OOTS AND W ORD -F ORMATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2. R OOT -D ERIVED V ERBS AND N OUN -D ERIVED V ERBS . . . . . . . . . 7.3. I N THE A BSENCE OF M ORPHOLOGY : T HE S EMANTIC P ROPERTIES OF D ENOMINALS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4. T HE R EMAINING P IECE : V ERB -D ERIVED N OUNS . . . . . . . . . . . . . 7.5. B ACK TO THE R OOT : T HE P HONOLOGICAL P ROPERTIES OF D ENOMINALS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.6. R OOTS : B ETWEEN THE U NIVERSAL AND THE L ANGUAGE S PECIFIC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

241 245 254 261 265 271

R EFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275 I NDEX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283

ACKNOWLEDGEMENTS This book presents a theory of the morphology and the lexical semantics of the Hebrew verb. It therefore owes a great deal to preceding work on morphology and on lexical semantics. I am particularly indebted to the work of Hagit Borer, Beth Levin and Alec Marantz, whose work served as a framework for the discussion. Borer’s work on the syntax–semantics interface, Levin’s work on the semantics of verb classes, and Marantz’s work on morphology were all inspiring and instrumental in developing the theory presented here. I have been fortunate to have fruitful discussions with all three, for which I am very grateful. I am equally grateful to the work of Outi Bat El and Edit Doron, highlighting the nature of the phonology and semantics of the Hebrew verbal system, respectively. If I argued against some aspects of their work, it is always out of great respect and admiration. Outi Bat El’s seminal work on the phonology of Hebrew was instrumental in forming my defense of the Hebrew root. I thank her for constructive criticism and for her enlightening comments. Edit Doron’s elegant account of the semantics and morphology of the Hebrew verbal system is an admiringly perfect system, and forms the background for much of this book. If only Hebrew verbs were as perfect! Many more people have helped in forming the ideas presented here. I wish to mention, in particular, Carlos Arregi, Lé´ na Baunaz, Rose Marie Déchaine, ´ Guy Deutscher, Miriam Engelhardt, Olga Fernandez-Soriano, Morris Halle, Daniel Harbour, Larry Hyman, Ali Idrissi, Idan Landau, Jean Lowenstamm, Victor Manfredi, Ora Matushansky, Léa V ´ Nash, Andrew Nevins, David Pesetsky, Jean Francois Prunet, Malka Rappaport Hovav, and Jan Wouter Zwart. Parts of this work were presented in several venues and conferences. For comments, criticism, input and thought provoking questions I wish to thank the audiences at Harvard, Stanford, MIT, UC Berkeley, Fondació´ n Ortega y Gasset, Madrid, Hungarian Academy of Sciences, Budapest, the University of Paris 8, the University of Venice, the University of Florence, the University of Geneva, the conference on the syntax of Semitic languages at Urbana-Champaign, the workshop on Root and Template morphology at USC, the Applicative Fest at MIT, the EPP and phases workshop at MIT, GLOW 23 Braga, in Portugal and GLOW 26 in Lund, Sweden.

vii

viii

A CKNOWLEDGEMENTS

I greatly benefited from questions and comments made by my students in the seminar of Semitic Syntax at Harvard University, the seminar on morphosyntax at the University of Geneva, and the class on comparative morphology at the New York Institute of Cognitive and Cultural Studies, University of St. Petersburg. Reviel Netz is to be thanked for careful proofreading and editing and general endless help with the manuscript. One person contributed to this book more than anyone else. Ur Shlonsky, my friend and colleague, shared with me his time, his insights, and his passion for Semitic morpho-syntax. His ideas and his inspiration are to be found throughout. This book would not have been possible without him. Finally, I am indebted to the Swiss National Science Foundation, for Research Grant No. 1214-057005.99, which made this work possible.

CHAPTER 1 ROOTS: WHERE SYNTAX, MORPHOLOGY, AND THE LEXICON MEET 1.1. 1.2. 1.3. 1.4.

Why Roots? The Decomposition Debate . . . . . . . . . . . . . . . . . . . . . . . . . 2 Distributed Morphology and the Syntax–Morphology Interface . . . . . 8 Hebrew Morphology: Roots and Patterns . . . . . . . . . . . . . . . . . . . . . . . . 12 The Argument for the Root: Structure and Scope of the Book . . . . . . . 14

This book provides an argument for a view of words as the syntactic combination of roots and morpho-syntactic features. In other words, it defends at least two controversial claims: that the theory of word-formation should make room for roots r as atomic lexical elements and that word-formation is a syntactic process. The first claim has its origins in the traditional grammar of Semitic languages but has recently come under attack, notably by Aronoff (1994) and Bat El (1994). The second claim has been made by a number of linguists with respect to some aspects of word-formation (see, e.g. Baker’s 1988 syntactic approach to incorporation). It is currently adopted in its most extreme form by the framework of distributed morphology, which I shall explore here, but is far from being the standard view in generative grammar. The aim of this book is threefold. First, it provides an analysis of Hebrew morpho-syntax incorporating and refining some of the intuitions of traditional Hebrew grammar. Second, it develops a theory of syntactic word-formation, illustrating in detail the nature of the root and the way it syntactically combines with grammatical elements. Finally, by spelling out in detail the way the Hebrew verbal system works, it provides an argument in favor of the main tenets of distributed morphology. This introductory chapter is organized as follows. Section 1.1 introduces the general debate concerning the decomposition of words into atomic units. Section 1.2 presents the theoretical framework of distributed morphology and Section 1.3 focuses in on the debate concerning the status of roots in a Semitic language like Hebrew. Following this theoretical background, Section 1.4 provides an outline of the argument of the book as a whole.

1

2

C HAPTER 1 1.1. WHY ROOTS? THE DECOMPOSITION DEBATE

Everyone knows that languages have “words”. Anyone who has tried to look into the matter also knows that the definition of a “word” is a very complex issue. Words may be very small—for example, the contracted negation n’t. Words W can also be very large. Examples of this are found in abundance: disentangle, abnormality and even antidisestablishmentarianism, to quote a textbook example. Words are often visibly complex. As traditional grammars invariably assert, W the word “disentangle” is composed of two prefixes—“dis” and “en”, and the stem “tangle”. Similarly, words such as “unbutton”, “reheat”, “precook”, or “caramelize” are composed of several units. In addition, words are often related to one another in a way that suggests they share some piece of their sound and meaning. Consider (1): (1)

a. b. c. d.

destroy, destruction, r destroying o write, writer, written t glory, r glorify, glorious fish (n), fish (v), fishing

Few would challenge the idea that words may be complex entities. What is debatable are the following two issues. First, in what way are words such as those in (1) above related? Is the relation between them just a matter of phonological similarity, or do they actually contain a common unit, or core—an entity that has to be registered separately from any of the language’s words? This is the question concerning the existence of decomposition. Second, how far does the decomposition of words go? Is it there only in cases such as “unbutton”, “precook”, and “decomposition”, or does it extend to all words? In other words, is “cat” decomposable too? The debate about the decomposition of words is not new. Rooted in analytic philosophy, it was introduced into linguistics with the work of generative semanticists, in particular by Postal (1969, 1970). In this form, the debate takes primarily not a morphological, but a lexico-semantic form. Postal (1969) proposed a decomposition of “pork” into “[MEAT] from [PIG]” in which the concepts [MEAT] and [PIG] are taken to be constitutive of the meaning of “pork”. Yet neither [MEAT] nor [PIG] are in and of themselves atomic elements of meaning in “[MEAT] from [PIG]” (the status of “from”, a relation, is more complicated) since they can be further decomposed into more basic elements. Indeed, later theories of decomposition, most notably Jackendoff (1972, 1990) and Pustejovsky (1995), have developed much more elaborate decomposition systems. Both these authors assume a certain meta-language, containing only the most primitive atoms of meaning, from which all other,

S YNTAX, M ORPHOLOGY Y,

AND THE

L EXICON

3

more complex elements, are built. The following decomposition is suggested, for example, by Jackendoff (1990: 80) for the verb “drink”: (2)

[V Pj > [ natural kinds) > verbs

While Hebrew morpho-syntax serves to make the phenomenon of MCM both more widespread and more visible, word-formation by the combination of root and environment is not specific to Hebrew. To see this wider, cross-

T HE C ONTENTS

OF THE

R OOT

57

linguistic motivation to our discussion, let us now take an example from English in Section 3.1.2, before proceeding further with the Hebrew data. A final clarification: this chapter deals with MCM—the phenomenon where several verbs and/or nouns are created from a single root. I take the multiple case as my subject here, as it provides us with a direct probe into the semantic relationship between the root and the words it gives rise to. I do not mean to imply that roots that create only a single word in the language necessarily have to have a more highly specified semantic content (though this may often be the case). It can also be that a root that potentially could give rise to MCM, simply happens to be assigned an interpretation in one context only. This is not a non-empirical claim, as in principle one could devise tests for such possible semantic ranges: even where a root is assigned only one meaning, there are potential words (especially using nominal patterns) one can create in Hebrew. It is quite feasible to ask speakers what range of meanings they are willing to associate with the new coinage (it is in fact often through this route that new words are introduced into the language). Of course, such tests are beyond the scope of this study, but their possibility shows that the semantic range of a Hebrew root is, in principle, more than just the sum of its actual verbal and nominal forms. In this chapter, however, my study takes those actual forms as its starting point. 3.1.2. Roots and Environments in English Word-formation To see the effect of the environment where the root is embedded, consider the case of English nominalizations as discussed by Marantz (1997). Marantz notes the following asymmetry between nouns and verbs. In a verbal environment roots always take an external argument, while in nominalizations, the availability of an external argument depends on √ properties of the root itself. Consider the difference between roots of the type grow and roots of the type √ destroy: (4)

a. b. c.

(5)

a. b. c.

John grows tomatoes. John’s growth of tomatoes. Tomatoes’ growth.

∗

The army destroyed the city. The army’s destruction of the city. The city’s destruction. √ √ Both grow and destroy take an √ external argument when they create a verb (4a, 5a). In a nominal environment, destroy allows an external argument (5b),

58

C HAPTER 3

√ although this argument is not obligatory, as shown in (5c). grow, w on the other hand, does not allow an external argument in its nominalization (4b). The dif√ √ ferent behavior of grow and destroy under nominalization is explained by √ the different semantic content of the two roots. grow belongs to the semantic class of “internally caused change of state” (Levin and Rappaport Hovav 1995). It refers to a change of state that is induced internally, by the same √ element that undergoes the change of state. destroy, on the other hand, is an “externally caused change of state”, that is, a change that is initiated by an outside causer, which does not undergo the change of state itself. According to Marantz (1997), external arguments in nominalizations are projected in the specifier of D and are interpreted as possessors. The ungrammaticality of “John’s growth of tomatoes” results, according to Marantz, from an incoherence between the semantic content of the root, i.e. an inherent change of state (growth) and the interpretation assigned to arguments at the specifier of Ds—that of possessors or originators of the event. Growth is induced internally, hence√ an external element, John, cannot be interpreted as its possessor, or initiator. destroy, on the other hand, refers to an externally caused change of state. The argument at the specifier of D is easily perceived as the originator of the destruction, since destruction is an event that must have some external causer. This example, then, suggests an observation and a methodology. The observation is that roots (and not only words) can have semantic contents inherent to them. The methodology is that the semantic contents of the root can be traced to the extent that we can find significant patterns in the different meanings of the same root as it appears in different environments. Yet, while Marantz’s example is a fine illustration of the possibility of studyY ing the semantics of roots, English, in fact, is not a promising language for such a study. In English there is usually at most one nominal and one verbal interpretation for each root. Furthermore, it is√ typical that the noun “growth” and the verb “grow”, which are derived from grow, w denote very close concepts. English nominalizations give us√an idea about coarse-grained semantic characterizations of roots (i.e. that grow is an internally caused change of state √ and destroy is an externally caused change √ of state), but not about the finergrained semantic distinctions that set apart grow from other roots of internally caused change of state. In general, each morpho-phonological realization of n or v serves as an environment where a root may receive an interpretation. Yet in English (taking now v as √ our example)√verbs select for √ a single verbal√morpheme: zero morphology ( fish), -ize ( caramel), -en ( weak), -ify ( ver), etc. That is, if a root takes-en in a verbal environment, it does not take -ify as well, and vice

T HE C ONTENTS

OF THE

R OOT

59

versa. Thus, there is weaken but not ∗ weakify, verify but not ∗ veren.1 The same also holds for English nouns: nominal suffixes tend to exclude one another. Roots most often select one of the many possible nominal suffixes. This, I assume, is a specific property of English word-formation, an arbitrary choice of English, regarding how many interpretations it assigns each root. In principle, there is no rule that prevents languages from assigning an interpretation to roots in three or more nominal environments. Many languages, Hebrew among them, have just this option. As we saw already in Chapter Two, Hebrew has 7 verbal patterns and over 50 nominal patterns. Many Hebrew roots appear in numerous nominal, verbal and adjectival patterns. This then makes possible the widespread phenomenon of MCM. 3.1.3. MCM in Hebrew: An Introduction My suggestion is that the Hebrew root, semantically just as phonologically, has a content. The root is associated with an idiosyncrasy—a meaning—going beyond semantic class alone. This meaning, however, is not fully specified. It only becomes fully determined in combination with nominal, verbal or adjectival patterns. Suppose we wish to resist this notion of a meaning which is “not fully specified”. This is not impossible. We can envisage two diametrically opposed views, avoiding such “not fully specified” meanings: (a) Roots have fully specified meanings. According to this view, a root has already a meaning associated with it, of the kind we find in dictionary entries. If different words associated with the root have different meanings, this may be because the meaning is derived from the meaning of the root in some well-defined way, or because the association of the root and the different words is illusory: they are mere homophones. (b) Roots have no meaning. According to this view, meaning occurs only with words: the root is a mere phonological tool for the production of words, without any semantic content. If different words sharing the same root also happen to share some semantic properties, this is a mere tendency, explicable on pure historical grounds.

1

As Beth Levin (p.c.) points out, some apparent counter-examples in English exist. For example, zero verbal morphology alternates with -ify (as in the case of to gas and to gasify) or with -ize (as in to winter and to winterize). It is interesting, however, that such cases involve multiple derivation from an existing noun, e.g. gas, rather than multiple derivation from a root.

60

C HAPTER 3

To make the alternatives more √ concrete, let us consider the example of the commonly cited Hebrew root ktb: (6) a. b. c. d. e. f. g. h. i. j. k.

Root √ √ktb √ktb √ktb √ktb √ktb √ktb √ktb √ktb √ktb √ktb ktb

Pattern CaCaC (v) hiCCiC (v) hitCaCeC (v) miCCaC (n) CCoCet (n) maCCeCa (n) CCaC (n) CCiC (n) CaCCaC (n) CaCCaCa (n) tiCCoCet (n)

Word katav (write) hixtiv (dictate) hitkatev (correspond with) mixtav (letter) ktovet (address) maxteva (writing desk) ktav (alphabet) ktiv (writing, spelling) katav (correspondent) katava (article) tixtovet (correspondence)

√ It is often claimed in traditional Hebrew grammars that the root ktb means something such as write. This, in a nutshell, is “theory (a)”, which claims that the root comes specified with a full dictionary meaning. We can already see that this theory rests uncomfortably with the evidence. The root actually means “to write” in one specific environment, that of the first verbal pattern, CaCaC. In other environments, the root acquires numerous nominal and verbal interpretations, all of which are indeed related to a lexical core, but usually not referring directly to the specific meaning “write”. Uncomfortable as it may rest with the evidence, theory (a) is historically important: it is the view of traditional Hebrew grammar. Thus, in Gesenius (1910), the root is taken as the lexical prime and all forms are derivationally reduced to the root, which forms its prime meaning in binyan 1; all other forms are then based on the meaning of the root in that binyan. (It is primarily because of the accidental route of the √ history of Hebrew linguistics that we so often think of the meaning of ktb as write). For authors such as Gesenius, the derivational analysis was clearly motivated by the model of the Indo-European classical languages, where the reduction of words’ meanings into the meaning of their constitutive roots was traditionally taken for granted. I leave aside the question of the validity of this approach to the IndoEuropean classical languages. It is enough to say that, by now, a gap has arisen between, on the one hand, Hebrew dictionaries and learning grammars, and, on the other hand, the consensus of Hebrew linguists. While dictionaries and learning grammars continue to teach the old model based on Gesenius (see Even Shoshan 1978, Glinert 1989), linguists working in the field since as early as the 1970s agree that the Hebrew system contains a significant element of

T HE C ONTENTS

OF THE

R OOT

61

opacity.2 Gesenius’ position lives on in the widespread tendency to take binyan 1 as, in some sense, more elementary than the rest. Phonologically, and in some cases even semantically, binyan 1 is still sometimes taken as the source from which the other binyanim are derived, or at least as the default form (see Ussishkin 2000). But overall, theory (a) in its most extreme form—with all word meanings derived from the meaning of more basic elements leading down to the root—is no longer tenable. Some linguists do pursue, of course, the possibility of deriving meanings from the meanings of more elementary units, in some subset of the system. To mention three examples: McCarthy (1979: 219–230), in a discussion of Arabic data, concentrates on a sub-group of de-verbal nouns (participles) to argue that some forms are indeed derived from others. Schwarzwald (1998) delineates the border between and inflection and derivation in Hebrew nominal forms, showing how certain meanings of nominal forms can be derived from other, more basic nominal forms. Bolozky (1978b) shows how word-formation can be understood in derivational terms for a certain class of denominal verbs. Typically, then, the subset taken where derivationality is shown to hold is within the domain above the root: in certain conditions, verbal or nominal forms give rise, in a predictable manner, to other verbal or nominal forms. We shall return in Chapter Seven to discuss the nature of word-formation above the level of the root: it is indeed, I argue, structurally different from word-formation from roots. The thrust of the literature, then, is toward theory (b), making the root semantically empty. Such precisely is the view adopted by, e.g. Nir (1976) and Ravid (1990). Here I wish to argue against such an analysis. It rests uncomfortably with the evidence just as theory (a) does. It is today more popular than theory (a), I believe, simply because it has the appeal of a revolt against tradition. More seriously, of course, theory (b) has the advantage of fitting in with a lexicalist approach where words, rather than morphemes, are taken as the semantic primes. My main contention is that such an account is untenable and, for this reason among others, lexicalist approaches should be avoided. What I present here is, in my view, simply the most direct analysis of the phenomenon, as follows: I take the root to be an underspecified (but not empty) semantic core, with numerous possible incarnations. Each incarnation, i.e. an actual word, has its specific phonological form and semantic content: (7)

2

{root}

→ incarnation 1, in the environment CaCaC → incarnation 2, in the environment CiCCeC → incarnation 3, in the environment hiCCiC . . . etc.

See Berman (1975), Schwarzwald (1975), Nir (1976), and Ravid (1990). The list can be extended as, in fact, this opacity claim is by now no longer controversial.

62

C HAPTER 3

In Hebrew, such incarnations of roots occur in the environment of nominal and verbal patterns, where “incarnation” means both semantic and phonological realization.3 This is the phenomenon of MCM. The fundamental test for “theory (b)” would come from the nature of MCM. To the extent that we find that MCM operates within clearly defined linguistic domains—MCM being sensitive to categories of linguistic theory, certain types of MCM being more productive than others—the more plausible it becomes to see MCM as part of grammar itself (and not the product of extra-linguistic, historical or even cognitive forces). This we shall see throughout the chapter. Let us note √immediately a certain constraint on the range of MCM. To start with: taking ktb in (6) above again, we can see that although all the words constructed√from that root share its core meaning, the interpretations assigned to the root ktb in different environments may vary. Some are related to the activity of writing (including the verb katav, write, and the noun katav, correspondent), and some—to the written signs themselves (e.g. ktav, alphabet). Some are obviously related to the core, but in a way that is less direct (e.g. ktovet, address, and mixtav, letter). The root coins a relation between a phonological core (three consonants) and a semantic core (a certain domain of meaning). √ Most crucially, the meanings assigned to the root ktb are constrained within √ that semantic domain. ktb never creates nouns and verbs that are related to other semantic domains, such as verbs of motion or abstract nouns of thought and contemplation. This observation is so obvious as almost to escape attention, yet its significance is great: almost without exception, the various words traditionally taken to be created from a single root are indeed related to the same semantic domain. This is the fundamental reason to believe in the existence of some semantic information associated√ with the root. Let us go back to the lexical core of ktb. Not all the words created from the root necessarily involve √ the activity of writing. It is in a specific verbal context, CaCaC, that the root ktb yields an activity verb, katav (write). The templatic meaning of “activity” is what the verb shares with many other activity verbs. The idiosyncratic part, that is, the fact √ that the activity involves producing meaningful signs, is given by the root ktb. In a nominal environment the root creates nouns that refer to entities rather than to activities: mixtav (letter), ktiv (alphabet), katava (article). √ If we abstract away the core shared by all the words created from the root ktb, we remain with the notion of “inscriptions”, or “signs”. Whether the root is incarnated as a noun related to inscriptions (e.g. a 3

Note that the environment of the pattern serves not only to fix the semantic interpretation of the root relative to that context, but also determines its phonological form. For example, those consonants that are spirantized in post-vocalic positions—/b/, /k/, and /p/—have their surface √ form determined by the phonological structure of the pattern. The consonant /b/ in the root xˇsb ˇ is realized as /b/ in xeˇsˇbon (bill, CiCCon) and as v in taxˇiv x (calculus, taCCiC). xˇ

T HE C ONTENTS

OF THE

R OOT

63

letter) or to a verb meaning “make inscriptions” (write), depends on the context in which the √ root is embedded. The root ktb serves, then, as a first example of how the underspecified lexical core can be assigned multiple interpretations. To further clarify the nature of MCM in Hebrew, I proceed in the rest of this section to consider a few more examples, while in the next section I shall explain my methodology for identifying the phenomenon. r √rgˇs and √xˇsb ˇ (8) a. b. c. d. e. f.

Root √ √rgˇs √rgˇs √rgˇs √rgˇs √rgˇs rgˇs

Pattern CiCeC (v) hiCCiC (v) CeCeC (n) hitCaCCut (n) CaCiC (adj.) CaCCan (n)

Word rigeˇsˇ (excite)4 hirgiˇs (feel) regeˇs (feeling) hitragˇsˇut (excitement) ragiˇs (sensitive) ragˇsˇan (sentimental)

a. b. c. d. e. f. g. h.

Root √ ˇ √xsb x sb √ ˇ ˇ √xsb x sb √ ˇ ˇ √xsb x sb √ ˇ ˇ √xsb xsb ˇ

Pattern CaCaC (v) CiCCeC (v) hiCCiC (v) hitCaCCeC (v) maCCeC (n) maCCaCa (n) CiCCon taCCiC

Word xaˇsav ˇ (think) xiˇsev ˇ (calculate) hexˇsiv ˇ (consider) hitxaˇsev ˇ (be considerate) maxˇsev ˇ (computer) maxˇsava ˇ (a thought) xeˇsbon ˇ (arithmetics/bill) taxˇsiv ˇ (calculus)

(9)

Observation 1: the templatic and the idiosyncratic These two roots both refer √ to some mental activity. The nature √ of that activity is different in each case: rgˇsˇ is related to feeling, while xˇsb ˇ is related √ to reasoning. Consider, first, in detail, the root xˇsb. ˇ This root has an inherent meaning of mental activity, which can then be realized as thinking (xaˇ ( av, P1), calculating (xiˇ ( ev, P3), considering (hexˇiv x , P5) or being considerate (hitxaˇev, xˇ P7). In nominal environments the root forms nouns that are the result of reasoning (taxˇiv x √, calculus) or instruments related to reasoning (maxˇev xˇ x , computer). The root rgˇs, ˇ by contrast, creates verbs and adjectives related to feeling (feel, excite, sentimental) and √ abstract nouns √ related to feeling (feeling, excitement). Now consider the roots xˇsb ˇ and rgˇsˇ together. Both belong to the same semantic class, of roots of mental activity. These roots never create verbs that ffall into other lexical classes, such as verbs of motion or verbs of location. 4

Biblical Hebrew has also the stative verb ragaˇ (P1), be in a state of excitation.

64

C HAPTER 3

However, each root preserves its idiosyncratic core in the words it forms. The verbs derived from the same root √ are all related to√the same, particular, type of mental activity: reasoning for xˇsb, ˇ feeling for rgˇs. ˇ The idiosyncratic core of the root is thus narrower than a semantic class. Observation 2: the arbitrariness of the environmental meaning The linguistic sign, the root, is arbitrary, and, similarly, the coinage of sound and meaning for a root in a particular environment is arbitrary. There is no way a to predict, given the root and the pattern, what meaning the √ noun or verb should have. For example, it could have been the case that xˇsb ˇ would be assigned the interpretation of “computer” in the environment of maCCaCa (i.e. the noun maxˇava x ) and the interpretation of “thought” in the environment of “maCCeC” (i.e. the noun maxˇev x ).5 The burden of the learning of meaning in Hebrew is thus placed not on the roots alone, but on the combinations of roots with environments, that is, patterns. English has a large number of roots, most of which are assigned meaning in one environment only. Hebrew has much fewer roots (about 1800 verb-creating roots, see Chapter Five), many of which are assigned an interpretation in several environments. From the learnability standpoint, the challenge for the learner is the same: learn many roots, each with one interpretation, or learn less roots, each with multiple interpretations. r √qcb (10) a. b. c. d. e. f. g.

Root √ √qcb √qcb √qcb √qcb √qcb √qcb qcb

Pattern CaCaC (v) hiCCiC (v) CaCCaC (n) CiCCa (n) taCCiC (n) CeCeC (n) miCCaC (n)

Word qacav (to ration) hiqciv (to allot) qacav (butcher) qicba (pension) taqciv (budget) qecev (beat, rhythm) miqcav (tempo)

Observation √ 3: the difficulty of pinning down the core meaning √ The root qcb seems to name a relation r . All the words created from qcb refer to dividing things or allotting portions. The verbs refer to the action of allotting. The nouns refer to the allotment itself (qicba, pension, and taqciv, budget) or to concepts related to allotment (qecev, rh rhythm). One noun refers to a person who divides meat into pieces, a butcher (qacav). This is a highly special√ ized meaning of qcb in a nominal environment CaCCaC: qacav (butcher) is 5

This arbitrariness holds in spite of certain regularities that are discussed in Section 3.2.1, e.g. that the pattern maCCeC is often associated with instruments.

T HE C ONTENTS

OF THE

R OOT

65

not anyone who divides anything, but someone who divides meat into portions. This shows us another important aspect of MCM. There may be different degrees of specialization of meaning in different contexts. In some contexts the root may acquire a highly specialized meaning (e.g. qacav, butcher). In others, the meaning assigned to the root may be closer to that of the underspecified core (e.g. qicuv, rationing), or to other words derived from that root. Even the degree of specialization of meaning assigned to the root is arbitrary. It cannot be predicted from the pattern, whether the meaning assigned to the root in its environment is very specialized or not. Likewise, no matter how specialized their meanings are, all the words derived from the same root retain the lexical core of the root. This core√ is what limits the degree of specialization of meanings assigned to the root. qcb can create the noun qacav (butcher), but it cannot be assigned the interpretation √ of katav (correspondent), which is the interpretation assigned to the root ktb in the same nominal environment, CaCCaC. r √sˇ mn (11) a. b. c. d. e. f. g.

Root √ ˇ √smn ˇ √smn smn √ˇ ˇ √smn smn √ˇ ˇ √smn smn ˇ

Pattern CaCaC (v) CiCeC (v) hiCCiC (v) CaCeC (adj.) CuCaC (n) CeCeC (n) CaCeCet (n)

Word saman ˇ (grow fat) simen ˇ (grease) hiˇsmin ˇ (fatten) samen ˇ (fat) suman ˇ (fat) semen ˇ (oil, grease) samenet ˇ (cream)

Observation 4: MCM and natural substances √ sˇmn is not the typical case. Many Hebrew roots that name a substance or material—sukar — (sugar), dvaˇ (honey), xol (sand) ʕec (wood)—are not assigned multiple interpretations.6 The √ “rawness” of the material of the root is particularly evident in the case of sˇmn. The root itself refers to some underspecified concept of “greasy, fatty, oily substance”. Crucially, the root does not specify the precise nature of that material. It is a potentiality, which may be incarnated in many possible ways. In different nominal environments, the root may create words such as sˇemen (oil), sˇamenet (cream) or sˇuman (fat), all of which share the property of being a fatty substance, but are obviously different w types of substance. More interestingly, even in a verbal environment, the root 6

Interestingly, the phonological form of these roots is syllabic, rather than segmental (see Discussion in Chapter Two). I assume that there is a correlation in Hebrew between phonological and semantic underspecification. Only segmental roots are assigned multiple interpretations in different environments. See the Discussion in Section 3.3.

66

C HAPTER 3

retains its potentiality. That is, it does not specify what fatty material the verb is related to. The verb hiˇmin iˇ (fatten) is related to accumulation of fat. On the other hand, sˇimen (grease) does not pose any requirements on the material used for greasing: this material could be oil, grease, fat, butter or any other fatty substance one can apply. r √bˇsl ˇ (12) a. b. c. d. e.

Root √ ˇ √bsl b sl √ ˇ ˇ √bsl b sl √ ˇ bsl ˇ

Pattern CiCeC (v) hiCCiC (v) taCCiC (n) miCCaCa (n) CaCeC (adj.)

Word biˇsel ˇ (cook) hivˇsil ˇ (mature, ripen) tavˇsil ˇ (a dish of cooked food) mivˇsala ˇ (brewery) baˇsel ˇ (mature, ripe)

Observation 5: Roots naming properties give rise to an MCM that is more difficult to characterize In this case, the shared lexical core is clear, yet the meanings assigned to the root are quite far apart, and no direct relation holds between them.7 There are two principal semantic domains, that of √ cooking, and that of maturing, into which the meanings assigned to the root bˇsl ˇ may fall. Each of the interpretations of the root in the environment of a pattern may fall into one domain or the other. Thus, hivˇ vsil (mature) and baˇsˇel (ripe) belong to the domain of maturing, while vˇ biˇsˇel (cook), tavˇ vsil (cooked dish), and mivˇ vˇ vsala (brewery) belong to the domain vˇ of cooking. It is, of course, arbitrary, which interpretation is coined with which pattern. 3.1.4. How to Define MCM? My claim is that a root acquires distinct meanings when it appears in distinct environments of nominal and verbal patterns. Put differently, there are several nouns and verbs, all of which are derived from the same root, but crucially, none of which is derivationally related to the others. We informally defined MCM above, but our discussion will not be complete if we do not provide a way to identify instances of this phenomenon. Identification of MCM in the nominal system is straightforward. In some cases, our interpretation will hinge upon our identification of a verbal form as formed directly from a root or, indirectly, from another verb (I will return to discuss this in detail in Chapter Six). For the 7

This may well be a case where historical considerations may legitimately be brought into the discussion: it seems that the words were coined within a conceptual world where ‘cook’ and ‘mature’ were not so far apart as they are for us.

T HE C ONTENTS

OF THE

R OOT

67

time being, let us sketch the problem with pattern 5, traditionally known as the “causative” pattern: (13) a. b. c.

Root √ √bwʔ √mcʔ ktb

Pattern 1 baʔ (come) macaʔ (find) katav (write)

Pattern 5 heviʔ (bring) himciʔ (invent) hixtiv (dictate)

A number of linguists have argued that cases such as the above are in fact instances of the causative alternation: bringing means “causing something or someone to come”, inventing means causing something to be found, dictating something to someone means that this person does some writing (for references see Reinhart 1996, Doron 1999, 2003, and discussion in Chapter Five). The same line of argumentation is sometimes used for what we would consider as the MCM of roots in patterns 3 and 7, the latter traditionally known as the “reflexive” pattern: (14) a. b. c.

Root √ √ʔ bd √pʔ r grˇs

Pattern 3

Pattern 7

ʔ ibed (lose) hitʔ abed (commit suicide) peʔ er (glorify) hitpaʔ er (boast)

gereˇsˇ (expel)

hitgareˇs (get divorced)

In these cases, one could argue that committing suicide is losing oneself, that boasting is glorifying oneself, and that a divorce is a mutual act of expulsion. In all these cases, the “reflexive” has acquired a meaning that is narrower than the standard reflexive meaning, as in “perform an action on oneself ”.8 Because two verbs derived from the same root are not necessarily, per se, an instance of MCM, we need a criterion to define and identify MCM of roots. In this work I use the following criterion: I look at the semantic entailments that hold between the two verbs derived from the root. Since roots have their meanings fixed in the environment of patterns, a root that participates in MCM is to be assigned two different interpretations in two environments. As shown above, these meanings can be highly specialized, and semantically far apart from one another. Consequently, when a root is assigned MCM in two verbal environments, we do not expect any obligatory semantic relation between the 8

I believe that this approach is influenced both by traditional characterization of the fifth pattern as “causative” and from a deterministic approach to the Hebrew verbal system, which requires semantic transparency for the combinations of roots and patterns (see discussion in Chapters Four and Five). In Chapter Four, it is shown that the pattern system is not deterministic. Multiple Contextualized Meaning for roots is an important property of the Hebrew verbal system, and any theory that ignores it and tries to provide a “neat”, transparent account of the system, runs into trouble when tackling the many cases of unpredictable meanings of roots in different patterns. Chapter Five shows that such traditional characterization of patterns is empirically untenable.

68

C HAPTER 3

two verbs, because the interpretations √are assigned to the root in each environment separately. This is the case of bˇsl ˇ in (12) above, which is assigned the interpretation of “cook” in pattern 3 (biˇsˇel) and “mature” in pattern 5 (hivˇ vsil). vˇ The two verbs are derived from the same root, but refer to two different events: cooking and maturing. (15) √

interpreted as “cook” in the environment CiCeC: /biˇsˇel/ bˇsl ˇ interpreted as “mature” in the environment hiCCiC: /hivˇsˇil/

(16) √

interpreted as “excite” in the environment CiCeC: /rigeˇsˇ/ rgˇs interpreted as “feel” in the environment hiCCiC: /hirgiˇsˇ/

Verbs that differ only with respect to their argument structure alternation behave V differently. Consider the following case:9 (17) √ rxc

interpreted as “wash—transitive” in the environment CaCaC: /raxac/ interpreted as “wash oneself” in the environment hitCaCeC: /hitraxec/

In this case, indeed, we do not see MCM (Crucially, as will be argued in Chapter Six, this is not an instance of word-formation from a root). The two verbs are an instance of an argument structure alternation: one is transitive, the other intransitive, and the action is understood as performed on oneself. The action or event itself is the same: washing. Argument structures are never encoded into the semantics of the root, but are always supplied by syntactic context. Once their argument structure and event focus are stripped away, both verbs have exactly the same semantic content, that of washing. This can be represented in terms of an entailment: (18) a. b.

raxac (wash) wash hitraxec (wash oneself) wash

The difference between the two verbs is only in the argument structure of the verb, and in transitivity. In other words, the verbs “wash” and “wash-oneself ” entail an hyper-set, “wash”, and are equivalent to the intersection of that hyperset with a syntactically available category, “±reflexive”. 9

There is only one apparent counter-example to our analysis, namely that of causative verbs of change of state, where word-formation from roots strongly tends to be associated with an argument alternation and not with MCM. I argue in Chapter Six that this follows from the lexical nature of such roots.

T HE C ONTENTS

OF THE

R OOT

69

The case of MCM is different. When a single root creates two verbs via MCM, then each pattern serves as the immediate environment for fixing the interpretation of the root. The interpretations roots take up in the environment √ of different patterns are often far apart, as shown above for the root rgˇs: (19) a. b.

rigeˇsˇ (excite, in pattern 3, CiCCeC) hirgiˇs (feel, in pattern 5, hiCCiC)

In this case the only way to obtain common entailments shared by the two verbs would be to strip away elements of meaning that clearly are not supplied by the syntax itself. Excite does not entail any hyper-set of feel f (or vice versa), unless we define ad-hoc, for the sake of the exercise, some semantic hyper-set of, say, “emotion”; and no syntactic process, intersecting with this hyper-set, would give rise to “feel” or “excite”. I believe that the entailment criterion successfully teases apart cases of argument structure alternation and cases of MCM. Consider gereˇ (expel, pattern 3) and hitgareˇ (get divorced, pattern 7), which were discussed in (14) above. Unf these verbs do not share the same like the case of wash and wash oneself, semantic entailments: getting divorced does not semantically entail expulsion in the same way that washing oneself semantically entails washing. Some cases are easier to identify as MCM of the same root. This is because the semantic common core is clearly evident, yet the two meanings are so far apart that no modification or forcing into a systematic argument structure alternation is possible: (20) a. b. c.

Root Pattern 1 √ bˇ sl ˇ – √ nbl naval (wilt)

Pattern 3 bisel ˇ (cook) nibel (soil, bad-mouth) √ srx sarax (become base) –

Pattern 5 hivˇsil ˇ (mature) – hisriax (stink)

In other cases, the meanings assigned to the two verbs are very close: (21) a. b. c.

Root Pattern 1 Pattern 3 √ smr ˇ samar ˇ (keep, guard) simer ˇ (conserve) √ slx ˇ salax ˇ (send) sileax ˇ (send away, y expel) √ pcc pocec (explode) –

Pattern 5 – – hifcic (bombard)

The fact that the meanings of these pairs are so close makes it tempting, perhaps, to try and relate such pairs of verbs derivationally. However, no systematic

70

C HAPTER 3

alternation known in the literature can relate such verbs. The entailment criterion identifies the two members as MCM. Although sˇamar (keep, guard, save) and sˇimer (conserve, preserve) are very close in meaning, none of these verbs includes the meaning of the other in the same way that hitraxec (wash oneself) entails the meaning of raxac r (wash—transitive). Consider the different entailments each of these verbs has: (22) a. b.

Dina Dina Dina Dina

sˇamra kept sˇimra preserved

et OM et OM

ha the ha the

pitriyot mushrooms pitriyot mushrooms

(22a) does not entail (22b): one may guard something without conserving it, ˇ and vice versa. The case of salax ˇ (send) and sileax ˇ is more intricate. Salax only refers to a verb of transfer, specifying some goal: sending someone or ˇ something from one place to another. Sileax, on the other hand, refers to the act of sending someone away from you, without a specific goal. Pocec P (explode) and hifcic (bombard) are also very close in meaning, but the first refers to an event of explosion, and the other—to sending the explosives toward their mission, i.e. a verb of transfer. Entailments differ on the two verbs: (23)

ha the aval but

tayas pilot ha the

hifcic bombarded pcaca bomb

et OM lo NEG

ha the hitpoceca explode

ʕ ir,

city

“The pilot bombarded the city, but the bomb did not explode” Conversely, a bomb can explode without there being a bombardment act of throwing explosives from above. Recall now my suggestion above, that proximity or distance between the meanings assigned to roots in different environments is itself arbitrary. Hence we expect to have, among the many cases of MCM, some where the verbs are very close in meaning, while still being instances of MCM. A final caveat in identifying MCM is the need to filter out those cases that involve homophonous roots, i.e. two roots that have nothing in common except for their being phonological homophones. In such cases, there is no relation between the two verbs that these roots create:10 (24) a. b. c. 10

Root √ √xlm √brʔ lxm

Pattern 1 xalam (dream) baraʔ (create) laxam (fight)

Pattern 5 hexlim (recover) hivriʔ (become healthy) hilxim (weld)

In distinguishing homophonous roots from MCM, I followed here the Even Shoshan Hebrew dictionary (1978), which points out different etymological sources for homophonous roots.

T HE C ONTENTS

OF THE

R OOT

71

Each of these roots happens to appear in a different binyan, which makes the two verbs appear similar to MCM. However, apart from the phonological similarity, no lexical core is shared by these two roots.

3.2. MULTIPLE CONTEXTUALIZED MEANING IN HEBREW 3.2.1. MCM in the Nominal System In Chapter Two, we briefly discussed Hebrew nominal patterns and illustrated how they make roots into nouns. Hebrew nominal patterns differ from Hebrew verbal patterns in one important respect. When a root appears in several verbal patterns, the result may be MCM—or it may be something else, namely, either an argument structure alternation or synonymy (see discussion in Chapter Five). When a root appears in several nominal patterns, however, the result must be MCM. We shall return to explain this asymmetry at the beginning of the next section. The nominal system therefore serves as a testing ground where we can consider the nature of MCM itself. The main thrust of this section is to show some of the regularities associated with MCM: its being more productive in some semantic areas, less productive in others; its tendency to follow certain models (e.g. with one contextualized meaning of the root being concrete, and the other—abstract). This regularity serves to make it more likely that MCM is a law-like, linguistic phenomenon and, as a consequence, to make it more likely that the root itself contains semantic information. Let us recall the nature of nominal patterns or misqalim iˇ . As we saw in Chapter Two, misqalim ˇ , unlike binyanim, do not carry any structural or semantic restrictions on the nouns appearing in them and, instead, they serve solely as environments where roots are assigned numerous interpretations as nouns. To see how different nominal patterns serve as contexts for the interpretation of the root, consider the following cases: (25) a. b.

Root Pattern Noun Interpretation √ √zrq maCCeC mazreq syringe zrq miCCaCa mizraqa fountain

a. b.

Root Pattern Noun √ str ˇ miCCaC miˇstar ˇ √ str ˇ miCCaCa miˇstara ˇ

Interpretation regime police

a. b.

Root Pattern √ brg CoCeC √ brg maCCeC

Interpretation screw screwdriver

(26)

(27)

Noun boreg mavreg

72

C HAPTER 3 (28) a. b. c. d. e.

Root √ √qlt √qlt √qlt √qlt qlt

a. b.

Root Pattern Noun Interpretation √ 11 towel √ngb maCCeCet magevet ngb maCCaC magav mop

a. b. c.

Root √ √rpʔ √rpʔ rpʔ

(29)

(30)

Pattern CeCeC maCCeC miCCaC taCCiC CaCeCet

Pattern miCCaCa CCuCa maCCeC

Noun qelet maqlet miqlat taqlit qaletet

Noun mirpaʔ a refuʔ a marpeʔ

Interpretation input receiver shelter, asylum (vinyl) record cassette

Interpretation clinic medicine12 cure

The examples above belong to the most common types of MCM in the nominal system. These include names of instruments (screwdriver, syringe, receiver), institutions (regime, clinic) or other man-made objects (screw, record, towel). MCM is not everywhere in use: certain semantic ranges, as we shall see below, do not accommodate it (most strikingly, natural kinds). Further, the examples above are also typical in that they contain 2–5 nouns each. MCM in the nominal system is thus characterized both quantitatively and qualitatively. Quantitatively, let us first make the comparison with verbs. Roots normally create two or three verbs; only a handful of them create four. In the nominal system there are over fifty patterns available, yet roots hardly ever create more than a dozen nouns. Indeed, the majority of roots that are assigned MCM in nominal environments acquire 2–5 interpretations. √ There are thus many potential nouns that are not materialized. The root zrq in (25), for example, appears in two nominal patterns only.13 Theoretically, it could have formed more words in other patterns (cf. the non-existent ∗ mizraq, ∗ tazriq, ∗ zraq). These are all potential words, although not actual Hebrew nouns. (Indeed, new occurrences of roots in nominal patterns are possible, and are being coined daily.)14 11

Note the assimilation of the root consonant n before g. Note also the noun rofeʔ, a doctor, from the same root. The form of this noun is derived from the verbal system—the Benoni (verbal noun, cf. Shlonsky 1997). 13 In the verbal system it creates two verbs, zaraq (throw, pattern 1) and hizriq (inject, pattern 5). Note also the verb-derived noun zriqa, w which has two interpretations: “a throwing” and “an injection”. √ 14 A case in point is qaletet, cassette, from qlt, which also creates the verb hiqlit, record, and the noun taqlit, record. 12

T HE C ONTENTS

OF THE

R OOT

73

Interestingly, in cases with more numerous words created through MCM, the interpretations assigned to the root are often far apart. Consider the following cases: (31) a. b. c. d. e. (32) a. b. c. d. e. √

Root √ √kns √kns √kns √kns kns

Pattern CeCeC CiCuC CCiCa CCeCet haCCaCa

Noun kenes kinus knisa kneset haxnasa

Interpretation conference convention entrance parliament income

Root √ √xbr √xbr √xbr √xbr xbr

Pattern CaCeC CiCuC CaCuCa maCCeCet CiCCa

Noun xaver xibur xavura maxberet xevra

Interpretation friend composition/adding group, band noteboook society, company15

The root kns is related to a location, being in a certain place. The interpretations assigned to it in different nominal patterns vary greatly: kenes (a convention), kneset (parliament, a convention of people), and haxnasa (income—that √ which is “incoming” to one’s abstract location of property). The root xbr w names a relation between two entities, and can form nouns as different as “composition” (putting things together), “friend” (someone close) or “notebook” (papers bundled together). Roots denoting a relation or location (a type of √ relation) are √ particularly likely to create multiple nouns. Compare the case of ʕ rk and rkz below: (33) a. b. c. d. e. f.

Root √ √ʕ rk √ʕ rk √ʕ rk √ʕ rk √ʕ rk ʕ rk

Pattern CeCeC maCCaCa maCCaC maCCeCet taCCuCa maCCoC

ʕ erex maʕ araxa maʕ arax maʕ arexet taʕ aruxa maʕ arox

Interpretation item/value campaign/battle array system exhibition pastry roller

a. b. c. d.

Root √ √rkz √rkz √rkz rkz

Pattern miCCaC CiCCuC taCCiC CaCCaC

Noun merkaz rikuz tarkiz rakaz

Interpretation center concentration concentrate coordinator

(34)

15

Noun

Gutturals such as x often give rise to phonological changes in the pattern vowels, in particular, from i to e.

74

C HAPTER 3

√

ʕ rk names order—but r does not specify what sort of order. It is therefore underspecified enough to take up meanings that differ √ greatly from one another, such as ʕerex (item) and ma ʕaraxa (campaign). rkz seems somewhat more √ specified than ʕ rk: it names a relation of being in the middle, or in the center. The three names it creates reflect the lexical core of the root. Another group likely to pick up multiple nominal interpretations is that of roots related to size:

(35) a. b. c.

Root √ √gdl √gdl gdl

Pattern CoCeC miCCaC CiCuC

Noun godel migdal gidul

Interpretation size tower growth/tumor

a. b. c.

Root √ √xsr √xsr xsr

Pattern maCCoC CiCaCon CoCeC

Noun maxsor xisaron xoser

Interpretation shortage drawback lack

a. b. c.

Root √ √drg √drg drg

Pattern CeCeC CaCCa maCCeCa

Noun dereg darga madrega

Interpretation standing rank step

(36)

(37)

√

gdl refers to a large size, or surplus (note that godel g simply means “size” but that size is measured by how large something is, rather than how small). √ xsr, on √the other hand, is the opposite case, creating nouns of lacking or deficiency. drg, finally, specifies a degree or ordering measurement. The interest of such roots is that, here, the meaning of the root is not just—as always— underspecified, but is also related to an abstraction. This makes the unpredictability of the concrete meanings assigned to the root even more apparent. What theory could ever predict that a concrete meaning of a root referring to size or surplus in a particular environment should be tower? Indeed, a common tendency in MCM is for the root to acquire two meanings, one of which is a concrete item, the other—abstract: (38) a. b.

Root √ ˇ √sbr sbr ˇ

a. b.

Root Pattern Noun Interpretation √ dxp CeCeC daxaf urge, impulse √ dxp maCCeC madxef propeller

(39)

Pattern Noun Interpretation CeCeC sever ˇ fracture maCCeC maˇsber ˇ crisis

T HE C ONTENTS

OF THE

R OOT

75

It is important to note that with certain other semantic ranges, MCM is less productive. Roots of activity or of manner of motion, for instance, usually do not create multiple nouns, and some do not create nouns at all:16 (40)

Root √ √rwc √dlg √tpr √qcc √sxh xrˇs

Pattern Noun a. maCCoC meroc (race) b. – c. CeCeC tefer (stitch) d. – e. – f. maCCeCa maxreˇsˇa (plow) √ An exceptional case for manner of motion roots is zxl, which creates the verb zaxal (crawl), as well as the following nouns: (41) a. b.

Pattern CaCaC CiCCeC CaCaC CaCaC CaCaC CaCaC

Verb rac (run) dileg (hop) tafar (sew) qacac (chop) saxah (swim) xaraˇs (plow)

Root Pattern Noun Interpretation √ zxl CeCeC zaxal caterpillar √ zxl miCCeCet mizxelet sledge

Those exceptional cases where activity roots do form multiple nouns usually involve instruments for performing an activity or the products or result of an activity: (42) a. b. c. d.

Root √ √nsr √ngb ʔ ph

Pattern maCCoC maCCaC maCCeC

Noun masor (saw) magav (mop) maʔ afeh (pastry) √ srg maCCeCa masrega (knitting needle)

Pattern CCoCet maCCeCet maCCiya CaCiC

Noun nesoret (sawdust) magevet (towel) maʔ afiyah (bakery) sarig (knitwear)

√ The most interesting example of an activity root that is assigned MCM is ʕ bd, which creates the verbs ʕavad (work), heʕevid (employ), and ʕibed (process), w as well as the following four nouns: (43) a. b. c. d.

Root √ √ʕ bd √ʕ bd √ʕ bd ʕ bd

Pattern CeCeC CCoCa maCCaCa CuCCa

Noun ʕ eved ʕ avoda maʕ abada ʕ uvda

Interpretation slave work, job laboratory fact

√ I ignore those nouns that are derived from verbs, e.g. dlg creates dilug, hopping, but this noun is derived not from the root, but from the verb dileg (Engelhardt 2000, Arad 2003).

16

76

C HAPTER 3

Before concluding our discussion of miˇsˇ qalim, the following distinction must be made. In addition to misqalim ˇ (nominal patterns), Hebrew employs suffixes for creating new words. It is important to emphasize here that nominal patterns differ from suffixes in the following way. Although suffixes may create new words in a way that seems similar to patterns (cf. Bat El 1997), in most cases they derive nouns or adjectives from existing words. Thus, for example, the suffix -ut, w which creates abstract nouns, attaches to yeled (child) and forms yaldut (childhood): (44) a. b. c. d.

yeled yald+ut

child childhood (note contraction of yeled before suffix) yald+ut+i childish (pertaining to childhood) yald+ut+i+yut childishness

Nominal patterns, on the other hand, always create nouns from roots r . There are practically no existing words that are put in nominal patterns, in the same way that existing words attach to suffixes.17 When the same root appears in different miˇsˇqalim (nominal patterns), the result is always different nouns formed from it. The relation between two miˇsˇqalim is not derivational in the same way the relation between two words, one basic and the other√suffixed, is derivational (e.g. child and childhood above). That is, when the root yld appears in two nominal patterns, the result is two meanings assigned to the root, with no derivational relation between them: (45) a. b. c. d.

yeled (child) yalid (native) leda (childbirth, note initial root consonant y drops) yeluda (birthrate)

In view of the difference between derivational suffixes and nominal patterns, there arises the need to distinguish between word-formation from roots r —lexical kernels—and word-formation from existing words. In Chapter Seven, this distinction will be discussed in detail and be motivated based on Hebrew and English data.18 17

The one exception I am aware of is nemeq (gangrene), which is created from the verb namaq √ (rot). The verb is created from the root mqq, in the second pattern, niCCaC. New words may also be formed in the prosodic structure of pattern, for example, mivreˇet ˇ , which w is a translation of brush, put in the pattern miCCeCet. Compared with verbs, however, which are often and productively formed from existing words, such cases are rare. 18 Any attempt to relate patterns derivationally—e.g. by arguing that miCCaCa is derived from miCCaC though the (independent) suffix -a—is bound to fail. For example, there is no way to derive miˇtara iˇ (police) from miˇtar ˇ (regime); rather, it is obvious that the two are derived from one root in two different patterns.

T HE C ONTENTS

OF THE

R OOT

77

To summarize, the following three conclusions can be drawn: (1) There is a specific meaning that is listed in the Encyclopedia (the depository where roots’ meanings are listed relative to different contexts) for each combination of a root and a pattern. This meaning is related to the root, but can be predicted neither from the root nor from the pattern in isolation. Such combinations have to be learned independently as words. (2) The other side of the semantic classification of the root is the semantic emptiness of the pattern: both roots and patterns cannot fit any neat analysis in terms of compositionality. There exist, of course, certain tendencies correlating nominal patterns to the type of nouns that appear in them. For example, most instruments appear in the pattern maCCeC (maˇspex, funnel, mavreg, screwdriver, maghec, iron) while many names of locations and institutions appear in miCCaCa (mifqada, headquarters, mirpaʔa, clinic, miˇtara ˇ , police). But it is not the case in Hebrew that certain patterns are associated with nouns of certain semantic types. For example, maCCeC also hosts, apart from the instruments noted above, masqeh ˇ (a drink), maˇsˇber (crisis), and masmer (a nail). On the other hand, there are instruments such as mazmera (pruning shears) and maqdexa (drill), which appear in maCCeCa. Similarly, in the nominal pattern miCCaCa one finds not only names of locations and institutions, but also miˇspaxa (family), mignana (defense), and mifraʕa (cash advance). There are names of places and institutions that appear in other patterns, such as misgad (mosque) and minzar (monastery) in miCCaC. In short, miˇsˇqalic patterns are not only syntactically empty (i.e. not correlating with any morpho-syntactic properties, as argued in the previous chapter), but also semantically empty. They are, in fact, no more than phonological environments required for the formation of words out of roots. Crucially, the root may be assigned a different interpretation in each such environment. (3) There is no neat analysis of roots and patterns in terms of compositionality. The important generalizations hold for the semantic ranges over which word-formation via MCM is preferred or avoided. The most common categories into which miˇsˇqalic (patterned) nouns belong are the following: r Instruments

(46) a. b. c. d.

√ √brg √sˇpk √qlt qdx

mavreg (screwdriver; cf. hivrig, to screw) maˇsˇpex (funnel; cf. safax ˇ , spill) maqlet (receiver; cf. qalat, absorb) maqdexa (driller, cf. qadax, drill)

78

C HAPTER 3 r Artifacts

(47) a. b. c. d. e.

√ √xbr √ktb √zxl √clm qlt

maxberet (notebook; cf. xiber, adjoin, connect) mixtav (letter; cf. katav, to write) mizxelet (sledge; cf. zaxal, to crawl) maclema (camera, cf. cilem, photograph) taqlit (vinyl record, cf. hiqlit, record)

r Institutions and names of locations

(48) a. b. c. d. e.

√ √rpʔ √prc √drˇs √sgd tbn

mirpaʔ a (clinic; cf. ripe ʔ, heal) mifrac (bay; cf. parac, break into) midraˇsˇa (college; cf. daraˇ, preach) misgad (mosque; cf. sagad, revere) matben (barn; cf. teven, hay)

taxˇsˇiv (calculus; cf. xiˇev, calculate) maˇsˇber (crisis; cf. sˇavar, break) zikaron (memory; cf. zaxar, remember) maxsor (shortage; cf. xasar, lack) godel (size; cf. gadal g , grow) miˇsʔ ala (wish, cf. sˇaʔal, wish)

r Abstract nouns

(49) a. b. c. d. e. f.

√ ˇ √xˇsb s ˇ br √ √zkr √xsr √gdl sˇʔ l

3.2.2. MCM in the Verbal System Both in the verbal and the nominal system of Hebrew there are semantic types of roots that are more likely—or less likely—to be assigned MCM. Furthermore, we find that the more underspecified the lexical core of the root is, the more likely it is to create multiple verbs. This uniform behavior across the verbal and nominal systems—otherwise quite distinct—points to the validity of both our initial assumptions: first, the roots have an inherent meaning, and, second, that this meaning is underspecified. In both systems, the common behavior is explicable in terms of the same process of roots interacting with patterns. Finally, the analysis of the verbal system suggests that roots, indeed, ffall into semantic classes, which will be further articulated in the discussion to follow. The differences that do exist between verbal and nominal environments may be attributed to two principal facts, one grammatical, the other one extra-grammatical. The grammatical fact is the difference between miˇsˇqalim and binyanim, noted in Chapter Two. Miˇsˇqalim are not associated with any

T HE C ONTENTS

OF THE

R OOT

79

grammatical features, serving only as nominal environments where roots are assigned an interpretation. Binyanim, on the other hand, are associated with voice properties, which means that the appearance of a root in two different binyanim is not necessarily an instance of MCM, but may be an argument structure alternation involving voice. The second, extra-grammatical fact, is the tendency noted in Section 3.2.1 for MCM to prefer naming objects and artifacts. As a consequence of this there are relatively few verbal meanings (compared with nouns) available to the root. This could well be due to the structure of our conceptual system rather than due to a purely linguistic fact. We shall talk more about this at the end of this section, once we reviewed MCM in the verbal system. A caveat is called for before we discuss in detail the lexical classes of MCM roots in the verbal system. Because the lexical core of the root is underspecified it is impossible to name precisely these lexical classes. The classes postulated here may therefore seem all-inclusive and slightly imprecise. This fact necessarily results from the claim that root meanings are underspecified: had it been possible to provide dictionary-like definitions of roots or root classes, this would undermine the underspecification of these meanings. The core of the following discussion is a detailed survey of root-classes acquiring MCM in the Hebrew verbal system. The roots analyzed here fall into the following principal groups: roots of mental activity/perception (50), roots of relations or configurations (55), roots of transfer and motion (57), roots of adding or subtracting (60), roots of getting/needing (61), roots of possession (62), roots of damaging (63), roots of cutting and dividing (64), roots of putting together (65), and roots related to temporal concepts (66). These groups are quite similar to the Levin’s (1993) English verb classes, which suggests that these groups have, indeed, semantic reality. Levin’s project w is to bring out the interaction between the lexicon and the syntax. She therefore attempts to characterize semantic classes that behave uniformly with respect to syntactic verb alternations. Such classes, it turns out, may either be “functional” (“change of state verbs”, “verbs with predicative complements”, “verbs of change of possession”) or idiosyncratic, similar to the root classes noted here (“verbs of combining and attaching”, “verbs of separating and dissembling”, “verbs of grooming”, “verbs of killing”). Our task in this chapter is different. We are interested in roots, not verbs, and our attempt is to characterize those semantic classes of roots that tend to be assigned multiple meanings in different verbal environment. Roots, unlike verbs, contain only idiosyncratic material. From this results, an important distinction between Levin’s verb classes and the root classes discussed here: root classes have no functional meaning (change of state, change of possession). Such meanings are not associated with roots, but with verbal elements, as we shall see in Chapter Six.

80

C HAPTER 3

The first group of roots assigned MCM in the verbal system includes roots of mental activity, or perception: r Roots of mental activity/perception

(50) a. b. c. d. e. f. g. h. i. j. k. l. m. n. o. p.

Root √ xsb ˇ

xasav ˇ (think, P1) xiˇsev ˇ (calculate, hexˇsiv ˇ P3) (consider, P5) √ ydʕ yadaʕ (know, P1) yideʕ a (inform, hodiʕ a P3) (announce, P5) √ btx batax (trust, P1) biteax (insure, hivtiax P3) (promise, P5) √ bwn hevin (understand, hitbonen (look, examine, P5) P7) √ bxn baxan (examine, P1) hivxin (discern, P5) √ xzh xazah (envision, P1) xizah (predict, P3) √ cipah (expect, P3) √cph cafah (foresee, P1) rch racah (want, P1) ricah (placate, P3) √ kxˇ s hikxiˇ s (deny, P5) hitkaxeˇs (disown, P7) √ s ˇ br savar (assume, P1) hisbir (explain, P5) √ hirgiˇs (feel, P5) √rgˇs rigeˇsˇ (excite, P3) zhr nizhar (be careful, P2) hizhir (warn, P5) √ bdh badah (make up, lie, P1) bidah (fabricate, perjure, P3) √ dmh damah (resemble, P1) dimah (imagine, envision, P3) √ nsh nisah (try, P3) hitnasah (experience, undergo, P7) √ nkr hikir (recognize, P5) hitnaker (alienate, P7)

The roots of this group name some core of recognizing, perceiving or feeling. Each of these roots creates two or three verbs, which, although semantically far apart, all belong to the domain of mental state or cognition. Crucially, while all roots are limited to a certain range of mental √ activity √ constrained by the lexical core of the root (cf. the discussion of rgˇs√and xˇsb ˇ above), they show no further constraints. This way, a root such as rgˇs can form both rigeˇ (excite, √ P3) and hirgiˇ i (feel, P5), which refer to different mental states. Similarly, bxn can create both baxan (examine, P1), which names a process of examination, and √ hivxin (discern, P5), which names the result of an observation. The root xˇsb ˇ creates three different verbs of mental activity— —xaˇav (think), xiˇev (calculate), and hexˇiv x (consider). xˇ

T HE C ONTENTS

OF THE

R OOT

81

It is important to note that MCM is possible for certain roots, but never obligatory: it is not the case that all the roots that name a mental state or an activity create multiple verbs.19 The group of roots of mental activity/perception allowing MCM may be defined more precisely. This becomes particularly evident if we look at a group of roots that is similar to the group above and yet seems to name a specific mental state, such as surprise, fear f , boredom or disgust. These roots usually create one verb (when the experiencer is the subject) or two verbs that form an instance of an argument structure alternation (when the experiencer is the object): (51) a. b. c. d. e.

Root √ √ʔ hb √pxd √sˇʕ mm √g ʕ l snʔ

Verb

Verb

ʔ ahav (love, P1)

paxad (fear, P1) hifxid (frighten, P5) sˇiʕ amem (bore, P3) hiˇsˇtaʕ amem (be bored, P7) higʕ il (disgust, P5) nigʕ al (be disgusted, P2) sanaʔ (hate, P1)

This difference can be seen in two ways. First, it seems likely that Hebrew chose to have many roots assigned fairly within the cate√ broad meanings √ gory of mental state, e.g. reasoning ( xˇsb), ˇ sensing ( rgˇs)—those ˇ roots in (50) above. At the same time, it also picked several salient mental states and coined roots naming, roughly, these mental states themselves—those roots in √ √ (51)—such as ʔ hb, snʔ , etc. Second, and correlated with the first, we may state the same generalization as follows: when a root creates several verbs via MCM, these verbs also tend to be far apart semantically. We do not find MCM verbs whose meanings are very close together as in the following hypothetical forms: (52) √ Root CCC

Verb Verb Verb like a lot, P1 like considerably, P3 like a little, P5

Stated either way, we motivate a distinction between two types of roots, whose core is more or less specified. And when the root names a specific state—of boredom, surprise or disgust—it is more difficult, conceptually, to assign it multiple meanings. Interestingly, the same contrast between the two classes of roots occurs also in the nominal system. Roots of perception√and mental state often form numerous nouns. Consider the nouns created by bxn: √ The root rch, for example, creates racah r (want, P3) and ricah (placate, P3). A root with √ closely related meaning, xˇsq, ˇ only creates xaˇsaq (desire, P1).

19

82

C HAPTER 3 √

(53) a. b. c. d.

bxn miCCaC CoCaC maCCeCa aCCaCa

mivxan (examination) boxan (quiz) mavxena (test-tube) avxana (diagnosis)

On the other hand, roots that name a specific mental state usually form one noun only. Often, this noun is from the verb (i.e. a nominalization), rather than the root: (54) a. b. c. d.

Root √ √sˇʕ mm √ptʕ √pxd gʕ l

Verb sˇiʕ amem (bore) hiftiʕ a (surprise) paxad ´ (fear) higʕ il (disgust)

Noun sˇiʕ amum (boredom) haftaʕ a (surprise) páxad ´ (fear) goʕ al (disgust)

The second large lexical group includes roots that name relations or configurations. Like mental states above, relations seem underspecified enough to allow two different interpretations for the root. r Roots of relations or configurations

Consider the following examples (note the large proportion of roots that name spatial relations): (55) a. b. c. d. e. f. g. h. i. j. k. l. m. n. o.

Root √ √gbl √gdr √kll √ngʕ √kns √ngd √lkd √ktr √yxs √nwx ˇ √spl rkz

gaval (border, P1) gider (fence, P3) kalal (include, P1) nagaʕ (touch, P1) nixnas (enter, P2) nagad (be opposite, P1) laxad (trap, snare, P1) kiter (lay siege, P3) yixes (ascribe, P3) nax (rest, lie, P1) hispil ˇ (humiliate, P5) rikez (concentrate— tr., P3) √ nwp henif (raise, P5) √ nsʔ nasaʔ (carry, P1)

√

srg

sarag (knit, P1)

higbil (limit, P5) higdir (define, P5) hixlil (generalize, P5) higiʕ a (reach, P5) kines (gather, P3) hitnaged (resist, P7) liked (unify, P3) hixtir (crown, P5) hityaxes (relate, treat, P7) hiniax (lay, P5) hiˇstapel ˇ (lower, P7) hitrakez (concentrate on work, P7) hitnofef (be afloat, P7) hitnaseʔ (be condescending, P7) histareg (be intertwined, P7)

T HE C ONTENTS

OF THE

R OOT

83

A small sub-group includes roots that name a certain order: (56) a. b.

Root √ √ʕ rk ʕ arax (lay, order) heʕ erix (estimate) pqd paqad (give order) piqed (command)

hifqid (deposit)

√ c. tqn tiqen (fix) hitqin (fit in, install) √ The root gbl creates√both gaval g (border, P1), a verb of spatial configuration, and higbil (limit, P5). ngʕ creates nagaʕ (touch, P1) and higiʕa (reach, P5). In many cases, the relation between the two verbs involves an element of metaphor: one verb is a verb of physical contact, the other—of metaphoric touching or relation. As with the former class, it seems that roots that are assigned multiple interpretations are those roots that name √ a certain type of relation or configuration, but not a specific configuration. ktr (55h), which names a relation of surrounding or encompassing, creates two different verbs, with highly specialized meanings—kiter √ (lay siege, P3) and hixtir (crown, P5). On the other hand, a root such as ʕ gl, which names a configuration of something round, creates only verbs relating to being round:ʕigel (make something round, P3) and its non-causative alternant, hitʕagel (be round or become round, P7).20 The third large group of roots with MCM in the verbal system is roots that name a transfer or motion. r Roots of transfer and motion

The semantic characteristics of this group tend to be especially fuzzy, as we see in the following examples: (57) a. b. c. d. e. f. g. h. i.

Root √ √zrq √bwʔ √xzr √ycʔ √bqʕ √npl √ptx √ntl kwn

zaraq (throw, P1) baʔ (come, P1) xazar (return, P1) yacaʔ (get out, P1) baqaʕ (break out, P1) nafal (fall, P1) patax (open, P1) natal (take, P1) kiven (aim, P3)

hizriq (inject, P5) heviʔ (bring, P5) xizer (court, P3) yiceʔ (export, P3) hivqiʕ a (break in, P5) hitnapel (attack, P7) piteax (develop, P3) hitil (assign, P5) hitkaven (mean, P7)

√ One counter-example exists: clb, which creates the noun clav (cross) and the verbs calav (crucify, P1) hicliv (cross, make a cross, P5) and hictalev (cross oneself, P7). The cultural markedness of this particular configuration (or object) for speakers of Modern Hebrew may help explain both the relative specificity of the root and its productivity.

20

84

C HAPTER 3 j. k. l. m.

√ √ngˇs √sbb ˇ √pˇst mlt

n. o.

√ √grˇs gereˇsˇ (expel, P3) mkr maxar (sell, P1)

p. q.

√ ˇ salax ˇ (send, P1) √slx msr masar (hand in, P1)

r. s.

√ √brx ʕ tq

nigaˇs (approach, P2) sovev (turn around, P3) pasat ˇ (take off, P1) nimlat (flee, P2)

barax (run away, P1) neʕ etaq (move, P2)

higiˇs (serve, P5) hesev (cause, P5) piˇset ˇ (simplify, P3) himlit (give birth, animals only, P5) hitgareˇs (divorce, P7) hitmaker (get addicted to. . . , P7) sileax ˇ (send off, drive away, P3) hitmaser (be dedicated to. . . , P7) hivriax (smuggle, P5) heʕ etiq (copy, P5)

The semantic field of “motion” or “transfer” seems to be inherently underspecified so that many roots that include a component of motion are capable of creating more than one verb. Unlike in the former groups, I do not assume that this group is semantically homogenous, and do not attempt to characterize each of its members. Rather, I take it to be a general group that may include several well-defined sub-groups. I will consider here two relatively well-defined subgroups. The first includes roots that name motion of material from a source, and the second includes roots that name motion that is directed to a certain goal. r Roots of motion from source

(58) a. b. c. d. e. f.

Root √ √qlx qalax (flow, P1) √ntp nataf (drip, P1) dlp dalaf (leak, P1)

qileax (shower , P3) hitif (preach, P5) hidlif (leak, reveal information, P5) √ ndp nadaf (reek, P1) hitnadef (evaporate, P7) √ spk ˇ safax ˇ (spill, P1) hiˇspix ˇ (ejaculate, P5) hiˇstapex ˇ (wax lyrical, P7) √ nbʕ navaʕ (spring, P1) hibiʕ a (express, P5)

In this group, too, the metaphoric relation between many of the verb pairs is evident. For example, nataf (drip) and hitif (preach), the latter being the metaphoric extension of spreading some material—or ideas. In other cases, both verbs specify a concrete motion of an entity from a source. The type of motion and the type of entity that moves differs, then, in each case, as in nadaf (reek) and hitnadef (evaporate).

T HE C ONTENTS

OF THE

R OOT

85

r Roots of directed motion or direction

The roots in this group create verbs that specify motion directed at a certain goal. An important fact is that the goal itself is not specified, and neither does the entity that moves: (59) a. b. c. d. e.

Root √ hinhig (lead, P5) √nhg nahag (drive, P1) nhl nihel (direct, manage, P3) hitnahel (move toward a goal, P7) √ pnh panah (turn, P1) pinah (evacuate, P3) √ ql ʕ qalaʕ (aim, P1) niqla ʕ (get into, P2) √ sˇqʕ sˇaqaʕ (sink, P1) hiˇsˇqiʕ a (invest, P5)

We have seen several examples supporting the interpretation of the meaning of the root as that of a templatic meanings (e.g. mental activity, motion) plus a certain idiosyncrasy. This idiosyncrasy might be fairly precise, allowing few if any MCM, or more broad, resulting in MCM in the verbal system. So far, however, we have seen classes of meaning that, themselves, were very wide in definition. It is therefore interesting to note the existence of several root classes whose templatic meaning is more sharply defined, and yet do allow MCM. w The following characterizations are to be taken as approximate labels rather than explanatory or precise definitions: roots of adding or subtracting, roots of getting/needing, roots of possession or letting go, roots of damaging, roots of cutting, dividing or allotting, roots of putting together and roots related to temporal concepts. r Roots of adding or subtracting

This semantic characteristic may seem somewhat odd, but the group itself is well defined. All these roots form verbs of adding or subtracting, augmenting or diminishing: (60) a. b. c. d. e. f.

Root √ √gdl gadal (grow, P1) skm saxam (integrate, P1)

gidel (raise, P3) sikem (sum up, P3) hiskim (agree, P5) √ spx safax (accummulate, P1) sipeax (annex, P3) √ rwx riveax (make spacious, hirviax (earn, win, P5) P3) √ ˇ sibeax ˇ (praise, P3) hiˇsbiax ˇ (make better, P5) √sbx rbh hirbah (augment, P5) hitrabah (reproduce, P7)

86

C HAPTER 3 g. h. i. j. k. l. m.

√ √xsr √qpd √xbl √nkh √lqh √pxt ʔ bd

xasar (lack, P1) qiped (cut off, P3) nexbal (be hurt, P2) nikah (deduct, P3) laqah (be impaired, P1) paxat (decrease, P1) ʔ ibed (lose, P3)

n. o. p.

√ √ptr ˇ √nsl krt

patar (exempt, P1) nisel ˇ (deprive, P3) karat (cut off, P1)

xiser (subtract, P5) hiqpid (be meticulous, P5) xibel (sabotage, P3) hikah (beat, P5) hilqah (hit, P5) pixet (devalue, P3) hitʔ abed (commit suicide, P7) piter (fire, P3) hiˇsil ˇ (take off, P5) hixrit (eliminate, P5)

r Roots of getting/receiving/needing/satisfying

(61) a. b. c. d. e. f.

Root √ √zkh zaxah (win, P1) √qlt qalat (absorb, P1) crk carax (consume, P1)

zikah (exonerate, P3) hiqlit (record, P5) hicrix (neccessitate, P5) hictarex (need, P7) √ lqx laqax (take, snatch, P1) hitlaqeax (catch fire, P7) √ hispiq (suffice, P5) √spq sipeq (provide, P3) qnh qanah (buy, P1) hiqnah (enable to acquire, P5)

r Roots of possession, holding or letting go

(62) a. b. c. d. e.

Root √ √xzq √nxl ˇ √slt smt ˇ √ smr ˇ

xizeq (strengthen, P3) naxal (get, acquire, P1) salat ˇ (rule, P1) samat ˇ (drop, P1) samar ˇ (guard, save, P1)

hexziq (hold, P5) hitnaxel (possess land, P7) hiˇslit ˇ (enforce, P5) hiˇsmit ˇ (omit, P5) simer ˇ (conserve, P3)

r Roots of damaging/piercing

(63) a. b. c. d. e. f.

Root √ √nqz √pcʕ √pcc prʕ

niqez (channel, P3) pacaʕ (wound, P1) pocec (explode, P3) paraʕ (undo, P1)

hiqiz (draw liquid, P5) hifciʕ a (pop out, P5) hifcic (bombard, P5) hifriʕ a (disturb, P5) hitpareʕ a (misbehave, P7) √ prq paraq (unload, P1) pereq (undo, P3) √ prr porer (crumble, P3) hefer (violate, P5)

T HE C ONTENTS

OF THE

R OOT

87

r Roots of cutting, dividing, defining and allotting

(64) a. b. c. d. e.

Root √ √xlq √prt √psq √prˇs qdˇs

xalaq (disagree, P1) peret (go into details, P3) piseq (punctuate, P3) paraˇs (quit, P1) qideˇs (sanctify, P3)

xileq (divide, P3) hifrit (privatize, P5) hifsiq (stop, P5) hifriˇs (set aside, P5) hiqdiˇs (dedicate, P5)

r Roots of putting together

(65) a. b. c.

Root √ √xbr xavar (gang up with, P1) xiber (connect, P3) ˇ (tie, P1) qiˇser ˇ (connect, P3) √qˇsˇr qasar mzg mazag (pour, P1) mizeg (fuse, amalgamate, P3)

r Roots related to temporal concepts

(66) a. b. c. d.

Root √ zmn zimen (summon, P3) hizmin (invite, P5), hizdamen (happen, P7) √ m sk ˇ nim sax ˇ (last, P2) himˇ six ˇ (continue, P5) √ ˇ sahah ˇ (stay, P1) hiˇstahah ˇ (linger, P7) √shh qdm qadam (precede, P1) qidem (advance, P3) hiqdim (be early, P5)

While it is striking that there are clear semantic characteristics of such roots, it is equally striking that there are also clear semantic characteristics of those roots that are not assigned MCM (or do so very rarely). This includes roots of activity, roots naming a property or a state, roots naming an entity, roots naming a psychological state, as well as all roots that name a manner component:21 21

There are, of course, counter-examples in each of these groups, as listed below: Roots of activity (i) a. b. c. d. e. i.

Root √ ktˇs √ nsp ˇ √ ʕ bd √ ʕ ml √ ycr √ xtb

kataˇs (crush down) nasaf ˇ (breathe out air, P1) ʕ avad (work, P1) ʕ amal (toil, P1) yacar (create, P1) xatav (chop off, P1)

Roots that name a property or a state

hitkateˇs (fight, P7) hitnaˇsef ˇ (breathe heavily, P7) ʕ ibed (process, P3) hitʕ amel (exercise, P7) yicer (manufacture, P3) xitev (carve, sculpt P3)

88

C HAPTER 3 r Activity roots

(67) a. b. c. d. e. f. g.

√ √srg √nqh √nsr √cyr √cbʕ √glp tyl

sarag (knit, P1) niqah (clean, P3) niser (saw, P3) ciyer (draw, P3) cavaʕ (paint, P1) gilef (carve, P3) tiyel (travel, P3)

r Roots that name a property or a state

(68) a.

√

qrr

qerer (cool—causative, P3)

b.

√ xmm

c.

√ kʕ r

d. e. f. g.

√ √zqn √chb √mtq ybˇs

h.

√ rtb

(ii) a. b. c. d. e. f. g. h.

Root √ yqr √ rph √ qˇsˇh √ tyb √ kbd √ ycb √ yˇsb ˇ √ ʕ md

hitqarer (cool— inchoative, P7) ximem (heat—causative, P3) hitxamem (heat— inch., P7) kiʕ er (make ugly, P3) hitkaʕ er (become ugly, P7) hizdaqen (become old, P7) hichiv (make yellow, become yellow, P5) himtiq (sweeten, P5) yibeˇs (dry—causative, P3) hityabeˇs (become dry, P7) hirtiv (wet—causative, P5) nirtav (become wet, P7)

yiqer (make expensive, P3) ripah (weaken, P3) hiqsah ˇ (make difficult, P5) tiyev (make something better, P3) kibed (honor, P3) nicav (stand upright, P2) yaˇsav ˇ (sit, P1) ʕ amad (stand, P1)

Roots that name a substance or an entity (iii) Root √ a. ʔ bq ʔ ibeq (dust, P3) √ b. sql saqal (stone, P1) √ c. smn ˇ simen ˇ (oil, P3) √ d. src ˇ sarac ˇ (swarm, P1) √ e. zrʕ zaraʕ (sow, P1) √ f. stl ˇ satal ˇ (plant, P1)

hoqir (cherish, P5) hirpah (let go, P5) hitqaˇsah ˇ (become hard, P7) heytiv (do good, P5) hixbid (burden, P5) yicev (stabilize, P3) yiˇsev ˇ (colonize, P3) ʕ imed (type-set, P3)

heʔ eviq (pollinate, P5) siqel (take out stones, P3) hiˇsmin ˇ (fatten, P5) hiˇsric ˇ (litter, P5) hizriʕ a (inseminate, P5) hiˇstil ˇ (transplant, P5)

T HE C ONTENTS i. j.

OF THE

R OOT

89

√

qpʔ hiqpiʔ (freeze—causative, P5) qafaʔ (freeze— inchoative, P1) √ hdq hideq (tighten—causative, P3) hithadeq (tighten— inch., P7)

r Roots naming a substance or an entity22

(69) a. b. c. d.

√ √zbl zibel (manure, P3, cf. zevel, manure) √mlx himliax (salt, P5, cf. melax, salt) ʕ bˇs heʕ eviˇs (become covered with mold, P5, cf. ʕoveˇ, mold) √ dˇsn diˇsˇen (fertilize, P3, cf. deˇen, fertilizer)

r Roots that name a psychological state

(70) a. b. c. d. e. f. g. h. i.

√ √pxd √gʕ l √ptʕ √rgz ʔ hb √

paxad (fear, P1) nigʕ al (be disgusted, P2) hiftiʕ a (surprise, P5) hitragez (get angry, P7) ʔ ahav (love, P1)

snʔ sanaʔ (hate, P1)

√ √sgd sagad (adore, P1) √ʕ rc heʕ eric (admire, P5) qnʔ qineʔ (envy, P3)

hifxid (frighten, P5) higʕ il (disgust, P5) hirgiz (anger, P5) hitʔ ahev (fall in love, P7) hisniʔ (make loathsome, P7)

r Manner of motion roots

22

(71) a. b.

√ √dlg dhr

dileg (hop, P3) dahar (gallop, P1),

c.

√ rqd

raqad (dance, P1)

d. e. f. g.

√ √pzz √krkr √zxl qrtʕ

hidhir (gallop— transitive, P5) hirqid (make dance, P5)

pizez (jump swiftly, P3) kirker (dance jumpily, P3) zaxal (crawl, P1) qirteʕ a (limp, P3)

As we shall see in Chapter Seven, almost all the verbs that are related to an entity are in fact derived from a noun rather than from a root. That is, when the root names an entity, it rarely creates verbs. Rather, it creates, first, a noun, from which a verb is later derived. This is why there is a relatively small number of cases of verb-creating roots in this semantic class.

90

C HAPTER 3 r Manner of speaking roots

(72) a. b. c. d. e.

√ √cʕ q √lxˇs √gmgm √sˇʔ g nhm

caʕ aq (shout, P1) laxaˇsˇ (w whisper, P1) gimgem (stutter, P3) sˇaʔ ag (roar, P1) naham (growl, P1)

There are, of course, many other roots that are not assigned MCM, and I shall not list them all here. Still, the examples given above are a large sample and, based on my corpus study (see Chapter Five), a representative one. The above analysis reflects the correlation between root classes and the availability of MCM. Having said that, the following qualification should be made: it is not possible to predict, for each individual root, whether it will actually create several verbs. Often, two roots that belong to the same semantic class differ √ with respect to their ability to be assigned MCM. For example, the root xˇsb, ˇ like many roots of mental state or mental activity, creates multiple verbs—xaˇ — av (think, P1), x (consider, P5). On the other hand, xˇ √ xiˇev (calculate, P3), and hexˇiv the root hrhr creates only one verb, hirher (think deeply, contemplate, P3). Both roots seem to belong to the same class—yet only one is assigned multiple √ interpretations. In this case, the difference√may be attributed to the fact that hrhr includes a manner component that xˇsb ˇ lacks. Hirher (contemplate) is thus a verb of process, which does not allow a sentential complement √ and specifies only the subject matter of thinking. On the other hand, xˇsb ˇ simply specifies a mental activity, with no manner component. Therefore, it can create four different verbs—all of which include a certain mental state or activity. Other √cases, however, are unlikely to be explained in such a way. For example, yˇsb, ˇ a verb √ of positional state, creates yaˇav (sit, P1) and yiˇev (to colonize, P3), but √ sˇkb, which belongs to the same group, creates only sˇaxav (lie, P1). Similarly, gˇsˇs, ˇ a root of physical √ contact, creates giˇsˇeˇ (search by touch, P3) and hitgoˇeˇ ˇ (fight, P7), but msˇ ˇs, ˇ which belongs to the same group, creates only one verb, miˇsˇeˇ (feel by touch, P3). Such differences between individual roots are predicted, since MCM is a possibility but is never obligatory. Before we conclude this section, an important note must be taken: MCM is defined as the interpretations assigned to roots r in different environments. Therefore, only roots are assigned MCM. It is therefore expected that verbs that are derived from existing nouns rather than from roots (i.e. denominal verbs) should not be able to be assigned MCM. Indeed, denominal verbs are

T HE C ONTENTS

OF THE

R OOT

91

tied in their interpretation to the noun from which they are derived, and are not assigned multiple interpretations: (73) a. b. c. d. e.

Base noun sabon (soap) telefon (telephone) √ taqciv (budget, from √ qcb) maxzor (cycle, from √ xzr) xeˇsˇbon (account, from xsb) ˇ

Verb siben (soap) tilfen (telephone) tiqcev (budget) mixzer (recycle) hitxaˇsben ˇ (settle accounts)

In Chapter Seven, we shall discuss in detail this difference between root-derived and noun-derived verbs with respect to the interpretations available to them. 3.2.3. MCM in Hebrew: Interim Conclusions Having examined MCM of Hebrew roots in the verbal and nominal system, let us summarize our main conclusions. 1. Roots fall into semantic classes

While such classes are not always easy to define—as resulting from the nature of the underspecification of roots meaning—there is a tendency for the same root classes to allow (or disallow) MCM in the verbal and nominal system. To take an example, roots of mental activity or perception tend to create multiple verbs and nouns: √ (74) bxn a. baxan (examine, P1) b. hivxin (discern, P5) c. mivxan (examination) d. boxan (quiz) e. avxana (diagnosis) f. mavxena (test tube) Similarly, roots of spatial relations and configuration create MCM in both the verbal and nominal system: (75) a. b. c. d.

gbl gaval (border, P1) higbil (limit, P5) gvul (a border) migbala (a limit)

92

C HAPTER 3

On the other hand, certain semantic classes seem to avoid MCM, for example, roots of manner of motion, roots of sound emission, and roots of manner of speaking: (76) a. b. c.

dileg (hop, P3), qirteʕ a (limp, P3), zaxal (crawl, P1) gaʕ ah (moo, P1), ciyec (squeak, P3), sˇaʔ ag (roar, P1) caʕ aq (shout, P1), laxaˇsˇ (w whisper, P1), gimgem (stutter, P3)

2. MCM is more common with nouns than with verbs

Consider now the roots classes that systematically fail to create multiple verbs. Such groups do not behave uniformly with respect to MCM in the nominal system. Roots of light emission, for example, normally form only one verb. In the nominal system, however, some of these roots form one noun only (77), while others are capable of forming several nouns (78): w √ (77) zhr a. zahar (glow, radiate, P1) b. zohar (glow) √ (78) qrn a. qaran (radiate, P1) b. qeren (beam) c. qrina (radiation) d. maqren (projector) Manner of motion roots are never assigned MCM in the verbal system. In the nominal system, some of these roots—rarely—create multiple nouns:23 √ (79) zxl a. zaxal (crawl, P1) b. zaxal (caterpillar) c. mizxelet (sledge) In other words, we see that MCM is more common in the nominal system than in the verbal system. There are three manifestations of this asymmetry. First, MCM is the main noun-formation strategy in Hebrew, while only 15% of verb√ creating roots are assigned MCM. Many roots— ypʕ below is a representative example—are assigned multiple interpretations in nominal environments, but only one interpretation in a verbal environment: 23

I do not count here nouns that are derived from verbs, e.g. riqud (a dance) and raqdan r (dancer), that are derived from raqad r (dance).

T HE C ONTENTS

OF THE

R OOT

93

mofaʕ (appearnace/show) hofaʕ a (performance) tofaʕ a (phenomenon) yiffʕ a (glow, radiance) hofiʕ a (to appear, P5) √ Other roots, such as sˇtr, create several nouns but not a single verb: √ (81) sˇtr a. miˇsˇtar (regime) b. miˇsˇtara (police) c. sˇoter (policeman)24 (80) a. b. c. d. e.

Second, those roots that are assigned MCM in the verbal system most often create no more than two different verbs. In the nominal system, on the other hand, there are many cases of roots forming four or five different nouns, and some roots create over a dozen different nouns. Third and finally, in most cases when a root is assigned an interpretation in one verbal environment, it is also assigned at least one interpretation in the nominal system:25 √ (82) cnx a. canax (to parachute, P1) b. micnax (parachute) The difference between the verbal and the nominal system is not only in actual words, but exists also with respect to our intuitions regarding possible words, serving further to underline the productivity of the nominal system. One often feels that coining roots with different √ nominal patterns yields what may be called possible words. Consider the root zrq: (83) a. b. c. d.

zaraq (throw, P1) hizriq (inject, P5) mizraqa (fountain) mazreq (syringe)

This noun is the present participle (Benoni) form of the non-existing verb ∗ sˇatar. There are numerous such Benoni forms of non-existing verbs in Modern Hebrew. 25 Although this generalization holds for most roots, there are still exceptions. A notable exception √ is sql, which creates the verbs saqal (kill by stoning, P1) and siqel (remove stones, P3), but √ does not create any nouns (e.g. ∗ seqel, a stone). Other roots, such as znx, create two verbs, zanax (abandon, P1), and hizniax (neglect, P3). In the nominal system this root does not create any nouns. The only noun that seems related to it, haznaxa (neglect), is derived from the verb hizniax (neglect), not from the root. 24

94

C HAPTER 3

It is easy to imagine more possible nouns derived from this root. For example, ∗ tazriq (the stuff that is being injected) or ∗ zraq (garbage, things that have been thrown away). It is much more difficult to imagine any more verbs derived from that root. If we put the root in patterns 3 or 7, getting ∗ zireq and ∗ hizdareq, respectively, it is very difficult to think of√a possible meaning for these verbs based on what we know about the root zrq. And indeed, there are many new Hebrew nouns that are formed by coining a new interpretation for a root in a hitherto unavailable pattern (in fact, this is the main strategy for creating novel nouns, apart from borrowing), and relatively few such verbs. As suggested above, this asymmetry seems to have two causes: morphosyntactic and conceptual. Conceptually, nouns tend to denote entities—abstract or concrete, while verbs denote states and activities. It is something about the conceptual structure of language that makes entities richer in range than activities. There are more entities around us to name than there are activities or states. So it is expected that it would be easier to coin new nouns than it would be to coin new verbs. So much for the conceptual reason. Syntactically, the Hebrew verbal system poses certain restrictions of transitivity that nominal patterns do not pose, which make it more difficult for roots to acquire multiple meanings in a verbal environment. Indeed, the majority of roots that pick up multiple interpretations in the verbal system create only two verbs, which normally appear in those patterns that are not used for argument structure alternations.

3.3. MULTIPLE CONTEXTUALIZED MEANING AND THE ROOT HYPOTHESIS In this chapter, we were concerned with an empirical phenomenon: MCM assigned to Hebrew roots. My presentation took Hebrew words to be composed of roots. This, as repeatedly mentioned, is not a universally accepted truth, but rather the issue at stake. So, let us consider the two competing theories, the Root Hypothesis and the Lexeme-based lexicon, against the background of the data presented in this chapter. In a lexeme-based approach, one extreme option is to view the clustering of Hebrew word meanings as a mere coincidence. Words that contain the same phonological material also happen to share some of their semantics. Each of the words containing similar consonants and related to the same semantic domain is an independent lexeme, and there is no underlying common root shared by all these words. Instead, √ there is just mass homophony in the Hebrew lexicon. Consider the root ktb again:

T HE C ONTENTS (84) a. b. c. d. e. f. g. h. i. j.

Root √ √ktb √ktb √ktb √ktb √ktb √ktb √ktb √ktb √ktb ktb

Pattern CaCaC (v) hiCCiC (v) hitCaCeC (v) miCCaC (n) CCoCet (n) maCCeCa (n) CCaC (n) CCiC (n) CaCaC (n) CaCaCa (n)

OF THE

R OOT

95

Word katav (write) hixtiv (dictate) hitkatev (correspond with) mixtav (letter) ktovet (address, writing) maxteva (writing desk) ktav (alphabet) ktiv (writing, spelling) katav (correspondent) katava (article)

Clearly, this extreme mass homophony is highly undesirable, and I am not aware of authors promoting it. Instead, the typical strategy adopted in lexemebased accounts is to isolate islands of apparent derivationality from word to word, and then try to generalize those islands over the Hebrew nominal and verbal systems as wholes (Bat El 1997). One such island of derivationality is denominal forms (noun-derived verbs), another is that of argument structure alternations within the verbal system. In the remainder of this book I proceed to delimit and qualify the range of such alternations and to show why they must not be generalized. But even now the following is clear: a derivational approach (deriving words from other words) is problematic for at least two reasons. First, as suggested in the preceding section, the derivational approach forces us to pick—arbitrarily—one surface form (typically, the verb in P1) as the basis for the semantic account. Second, almost by definition, a derivational approach cannot account for the semantic distance between words of the same group, but must explain them away as homophones. √ On the other hand, if we assume that these words all share the root ktb, whose core is underspecified, and, furthermore, assume that the root is assigned w multiple interpretations in the environment of different patterns, then the correlation between these words is straightforward. Since they are all derived from the root, and the root has an underspecified core meaning, then their meanings are expected to share a certain core. But these meanings should not be derivable from any existing word. The variety of interpretations on the one hand, and the common core on the other hand, both follow elegantly from the Root Hypothesis.26 26

We should note that there exists a formulation of Bat El’s idea under which the difference between the two approaches is notational. If we assume that there exists a “base lexeme”, e.g. √ /katav/, which serves as the basis for all the words derived from (the hypothetical root) ktb. This lexeme is phonologically modified, yielding nouns and verbs with different interpretations.

96

C HAPTER 3

Roots are underspecified cores. How underspecified? I believe that here there may be a large variety between individual roots in their degree of underspecification. We saw that while some roots name a general fatty substance, which is then incarnated as oil, cream or grease, others name a specific substance (sugar, butter). Roots may be assigned multiple interpretations in different environments. How far apart may these interpretations be? Again, there may be a large variety between individual roots. There are cases where the core is to a large degree underspecified, and the root is assigned interpretations that are highly specialized, and semantically far apart, as in (85): (85) a. b.

Root √ piteax (develop, P3) √ptx patax (open, P1) ycʔ yacaʔ (get out, P1) yiceʔ (export, P3)

√ √ In other cases, such as (86), the core is quite evident ( bˇsl—maturing, ˇ rgˇs— mental state), yet the root is free to be assigned interpretations that are far apart: (86) a. b.

Root √ ˇ bisel ˇ (cook, P3) hivˇsil ˇ (mature, P5) √bˇsl rgˇs rigeˇsˇ (excite, P3) hirgiˇs (feel, P5)

But roots are also free to be assigned interpretations that are very close—yet distinctly different: (87) a. b.

Root √ ˇ samar ˇ (guard, save, P1) simer ˇ (conserve, P3) √smr pcc pocec (explode, P3) hifcic (bombard, P5)

The Root Hypothesis provides a unified explanation to word-formation phenomena in Hebrew. By showing how different words derived from the same root retain not only the phonological, but also the lexical core of the root, we lend further support to the existence of the root.

The lexeme /katav/ is the semantic and phonological nucleus that serves as a basis for the derivation. Under such a formulation, the difference between the two approaches boils down to the phonological representation of the root: are only consonants represented, as traditionally assumed, or are the vowels represented in the mind as well? Of course, in order for the two hhypotheses to be compatible, one has to retain the idea that the base lexeme is semantically and phonologically underspecified, and that there exists a clear distinction between word-formation from underspecified lexemes and word-formation from existing words (see discussion in Chapter Seven).

T HE C ONTENTS

OF THE

R OOT

97

Interestingly, the discussion in this chapter motivates not only the existence of the root, but also that of the pattern. Indeed, the discussion in this chapter fully exposes the role of patterns. Patterns, whether nominal or verbal, were shown here to serve as the immediate environment where the phonological and semantic incarnation of the root is given. All the possible interpretations are available only in the environment of certain patterns, and each interpretation is assigned to only in a particular pattern. Consider the following: √ (88) qlt a. CaCaC (v) qalat (to absorb, receive) b. hiCCiC (v) hiqlit (to record) c. miCCaC (n) miqlat (a shelter) d. maCCeC (n) maqlet (a receiver) e. taCCiC (n) taqlit (a record) f. CaCCeCet (n) qaletet (a cassette) g. CeCeC (n) qelet (input) √ The root qlt, whose core is “taking in”, or “absorption”, is assigned the meaning of “receiver” only in the environment of the nominal pattern maCCeC. In the environment of miCCaC—whose phonological form differs only minimally from maCCeC, the root is assigned the interpretation of “shelter”. Recall that the consonantal root itself not only lacks a fixed semantic interpretation, but is also unpronounceable. The same level–that of the pattern—decides three different things: the grammatical category of the (category-neutral) root, its semantic interpretation relative to that environment, and its phonological form. 3.3.1. The Existence of the Root: The Semantic Evidence I suggested distinguishing between coarse-grained properties of roots, or between coarse-grained root classes, and between the idiosyncratic lexical properties of roots. Coarse-grained root classes are types of denotations of roots, i.e. what the root names. One such possible formulation of coarse-grained root classes is tentatively suggested by Marantz (2000): (89) a. b. c. d.

state/property manner relation entity

These categories include types of denotations. Each such group includes many roots. The idiosyncratic part of roots’ meaning is given by their lexical and

98

C HAPTER 3

√ √ phonological core. For example, both xˇsb ˇ √and zkr are roots of perception √ or mental activity, but it is the ˇ or zkr, that determines the particular √ core, xˇsb activity. It is the fact that xˇ sb ˇ is related to reasoning, rather than to feeling, √ and that zkr is related to memory, rather than to hating, that determines the range of interpretations assigned to the verbs created from each root. Similarly, √ √ ybˇs and ʔ dm are both roots that name a property, but it is the core that √ determines whether this property is being dry ( ybˇ s, ˇ which creates yibeˇ, ˇ dry, √ P3) or being red ( ʔ dm, which creates heʔedim, redden, P5). Each root has its own idiosyncratic lexical core, but when we look at a large group of roots, it is evident that there are semantic groups of roots whose characteristics are finer-grained than those suggested by Marantz; it appears that Levin’s (1993) semantic characterization of English verbs’ classes is fruitful when considering groups of roots as well. Such groups include roots of menw tal state or mental activity, roots of motion from source, roots of holding or possession, roots of damaging and piercing, etc. There seems to be a relation between the idiosyncratic meaning of the root, and the availability of different semantic interpretations for that root. It is not surprising that such finer-grained distinctions play a role in deciding the availability of multiple interpretations for the root. My assumption is that the ability to conceptualize multiple words made from a root is not purely linguistic, but also interfaces with cognitive structure, conceptual structure, etc. Therefore, it is that part of the root that interfaces with the conceptual structure which is relevant for the availability of MCM. At the same time, it is remarkable that there are no roots whose core is so general that it equals no more than a fine-grained class of the type proposed by Levin (1993). One could envisage a root that created words, all of which referred to damaging or piercing, without any further constraint on the range of meanings available to those words. Such roots are not found in Hebrew. Roots always contain some lexical substance, or idiosyncrasy. It appears that the semantic content of the root must be at least the following: (90) Minimal semantic content of the root: {lexical class + idiosyncrasy} The lexical class to which a root belongs constrains the interpretations it may be assigned (i.e. roots naming mental states will not create manner of motion verbs). The idiosyncrasy further specifies a meaning for that root within the lexical class, and distinguishes it from other roots that belong to that class.27 27

To put this in suggestive, tentative terms: it is as if the mind, in acquiring its basic meanings in its language, seeks out idiosyncratic meanings, relying on universally available conceptual categories, so that the roots found by the mind naturally have the form of an idiosyncratic meaning bounded within a lexical class.

T HE C ONTENTS

OF THE

R OOT

99

One issue that arises from the discussion of the semantic content of roots concerns the difference between individual roots. One of my conclusions, based on an examination of the Hebrew data, is that although all roots are semantically and phonologically underspecified, there may be degrees of underspecification: it may be the case that some roots are more underspecified than others. And the degree to which a root is underspecified may be arbitrary, just like the lin√ guistic sign itself. For example, xˇsb ˇ is related to an underspecified activity of reasoning, but does not specify a precise activity. In different verbal environments the root may be assigned the interpretation of xaˇav√(think, P1), xiˇev (calculate, P3), or hexˇiv x (consider, P5). On the other hand, pxd names xˇ a specific mental state, that of fear, and can only create verbs that are related to that particular state: paxad (fear, P1) and its causative counterpart hifxid (frighten, P5). My suggested principle—almost a truism—is that there is a correlation between semantic underspecification of the root and its ability to take on multiple interpretations. Roots are everything that is idiosyncratic about language. Each root specifies some idiosyncratic core that differs from other cores, or roots. It is plausible that the degree of underspecification itself is idiosyncratic, depending on the specific properties of each root. In Chapter Two, we made the claim that a certain set of Hebrew roots are phonologically more specified than others. Those are the syllabic roots, which, unlike consonantal roots, contain vowels in them, and so do not give rise to context-dependent phonological effects such as spirantization: (91) a. b. c.

Syllabic roots festival (festival, ∗ pestival) xrop (snooze, ∗ xrof) vandal (vandal, ∗ bandal)

Consonantal roots √ xsb: ˇ xaˇsav ˇ (think, P1, ∗ xaˇsˇab) xeˇsˇbon (bill, ∗ xeˇsˇvon) taxˇsˇiv (calculus, ∗ taxˇsˇib)

Even within consonantal roots there are roots with more or less √phonological content. Certain√roots contain four, rather than three consonants: trgm (tirgem, √ translate, P3), prns (pirnes, support, maintain, P3), blbl (bilbel, confuse, mix, P3).28 √ On the √ other √ hand, there are roots whose middle consonant is a glide— √ qwm, rwc, s ˇ wt, and roots whose final two consonants are identical: sbb, √ √ xmm, qll. Such roots are often regarded as phonologically deficient, and it has been argued (McCarthy 1979) that they contain two consonants only. So,

28

Note that these are consonantal roots, not syllabic nouns whose stem was modified. I return to discuss the evidence for root-derivation vs. noun-derivation in Chapter Seven.

100

C HAPTER 3

even with respect to the sheer amount of phonological material, roots differ from one another.29 I do not make here the stronger claim (which is not, on its face, impossible, but does require further research) that richer phonological information corresponds to richer semantic information (e.g. more semantic information in syllabic roots). Instead, I make the weaker correspondence claim: both domains, the phonological and the semantic, display different degrees of non-specificity. Thus, within each coarse-grained category of roots, there may be more or less specification regarding the semantic content of the root. This was illustrated above with respect to the category “state”: some roots name a specific mental √ state (fright), while others only specify “something to do with feeling” ( rgˇs, which creates rigeˇ, excite, and hirgiˇ w i , feel). The same contrast exists between √ roots that name material or entities. To take a minimal pair, compare sˇmn, which names some fatty material, but does not √ w specify which (and creates sˇemen, oil, sˇuman, fat, f and s ˇ amenet , cream), with xemʔa—which only refers √ √ to butter, or melax (from √mlx), which only refers to salt. A similar contrast √ holds between the root sukar, w which names an entity (sugar), and the root mtq, which names a property (being sweet) and creates himtiq (sweeten, P5), matoq (sweet), and mamtaq (candy). While sukar only refers to the material known as sugar, mamtaq can be anything sweet—hard candy, chocolate, marshmallow, etc. What is clear is that all types of roots must have a certain amount of phonological and semantic content. This is given by our definition of roots as coinages of lexical and phonological cores: (92) Root: {minimal phonological content, minimal semantic content} (possibly more of each) The minimal phonological content is, in Hebrew, an ordered sequence of segments; the minimal semantic content is, in Hebrew—as suggested in (90) above—{lexical class + idiosyncracy}. My assumption is that roots, universally, have some phonology and some semantics. I believe that it is possible that the amount of semantics and phonology may differ from one language to another. This raises a new series of questions. What is the minimal amount of semantics or phonology that a root must include, universally, below which it does not have enough “material” to qualify as a root? Conversely, what is the largest amount of semantic or 29

Some linguists have argued that “deficient” roots (with middle glides) do not have a psychological reality, as speakers are often unable to identify them within words (see Ravid 1990 and references therein). I believe that these cases exhibit phonological peculiarities, specific to these roots, which yield unpredictable forms in certain phonological contexts. See Chapter Seven for discussion.

T HE C ONTENTS

OF THE

R OOT

101

phonological material, or the largest degree of specification, that a root may have, above which such specificity tends to go not with a lexical atom, but with a composite entity (e.g. noun)? Those questions are beyond the scope of this study. I offer the discussion of Hebrew as a suggestive case for such an inquiry and merely add here a brief comparison between Hebrew and English. Certain English roots are assigned MCM in different environments. In par√ √ √ ticular, Latinate roots such as mit, fer, and tend take up different interpretations in combination with different prefixes. A similar phenomenon is phrasal verbs, where verbs acquire a variety of interpretations in combination with prepositions: take up, take off, take to, etc. The extent of the phenomenon, however, is much more limited in English than in Hebrew. In addition, it is limited to a group of verbal elements. English roots do not form a variety of nouns in the same way Hebrew roots do. Suppose, as I will suggest in Chapter Seven, that the variation between languages lies not only in their initial root inventory, but also in the number of interpretations assigned to each of their roots in different environments. Some languages, like Hebrew, may assign as many as fifteen different interpretations to the same root. Other languages, like English, may assign each root one, or at most two interpretations. A further question is whether this difference in the productivity of wordformation from roots might be related to a true semantic difference between English and Hebrew. Could Hebrew roots be more often of “fuzzy” kind, so that they lend themselves more easily to be assigned MCM? Phonologically, Hebrew consonantal roots are more underspecified than English roots: they only get their phonological form in the environments of patterns. It may be the case that they are also semantically more underspecified. Thus the suggestion above is not impossible. In this book, I leave it as a suggestion only. 3.3.2. The Relation between Roots and Words We discussed above the relation between the content of the root and its ability to lend itself to MCM. Can there be certain characteristics of the nouns and verbs created by MCM of roots? In other words, is there anything common to the interpretations assigned to roots in different environments? We can now draw the conclusion that MCM has a preference for naming the following types: r r r r

Instruments Objects and artifacts Locations and institutions Abstract nouns

102

C HAPTER 3

The actual interpretation of a noun formed by MCM depends both on the root and on the choice made by the language. Roots that denote a certain √ manner are likely to create nouns that refer to instruments. For example, ghc creates maghec (iron). In a verbal environment, such roots create manner verbs: gihec (iron, P3). But, as mentioned already, there √ is a large degree of freedom in the assignment of interpretation. The root zll, which names a manner of eating (it creates the verb zalal, devour, in pattern 1), creates a noun that denotes a place: mizlala (eatery, fast-food place). It does not create a noun that serves as an instrument for eating (e.g. ∗ mazlel). Instruments are often related to manner verbs, and denote an instrument that is used for performing the activity denoted by the verb. Artifacts are often the result r of the activity denoted by the verb that is derived from the same root. √ Thus, mixtav (letter) is the result of katav (write, P1), both derived from √ktb. Mexqar (research) is the result of xaqar (research, P1), both derived from xqr. What is striking is that names for natural kinds (animals and flowers), kinship terms or basic food materials are rarely created by MCM for roots. That is, they are usually created by roots that form one noun only. Consider the following examples:30 (93) a. b. c. d. e. f. g. h. i.

kelev (dog) xatul (cat) sus (horse) xamor (donkey) parah (cow) pil (elephant) ʔ arnav (rabbit) ʕ ez (goat) sˇafan (hare)

The only exceptions I am aware of, where animal names √ are formed by a root that creates further words, are zaxal (caterpillar, from zxl, which creates zaxal, √ crawl, P1) and zamir (nightingale, from zmr, which creates zimer, sing, P3). Both these animals seem to be named by their attributes (crawling and singing, respectively), extending an existing root (see Brown 1999). Food materials (94) and kinship terms (95) tend to follow a similar pattern to animlas:31 30

Most of these words (cow, dog, horse, donkey, elephant) exist in Biblical Hebrew. They have therefore been a part of the language for a long period, not recent borrowings. √ 31 A notable exception to this is semen and sˇamenet (oil and cream, respectively, from sˇmn).

T HE C ONTENTS

OF THE

(94) a. b. c. d. e. f.

lexem (bread) basar (meat) xemʔ a (butter) riba (jam) sukar (sugar) gvina (cheese)

(95) a. b. c. d. e.

ʔ aba (father) ʔ ima (mother) ʔ ax (brother), ʔ axot (sister)

R OOT

103

dod (uncle), doda (aunt) saba (grandfather), savta (grandmother)

Again, the existence of MCM serves as a probe into the way in which languages realize different concepts. In a language like English, where most roots create one word only, there is no apparent difference between nouns that denote artifacts and those that denote animals or food materials. In Hebrew there is an evident morphological difference between the two types of nouns. Although it is possible, in principle, to create names for animals from roots that are underspecified to a large degree, this is a strategy the language simply avoids. √ Why don’t we have a word for dog that is derived from the root s ˇ mr (which √ creates sˇamar, guard, P1), or from the root nbx (which creates navax, bark, P1)? Why did Hebrew borrow √sus (horse) from Akkadian when it could have created a word from the root rkb (which creates raxav r , ride, P1)? On the other hand, Hebrew frequently and productively creates names for places, institutions, instruments and artifacts from highly underspecified roots. What this indicates is that there seems to be a fundamental difference between nouns such as horse, dog, father or bread on the one hand, and nouns such as letter, screwdriver, sledge, clinic, mosque, computer and abstract nouns such as memory, calculus, feeling, f or knowledge on the other hand. In a language such as Hebrew, which allows MCM of roots, nouns of the latter type tend to be derived from underspecified roots, which create numerous nouns and verbs. Nouns of the former type, on the other hand, are almost without exception created from roots that are not assigned a range of meanings via MCM. Such roots denote an entity—and a very specific one: a horse, a dog, etc.32 As mentioned already above, it should be noted that this distinction between natural kinds and artifacts in Hebrew has been observed in a number of unrelated languages. Brown (1999) has studied the acculturation of loan words in over 32

Note that the semantic high specification of such roots is independent of their phonological specifications. Some of these roots are consonantal (i.e. more underspecified phonologically), √ e.g. klb, which creates kelev (dog) in the pattern CeCeC.

104

C HAPTER 3

seventy native-American languages, and found that natural kinds are much more likely to be borrowed than artifacts and instruments. The latter tend to receive a native name (often a compound related to their function), while the former are borrowed. We may tentatively approach this pattern as follows. Recall our interpretation of the minimal semantic content of the root: {lexical class + idiosyncracy}. As the mind looks for the idiosyncrasy to associate with a root, it may identify or may not identify it as a natural kind. Such an identification of a natural kind would turn the semantic content of the root into {lexical class + natural kind}, and thus would make the root highly specified: the natural kind, effectively, occupies the “idiosyncrasy” position and prevents it from being associated with any weaker semantic content, in this way blocking the way for MCM. 3.3.3. Do Roots Fall into Semantic Classes? Using MCM as a probe into root semantics, our initial investigation of Hebrew roots has identified a number of semantic classes. First, there are those classes that serve as good candidates to be assigned MCM: r r r r

Roots of perception or mental activity Roots that name a relation Roots of motion Roots of positional state

There are other root classes, which are unlikely candidates for MCM, and tend to create one or two words only: r r r r

Roots of manner (manner of motion, manner of speaking) Roots of sound, light, and smell emission Root that name a property or a state (with a few exceptions) Roots that name a mental state

Some of these characterizations refer to the general type of the root—whether it names a state or a manner. Some of these characterizations are more fine-grained. For example, roots that name a state do√not tend to create multiple verbs, but roots that name a positional state do (e.g. yˇsb, ˇ which creates yaˇav, sit, and yiˇev, colonize). An important difference between verb classes and root classes is the following. In the characterization of verbs, one often uses both characterizations of templatic meaning and idiosyncratic meaning. Levin (1993), for example, has classes of “change of state verbs” alongside groups such as “gorge verbs” (manner of eating), “kill verbs”, etc. Roots, by definition, contain only idiosyncratic meaning. They are devoid of all functional material that is contributed

T HE C ONTENTS

OF THE

R OOT

105

by the verbal head, tense, aspect, etc. The root classes that arise from this preliminary investigation are precisely those classes one expects, given the nature of roots. They include basic semantic types (manner, state, relation, etc.), and some finer-grained specifications (positional state, mental state). They do not include, of course, any functional information (e.g. roots that form causative verbs, as causation is given by the verbs). The question is, then, where such information derives from. One simple hypothesis would be to equate verbal patterns directly with functional informah tion: each binyan standing for a functional head such as “causative”, “reflexive”, etc. As already seen from the evidence accumulated so far by MCM, it is unlikely that patterns—verbal or nominal—can be analyzed as having such a rich functional information. Rather, as we shall see in the following chapters, the relation between morpho-syntactic features and their phonological exponents is, to a large degree, indirect.

CHAPTER 4 THE MORPHOLOGICAL CONSEQUENCES OF MCM: AN INTERMEDIATE SUMMARY 4.1. Binyanim and their Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 4.2. Roots across Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 4.3. Regularity and Irregularity Predicted and Explained . . . . . . . . . . . . . . . 115 We concentrated, so far, on two aspects of the Hebrew verbal system: the obligatoriness of verbal pattern morphology in Chapter Two, and the lexical contents of the root itself in Chapter Three. In this, and the following two chapters, I move on to analyze the interaction of roots and patterns. The discussion will be organized around two central questions. The first regards the properties of individual binyanim. Hebrew has five non-passive binyanim, and binyan morphology is obligatory for all verbs. What is the significance of a root being assigned to one of the five binyanim? The question that arises is the following: r Is there a direct correlation between semantic and syntactic properties of

verbs, and their morphological form, so that binyanim can be characterized in semantic terms (e.g. reflexive, causative)? The answer—to anticipate the discussion in this and the following chapters—is negative. As we shall see, the association of patterns with semantic roles, which is advocated in some traditional grammars, cannot be maintained. While certain tendencies of certain verb types to appear in certain patterns do exist, there is no strict relation between any single binyan and any single role. The second question regards not individual patterns, but the pattern system as a whole. While we have to give up the neat association of individual patterns with individual semantic roles, this need not force a rejection of the notion of a regularity in the verbal system. Instead of being stated for individual binyanim,

106

T HE M ORPHOLOGICAL C ONSEQUENCES

OF

MCM

107

regularity might be found in the system of relations r between binyanim. This is our second question: r What generalizations and regularities, if any, can be found in the pattern

system? As we shall see, there is a large degree of regularity in the pattern system, and at the end of Chapter Six we shall see a complete mapping of a system of rules for assigning roots to binyanim. At the same time, I predict that complete regularity of the system is impossible. This is not because binyanim do not possess any roles, but because they possess more than a single role: each binyan serves several different semantic roles, according to the several different relations it bears to other binyanim. In Chapters Five and Six, I will show the fine-grained structure of this overlaying pattern of roles. But already given the data covered so far, we can see why Hebrew cannot have a neat system of individual roles for individual binyanim, based just on the coarse-grained distinction between MCM and other roles. Thus, the inherent irregularity of Hebrew verbal morphology can be seen as the morphological consequence of MCM. To be more precise, the inherent irregularity of Hebrew verbal morphology is a consequence of the combination of the two facts that were at the center of Chapters Two and Three: the obligatoriness of pattern morphology and the ability of roots to create multiple verbs when appearing in several patterns. The two, combined, result in a system where pattern morphology, over and above its systematic functions, also serves as the host for the fundamentally unpredictable MCM. The organization of this chapter is as follows: Section 4.1 provides data that illustrate the inadequacy of associating verbal patterns with semantic roles. Section 4.2 presents an overview of the systematic aspects of the association of roots with patterns, paying special attention to argument structure alternations. Finally, Section 4.3 shows why the combination of obligatoriness of pattern morphology and the ability of roots to create multiple verbs guarantees a degree of irregularity in the pattern system.

4.1. BINYANIM AND THEIR PROPERTIES Given that there are several different verbal patterns (binyanim), it would obviously be desirable to find certain regularities between the morphological form of a verb—binyan—and its syntactic and semantic properties. Otherwise, had there been a complete irregularity in the Hebrew verbal system, speakers would have had to learn separately each combination of a root with a binyan, and this,

108

C HAPTER 4

while not impossible, is also theoretically undesirable, at least from the acquiw sition standpoint. Let us consider the individual binyanim in more detail. In Chapter Two, we noted the well-known observation that two binyanim, P4 (CuCCaC) and P6 (huCCaC), are uniquely associated with passive verbs.1 These patterns are dependent on their active counterparts, CiCCeC and hiCCiC, morphologically, syntactically, as well as semantically. Morphological dependence means that the form of the passive verb is determined by that of the active verb. That is, the active form of a passive verb in P4 (CuCCaC) is always in P3 (CiCCeC), and the same holds for P6 (huCCaC) and its active counterpart P5 (hiCCiC). We never find pairs that cross those binyanim diagonally:2 (1) a. b.

xinex → xunax (∗ huxnax) educated was educated hicmid → hucmad (∗ cumad) attached was attached

By syntactic dependency, I mean that the existence of an active form is a necessary condition for the existence of the passive one. There is never a passive form in the absence of an active counterpart. Furthermore, the properties of the active form, in particular transitivity, determine whether a passive form is available.3 Finally, the semantic relation of active and passive verbs is expressed 1

I refer to these patterns as to individual binyanim, in accordance with the traditional view of Hebrew morphology. Note that, as pointed out in Chapter Two, these patterns can be seen as a modification of their corresponding active patterns, through insertion of different voice exponents. 2 The opposite is nearly, but not always, the case. A very small number of verbs, whose active form appears in pattern 3 have pattern 7 passives, rather than the expected pattern 4 (see discussion in Chapter Five): ∗ (i) a. biqeˇs hitbaqeˇs buqaˇs asked (for something) was asked ∗ b. qibel hitqabel qubal received/accepted was received/accepted It should be mentioned that these forms do not have all the characteristics of passives: for example, hitbaqeˇ is felicitous only with an infinitive form (hu hitbaqeˇ lalexet, he was asked to leave), and does not allow the logical object of the active verb to appear. This could count as an argument that these forms are not true passives, after all, as they are not syntactically derived from the active. Note, though, that the active verbs in P3 do not have any other passive form in P4. Mention should also be made of some other idiosyncracies, such as the passive alternation between sˇar (‘sang’, pattern 1) and huˇar (‘was sung’ pattern 6). 3 There are some exceptions to this generalization, but they invariably involve Benoni (participial) adjectives in pattern 4, such as metoraf, f ‘mad’ mexucaf, ‘impudent’, mexuyat, ‘tailored, elegant’. These are not derived from any active or passive verb. See Shlonsky (1997) for a study of the Benoni form.


OF

MCM

109

by the fact that they both refer to the same event (cf. Section 3.1.3 in the previous chapter: both JJohn read the book and the book was read by John refer to a reading event). Passive forms never have a special or idiomatic meaning that is not available in the active. Next, there are two binyanim, P2 (niCCaC) and P7 (hitCaCCeC), that only host non-transitive verbs. This means that the verbs appearing in them may be intransitive, or take an indirect object or a preposition, but they never take direct objects (ʔet introduces definite direct objects): (2) a. b. c. d. e. f.

histakel be /ʕ al looked in /at nixnas la-bayit entered to-the house nilxam be fought with/against hictaref le joined to hitxabeʔ mi hide from hiˇsˇtameˇs be. . . used (of)

/∗ ʔ et /∗ et ha bayit /∗ ʔ et /∗ ʔ et /∗ ʔ et /∗ ʔ et

The only property P2 (niCCaC) and P7 (hitCaCCeC) share is their transitivity restriction. They do not share any other semantic property—a fact of some importance, to which I return below. Passive verbs are always intransitives, and consequently, nothing prevents passive forms from appearing in these two binyanim. Thus, while neither P2 nor P7 is inherently passive, they host the passive forms of some roots: P7 serves as the passive for some P3 verbs, while P2 serves as the passive form for active verbs in P1:4 (3) a.

4

Active (P3) qibel received biqeˇsˇ asked for ʔ ifˇ fser fˇ made possible Active (P1)

/ / / / / /

Passive (P7) hitqabel, was received hitbaqeˇs was asked hitʔ afˇ fser became possible Passive (P2)

At this point, the question arises, if intransitivity is all that is specified for patterns, why can’t patterns 1, 3, and 5 host passive verbs? As we shall see in Chapter Six, patterns 1, 3, and 5 are exclusively reserved for verbs derived from roots r . Passives, on the other hand, are verbs derived from (active) verbs, and therefore they do not appear in these patterns.

110

C HAPTER 4 b.

harag killed ʔ axal ate sadad ˇ robbed

/ / / / / /

neherag was killed neʔ exal was eaten niˇsdad ˇ was robbed

The remaining three binyanim, P1(CaCaC), P3 (CiCCeC), and P5 (hiCCiC), can host both intransitive and transitive verbs: (4) a. b. c.

P1: yaˇsˇan (sleep), ʔ axal (eat), caxaq (laugh) P3: tiyel (travel), gihec (iron), diber (talk) P5: hitxil (begin), higdil (enlarge), himtin (wait)

Following Berman (1978), I claim that the restriction on transitivity is the only property encoded in the binyanim themselves (as will be illustrated below, there are certain properties characteristic of the binyan system, rather than of individual binyanim). In traditional grammar, as well as in some more recent theories, binyanim are sometimes characterized by their supposed semantic or syntactic role, e.g. “reflexive”, “causative”, etc. (see Ussishkin 2001, and, to some extent, Doron 1999, 2003, as well as references in Chapter Five). However, these labels at best point to tendencies. For example, the so-called “causative” binyan, P5, has many non-causative verbs appearing in it. Conversely, there are many causative verbs that do not appear in this binyan, but in P1 and P3: (5) a. b. c.

Non-causatives in P5 himtin (wait) hizmin (invite) hexlit (decide)

Causatives in other binyanim ximem (heat) (P3) patax (open) (P1) sˇavar (break) (P1)

Similarly, P7 (hitCaCCeC), which is often called “reflexive”, hosts many nonreflexive verbs: (6) hitCaCCeC a. b. c.

hiˇsˇtameˇs (use) hiˇsˇtaʕ el (cough) hitʕ alef (faint)

So, identifying binyanim with lexical or semantic roles such as reflexive/reciprocal or causative/inchoative cannot be maintained. Binyanim are specified for the syntactic property of transitivity, but not for any semantic properties such as causativity or reflexivity: there is a [−transitive binyan], but


OF

MCM

111

no [+causative binyan].5 All of this will be illustrated in greater detail in the following chapter.

4.2. ROOTS ACROSS PATTERNS So far we discussed the properties of individual binyanim. Now, an important property of the Hebrew verb system is that most roots appear in more than a single binyan. One such case of multiple appearances of roots was discussed in Chapter Three—MCM. In this case, a root creates two different verbs by appearing in two different binyanim: √ (7) Root: bˇsl ˇ a. P3 biˇsel ˇ (cook) b. P5 hivˇsˇil (mature) √ (8) Root: bxn a. P1 baxan (examine) b. P5 hivxin (discern) √ (9) Root: btx a. P1 batax (trust) b. P3 biteax (insure) c. P5 hivtiax (promise) These three examples are characteristic of MCM: the verbs the root creates appear in patterns 1, 3, and 5 (see footnote 58). As we shall see in Chapter Five, which offers a quantitative analysis of roots across patterns, these three patterns w most commonly mark MCM. Interestingly, they are also precisely those patterns that do not pose any transitivity restrictions on the verbs they host.6 Cases of MCM in other patterns are not unknown, but they make a smaller proportion of the total of MCM verbs (see Chapter Five for the complete data): √ (10) Root: grˇs a. P3 gereˇsˇ (expel) b. P7 hitgareˇs (divorce) 5

Interestingly, no binyanim are associated exclusively with transitive verbs: it is intransitivity, rather than transitivity, that is coded in the binyan system. I leave open the question whether this is a cross-linguistic tendency (Romance languages also morphologically mark intransitivity, by the pronominal clitic si/se / ), or whether some languages may choose to mark transitivity. In Chapter Six I will discuss some of the syntactic consequences of the transitivity restriction on binyanim. 6 As will be argued in Chapter Six, patterns 1, 3, and 5 are those patterns that may only host root-derived verbs. This is in opposition to patterns 2 and 7, which may host either root-derived verbs or verb-derived verbs, such as reflexives or passives. MCM, by definition, is derived from roots, and we should therefore expect a large number of its tokens to be marked by these patterns that only host root-derived verbs.

112 (11)

C HAPTER 4 Root: ʕ lm a. b.

p2 neʕ elam (disappear) p7 hitʕ alem (ignore)

As we already began to see, MCM is not the only case where a single root appears in several binyanim. In other cases, a root appearing in two binyanim creates two verbs with similar meanings—which differ only with respect to the prism through which the event is perceived. These are known in the literature as argument structure alternations. In such cases, the appearance of a single root in two different binyanim corresponds to a regular semantic and syntactic change in the meaning of the verb it forms, that is, a change in the structure of the arguments. The main alternations in Hebrew include active and passive forms, transitive–reflexive, inchoative–causative, and inchoative–stative. Examples of each type are given below: (12)

Root √ a. √smr ˇ b. √bˇsl ˇ c. cmd

Active samar ˇ (guard, P1) bisel ˇ (cook, P3) hicmid (fasten, P5)

Passive niˇsmar ˇ (guard—pass., P2) buˇsal ˇ (cook—pass., P4) hucmad (fasten—pass., P6)

(13)

Root √ a. √rxc b. √glx c. srq

Transitive raxac (wash, P1) gileax (shave, P3) sereq (comb, P3)

Reflexive hitraxec (wash—refl., P7) hitgaleax (shave—refl., P7) histareq (comb—refl., P7)

(14)

Root √ a. √qpʔ b. √bhr c. √bhl d. xmm

Inchoative/Non-causative qafaʔ (freeze, P1) hitbaher (clear, P7) nivhal (get frightened, P2) hitxamem (get hot, P7)

Causative hiqpiʔ (freeze, P5) hivhir (clear, P5) hivhil (frighten, P5) ximem (heat, P3)

(15)

Root √ a. √yˇsb ˇ b. √ʔ hb c. ʕ md

Stative yaˇsav ˇ (sit, P1) ʔ ahav (love, P1) ʕ amad (stand, P1)

Inchoative/Non-stative hityaˇsev ˇ (sit down, P7) hitʔ ahev (fall in love, P7) neʕ emad (stand up, P2)

These alternations are related to the event structure of the verb.7 Unlike MCM, the root does not create verbs with different semantic interpretations when participating in such alternations. The difference between the alternants is expressed in the event type: a state ((yaˇav, be seated) vs. a change of state, or an inchoative (hityaˇev, sit down), an externally caused event (e.g. ximem, 7

Some less common alternations exist as well, especially with manner of motion verbs, such as halax/hithalex, ‘walked/walked around’ ʕaf/hitʕofeff, ‘flew, flew about’.


OF

MCM

113

heat—transitive, hilbiˇ, dress—transitive) vs. a spontaneously occurring (hitxamem, was heated) or self-induced event (that is, reflexive, as in hitlabeˇ, get dressed).8 This difference in event structure also correlates with a difference in argument structure. Passives and reflexives are obligatorily intransitives, with the external argument not projected (in passives) or identified with the internal argument (in the case of reflexives). In causatives, an argument is added—the causer of the event. These alternations are also morphologically regular to a high degree. We noted above that passives depend morphologically on actives: the passive of P3 is in P4, never in P6, etc. The same regularity exists in the other alternations as well. For example, in the transitive–reflexive alternation the reflexive alternant is always in P7: (16) a. b. c.

Root √ √rxc √glx lbˇs

Transitive raxac gileax hilbiˇs

Reflexive hitraxec hitgaleax hitlabeˇs

Pattern 1–7 3–7 5–7

Interpretation wash shave dress up

The alternation between a stative and an inchoative (coming into a state) verb has the stative alternant in P1, and the non-stative alternant either in P2 or in P7: (17) a. b.

Root Stative Inchoative Patterns √ yˇ sb ˇ yaˇsav ˇ hityaˇsev ˇ 1–7 √ ʕ md ʕ amad neʕ emad 1–2

Interpretation sit (down) stand (up)

Finally, the causative–inchoative alternation exhibits a larger degree of freedom in morphological marking: (18) a. b. c. d. e. f.

Root √ √qpʔ ˇ √sbr √xmm √mss √bhr ʔ dm

Inchoative qafaʕ niˇsbar ˇ hitxamem namas hitbaher heʔ edim

Causative hipqiʕ savar ˇ ximem hemes hivhir heʔ edim

Patterns 1–5 2–1 7–3 2–5 7–5 5–5

Interpretation freeze break heat melt clear redden

In this case, we cannot point out a specific pattern that is uniquely associated with the causative or with the inchoative alternant. Rather, there are several possible patterns of alternations. If the inchoative alternant is in P1, then the causative alternant is in P5. If the inchoative alternant is in P7, then the causative 8

See Levin and Rappaport Hovav (1995) and Reinhart (1996) for a discussion of this kind of alternations.

114

C HAPTER 4

one is in P3 (or, for a small group of roots, in P5). However, in spite of this relative freedom, there is a strong constraint on the marking of alternations, which seem never to be violated. This constraint, pointed out by Doron (1999, w 2003), concerns the uni-directionality of the alternations with respect to the binyan system. Alternations always follow a certain direction. That is, when the non-causative verb appears in P1, its causative counterpart appears in P5, and never the other way around. There is never a causative alternant in P1 whose inchoative alternant is in P5, and never a causative alternant in P7 whose w inchoative alternant is in P3: (19) a. b.

∗ ∗

ʕ amad (make stand up, P1)

ximem (get hot, P3)

∗ ∗

heʕ emid (stand up, P5) hitxamem (heat, P7)

In the case of P7, we could argue that since causatives must be transitive, and P7 verbs are never transitive, no P7 can host a causative verb. Thus, the impossibility of (19b) follows straightforwardly from the inherent intransitivity of pattern 7. The relation between P1 and P5, however, cannot benefit from such a principled explanation since both patterns can host transitive verbs, and P1 may host a causative verb whose inchoative counterpart is in P2 (ˇsavar/ r niˇsˇbar “broke/got broken”). Thus, we cannot explain the association of roots with binyanim in terms of transitivity constraints alone. A possible way to explain the remaining case of binyan 1 is to claim that it is a default form (the idea that binyan 1 is a default is entertained in many theories, especially those that aim to derive all other verbal forms from it; see, for example, Ussishkin 2001). If we pursue this idea, we may be able to explain the fact that causatives should always appear in binyanim 1, 3, and 5 and not 2 or 7. But there is another, more interesting fact to explain, apart from the uni-directionality of the alternations. The marking of alternations follows certain paths. Either the two verbs are marked by patterns 3 and 7, or by patterns 1 and 5, or by 2 and 5, etc., but there are no causative–inchoative alternations in patterns 2 and 3, or in patterns 1 and 7. In other words, although the Hebrew verbal system makes available numerous potential ways for marking verb alternations, many of such possible ways are not found in practice. It is particularly interesting to note that there are no argument structure alternations between P3 and P5. Thus, we never have a causative in P3 whose inchoative alternant is in P5, or the other way a around.9 9

As we shall see in Chapter Six, alternations involving transitivity very strongly tend to be marked so that the intransitive verb appears in one of the patterns specified for intransitivity. √ A root that happens to form an intransitive verb, e.g. tyl, may appear in any binyan (tiyel, in CiCCeC), and is not confined to the two non-transitive binyanim. But if the root participates in a transitivity alternation then this alternation must be marked, and the non-transitive alternant must invariably appear in one of the two non-transitive binyanim.


OF

MCM

115

Recall our discussion of MCM above, where we noted that patterns 3 and 5 are most commonly associated with MCM. Thus, another source of regularity is added: the two patterns most commonly associated with MCM are those patterns where no argument structure alternations appear. How does the Hebrew binyan system accommodate both MCM and argument structure alternations? It is a fundamental fact that there are many more binyan slots than there are argument structure alternations. Even if a root creates four different verbs participating in an alternation, there are three more empty binyan slots that remain. Those may be used to form MCM. The prediction is that whenever a single root appears in five or more binyanim, then at least some of these appearances will be instances of MCM. This, indeed, seems to be the case, as illustrated in (20) and (21) below: √ (20) xˇsb ˇ a. CaCaC xaˇsˇav (think) b. niCCaC nexˇsˇav (think—passive) c. CiCCeC xiˇsˇev (calculate) d. CuCCaC xuˇsˇav (calculate—passive) e. hiCCiC hexˇsˇiv (consider) f. huCCaC huxˇsˇav (consider—passive) g. hitCaCCeC hitxaˇsˇev (be considerate) √ (21) ydʕ a. CaCaC yadaʕ (know) b. niCCaC nodaʕ (become known) c. CiCCeC yideʕ a (let-know) d. CuCCaC yudaʕ (let know—passive) e. hiCCiC hodiʕ a (inform) f. hitCeCCeC hitvadeʕ a (get acquainted with) We saw that the Hebrew verbal system has a large degree of regularity. The interesting question is, why is the system not completely regular? 4.3. REGULARITY AND IRREGULARITY PREDICTED AND EXPLAINED Consider the main facts we discovered so far regarding the Hebrew verbal system. In Chapter Two, we analyzed Hebrew verbs as composed of roots and binyan morphology. I argued that the obligatory binyan requirement for all verbs results from the need of morpho-syntactic features, in this case voice, to be phonologically spelled out. In the following two chapters, we examined the cases of roots that appear in more than one binyan. Chapter Three offered an

116

C HAPTER 4

analysis of Multiple Contextualized Meaning, the formation of several different verbs out of one root. This chapter reviewed argument structure alternations. Putting together our findings, consider now the consequences for the regularity in the binyan system. The fact that every verb must be in the form of a binyan entails that all argument structure alternations—active and passive, causative and inchoative, etc.—must be mediated through the binyan system. That is, if the same root forms both an inchoative and a causative verb, both verbs will appear in (different) binyanim. There is no way in Hebrew, as opposed to, say, Classical Greek, to form a passive or a reflexive verb by adding an independent morpheme or a suffix to the verb. Because all verbs must be in the form of a binyan, Hebrew verbs simply cannot alternate outside the binyan system. For the very same reason, all MCM assigned to roots must also be expressed through the binyan system: binyanim are the only way to form verbs from roots. While argument structure alternations have a lot of semantic, syntactic, and morphological regularities, MCM is by definition “irregular”. The meanings assigned to roots in the environment of different patterns, while constrained by the lexical core of the root, were shown to be arbitrary to a large degree and non-compositional, that is, not predictable from the combination of the root and the pattern. The binyan system thus has a dual role: it marks argument structure alternations and, simultaneously, forms multiple verbs from a single root. From the two facts, that (1) all verbs must be in the form of a binyan, and that (2) Hebrew allows Multiple Contextualized Meaning for roots, follows the degree of regularity and of irregularity found in the Hebrew verbal system. Put in other words: the binyan system displays a syncretism in that it hosts both alternations and MCM. Seen in this way, the very unpredictability of the Hebrew verb system is ultimately made predictable. So, when dealing with the Hebrew verbal system, one must expect both a degree of regularity and a degree of irregularity. In this interim summary, I pointed out the regularity and irregularity as a consequence of the coarsegrained distinction between MCM and the other, more regular phenomena of Hebrew verbal morphology. In the following two chapters, I proceed to disentangle the finer-grained structure of Hebrew verbal morphology, analyzing the entire system of overlaying morphological roles associated with binyancombinations. It is at this point that a quantitative analysis becomes crucial. How many instances of MCM and how many instances of argument structure alternations does the system have? In particular, which alternations exactly does the system have, and how do they pattern with the different binyanim? This quantitative analysis will not only sharpen our account, but also allow us to consider the empirical adequacy of theories of the Hebrew verbal system, and emphasize the


OF

MCM

117

empirical problems that have to be addressed by any such theory. This empirical investigation is the subject of Chapter Five. Following this, in Chapter Six, I fill out the sketch provided here, as a full theory of the Hebrew verbal system, as hosting both alternations and MCM, in its rich structure of overlaying roles, and—a distinction we shall find to be even more fundamental—as containing both root-derived and verb-derived verbs.

CHAPTER 5 ROOTS ACROSS PATTERNS IN HEBREW 5.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 5.2. Roots across Patterns: Types and Tokens . . . . . . . . . . . . . . . . . . . . . . . . . 121 5.3. The Relation between Verb Alternations and Morphological Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 5.4. A Morphological Account of the Hebrew Verbal System: Binyanim as Inflectional Classes (Aronoff 1994) . . . . . . . . . . . . . . . . . . 150 5.5. A Syntactic Account of the Hebrew Verbal System: Binyanim as a Spell-out of Syntactic Heads (Doron 1999, 2003) . . . . . . . . . . . . . . . . . 157 5.6. Binyanim and the Typology of Verb Alternations . . . . . . . . . . . . . . . . . 173 5.1. INTRODUCTION Most Hebrew roots appear in more than one verbal pattern. Two extreme approaches could be adopted for treating the multiple occurrences of roots in different patterns. I will term these approaches “Determinism” and “Nondeterminism”. To see how a deterministic system would work, imagine a language with five verbalizing suffixes that attach to roots. For each root, the suffixation yields verbs whose meanings are the following: (1)

Verbalizing morpheme -ma -boo -pi -to -ge

Verb meaning perform the action for a while perform the action to completion perform the action for the benefit of someone else perform the action with an instrument cause someone to perform the action

Such a system (which is approximated in many agglutinative languages) is characterized by two properties.

118

R OOTS A CROSS P ATTERNS

IN

H EBREW

119

First, it has little if any gaps. Every verb can be conceived of as being performed for a while, with an instrument, or for the benefit of someone else. Some verbs, such as “run” or “laugh”, may not lend themselves easily to a completive interpretation, due to their inherent semantic properties. Other verbs, such as “arrive” or “win”, may be difficult to construe with a durative interpretation (perform an action for a while). But the majority of roots would combine with all five suffixes, creating verbs of different types (indeed, in languages with rich verbal prefixes, such as those of the Bantu family, most roots create both causative and applicative verbs). The second property of such a system is that the contribution of each verbalizing suffix is semantically transparent. That is, for each combination of a root and a suffix, we know what kind of event is denoted by the verb. Now consider a non-deterministic system. In its most extreme form, verbalizing suffixes not only do not have a fixed contribution, but they do not equally attach to all roots. For each root one has to learn, first, what verbal suffixes attach to it, and, second, the interpretation assigned to each combination of a root and a suffix. With these two extremes in mind, let us examine the Hebrew verbal system. W We saw in Chapter Four that the system has both a large degree of regularity (in argument structure alternations), and a considerable degree of irregularity (gaps, MCM). When faced with those empirical facts, linguists working on the Hebrew tend to adopt some version of either determinism or non-determinism.1 Those advocating a deterministic approach consider the root itself as lexically fully specified. The contribution of the binyan is no more than that of other inflectional categories such as tense or person. Hence, the semantic result of the combination of the root and the binyan is always predictable. Such a view was adopted already for Biblical Hebrew by Gesenius (1910), and is followed by authors such as Waltke and O’Connor (1990).2 We already saw that a deterministic approach encounters several difficulties. To start with, I argued in Chapter Three against a fully specified semantic content of the root. Further, following Berman (1978), we can notice many irregularities in the relations of verbs across patterns. As will be illustrated

1

A notable exception is Berman (1978), who recognizes both the regularity and irregularity of the system, but does not offer a theoretical account deriving it. Glinert (1989) also notes that binyanim cannot be equated with semantic or syntactic roles, but, unlike Berman, does not acknowledge the highly specialized meanings roots acquire in the environment of different binyanim. 2 This approach is also applied to Modern Hebrew, particularly in grammars and textbooks for teaching Hebrew (e.g. Rosen 1962). Some observations on inherent meanings of binyanim can be found in Goshen-Gottshtein et al. (1967) and Sadka (1975).

120

C HAPTER 5

below, less than a third of Hebrew verb-creating roots participate in systematic alternations of argument structure, and the semantic relation between two verbs derived from the same root may be quite tenuous. As for binyanim themselves, we saw in Chapter Four that it is impossible to identify their semantic contribution as “causative” (for binyan 5) or “reflexive” (for binyan 7). Apart from the passive patterns, 4 and 6, all other binyanim may have many different types of verbs in them. We shall consider such difficulties for a deterministic approach later in this chapter. The non-deterministic approach, on the other hand, advocates total irregularity in the binyan system. That is, the lexical prime is not the root, but the “word”, or the combination of a root and the binyan (Ornan 1971, Schwarzwald 1974). Such an approach does not have a problem with gaps in the system, with MCM or with synonyms. If the system is arbitrary, then each root has to be learned individually with the patterns in which it appears and the meanings it acquires when appearing in these patterns. However, as already Berman (1978) has noted—and as argued in Chapter Three—such an approach fails to capture the lexical core shared by verbs containing the same root (to repeat our example—xaˇ — av, think, xiˇev, calculate, and hexˇiv x , consider). Furtherxˇ more, it fails (or, in fact, does not attempt) to capture the degree of regularity that exists in the system, and which seems to form part of the knowledge of language that Hebrew speakers have (e.g. the fact that most verbs in pattern 2 are related to verbs in pattern 1, either as passive of active forms in pattern 1 or as non-causative forms of causative verbs in pattern 1, remains accidental). Theories advocating complete regularity or irregularity are bound to fail empirically. It is a fact that there is a considerable degree of chaos in the Hebrew verbal system, alongside a considerable degree of systematicity. When attempting to provide a theoretical account of the system, one can go in two possible directions. The first line of research is to abstract away from the chaos and study the systematic aspects of the system, simply by studying a sub-system of Hebrew. Once we do that, we can try to uncover the order behind what appears like chaos, and provide a unified account of the system. In other words, we can try to get as close as we can to a deterministic system. This research program is adopted by Doron (1999, 2003). The second line of research is to find a principled reason why there should exist such chaos in the system. Once we understand why the chaos is an inherent part of the system, we can understand how the system works, and characterize its systematic and nonsystematic aspects. This is the approach taken by Aronoff (1994), who focuses on the dual role of the verbal patterns in terms of derivational and inflectional morphology (see discussion of both Aronoff and Doron in Sections 5.4 and 5.5).


IN

H EBREW

121

In this book, I take the second direction (while adopting the insights of Doron 1999, 2003). I argued in Chapter Four that the chaos in the binyan system is not a coincidence. It results from the fact that the system has a dual role: to mark argument–structure alternations and to create multiple verbs from the same root (MCM). Since all Hebrew verbs must appear in a binyan (as argued in Chapter Two), this entails that the system should contain, simultaneously, a degree of regularity and of irregularity. Once we assume that there is some irregularity in the system, the extent of the irregularity becomes crucial. It makes a great difference for the language learner if this irregularity concerns 80% of the roots or only 5% of them. In a fully rule-regulated system, the abstract rules matter. In a partially stochastic system, the learner must rely not only on rules, but also on quantitative tendencies. It is not my intention to provide a computational, stochastic account of the learning of the Hebrew verbal system. However, the sense of relative numbers should inform the priorities as we disentangle, in the following two chapters, the two aspects of the Hebrew verbal system—the regular and the irregular. In Sections 5.2 and 5.3, “types and tokens” and “verb alternations and morphological form”, I offer a quantitative examination of the binyan system, based on a corpus study. Section 5.2 introduces the corpus, while Section 5.3 describes the corpus findings in detail. Following these findings, I discuss, in the remaining three Sections 5.4–5.6 three theoretical approaches that serve, further, to refine my analysis. Aronoff (1994) and Doron (1999, 2003) are discussed in Sections 5.4 and 5.5, respectively. I show what empirical aspects of the Hebrew verbal system these theories fail to capture, but, at the same time, I also outline the pieces of the puzzle that Aronoff and Doron contribute. A further piece is a cross-linguistic typology of verb alternations, discussed in Section 5.6 (especially following Haspelmath 1993). With those pieces in place, I move on in Chapter Six to expand my sketch from Section 5.3 into a theory of roots and patterns in the Hebrew verbal system.

5.2. ROOTS ACROSS PATTERNS: TYPES AND TOKENS This section is based on a study of a corpus of Hebrew roots, which I constructed specifically for that purpose. It contains those roots listed in the Even Shoshan Hebrew dictionary (1978) which create verbs, and which are actually used in present day Hebrew (with the addition of some roots that have become a part of the vocabulary since). My purpose was to create an inventory of verb-creating roots that an educated speaker of Israeli Hebrew has. Leaving out the more marginal slang verbs and some frozen bookish forms (in particular, those that only appear with one lexicalized object), this leaves us with a corpus 1812

122

C HAPTER 5

verb-creating roots.3 This is in itself a significant result: the number of verbcreating roots in Hebrew is surprisingly low (note that about 15% of these roots create two or more different verbs through MCM, as will be illustrated in detail below).4 I showed in Chapter Two that Hebrew is more reluctant than English in admitting new verbs, a fact we associated with the morpho-phonological constraints on the form of Hebrew verbs. But it seems that, independently of such constraints, the language does not allow many verbal forms that exist in other languages such as English. Comparing the two languages with respect to one lexical verb class shows the vast difference between them. Hebrew, for example, has a handful of verbs of removal (qilef, f peel, xalav, milk, gilʕen and girʕen—both meaning pit and probably related), whereas English has over 40 (cf. Levin 1993: bone, core, rind, stalk, stone,skin, peel, etc.). No morphophonological constraints could explain why there is no Hebrew parallel to stalk, meaning “remove the stalk”. Manner of motion verbs in English and Hebrew show the same contrast. While Hebrew has only a few manner of motion verbs, English has over a hundred. Manner components that are lexicalized into verbal forms in English (hop, jog, race r , sleepwalk, tiptoe—cf. Levin 1993) either do not exist in Hebrew, or are expressed by adverbial elements (“walk on tiptoes”). This is the first significant fact provided by the corpus: Hebrew has relatively few (pure) verbs. In addition to listing each root according to the patterns in which it appears, the corpus also sorts out roots according to the relation r between their different appearances. This categorization provides another important result: the appearances of the same root across different patterns fall into a limited, well-defined set of categories. In this corpus, I identify five such types: r r r r r 3

Single occurrence (SO) Passive only Alternation Multiple contextualized meaning (MCM) Synonyms

Although many slang verbs are included in the corpus, I left out some old words from 1970s and 1980s that are out of use in current Hebrew (e.g. hitparxeax, behave wildly). For the same reason, I did not include some of the most recently coined verbs, such as cˇ itet (to chat on the internet) or dijah (to DJ). Several roots that used to create verbs in Biblical Hebrew create only adjectives in Modern Hebrew, and were therefore left out. Similarly, I decided not to include roots that only appear with one lexicalized object (e.g. maxaʔ kapayim, applaud). 4 McCarthy (1979) found about 4000 roots for Classical Arabic. Even this number is considerably smaller than the potential number of roots made available by the language (the possible combinations of segments, modulo phonotactic considerations). The number of Hebrew roots would have increased considerably if we included all roots found in Biblical Hebrew (although it would be nowhere near 4000), but my analysis is based on synchronic knowledge of speakers rather than a diachronic corpus.


IN

H EBREW

123

Single occurrence roots are those roots that appear in one pattern only. Examples of such roots are as follows: (2) a. b. c.

Root √ √clʕ √gxk sˇʕ l

Binyan CaCaC (P1) CiCCeC (P3) hitCaCCeC (P7)

Verb calaʕ (limp) gixex (chuckle) hiˇta ˇ ʕel (cough).

The second group, which I termed “passive only”, are roots that appear in one active pattern and its corresponding passive pattern, creating two verbs—one active, another passive: (3) a. b. c.

Root √ √ʔ rz √qrcf svh

Active

Passive neʔeraz (pack—passive, P2) qircef (scrub, P3) qurcaf (scrub—passive, P4) hisvah (camouflage, P5) husvah (camouflage— passive, P6).

ʔaraz (pack, P1)

The third group, argument–structure alternations (or, for brevity’s sake, alternations), includes roots that appear in two or three different patterns, where each appearance corresponds to a systematic alternation of argument structure. Examples are given below: (4) a. b. c.

Root √ prd

Verb1 Verb2 V nifrad (separate, intr., P2) hifrid (separate—trans., P5) √ √xmm hitxamem (heat, intr., P7) ximem (heat—trans., P3) ʕ md ʕ amad (be standing, P1) neʕ emad (stand up, P2)

The fourth type of relation between two occurrences is multiple contextualized meaning (MCM), as discussed in Chapter Three: (5) a. b. c.

Root √ √bxn ˇ √xsb grˇs

Verb1 baxan (examine, P1) xasav ˇ (think, P1) gereˇsˇ (expel, P3)

Verb2 V hivxin (discern, P5) xiˇsev ˇ (calculate, P3) hitgareˇs (get divorced, P7).

Finally, a relatively small group of roots creates synonymous or nearly synonymous verbs in different patterns:5 5

Only a few homonyms exist in Hebrew, and these can usually be traced historically as belonging √ to different roots (e.g. xlt1 , which w creates hexlit (decide) in pattern 1, is related to the Aramic √ √ root xlt, and the root xlt2 , w which creates xalat (brew) in pattern 5, is related to the Arabic √ root xlt). I do not list them here as a group, but take each homonym as a separate root.

124

C HAPTER 5

(6) a. b. c.

Root √ √qnʔ √gwr kpl

Verb1 qineʔ (envy, P2) gar (live in, P1) kafal (multiply, P1)

V 2 Verb hitqaneʔ (envy, P7) hitgorer (reside, P7) hixpil (multiply, P5)

The very existence of this final category seriously undermines a deterministic account of the system. If the contribution of the binyan is fixed and determined, how come the same root creates two synonymous verbs in different binyanim? One could argue that synonyms make only a small percentage of the corpus, but they are certainly not negligible, making almost 7% of roots (see discussion below). Furthermore, what about MCM? Fifteen percent of the roots create two or more different verbs when appearing in different patterns. In such cases, it is evident that the binyan itself does not have a fully predictable role (except for the transitivity restrictions on patterns 2 and 7, which always hold). Moving now to alternations, as was illustrated in Chapter Four, causative verbs may appear in patterns 1, 3, or 5, inchoatives verbs appear in patterns 1, 2, 5, or 7, and reflexives—in patterns 2 or 7. The binyan system is thus far from being deterministic. Recall that in a deterministic system we predict little or no gaps. The corpus shows that the binyan system has many gaps, that is, binyan slots that are not used for creating actual verbs. Twenty seven percent of the roots appear in one pattern only—leaving six slots empty. Another almost 30% of the roots appear in two patterns only, active and passive, leaving five slots empty. Among the 30% of roots that alternate, most create no more than two verbs (three, if we count passives), leaving four slots empty. Only 34 roots—less than 2%!—create three verbs (four, if passives are included) when participating in argument–structure alternations. And even this still leaves three slots empty. A handful of verbs appear in five, six, or even seven patterns—creating both argument–structure alternations and MCM (or synonyms). But the majority of verbs appear in one or two patterns only. These gaps in the system reflect two features. First, this is not a deterministic system. It is not a coincidence that so many roots appear in one or two patterns. The contribution of the binyan is not semantically and syntactically transparent, and there is no way to learn verbs, combinatorically, from their roots and patterns. In other words, there is no way for a single root to appear in all the binyanim, creating imperfective, applicative, perfective verbs, etc. Rather, the number of semantic and syntactic relations encoded in the Hebrew verbal system is small, and it is up to the semantics of each root to decide whether it can participate in argument structure alternations or be assigned multiple contextualized meaning. Second, the system is “poor”: Hebrew makes do with a small number of verbs. For each root, there are five possible morphological verbal


IN

H EBREW

125

forms (the five active binyanim). Hebrew could, potentially, become richer in verbs by filling in these empty binyan slots through massive MCM or even massive homophony. Although this would make the binyan system gap free, it would also further erode its regularity. So, the degrees of regularity and of completion of the system are, in fact, inversely correlated. Finally, looking at the corpus, it becomes obvious that except for the exclusively passive patterns 4 and 6, in each binyan there may be instances of MCM, argument–structure alternation or synonyms. Here is where the quantitative aspect matters: we may note that a verb in pattern 2 is much more likely to be a passive of a verb in pattern 1 than a synonym or a MCM with another binyan. Such forms as the latter are less likely, yet possible, marking the system as non-deterministic. Before we move on to the analysis of the corpus, I would like to make a note about the listing of roots in the corpus: the categories single occurrence and passive only are, by definition, exclusive of all others. The other categories, alternation, MCM, and synonym, do not exclude one another, nor do they exclude passives. About 10% of the roots undergo an argument–structure alternation in three patterns, and are assigned MCM or create a synonym in another pattern. Consider the following example: (7) a. b. c. d. e.

Root √ √xzq √xzq √xzq √xzq xzq

Pattern CiCCeC CuCCaC hiCCiC huCCaC hitCaCCeC

Verb xizeq (strengthen—transitive, P3) xuzaq (strengthen—passive, P4) hexziq (hold, P5) huxzaq (hold—passive, P6) hitxazeq (strengthen—intransitive, P7)

√ Roots such as xzq are listed in the corpus as both alternation and MCM (I do not specify their passive forms separately, for reasons that will become clear in Chapter√ Six). Roots that create two different verbs, each with a passive form, such as ktr below, are listed as MCM: (8) a. b. c. d.

Root √ √ktr √ktr √ktr ktr

Pattern CiCCeC CuCCaC hiCCiC huCCaC

Verb kiter (circumvent, P3) kutar (circumvent—passive, P4) hixtir (crown, P5) huxtar (crown—passive, P6)

In the tables below roots are listed according to the five categories discussed above: A—an argument structure alternation. MCM—multiple contextualized meaning of the root.

126

C HAPTER 5

SO—single occurrence of the root. 1P, 3P, 5P—a root occurring only in pattern 1, 3, or 5, together with its matching passive pattern (2, 4, 6, etc.). Syn—a root that creates synonymous verbs in two different patterns. Table 5.1. Occurence of roots across patterns: types and numbers SO1

SO2

SO3

SO5

147 14 148 80 (8.1%) (0.77%) (8.1%) (4.4%)

1P

3P

5P

201 226 111 (11%) (12.5%) (6%)

Total (SO 1P) 348 (19.2%)

SO7

A

CM

100 (5.5%)

542 (30%)

284 123 (15.5%) (6.8%)

1 Total + (SO 3P 374 (20.6%)

3 Total + (SO 5P 196 (10.5%)

Syn

5 Total SO Total SO + + Passive 489 (27%)

1027 (56.7%)

Consider table 5.1, where the initial results of our root corpus are summarized. The total number of root is 1812—but the total number of categories listed is 1976. One hundred and sixty four roots (9%) have more than one entry, usually creating a combination of an alternation, MCM, and synonym. The findings of the corpus, as summarized in table 5.1, show how far the binyan system is from being deterministic. Twenty seven percent of the roots are single occurrence roots. Another almost 30% of the roots create only active and passive forms of the same verb. That is, more than half of the roots appear in one or two patterns only, leaving five or six gaps. Only 30% of the roots participate in argument–structure alternations, which do not occupy more than two or three slots. A little over 15% of the roots create different verbs through MCM, and almost 7% of the roots create synonymous verbs when appearing in different patterns. In other words, more than a fifth of the roots create multiple verbs whose meanings are in no way predictable from the binyan in which they appear. Consider the formation of passives. Hebrew only has passive forms of verbs with direct objects. It does not have impersonal passives or and does not allow passivization of prepositional objects, in the same way English does.6 Yet there 6

The only exception to this is the locative preposition be (in/at). Certain verbs that take this preposition may be passivized, for example tamax be. . . (support someone) and nitmax ʕal yedey. . . (be supported by).


IN

H EBREW

127

are quite a few Hebrew verbs that fail to form a passive form for no apparent reason. For example, ripeʔ (heal, P3) takes a direct object, but does not have a passive form ∗ rupaʔ in P4. Similarly, ʔereax (host, P3) takes a direct object but does not passivize: and there is no ∗ ʔurax (only hitʔareax, be a guest, in P7). Many other verbs allow passivization, but pose certain restrictions on their passive forms. In particular, durative, habitual, gradual, and imperfective readings are much preferred to perfective or resultative readings. Therefore, passive forms of many verbs sound odd without some adverbial modification. The availability of the passive also has to do with the lexical properties of the verb. Verbs such as ʔaxal (eat, P1) are more naturally construed as a passive than zalal (eat up, devour, P1), which is more perfective. It is not the case that such verbs lack passive forms completely, but they require more context, and are less widely used. To achieve a felicitous reading of the passive of zalal (devour, P1), adverbial modification regarding completion, as well as reference to the external argument, are required: (9) a.

b.

c.

ha xatifim neʔ exlu the snacks eat—passive “The snacks were eaten.” ?ha xatifim nizlelu the snacks devour—passive “The snacks were devoured.” ha xatifim nizlelu tox sˇniyot the snacks devour—passive in seconds al yedey ha yeladim by the children. “The snacks were devoured by the children within seconds.”

5.3. THE RELATION BETWEEN VERB ALTERNATIONS AND MORPHOLOGICAL FORM We now know the share of each of the five groups in the pattern system. Given the many gaps it has, the system is rather far from being deterministic. But now the question arises, whether within each of these groups there is morphological regularity. That is, even if argument structure alternations make only less than a third of the roots, do these alternations follow strict rules, e.g. appear in certain patterns and not others? This question also bears on the learnability of the pattern system. Language learners discover that the verbal system of Hebrew has a degree of irregularity in it. But what is the degree of irregularity? For example, do roots that are assigned MCM appear in any given two patterns, or

128

C HAPTER 5

do they tend to appear only in certain combinations of patterns? If the latter is the case, then this already provides valuable information for learners: they can expect certain pairs to be an instance of MCM, but not others. If, on the other hand, there are no particular patterns associated with MCM, then the learner’s task in considerably more difficult. Similarly, knowing that the Hebrew verbal system has many systematic alternations of argument structure would be of little help if no morphological cues would be available to identify such alternations. If any two patterns could be an instance of any alternation, learning the binyanim as a system would become meaningless. On the other hand, if each syntactic alternation is expressed through its specific patterns, then the system becomes close to deterministic, in spite of its gaps.7 In what follows, I show that there is a large degree of morphological regularity within each class of roots. These regularities narrow down considerably the task of the language learner. Passive and argument–structure alternations tend to appear only in a certain groups of patterns, while MCM and synonyms tend to appear in another group (The two groups do overlap, but not completely). On the other hand, there are other factors that bring irregularity into the system: certain combinations of patterns may mark either argument structure alternations or MCM, alternations themselves do not always conform to the same morphological pattern and the same syntactic alternation may be morphologically marked in two different ways. The fundamental issue, then, is that of a many–many relation between morphological marking and syntactic structure: the same morphological patterns marks more than one syntactic structure, and the same pairs of patterns mark more than one syntactic relation between two verbs (e.g. causative–inchoative and MCM). This property of the system will serve as a starting point of our analysis in Chapter Six. 5.3.1. Single Occurrence (SO) Roots Single occurrence roots appear in patterns 1, 2, 3, 5, and 7—that is, all the patterns except the two that are used exclusively as passives. Table 5.2 presents the number of SO roots across these patterns, and their percentage of the total number of roots in the corpus. Patterns 1 and 3 have the largest number of SO roots. Of the 489 single occurrence roots, 295 (60.3%) appear in patterns 1 or 3. Put differently, these roots 7

Obviously, a full description of the problem facing the language learner would have to take account also of the frequencies of root-tokens as they appear in speech. While such an analysis is beyond the scope of my study, I point out where certain verb forms tend to be ‘bookish’ and so less important for language acquisition. Usually, however, the phenomena I describe do not correlate with specific registers, and nearly all are present in the learner’s environment.


IN

H EBREW

129

Table 5.2. Distribution of single occurrence roots 1

2

3

5

7

SO 147 (8.1%) 14 (0.77%) 148 (8.1%) 80 (4.4%) 100 (5.5%)

include 16% of the entire root corpus. Patterns 5 and 7 also contain a considerable number of SO roots: about 9% of the roots in the corpus appear exclusively in pattern 5 or 7. There are thus fairly good chances that a verb appearing in patterns 1, 3, 5, or 7 is a single occurrence and does not have any corresponding forms derived from the same root. This morphological information, in combination with syntactic information (can the verb be passivized? Is it likely to be causativized?), can be exploited by speakers to estimate whether the verb is likely to be a single occurrence or whether it may have some corresponding alternants. Pattern 2, on the other hand, contains only a handful of SO roots (14) which makes less than 1% of the corpus. These 14 verbs also tend to be bookish, and not constitute a part of the early, daily vocabulary.8 We can conclude that children learning Hebrew do not face the question whether verbs appearing in pattern 2 are single occurrence or not: they are practically always passives or part of an argument structure alternation (see discussion of these categories in what follows). w The group of SO roots is not semantically or syntactically homogeneous. My criterion here is purely morphological: this group includes the roots that create only one verb. A detailed analysis of types of SO roots is beyond the scope of this work, but we should note that at least for some such roots there are clear semantic grounds for their being SO. Roots of activity or of manner of motion usually appear in one binyan only, where they create an activity verb or a manner of motion verb. They cannot be causativized: causativization of action verbs is rare in Hebrew.9 They also cannot be passivized: passivization requires a direct object, and these verbs do not normally take a direct object. They cannot participate in other alternations involving a direct object, such as reflexivization or de-transitivization, because they do not have a direct object in the first place. Finally, as we noted in Chapter Three, such roots are unlikely to 8

The two main exceptions are nehenah, enjoy, and nitqal, bump into, which are widely used in everyday language. Examples of SO roots in pattern 2 are nexpaz (make haste, literary), namaq (rot, bookish), nixlam (be ashamed, bookish), neʔenaq (groan), nasog (retreat), and neʕedar (be absent). 9 Hebrew allows causativization of certain action verbs, such as raqad r (dance), rac (run), ʕaxal (eat), or caxaq (laugh). In this respect, it is more permissive than English, where such verbs may only form periphrastic causatives. However, the process is not fully productive (see Arad 2002).

130

C HAPTER 5

create multiple verbs through MCM. To this group belong also roots of manner of speaking and of sound and light emission. Such roots create verbs without a direct object, so transitivity alternations do not apply to them. They cannot be construed as having an external causer, and so they do not causativize, and, finally, they do not tend to create multiple verbs through MCM. On the other hand, there are many roots that are included in the SO group because of diachronic reasons. Biblical Hebrew had many verbs denoting a state. Modern Hebrew did not retain most of these stative verbs, except in their adjectival form that is derived from the present tense of the verb (Benoni). Such roots used to create argument structure alternations in Biblical Hebrew, with the stative verb in pattern 1 or 2 and its causative counterpart in pattern 5. Modern Hebrew typically only retained the causative verb, which is listed in the corpus as a single occurrence verb: (10) √ Root √cmʔ √lʔ h qll

Stative form camaʔ (P1, be thirsty) nilʔ ah (P2, become tired) qal (P1, be light, easy)

Causative form hicmiʔ (P5, cause thirst) hilʔ ah (P5, tire) heqel (P5, make easier)10

Another group of SO roots creates verbs that could in principle passivize, but for which the passive form is not in fact in use. For example, teʕev (loath, pattern 3) does not have a passive form, although the verb takes a direct object, and there is no syntactic reason why it could not passivize:11 (11) a. b.

Dani teʕ ev tered Dani loathed spinach. ∗ tered toʕ av ʕ al yedey Dani Spinach loath—passive by Dani

Note that other verbs of this semantic type—such as love—do have a passive form. As noted above, the availability of passive is not completely regular, in the sense that, from the presence of the active, the presence of a passive form cannot be predicted.12 Finally, let us note again that the share of SO in the corpus would have been higher if we included more slang verbs and more marginal, frozen verbs. Not surprisingly, these tend to appear in one pattern only. √ These verbs are made from the root qll, which yields contracted phonological forms (∗ qalal qal, ∗ hiqlil heqel). 11 There exists an adjective metoʕav, loathsome, which is in the form of the present passive participle, but crucially this is not a verbal form. 12 See Doron’s (2003) observation, that passives require an Agentive logical subject in the active counterpart. The verbs cited above have an Experiencer subject. 10


IN

H EBREW

131

5.3.2. Passive in the Hebrew Verbal System

Table 5.3. Marking of passive forms 1–2 Active–passive

279

3–4 275

5–6

3–7

213

13

6

Other 314

The corpus lists separately the roots that form only active and passive forms of the same verb. But many other roots form passives as well, alongside other types of alternations (e.g. MCM). Let us now consider the morphological realization of all the passive forms that appear in the binyan system, as summarized in table 5.3. This table summarizes all the tokens of active and passive pairs in the Hebrew verbal system.15 The degree of regularity exhibited by Hebrew passive forms is striking. It is not an exaggeration to say that active–passive forms show the largest degree of regularity in the Hebrew verbal system. Almost without exception, the morphological form of the passive is predictable from that of the active, and vice versa, a fact that considerably facilitates the task of learning the language. Coming across an active verb, the child knows immediately what would be the morphological form of the passive verb, assuming one exw ists; coming across a passive verb, the child infers both the existence and the morphological form of the corresponding active. Onto the syntactic and morphological regularity between actives and passives is added a semantic regularity (that is, passives never acquire a special meaning independent of the active). This gives the child a valuable key into a considerable part of the Hebrew verbal system. Note, now, that while patterns 3 and 5 have two patterns that are exclusively used for their passive forms, the passive forms of verbs in pattern 1 appear in pattern 2, which is used for √ A small number of roots have an active form in pattern 3 and a passive one in pattern 7: ʕ kzb; √ √ √ √ ʕ ffˇsr; bqˇs; zmzm; cwh; qbl. The forms in pattern 7 are true passives, and allow a by phrase. A number of other verbs in pattern 3 have a passive-like form in pattern 7, sometimes in addition to a passive form in pattern 4—but such forms do not allow a by phrase. √ √ 14 xˇsb ˇ and yˇsʕ have their active forms in pattern 5 (hexˇiv x , consider, hoˇi xˇ ˇ ʕa, save) and their passives (nexˇav x , be considered, noˇaʕ, be saved) appear in pattern 2 rather than the in pattern √ 6, as expected. The root sˇyr creates an active form in pattern 1 (sˇar, sing) whose passive form in pattern 6 (huˇar, was sung), rather than the expected pattern 2. 15 Passive forms of verbs that are part of an argument structure alternation are not listed here. Note, though, that they behave similarly to these forms from a morphological standpoint: active forms in patterns 1, 3, and 5 have their passive form in patterns 2, 4, and 6, respectively. 13

√

132

C HAPTER 5

other alternations (argument structure alternations, MCM, synonyms).16 However, from a purely quantitative point of view, the number of verbs in pattern 2 that are passives of verbs in pattern 1 is much higher than those that form synonyms, MCM, or other alternations: 279 passive forms as opposed to about 40 synonyms, 14 tokens of SO roots and 25 instances of MCM. Almost 80% of the verbs appearing in pattern 2 are passive forms of verbs in pattern 1. The ffact that pattern 2 is not exclusively passive has a further significance. While for certain agentive roots pattern 2 is indeed used only as the passive of pattern 1 (e.g. neʔexal, eat—passive, nixtav, write—passive), for verbs of change of state, on the other hand, pattern 2 is ambiguous between a passive and an inchoative reading. For example, sˇavar (break—transitive, P1) has a form in pattern 2, niˇsˇbar, which w is ambiguous between an inchoative reading, of coming into a state (something breaking spontaneously) and a passive reading. Only adverbial elements such as by phrases can distinguish between the two. Such ambiguity does not occur with verbs in pattern 3, whose passive and inchoative forms appear in patterns 4 and 7, respectively. The regularity and irregularity in binyan 2 are thus derived by event-structure properties of the corresponding verb in binyan 1. Even the few exceptional passive forms in Hebrew have a large degree of regularity in them. All the active verbs in pattern 3 whose passive form is not in pattern 4 have their passive in pattern 7. All the irregular passive forms appear either in patterns 2 and 7 (inherently intransitive patterns), or in pattern 6 (which is inherently passive). That is, no passives appear in patterns 1, 3, or 5, those patterns that may host transitive verbs. To summarize: although the availability of passive forms in Hebrew is somewhat irregular, passive morphology is nearly completely regular. From a given w active form, we can tell what the passive form would be. From a given passive form, we can tell what the active form is. 5.3.3. Argument Structure Alternations in the Hebrew Verbal System Thirty percent of the roots in the corpus participate in argument structure alternations. As we saw in Chapter Four, the actual number of alternations marked by the Hebrew verbal system is relatively small. The three major alternations we shall be concerned with here are the following: r The causative alternation (an alternation between a causative and a non-

causative verb) 16

Hebrew had, at an earlier stage, a pattern that was used exclusively as the passive of pattern 1: CuCaC. This pattern was lost at an early stage; only its traces exist in the bible, and it does not exist in Modern Hebrew.


IN

H EBREW

133

r The stative alternation (an alternation between a stative and a non-stative

verb)

r The reflexive alternation (an alternation between a reflexive and a transitive

verbs) All these alternations involve changes in the event structure of the verb, syntactically expressed by a change in the transitivity of the verb. As will be illustrated below, most alternations involve two verbs derived from the same root (three verbs, if passive forms are counted). This means that even those roots that alternate leave many gaps. This corroborates our initial assumption, that the binyan system is non-deterministic: the syntax allows a limited number of alternations, and this number is smaller than the number of morphological patterns (of course, argument structure alternations and MCM are not exclusive of one another, as we saw in Chapter Four). The corpus reveals an important fact concerning patterns 1, 3, and 5. These patterns may host either transitive or intransitive verbs. In principle, then, there could be transitivity alternations with verbs appearing in these patterns. However, the only such case is a causative alternation where the non-causative verb appears in pattern 1 and the causative appears in pattern 5.17 Interestingly, the majority of transitivity alternations involve pattern 7, and to a lesser extent pattern 2, both of which mark the intransitive alternant. In the remainder of this section, I will review the morphological form of verbs alternations, according to three different groups. The first, and by far the largest group of alternating roots, contains roots that alternate in two or three different patterns (passive forms counted). The second group, much smaller, contains roots that alternate in four patterns. Finally, there is a small group of roots that undergo certain alternations that do not fall under any of the principal argument structure alternations related to event structure, and whose members will be discussed individually. 5.3.3.1. Two and three-place alternations The following tables list the number of roots that participate in argument structure alternations, according to the patterns in which the different verbs they create appear. In other words, I present here the materials required to study the question “what do ordered relations of binyanim mark?”. For example, does the sequence {P3, P5} mark causativization? In the preceding section it became clear that the sequences {P1, P2}, {P3, P4} and {P5, P6} did, at least among 17

There are only two causative alternations between patterns 1 and 3, samax (rejoice, P1) and simeax (make happy, P3), and lamad (learn, P1) and limed (teach, P3), whose status as a causative is not clear. No causative alternations exist between patterns 3 and 5.

134

C HAPTER 5 Table 5.4. Two-place alternations (plus passive) 1–2–3 1–2–7 1–3–4 1–5–6 1–5–7 2–5–6 3–4–7 5–6–7 2–3–4

Cos, caus, pass. Caus. pass. stative Non-caus. caus, pass. Trans. reflex T (recip) pass. Active pass., impers. pass. Trans., intr., T pass. Stative, caus., reflex. Non caus., caus., cos State, pass. of state, cos Active pass. 130 caus.

218 119 121

8

17

104

11

10

4

13

120

53

1122

55

123

74

6

8

124

525 126 127 528 129

others, mark passivization. To anticipate: the very difficulty of reading any such simple pattern of marking off the tables is significant. Two points should further be clarified with regard to the listing of roots. First, I broke the roots into two groups, given in tables 5.4 and 5.5, according to whether there is also a passive form. For the sake of looking at the alternations w 18 19 20 21 22 23 24 25 26 27 28 29 30

√ √ hpk, cbr. √ sˇkk. √ gˇsm. √ xrz. √ √ √ √ √ √ √ √ √ √ √ ˇ ngr, ncl, ntz, sbb, swt, prd, cmd. dps, kns, knʕ , kˇsl, √ ysp. √ ntq. √ √ √ √ √ bcʕ , kbd, mnh, ngn, nhl. √ glm. √ cdq. √ √ √ √ √ kʕ s, chl, cxq, rgz, sˇlt. √ ʕ hb. √ lmd.


IN

H EBREW

135

Table 5.5. Two-place alternations (no passive) 1–2 Trans.–refl. (recip.) T Cos (inch.) -caus. Non-caus. -caus. Tr. -non-tr. Stative -cos Action -cos Stative-caus. Caus–inch. Body-part

1–3

1–5

1–7

21

2–5 7

2

15

1

3 2 6

3–7

5–7

6 27 5 5 2

2 2 3

5

2 1131

3

2

2

themselves, we could combine the two tables together, since the formation of passive does not indicate any particular semantic or syntactic properties of verbs (except that of having a direct object). I set apart the two tables to give the reader an idea of the number of gaps occurring in each root group. Second, the different rows in each table should be clarified: I discuss here the causative alternation, an alternation between a causative and a non-causative verb. In the tables, I broke the alternation verbs into inchoative verbs (change of state verbs, verbs of becoming, verbs of falling into a state) and “other” non-causative verbs that participate in the causative alternation. The alternation itself is the same, one root combining with a causative or a non-causative verb. The distinction between inchoative verbs and other verbs is relevant for the debate regarding the transparency of verb morphology (see in particular Section 5.6), so I kept this distinction in our tables. Consider now the tables and the shorthand for the categories appearing in them: r r r r r r r r

31

(in)trans.—(in)transitive refl.—reflexive recip.—reciprocal inch.—inchoative caus.—causative cos—change of state impers.—impersonal pass.—passive

This group includes those roots that create both inchoative verbs and causative verbs in pattern 5: hichiv (become yellow, make yellow), heʔedim (redden), hiˇxir ˇ (blacken), hizkin (grow old, make older), hiˇmin iˇ (grow fat, fatten).

136

C HAPTER 5

The tables above sum up the alternations in the Hebrew verbal system. Syntactically, nearly all these alternations involve a change in transitivity, reflected semantically by a different event structure: r Causative vs. non-causative:

(12) a. b. c.

Non-causative caxaq (laugh, P1) niftax (open, P2) hitqapel (fold, P7)

Causative hicxiq (make laugh, P5) patax (open, P1) qipel (fold, P3)

r Stative vs. non-stative:

(13) a. b. c.

Stative ʕ amad (be standing, P1) ʔ ahav (love, P1) yaˇsˇav (be sitting, P1)

Non-stative neʕ emad (stand up, P2) hitʔ ahev (fall in love, P7) hityaˇsev ˇ (sit down, P7)

rT Transitive vs. reflexive (or reciprocal):

(14) a. b. c. d.

Transitive Reflexive gileax (shave—transitive, P3) hitgaleax (shave—reflexive, P7) raxac (wash—transitive, P1) hitraxec (wash—reflexive, P7) xiten (marry—transitive, P3) hitxaten (get married, P7)32 niseq ˇ (kiss—transitive, P3) hitnaˇseq ˇ (kiss—reciprocal, P7)

The causative alternation and the reflexive alternation involve a change in the number of verb arguments: one alternant (reflexive, non-causative) is intransitive, while the other is transitive. The stative alternation, on the other hand, typically retains the number of arguments in both alternants.33 All these alternations belong to a relatively low domain in the VP, the domain hosting verb arguments. Apart from passives, which were discussed separately above, the Hebrew verbal system does not encode any alternations involving higher domains, such as grammatical aspect, high applicative heads (see Pylkkänen 2002) or analytical causatives (causatives involving a separate causative verb or a higher causative morpheme, see Arad 2002). As is evident from the tables, the causative alternation is by far the most common alternation marked by the Hebrew verbal system. The transitive–reflexive 32

I did not separate reflexive and reciprocal verbs. I believe that both involve the same syntactic process, and they differ only in the way in which the event they describe is prototypically performed. One can shave oneself, but not marry oneself. 33 Some changes in transitivity may occur: ʔahav (love, P1) takes a direct object, while hitʔahev (fall in love with, P7) takes a prepositional complement.


IN

H EBREW

137

Table 5.6. Four-place alternations 1–2–5–6 1–3–4–7 1–5–6–7 2–5–6–7 1–2–5–6–7 Non-causative (cos included) 1734 -passive, causative, passive of causative Stative, causative, passive 635 of causative, cos Stative, causative, passive of causative, reflexive Transitive, reciprocal, T 139 causative, passive of causative Transitive, passive, T causative, passive, reflexive

136

637 138

240

alternation is also frequent. Note, though, that many of the roots listed under this group are not “true” reflexives, such as hitraxec (wash oneself) or hitgaleax (shave). Rather, in many cases it is clear that the two verbs are alternating, but the intransitive is not interpreted as “performing an action on himself”. Examples are zerez (hurry someone up, P3) and hizdarez (make haste, P7), or siken (put in danger, risk, P3) and histaken (take a risk, P7). Finally, the type of alternation termed here “body-part” alternation, is a special case of the reflexive alternation. The verbs that participate in it can be construed as involving the whole body of the participant, or a part of the body. In cases where the whole body is involved, the verb is understood as reflexive. When only a part of the body is involved, it is understood as a transitive: (15) a.

b.

34 35 36 37 38 39 40

Dani hizdaqef Dani rose “Dani rose.” Dani zaqaf roˇsˇ / ecbaʕ . Dani raised head / finger “Dani raised his head/his finger.”

√ √ √ √ Examples are zkr, sˇkx, ʔ kl, ʕ br. √ √ √ √ √ √ dlq, zkr, xrd, ʕ md, rʕ d, sˇkb. √ ybˇs. √ √ √ √ √ √ ʔ rk, blt, yˇsb, ˇ lht, rtx, sˇkn. √ bdl. √ pgˇs. √ √ lbˇs, pˇst. ˇ

138

C HAPTER 5

5.3.3.2. Four-place alternations While the majority of argument structure alternations involve two or three verbs from the same root, there are several cases involving four (and in two cases five) verbs derived from the same root. These cases are listed in table 5.6: The majority of these roots create two active verbs—one non-causative and one causative (I grouped here change of state verbs with other non-causative verbs)—each with a passive form. To take an example, consider the root √ ʔ kl:41 (16) a.

b.

c.

d.

41

Ha yeled ʔ axal daysa. ate (P1) porridge The child “The child ate the porridge.” Ha daysa neʔ exla (ʕ al yedey ha yeled). The porridge was eaten (P2) (by the child). “The porridge was eaten (by the child).” Ha ʔ em heʔ exila ʔ et ha yeled (be daysa) The mother fed (P5) OM the child (with porridge) “Mother fed the child (porridge).” Ha yeled huʔ axal (ʕ al yedey ha ʔ em be daysa) The child was fed (P6) (by the mother with porridge). “The child was fed (porridge by the mother).”

The roots that undergo this alternation include: √ ʔ kl ʔ axal (eat, P1), heʔ exil (feed, P5) (i) a. √ b. xdr xadar (penetrate, P1), hexdir (stick something in, P5) √ c. xkr xaxar (lease, P1), hexxir (lease to someone, P5) √ d. xrb xarav (fall down, P1), hexriv (destroy, P5) √ e. xrd xarad (worry, P1), hexrid (make worry, P5) √ f. xˇsd ˇ xasad ˇ (suspect, P1), hexˇsid ˇ (make suspicious, P5) √ g. xtm xatam (sign, P1), hextim (have someone sign, P5) √ h. tbl taval (be in water, P1), hitbil (dunk, P5) √ i. tbʕ tavaʕ (drown, P1), hitbiʕ a (drown, capsize, P5) √ j. nʕ l naʕ al (put a shoe on, P1), hinʕ il (put a shoe on someone, P5) √ k. spg safag (absorb, P1), hispig (make absorb, P5) √ l. ʕ bd ʕ avad (work, P1), heʕ evid (enslave, make toil, P5) √ m. ʕ br ʕ avar (pass, P1), heʕ evir (pass, P5) √ n. sˇʔ l sˇaʔ al (ask, borrow, P), hiˇsʔ il (lend, P5) √ o. sˇkx saxax ˇ (forget, P1), hiˇskeax ˇ (make forgotten, P5) √ p. skr saxar (rent, P1), hiskir (rent out, P5) √ q. sˇmʔ sˇamaʕ (hear, P1), hiˇsˇmiʕ a (make heard, P5) √ r. trm taram (donate, P1), hitrim (fund-raise, P5)


IN

H EBREW

139

The morphological regularity of this alternation is evident. In all the 17 roots that undergo this alternation, the non-causative alternant appears in pattern 1, the causative one—in pattern 5, and their passive forms appear in patterns 2 and 6, respectively. This group is considerably smaller than those roots that only form three verbs: non-causative, causative, and passive of causative. This may be because of the (universal) syntactic difficulty to causativize a transitive verb. Hebrew is more permissive than French or English in creating lexical causatives from transitive verbs (Arad 1998, 2002). Note, however, that there are often differences in interpretation between such causativized pairs. The causative of xatam (sign) is hextim, meaning “have someone sign something”; the causative of taram (donate) is hitrim, which w means get donations from someone, or fundraise. Heʕevid, the causative of ʕavad (work) means “have someone toil for you”, or “enslave”. In other cases, there is a change in the argument structure: xaˇsˇad means to suspect, while hexˇ xsid means make someone look suspicious. xˇ The subject of xaˇsˇad is not the object of hexˇ xsid. Similalry, safag means to absorb xˇ (mostly into oneself) while hispig means to absorb something into something else (cloth, sponge, etc.). The object of hispig is the substance absorbed, not the subject of safag. The second largest group of roots participating in four-place alternations are those that create a stative verb, a change of state (inchoative) verb, a causative verb (i.e. causative of a√state or a change of state), and a passive of that causative. For example, the root sˇkb creates sˇaxav (be lying, P1), niˇsˇkav (lie down, P2), hiˇ √sˇkiv (make someone lie down, P5), and its passive huˇsˇkav (P6). Another root, lht, creates lahat (be hot, P1), hilhit (make hot, P5), and its passive hulhat (P6), as well as hitlahet (become hot, P7). As made evident by table 5.6, unlike the group of two actives and two passives, here there is a considerable variety in the morphological expression of the alternation. An equal number of roots (six) have their change of state verb in pattern 2 or 7. However, there is a great tendency for the non-causative verb to appear √ in pattern 1 and for the causative verb to appear in pattern 5. Only one root, ybˇs, ˇ appears in a different morphological form altogether: yavaˇ (be dry, P1), yibeˇ (dry, P3), and its passive yubaˇ (P4) and hityabeˇ (become dry, P7).42 √ The other four-place alternations contain only one or two roots. bdl creates a stative verb, nivdal (be different, P2), a causative (hivdil, distinguish, differentiate, P5), and its passive, huvdal, in P6, as well as a reflexive √verb, hitbadel (make oneself different, seclude oneself, P7). A single root, pgˇs, ˇ creates a transitive verb (pagaˇ ( , meet someone, P1), a reciprocal (nifgaˇ, meet, as in 42

Some speakers do not accept the stative form yavaˇ in pattern 1. For them, this root belongs to a standard, three-place alternation of change of state in patterns 3, 4, and 7.

140

C HAPTER 5

“John and Mary met”, P2), a causative (hifgiˇ, ˇ have someone meet someone, P5), and passive of that causative (hufgaˇ, P6). Even in this singular alternation, the relation between the patterns is maintained. The non-causative verb appears in pattern 1, the causative verb—in pattern 5. Note the absence of a passive form off pagaˇ, meet. As noted above, there are quite a few verbs such as “meet”, which do not have passive forms in spite of there being no syntactic constraint w on their passivization.43 What is interesting in this case is that pattern 2, which is mostly used as the passive of pattern 1, hosts, in this case, the reciprocal r 44 form of the verb. √ √ Finally, two roots, lbˇs and pˇst, ˇ create five different verbs: lavaˇ (put on clothes, transitive, P1) and its passive nilbaˇ (whose subject is the clothes put on, P2), hilbiˇ (dress someone, P5) and its passive hulbaˇ (whose √ subject is the person being dressed, P6), and hitlabeˇ (dress oneself, P7). pˇst ˇ similarly refer to taking off clothes: paˇat ˇ (take off clothes, P1), its passive nifˇat f (P2), fˇ the causative hifˇit f (take someone’s clothes off, P5), its passive hufˇat fˇ f (P6) and fˇ the reflexive hitpaˇet ˇ (take one’s clothes off, P7). Obviously, such cases are limited because they are subject to constraints posed both by causativization of transitives and by the formation of reflexive verbs, which, in Hebrew, is not completely productive. Summing up the preceding two subsections, three observations are immediately noticeable. First, the main types of alternations do display meaningful tendencies (if not strict rules). Second, those tendencies derive mostly from the transitivity restrictions on binyanim 2 and 7, and the fact that binyanim 3 and 5 do not mark the non-transitive alternants. Third and finally, in the examples discussed above we often see a semantic freedom in the interpretation of each alternant, which goes beyond the event alternation in its pure form and is reminiscent, in fact, of MCM. I will return to these observations in Chapter Six. 5.3.3.3. Irregular alternations Finally, there is a small group of roots that participate in what I term “irregular alternations”. In such cases, the verbs created by the root do not belong to any 43

In the case of ‘meet’ there may be a semantic reason for the lack of passivization: the verb is reciprocal, so there is no apparent semantic distinction between the subject and the object. Semantically, the object is not an undergoer, but is more like an external argument. 44 Unlike reflexives, which strictly appear in pattern 7 (pace Doron 1999, who takes nidxaf (push into) and nidxaq (thrust in, shove in) in pattern 2 to be reflexives), verbs with reciprocal meanings may be found in several binyanim. Hagit Borer (p.c.) observes nilxam (fight with each other, P2), rav (fight with each other, P1), hiˇsˇlim (make peace with each other, P5). These verbs are different from nifgaˇ (meet-reciprocal) in that they do not have any transitive verb corresponding to them. The reciprocality comes from the root—and the verb it creates—rather than from a syntactic process of reflexivization or reciprocalization.


IN

H EBREW

141

of the argument structure alternations above. Such cases are rare, as is evident from table 5.7: Table 5.7. Irregular alternations

Airr

1–2

1–3

1–5

1–7

1–2–5

1–3–7

2–3–4

2–5 3–7

12

1

2

6

1

1

1

1

11

Since these alternations are irregular, it is difficult to discuss them in a general manner. I examine each group separately, noting the types of alternations it has. The largest group, with 12 roots, appear in patterns 1 and 2. These roots can be divided into several sub-groups. The first group contains verbs that are unaccusatives in pattern 2, and transitives (not necessarily causatives) in pattern 1. These roots undergo alternations such as yalad (give birth, P1) and nolad (be born, P2); gamal g (wean, P1) and nigmal (the same event, from the baby’s point of view, P2); hayah (be exist, P1) and nihya (become, P2); and g gamar (finish, P1) and nigmar (terminate, run out, P2). The second group contains verbs of motion through contact which are transitive in pattern 1, and in pattern 2 could be interpreted as either passives of pattern 1 or as intransitives: daxas (thrust in, shove in, P1) and nidxas (squeeze in P2); daxaf (push, P1) and nidxaf (push into, P2); daxaq (shove P1) and nidxaq (squeeze in, P2); and laxac (press, pressurize P1) and nilxac (press against, or become pressed, P2). The third group includes verbs of contact. Here, there is a difference in the degree of force exercised, or in the intentionality of using it. Compare, for example, xavat (hit, as in JJohn hit the wall, P1) and nexbat (hit, as in The ball hit the wall, P2); ʔaxaz (hold tightly, P1) and neʔexaz (cling √ to, P2); and safax (absorb, P1) and nispax (get accreted, P2). Finally, one root, yʕ c, creates yaʕac (give advice, with sentential complement) in P1 and noʕac (seek someone’s advice, with a prepositional complement) in P2. An almost equally large group of roots alternates in patterns 3 and 7. Here it is difficult to find sub-groups. Some negative generalizations hold: the verbs in pattern 7 are never transitives, the verbs in pattern 3 are not causative, and finally, the verbs in pattern 7 are often stative. Otherwise, the alternations are quite free. Consider the following verbs:ʔixer (be late, said of a person, P3) and hitʕaxer (be late, said of a time or an event, P7); siyeʕa (assist, P3) and histayeʕa (become possible, P7); simeˇ ˇ (use, as in The pen uses for writing, P7) and hiˇtameˇ iˇ (use, as in JJohn used the pen); nixem (console, P3) and hitnaxem (find comfort in something, P7); nibeʔ (prophesize, with sentential complement, P3) and hitnabeʔ (be occupied in making prophecies, no sentential complement, P7); xinex (educate, P3) and hitxanex (study at an institution, P7);

142

C HAPTER 5

biyeˇ (shame, P3) and hitbayeˇ (be ashamed, P7); berex (bless, give a blessing, P3) and hitbarex (be blessed or equipped with. . . , P7); zigzeg (to zigzag, P3) and hizdagzeg (be in the form of zigzag, P7); and nifnef (wave, as in JJohn waved his hand, P3) and hitnafnef (wave, as in The flag waved, P7). We noted above that the relation between certain pairs of causative and noncausative verb is not completely transparent, and is reminiscent, of some cases, of MCM. Irregular alternations are even more reminiscent of MCM, and, in ffact, in some cases it is very difficult to decide whether two verbs are an instance of an irregular alternation or of MCM. I used here the same entailment criterion applied in Chapter Three, to tease apart alternations, as irregular as they are, and MCM. In Chapter Six, I will return to the proximity between alternations and MCM. The other pattern√combinations involved in this group contain few roots each. Consider, first, ʔ bd, which appears in patterns 1 (ʔavad d) and 3 (ʔibed d): (17) a. b.

Ha sefer the book Dani Dani

ʔ avad

(le Dani) was lost (to Dani) ʔ ibed et ha sefer lost OM the book.

In spite of the English glosses, the form in P1 is not passive, but unaccusative. The relation between ʔavad (become lost) and ʔibed (lose) is subtle, and is clearly manifested by the different argument structures: the lost object, which is the subject of ʔavad, is the object of ʔibed. Yet Y this alternation is clearly not a causative alternation: the√transitive form in pattern 3 is not causative. √ Two roots, ndp and ntp, alternate between patterns 1 and 5, creating verbs of liquid and smell emission: (18) a.

b.

Reax dagim nadaf me ha sir. Odor fish smelled from the pot. ‘There was fish odor from the pot’. Ha sir hidif reax dagim. The pot smelled odor fish. ‘The pot smelled of fish odor’.

This alternation involves the source of emission and the emitted thing, as expressed in the different argument structures. Note that this alternation does not involve event structure—both verbs are statives—nor does it affect the valency of any of the alternants. This alternation is reminiscent of swarm alternation in English (interestingly, English, too, has very few verbs that participate in this alternation: swarm, teem, throng, bustle—see Levin 1993).


IN

H EBREW

143

√ Consider, next, the alternations between patterns 1 and 7. One of these roots, xˇsq, ˇ creates an alternation between a nominative subject with a preposition ( saq (xa ˇ be, to desire) and a dative subject with a nominal or sentential complement (hitxaseq ˇ lo, he felt like. . . ). As in English, the first one is bookish, while √ the second is colloquial.45 Another root, nˇsm ˇ alternates between an activity in pattern 1 (naˇam, breathe—including taking a single breath) and an intensive activity in pattern 7 (hitnaˇem, breathe heavily). Two roots acquire a sense of √ activity in pattern 7: lxˇs creates a verb of communication in pattern 1, taking a sentential complement (laxaˇ, whisper), w and a verb of manner of speaking in pattern 7 (hitlaxeˇ, whisper w to each√ other, tell secrets), which does not allow a sentential complement. Similarly, dwn creates dan, discuss (a problem, an issue, etc.) in pattern 1, and hidayen (discuss for hours, subject matter √ no√ √ of discussion specified) in pattern 7. Finally, √ three roots, hlk, rwc, and ʕ wf, create motion verbs in both patterns ( ʔ wp appears in pattern 3 as well). The verbs in pattern 1 may be interpreted as either verbs of directed motion, when they take a directional PP indicating the goal of the motion, or as verbs referring to motion itself: (19) a. b.

Dani Dani Dani Dani

halax walked halax walked

/ / / /

rac ran rac ran

la xanut to the shop ba gina in the garden.

In pattern 7, on the other hand, these verbs may not take a directional PP. They refer to the manner of motion itself—walking, running, or flying, and have a strong taste of durativity, run about, or walk about: (20) Dani hithalex / hitrocec ba gina / ∗ la xanut. Dani walked about / ran about in the garden / to the shop. √ Turning to mʔ s, this root creates three psychological verbs in patterns 1, 2, and 5 (because of its highly specialized meaning, we shall not attempt to translate it precisely in transliteration): (21) a. b. c.

45

Dani maʔ as(P1) ba daysa. “Dani renounced the porridge.” ha daysa nimʔ asa (P2) ʕ al Dani. “Dani is sick of the porridge.” Ha tabax himʔ is (P5) et ha daysa ʕ al Dani “The cook made Dani sick of the porridge.”

The dative construction, hitxaˇseq, may be a borrowing from Russian, where the verb ‘want’, xotet’, is used with a reflexive marker—sja — and a dative subject to express the same meaning.

144

C HAPTER 5

Pattern 5 here (21c) is clearly a causative, but the relation between maʔas (P1) and nimʔas (P2) is clearly like that between love and appeal in English. In pattern 1, the experiencer (Dani) is the subject, and the subject matter of emotion (the porridge) is the object, while in pattern 2 the subject matter of emotion (the porridge) is the subject and the experiencer (Dani) is the object. √ A unique pattern of alternation is exhibited by the root ʔ lc, in patterns 2 and 3: (22) a. b.

Dani neʔelac Dani had to Ha menahel ʔ ilec The boss forced

lifroˇs. retire. et Dani OM Dani

lifroˇs. to retire.

The verb ʔilec, in pattern 3, seems like the causative of the stative verb neʔelac, in pattern 2. Note that this is the only case of an alternation between patterns 2 and 3. Still, even in such a rare case, pattern 2 hosts √ the stative verb while pattern 3 hosts the causative one. Finally, one root, sˇqp, undergoes another unique alternation in patterns 2 and 5: (23) a. d.

Ha nof niˇsˇqaf The view was visible Dani hiˇsˇqif al Dani observed on

me ha xalon. from the window. ha nof. the view.

I believe that the most important fact is that such cases of idiosyncratic alternations do exist. Although most of the alternations in the Hebrew verbal system involve causativization and reflexivization, regular processes well-attested in many languages, there are also quite a few instances of alternations that fall outside the scope of these canonical alternations. This further indicates that the Hebrew verbal system is not deterministic: otherwise, how come it has a number of irregular alternations, each shared by few—perhaps only one or two—roots?

5.3.4. MCM in the Hebrew Verbal System Table 5.8 summarizes the combinations of patterns in which multiple contextualized meaning of roots is realized.


IN

H EBREW

145

Table 5.8. MCM in the Hebrew verbal system 1–3 MCM 48

1–5

1–7

3–5

5–7

2–3

2–5

2–7

1–2

3–7

49

33

44

17

7

4

4

5

30

1–2–3 1–3–5 1–3–7 1–5–7 3–5–7 2–3–7 2–5–7 1–3– 1–2– 1–2– 5–7 3–5 3–5 MCM 1

7

8

11

12

2

2

5

1

1

The first notable fact that table 5.8 shows is that the majority of roots that are assigned MCM—241 (82%)—create only two verbs. Examples are in (24): (24) a. b. c.

Root √ √ptx √brx bˇsl ˇ

Pattern 1 patax (open) barax (flee) –

Pattern 3 piteax (develop) – bisel ˇ (cook)

Pattern 5 – hivriax (smuggle) hivˇsil ˇ (mature)

Another 43 roots create three verbs, as in the following examples: (25) a. b.

Root √ btx

Pattern 1 Pattern 3 Pattern 5 Pattern 7 batax (trust) biteax hivtiax – (insure) (promise) √ mcʔ macaʔ (find) – himciʔ hitmaceʔ (invent) (be savvy)

Few roots create four verbs: (26) a. b.

Root P1 P3 P5 √ xˇsb ˇ xaˇsˇav xiˇsev ˇ hexˇ xsiv xˇ (think) (calculate) (consider) √ yideʕa hodiʕa ydʕ yadaʕ (know) (notify) (inform)

P7 hitxaˇev (be considerate) hitvadeʕa (get acquainted)

These data are significant, because they limit the extent of MCM—and of irregularity—in the Hebrew verbal system. Only 15% of the roots in the corpus participate acquire multiple meanings, and most of them only create two verbs. Note that this fact cannot be due to the lack of available patterns (assuming, that is, some hypothetical ranking where alternation would rule out MCM). In most cases, there are quite a few patterns that are not occupied, that could, in principle, be used for MCM.

146

C HAPTER 5

Consider now the morphological manifestation of MCM. The most common patterns for MCM are 1–3, 1–5, 1–7, 3–5, 3–7, and 5–7. Some of these combinations are specific to MCM (as well as to synonyms, as we shall see in what follows) and do not mark argument structure alternations. These include 1–3, 3–5, 2–3, and 2–7 (as well as some combinations with very few patterns, such as 1–2–3). About 35% of the roots that are assigned MCM appear in these combinations unique to MCM and synonyms. In other words, most MCM roots appear in the same pattern combinations as argument structure alternations. There are four large groups that fall into this category: 1–5, 1–7, 3–7, and 5–7. And, of the total of roots that are assigned MCM, 45% appear in those patterns. I argue that this is the reason why MCM, in spite of being quantitatively considerably smaller than argument structure alternations, has a pervasive effect on the system. The learner’s task is twofold: first, find out whether two verbs derived from the same root are an instance of MCM or of a systematic alternation. Second, if this is indeed MCM, find out the two meanings associated with the root. On the other hand, had MCM and argument structure alternations occupied completely different patterns, then the system would have come closer to the deterministic model described at the beginning of this chapter, and the learner’s task would have been considerably easier. Even with this conflation between argument–structure alternation and MCM, there is quite a lot of regularity in the expression of MCM. Two patterns, 4 and 6, are completely excluded from this alternation. Pattern 2 is quite marginal. The greatest majority of roots appear in patterns 1, 3, 5, and 7, and of those, a large number appears in the combination of patterns 1 and 3 or 3 and 5—a combination of patterns that happens not to host any alternations. This tendency, to have many cases of MCM in these two non-alternating combinations, considerably simplifies the Hebrew verbal system from the point of view of the learner. 5.3.5. Synonyms in the Hebrew Verbal System Synonymous verbs formed in different patterns by the same root can be divided into three sub groups. The first group contains truly synonymous verbs. The only difference between them is register variation. Almost without exception, one verb is bookish while the other is colloquial. For example, qineʔ (pattern 3) and hitqaneʔ (patw tern 7) both mean envy, but the former is the standard colloquial verb, while the latter is only found in high literary language. Interestingly, it is not the case that within synonymous pairs, the same √ patterns are associated with the bookish or colloquial form. In√the case of qnʔ above, the bookish form is in pattern 7, but in the case of glh pattern 2 is the more bookish (niglah, appear, be


IN

H EBREW

147

discovered) and its synonym in pattern 7, hitgalah, is colloquial. Even here, synonymy is not total: for some pairs, there is a semantic narrowing in one member. For example, palat (P1) and hiflit (P5) both mean “let out something”. Hiflit, however, apart from being bookish, tends to refer to letting out sounds—a sigh, a cry, etc., while palat does not obey such restrictions. The second group involves synonymous verbs with subtle differences in their aspectual meaning or in their external argument (they thus resemble alternations). For example, neʔenax (pattern 2) and hitʔaneax (pattern 7), from √ ʔ nx, both mean “sigh”. But neʔenax usually means “let out a sigh”, while hitʔaneax means “sigh repetitively”. Note that apart from this aspectual difference, there is also a register variation between the two verbs. neʔenax is colloquial, while hitʔaneax is bookish. In other cases, the two verbs have different types √ of external arguments. Thus, cafan (pattern 1) and hicpin (pattern 5), from cpn, both mean hide. But the subject of cafan can be inanimate (e.g. ha ʕatid cofen haftaʕot, the future hides surprises), while that of hicpin must be animate and intentional, actively hiding an object. In other cases, the two verbs differ with respect to their internal arguments. Consider the following pairs: (27) a.

b.

∗ Hu nasaʔ (ʔ iˇsˇah). he married a woman “He took a wife.” Hu nisaʔ he married “He got married.”

nasaʔ (marry), in pattern 1, obligatorily takes a direct object, while nisaʔ (marry), in pattern 2, optionally takes an oblique object (recall—verbs in pattern 2 never take direct objects). The third group of synonyms has pairs of verbs with very close meanings, yet not totally synonymous. It is important that, in such cases, the border between synonymy and MCM becomes truly blurred. Such verbs √are often lexicalized as two different verbs in English. For example, the root gwr creates gar (live in a place) in pattern 1 and hitgorer (reside) in pattern 7. Both verbs obligatorily take a locative PP, but gar allows this PP to name a country, a city or a type of domicile, while hitgorer strongly prefers a type of domicile (although it may admit other types of PPs as well). In this case, too, there is a register variation between the two verbs: gar is colloquial, hitgorer is bookish. Other such verbs include hegen (defend) in pattern 5 and gonen g (protect) in pattern 3, both√from √ gnn; xalaq (share), in pattern 1, and xileq (divide), in pattern 3, from xlq; √ sˇamat (drop off), in pattern 1, and hiˇmit iˇ (omit), in pattern 5, from sˇmt.

148

C HAPTER 5

From a quantitative point of view, synonyms are the smallest group represented in the corpus, 6.8%. Their occurrences across different patterns are summarized in table 5.9: Table 5.9. Synonyms in the Hebrew verbal system 1–2 1–3 1–5 1–7 2–5 2–7 3–5 3–7 5–7 1–2–5 1–2–7 1–3–5 1–5–7 2–3–7 Syn 7

13 20 21 1

29 16 5

6

1

1

2

1

1

As with MCM, the great majority of these roots—118 (95%)—create two synonymous verbs. Very few roots (six) create three verbs.46 The most common combinations of patterns in which synonyms appear are 2–7, 1–7, and 1–5. It is significant that the most frequent combination, 2–7, is almost exclusively associated with synonyms: there are four tokens of MCM and no argument structure alternations appearing in these two patterns, as opposed to 29 tokens of synonyms. In this very limited sense, then, synonymy and MCM display a certain tendency for complementary distribution. However, other combinations, such as 1–3, 1–7, and 3–5, are often associated with MCM, or with argument structure alternations (1–5, 3–7, 5–7, and, to some extent, 1–7). Still, note that since synonyms almost invariably belong to different registers they are not present at the first stages of acquisition and do not pose a serious obstacle for the learner. Children learn qineʔ (envy, P2) long before they come across hitqaneʔ (envy, P7); the first is used in everyday speech, the second is only found in written language. 5.3.6. Summary The Hebrew verbal system exhibits both a degree of regularity and a degree of irregularity with respect to the syntax and semantics of root alternations. This is true also with respect to the morphological realization of these alternations. The regular aspects of Hebrew verb morphology are as follows: (1) The morphological form of the passive is almost without exception predictable from the morphological form of the active (and—without any exception—vice versa). √ √ √ √ These roots include psq in patterns 1–2–5, kbh in patterns 1–2–7, zlp and yʕ d in √ √ patterns 1–3–5, zhr in patterns 1–5–7 and lwh w in patterns 2–3–7.

46


IN

H EBREW

149

(2) Transitivity alternations involve, in most cases, one of the nontransitive patterns, 2 or 7.47 (3) The system tends to use different combinations of patterns for different purposes. For example, there are no argument structure alternations between patterns 3 and 5 (and practically no alternations between patterns 1 and 3). When a root appears in these patterns, the result is MCM or synonyms. The combination of patterns 2 and 7 is almost exclusively reserved for synonyms. On the other hand, as noted throughout the book, there is no exclusive mapping between the semantic and syntactic properties of verbs and their morphological realization. This holds also for roots across patterns. The same combination often represents different relations between verbs derived from the same root. There are 71 instances of the causative alternation in patterns 1 and 5. There are 49 instances of MCM in patterns 1 and 5 as well, and 20 instances of synonyms. In some cases, tendencies—no more—can be observed. There are 204 instances of the causative alternation in patterns 3 and 7. Compare this to 80 instances of a reflexive alternation and 40 instances of MCM. Finally, the same argument structure alternation may be realized in different verbal patterns. There are some reflexive verbs in pattern 2, not only in pattern 7. The causative alternation and the stative/non-stative alternation may be morphologically realized in as many as four different ways. In general, then, there is no direct mapping from syntactic alternations to groups of binyanim (that is, no rules of the form “the causative–inchoative alternation is mapped into binyanim 3–7”). It is only by looking at a large corpus of Hebrew roots that the real regular and irregular aspects of the system can be uncovered. Note, for example, that there are many more instances of the causative alternation in patterns 3 and 7 (159) than in patterns 1 and 5 (61), although, traditionally, it is pattern 5 that is dubbed “causative”. Pattern 7, traditionally known as “reflexive”, contains reflexive verbs, but these are few (74), compared with the number of inchoative verbs that appear in this pattern (159). The exhaustive, quantitative survey presented in this chapter serves both to identify the real tendencies, and to show that they are, indeed, no more than tendencies (with the exception of passivization). The data presented above are the empirical basis for a theoretical discussion of the Hebrew verbal system. In what follows, I discuss two recent analyses of the Hebrew binyan system, Aronoff (1994) and Doron (1999, 2003). These 47

The only exception to this is the causative alternation in patterns 1 and 5, as well as a small group of roots that create both a causative and an inchoative verb in pattern 5 (e.g. hilbin— whiten/become white). w

150

C HAPTER 5

theories are complementary both in their orientation and in their view of morphology. Doron aims to uncover the order in the chaos while Aronoff aims to explain the reason for the chaos; Doron regards morphology as a direct and transparent spell-out of syntactic structure, while Aronoff treats morphology as an independent module. In Sections 5.4 and 5.5, I examine each of these theories on the background of our root corpus, and see how they fare both theoretically and empirically. Finally, in Section 5.6, I discuss some recent typological work on verb alternations. I argue that while the alternations in the Hebrew verbal system conform to patterns found in many languages, there are no grounds for identifying binyanim with “basic” or “causative” alternants. 5.4. A MORPHOLOGICAL ACCOUNT OF THE HEBREW VERBAL SYSTEM: BINYANIM AS INFLECTIONAL CLASSES (ARONOFF 1994) 5.4.1. Foundations Aronoff’s (1994) discussion of the Hebrew verbal system is part of a research program that advocates an autonomous status of morphology, independent of other grammatical components such as phonology and syntax. He is therefore interested, first and foremost, in the morphological form of the binyanim. Aronoff starts from the apparent paradox, that binyanim act simultaneously as derivational and inflectional morphemes. The derivational aspect of binyanim is apparent in their gaps and the non-predictable meanings assigned to the verbs that appear in them, the inflectional aspect is their being obligatory for all verbs. Aronoff then argues that this is really not a paradox, since derivation and inflection are not two types of morphology, but rather, two uses of morphology: “inflection is the morphological realization of syntax, while derivation is the morphological realization of word-formation” (Aronoff 1994: 126). Most crucially, Aronoff assumes that the same morphology can realize both inflectional and derivational processes. The system works as follows. Abstract lexeme-formation rules assign their output verb stem a membership in some inflectional class (binyan). The inflectional class itself, that is, the binyan, determines the inflectional paradigm of that verb stem: Lexeme-formation rules √ assign membership √CCC binyan 1 √CCC binyan 3 CCC binyan 5

Binyanim determine paradigm

CaCaCti, CaCCta, CaCaC. . . CiCaCti, CiCaCta, CiCeC. . . hiCCaCti, hiCCaCta, hiCCiC. . .


IN

H EBREW

151

The binyan system, according to Aronoff, thus has a dual role: “it serves as the abstract morphological mark of the lexeme formation rule and, as an inflectional class, it dictates the phonological form of the verb.” (ibid: 127). Note that the lexeme formation rules are abstract. That is, the assignment of verbal lexemes to different inflectional classes has no direct overt phonological reflection. The only overt reflection is the full inflectional paradigm, which is determined by the binyan. Hebrew lexeme formation rules thus carry no morphology of their own, which is why the initial “paradox” regarding the binyan system arose in the first place.48 Binyanim are thus similar in all respects to inflectional classes in other languages (e.g. Latin verb conjugations). What is special about Hebrew is that the verbal inflectional class, binyan, determines not the inflectional morphology of the stem, but rather, the form of the stem itself. In languages with noun classes (e.g. Latin, Russian), the noun class determines the phonological realization of case and number features, so that two nominal stems that belong to different inflectional classes may take different suffixes to mark their dative plurals or genitive singular. In Hebrew, all verbs take similar suffixes of gender, number, and person, regardless of their binyan. It is the verb stem itself that differs according to the binyan. This is why binyanim may not seem like an obvious case of inflectional class, in addition, of course, to their being related to lexeme-formation. We can therefore represent the binyan as function. The input of this function is a root (assigned to it by lexeme-formation rules) and its output is a verb stem of a certain form, which can then be inflected according to the inflectional paradigm of Hebrew verbs: √ (28) root, lexeme-formation rules binyan inflectional paradigm Only the second effect of the binyan, assigning membership in an inflectional class, has overt morphological manifestation. Based on his analysis of Hebrew binyanim, Aronoff suggests that the binyan is indeed an arbitrary rule, or a function that maps roots into stem classes. He rejects McCarthy’s (1979) claim that the binyan has some independent morphological content, the prosodic template: although most binyanim can be equated with prosodic templates, there are two binyanim—1 (CaCaC) and 2 (niCCaC), which do not retain their prosodic templates throughout the paradigm. These w binyanim have different templates in the past tense (CVCVC for pattern 1 and CCVC for pattern 2) and in the future tense (CVCC for pattern 1, CVCVC for

48

As Aronoff shows, it is not a Hebrew specific property that lexeme-formation rules lack overt phonological manifestations. He discusses a similar case of Latin intensive verb-formation.

152

C HAPTER 5

pattern 2). Therefore, a context-sensitive mapping rule is required in order to determine the prosodic form of the binyan. Because of the cases of binyan 1 and 2, Aronoff suggests to treat binyanim in general as functions, rather than as phonological entities. Aronoff further notes that the only two binyanim where stem vocalism changes are those two binyanim where the prosodic template is changed, that is, patterns 1 and 2 (changes in vocalism in other binyanim are ascribed by Aronoff, following Prince (1975), to phonological rules operating on vowels in Hebrew). This is accounted for if we treat the prosodic template and the stem vocalism as a single unit, which Aronoff calls “stem template”. This stem template contains a prosodic template with filled vowels, which determines what possible stems are available in Hebrew. Patterns 1 and 2 thus have two w possible stem templates. This way, there is no need to treat the prosodic template and the stem vocalism separately, or to explain their covariance. Finally, note that the account above holds for patterns 1, 2, 3, 5, and 7 only. Passive patterns (numbers 4 and 6) are different in this respect. They are not independent stem templates, but rather, are derived from the active stem templates through deletion of the stem vocalism and its replacement with the stem vowels characteristic of passives—u/a.

5.4.2. Results The main advantage of Aronoff’s proposal is the elegant way in which it captures the dual role of the binyan morphology. It can therefore account both for the regular aspects of the system (which are assigned to its inflectional role) and for its irregular aspects (which are instances of word-formation). Consider the theory on the background of our findings above. The fact that there are so many gaps in the binyan system—single occurrence roots, roots appearing in two binyanim (active and passive) only, etc.—is explained, and even expected, by the fact that binyan morphology is the overt output of a lexeme-formation rule: word-formation is not completely productive, and therefore we cannot expect each root to form seven different verbs. The fact that binyanim are the overt manifestation of lexeme-formation rules also explains the highly specialized, non-transparent meanings that many roots acquire in different binyanim, and, in particular, the fact that a single root may be assigned numerous interpretations in the environment of different binyanim. Word-formation, by definition, creates words with lexicalized meanings. W The derivational nature of the binyan is also compatible with the observation arising from the corpus that the roots creating verbs of certain semantic and syntactic type (e.g. inchoatives) may select for different binyanim (e.g.


IN

H EBREW

153

inchoatives may appear in patterns 1, 2, 5, or 7). Just like English, stems may select for their adjectival suffixes (e.g. ment-al vs. glori-ous), Hebrew roots may select for their verb-forming environment (binyan). On the other hand, the fact that the binyan is an inflectional class, giving “instructions” regarding the inflectional paradigm of the verbs that fall into that class, explains why for any given verb there exists a complete paradigm with all person, number, gender, and tense forms. To summarize, Aronoff’s theory successfully accounts for some of the most important phenomena occurring in the Hebrew verbal system. The derivational aspect of binyanim explains the gaps and highly specialized meanings assigned to roots (including MCM). The inflectional aspect explains the obligatoriness of the binyan. Since binyanim are obligatory, then both coinage of new verbs and argument structure alternation of verbs should occur within the binyan system. The regularity and irregularity attested by our corpus are thus predicted. So much of Aronoff ’s account ought to be preserved by any theory of the Hebrew verbal system.

5.4.3. Problems In Aronoff ’s system, binyanim serve simultaneously for word-formation and for assigning inflectional classes to verbs. Both word-formation and inflection are, under this analysis, morphological processes. This is consistent with Aronoff ’s treatment of binyanim as purely morphological devices, which is part of the overall theoretical perspective he adopts, “Morphology by itself ”. The questions I raise regarding Aronoff’s theory are the following. First, by treating morphology by itself, how many empirical phenomena are left unexplained just because they fall out of the scope of pure morphology? Second, are binyanim indeed instances of word-formation, or is there a syntactic component to verb-formation that is ignored? Finally, suppose we limit ourselves to the purely morphological aspect of the binyan system: can we really explain binyan morphology by itself? Consider the first question. While there is a lot to be said about the purely classificatory properties of the binyan system, it is also evident that binyanim interact with the syntax in an important way that cannot be ignored. Any theory has to capture the fact that binyanim 2 and 7 do not mark transitive verbs. Furthermore, it also has to capture the fact that transitivity alternations involve (with the exception of the pair 1–5) one of the non-transitive patterns, 2 or 7. This fact is significant, yet falls outside the scope of Aronoff’s theory: there must be, pace Aronoff, some mapping rule from syntax to morphology that should prohibit verbs with direct objects from being mapped into these binyanim.

154

C HAPTER 5

Similarly, the important generalizations about the relations between the binyanim are problematic for Aronoff. As noted above, although roots may, to some extent, select for their patterns, there is a systematic relation between certain patterns, so that argument structure alternations tend to appear in certain pairs or triplets. Thus, the causative–inchoative alternation appears in patterns 1 and 5, 1 and 2, 2 and 5, 5 and 7, or 3 and 7. Furthermore, as noted in Chapter Four, the relation between pairs of binyanim observes strict directionality. It is always the causative verb that appears in pattern 5, and the inchoative in pattern 1—never the other way around. These generalizations add a great deal of systematicity to the binyan system, but Aronoff ’s theory is not designed to deal with them. If we treat morphology by itself, the interaction between syntactic structures and binyan morphology remains arbitrary. On the other hand, if we take morphology to be a reflection of syntactic processes, the relation between binyanim and the syntactic properties of verbs appearing in them is not only not accidental, but expected. The second question I raised is related to an ongoing theoretical debate: are verbs lexical atoms, or are they produced by the syntax? The approach adopted in this book does not distinguish between “word-formation” and syntactic computation, so that all combination of roots into verbs, phrases, and sentences, is performed by the syntax. Aronoff assumes that all verbs fall under the domain of “word-formation”. In fact, most current theories assume that processes such as causative and reflexive formation are performed by the syntax.49 The only case Aronoff takes to be a syntactic process is passives, which, according to him, are created on the basis of active binyanim. The discussion of passives brings us to the last issue. Aronoff is not interested in the morpho-syntactic aspect of the binyan system, and aims to treat only the purely morphological aspects of the system. Ignoring, then, the interaction between the binyan system and the syntax, and concentrating solely on the morphology of binyanim, the question is, does Aronoff’s theory offer an adequate account of these morphological phenomena? In other words, can morphology be treated by itself? I argue that at least as far as the binyan morphology is concerned, we run into some empirical problems if we treat morphology by itself. Let us look at Aronoff’s treatment of the morphology of passives. Aronoff agrees that passives are the output of syntactic rules. Regarding the actual syntactic process involved in passive formation, he claims that “For my purposes, the question is just not interesting” (ibid: 134). Since syntax and morphology are autonomous, passive morphology can be treated as a purely morphological phenomenon.

49

cf. Borer (1998), Hale and Keyser (1998) for causativization.


IN

H EBREW

155

However, Marantz (2000) notes the following fact. Based on an observation made in Arad (1999a), he shows that the choice of binyan for causative, inchoative, or other types of verbs is, to a certain extent, up to the root (cf. Chapter Four). Inchoative verbs can thus appear in pattern 1, 2, 5, or 7: (29)

Root √ √qpʔ √mss √chb xmm

a. b. c. d.

Pattern 1 2 5 7

Inchoative verb qafaʔ (freeze) namas (melt) hichiv (become yellow) hitxamem (become hot)

The interesting fact is that in passives, no such choice is available. Passive morphology always depends on the active morphology: (30) Active 1 2 5

Passive 2 4 6

Here, there is no room for idiosyncracies of the root. No root may decide to form an active verb in pattern 3 and a passive in pattern 6. In Aronoff’s system, where morphology is autonomous, this property of passives remains arbitrary. However, if we take into consideration the syntactic properties of passives, which distinguish them from other verbs, their behavior is easily explained. Marantz ascribes this difference to the structural difference between passives and other verbs. In causatives, inchoatives, and other types of active verbs, the root in a local relation with the verbal head that merges with it: (31)

vcausative v

vinchoative √CCC

v

√CCC

In this case, the roots are in a local configuration with the verbal head that merges with them. Anticipating the discussion in Chapter Six, note that the roots may locally select for their verbal head, or binyan. The fact that causative and inchoative verbs may select for different patterns, up to the root, is not surprising. Now consider passives. Following Kratzer (1996), Marantz assumes that passives involve a Voice head, which is responsible for the projection (or lack

156

C HAPTER 5

thereof) of the external argument, and which is located above the head v, which verbalizes the root: (32)

voiceP

voice (passive)

vP v

√root

The voice head involved in passive formation merges with the vP; there is no local relation between the voice head and the root. The choice for binyan has already been made when the root has merged with the v head. Voice head merges with a vP, namely, with something whose morphological form has already been determined. The voice head has no access to the root, and therefore the root cannot decide to select a different passive morphology. Rather, the morphology of passives involves a modification of the already existing active morphology. If we take into consideration the syntactic structure of passive, it is predicted that passive morphology will be regular, while the morphology of non-passive verbs will be, to some extent, dependent on idiosyncratic properties of the root. Recall that Aronoff assumes that the prosodic template and the stem vocalism form a single unit, which he calls the stem template. The only case where vocalism is not part of the stem template is passives. Here, Aronoff argues that the stem vowels are deleted and replaced by the vowels representative of passives. Under a purely morphological treatment, the fact that vowel deletion occurs only in passive patterns is at best accidental. On the other hand, this fact is immediately derived if morphology is taken to as the reflection of syntactic structures, and if we take into account the syntactic structures of different verbs. An additional issue regarding the theoretical foundations of Aronoff is the status of inflection. Aronoff uses the hypothesis that binyanim are inflectional classes to explain the obligatoriness of binyanim for all Hebrew verbs. In Chapter Two, I questioned the validity of this explanation. It was shown that the obligatoriness of inflection predicts that there should be non-inflected Hebrew verbs that cannot be assigned to any inflectional class, just like there are non-inflected Russian nouns that cannot be assigned to any nominal inflectional class. The ffact is that there are no such verbs. In order to account for the Hebrew data, we need extra assumptions about the nature of inflection, which do not work for Russian, thus considerably weakening the validity of the explanation. Finally, it was pointed out that the inflection account creates a false asymmetry between Russian and Hebrew, while failing to expose the true asymmetry between nouns and verbs, an asymmetry that both Russian and Hebrew obey. I


IN

H EBREW

157

therefore believe that although Aronoff’s intuition regarding the dual role of the binyan system is correct, his account cannot stand as is does, stated purely in terms of morphology. A morpho-syntactic account, incorporating the valid part of Aronoff (1994), is offered in Chapter Six.

5.5. A SYNTACTIC ACCOUNT OF THE HEBREW VERBAL SYSTEM: BINYANIM AS A SPELL-OUT OF SYNTACTIC HEADS (DORON 1999, 2003) 5.5.1. Foundations Doron’s (1999, 2003) account of the Hebrew binyan system is almost the opposite of Aronoff ’s. The difference is manifest, first of all, in the different question asked by each of these authors. While Aronoff concentrates on the morphological form of binyanim, Doron is interested in how the binyan morphology reflects different syntactic structures. The difference is also evident in the general approach. While Aronoff proposes a reason why the binyan system is non-deterministic, Doron aims to get as close as possible to a deterministic system, where the relations between the appearances of a root in different binyanim are systematic. Doron adopts the Root Hypothesis. In cases of verb alternations, she assumes that both verbs are derived from the same root, rather than one verb being derived from the other. A verb is thus a complex entity, made from a root combined with different semantic–syntactic heads, while binyan morphology is a reflection of such verbal heads. Doron’s main claim regards the syntactic information most saliently contained in the verbal system. She argues that the binyan system encodes two aspects of verbs’ meaning, voice and agency. The agency dimension is related to the thematic role of the verb’s external argument. According to Doron, there are three possibilities for this role: agent, cause, or unmarked. These options correspond to the three binyanim that express active voice: (33) The active voice a. The simple template b. The intensive template c. The causative template

CaCaC CiCCeC hiCCiC

(Doron 2003: 17)

Pattern 1 (CaCaC) is associated with an unmarked external argument. Pattern 3 (CiCCeC) is associated with an external argument which is an actor. This interpretation is given by the functional head ι, which w combines with the root and lends the event an intensive interpretation. Pattern 5 (hiCCiC) is associated

158

C HAPTER 5

with an external argument which is a cause, introduced by the functional head γ . This head may attach above the root or above little v. The external argument it introduces is a cause.50 If the verb derivation contains no extra head, the external argument is unmarked, and the verb appears in pattern 1: v

(34) y

v √CCC

v unmarked

(35)

v y

v ι

v ι

intensive

√CCC

γ

(36)

γ

z γ

v v

y v

√CCC

causative

The functional heads ι and γ thus modify the basic event denoted by the root. ι directly combines with the root, while γ combines with the root and its arguments (this, Doron argues, explains the fact that verbs in pattern 3 tend to have a more idiosyncratic interpretation than those in pattern 5: the closer to the root the functional head attaches, the more idiosyncratic the meaning of the verb may be). In both the cases, the external argument is interpreted according to the denotation of the root plus the modifying head. 50

The definition of an actor remains somewhat vague throughout the discussion, mostly negatively defined: Doron claims that it cannot be equated with agency or animacy.


IN

H EBREW

159

To support this claim, Doron argues that whenever a single root appears in two or three of the active patterns, the result corresponds to different perspectives of the same event. In one, the external argument is unmarked with respect to agentivity, in one it is an agent or actor and in another—a causer: (37)

Root √ √ycr √qcc ˇ √sbr y √ ʕc √cph rqd

Simple (P1) yacar (create) qacac (cut, chop) savar ˇ (break) yaʕ ac (advise) cafah (watch) raqad (dance)

Intensive (P3) yicer (manufacture) qicec (cut, chop—intensive) siber ˇ (actively break) yiʕ ec (advise) cipah (expect) riqed (dance)

(38)

Root √ √qpʔ √gdl √rtx √rqd ʕ bd

Simple (P1) qafaʔ (freeze—intr.) gadal (grow—intr.) ratax (boil—intr.) raqad (dance) ʕ avad (work)

Causative (P5) hiqpiʔ (freeze—tr.) higdil (increase) hirtiax (boil—tr.) hirqid (make-dance) heʕ evid (employ, make work)

The verbs in pattern 3 have an external argument interpreted as an actor, while the external argument of the verbs in pattern 1 is unmarked. For example, yacar, in pattern 1, means “create”. Such creation may be spontaneous or accidental. On the other hand, yicer, in pattern 3, means “manufacture” and entails that the external argument is directly involved in the creation. Similarly, the verbs in pattern 5 are causative, and their external argument is interpreted as a cause. This is particularly evident if we look at the pairs of verbs in patterns 1 and 5 above. Raqad (pattern 1) means “dance”, while hirqid (pattern 5) means “make someone dance”. It is important to note that Doron is only committed to the interpretation of the external argument of each verb in the alternation. She does not argue that the meanings of the intensive and causative verb should be transparently related to that of the unmarked one, nor that the unmarked case itself should be transparent. Now consider the other dimension of verb meaning encoded in the binyan system, voice. The voice dimension has two parts: passive and middle. Both involve a modification of the argument structure of the verb, particularly the external argument. Passive voice is derived by a head π that modifies the syntactic active head. This correctly captures the fact that passives always have a corresponding active verb, to which they are tied syntactically and semantically. A further advantage of this, according to Doron, is that we derive the absence of a pattern which uniquely marks the passive of pattern 1 (or, as Doron puts it, we derive the lack

160

C HAPTER 5

of a passive template for binyan 1). Since the simple template (pattern 1) does not have a syntactic head, but is the phonological realization of a lack of such head, no head π may modify it.51 An important observation made by Doron is that passive verbs require a more agentive external argument than their active counterparts (this will be illustrated in Section 5.5.2). She suggests that this observation is theoretically captured if we assume that π modifies the eventuality by lexically determining the external argument as an actor. That is, π adds an extra requirement on the external argument of the verb.52 The highest v head is missing in passives, so that the external argument is not projected syntactically. The middle voice is morphologically realized by patterns 2 and 7, corresponding to patterns 1 and 3 (Doron 2003: 55): (39) a. b. c.

The active voice The simple template The intensive template The causative template

The middle voice CaCaC niCCaC CiCeC hitCaCeC hiCCiC –

Unlike passives, which are only created when a correlating active verb exists, middles may be formed independently of transitive verbs. The examples cited by Doron are hitʕalef (faint), hiˇta ˇ ʕel (cough), hitvakeax (argue), hitxaret (regret), and hitʕaqeˇ (insist)—all of which appear in pattern 7 and lack a transitive counterpart. Doron does not discuss the semantics of these verbs (they are considered as middles, in fact, only by virtue of the fact that they appear in the pattern labeled as “middle”). I return to this in Section 5.5.3. Semantically, middle formation involves a modification of the external argument, and a derivation of an intransitive verb. As noted above, this verb may be derived from the root r , not necessarily from a transitive verb. Syntactically, middles involve a head µ. This head may either modify the root (if no agentivity head ι or γ is present), as in the case of pattern 2, or it may modify the active head ι, in w which case the verb appears in pattern 7 (Doron 2003: 53 59): 51

This argument confuses synchronic and diachronic information. As noted above (and as is acknowledged by Doron), pattern 1 had a corresponding passive pattern at an earlier stage of the language. It also ignores the fact that syntactically, verbs appearing in pattern 1 do have a passive form, in pattern 2. 52 In particular, Doron notes that while causatives often allow non-agentive causers, their passive forms may require a volitional, intentional, animate external argument. 53 The fact that µ may modify the root (or the root plus the intensive head ι) but not the root together with its arguments, derives, according to Doron, the fact that causative verbs do not have a middle voice, since the head γ always attaches to the root together with its arguments.

R OOTS A CROSS P ATTERNS (40)

µ µ

H EBREW

161

µ µ

R x

IN

˘ √sbr

ι ι

R x

/nisbar/(break-intras., P2) ˘

√prq

/hitpareq/(fall apart, P7)

Note that Doron considers verbs in pattern 2 as middles although, as noted above, they are similar to passives in every way (the only difference from “true” passives is that pattern 2 is not exclusively used for passive verbs). This is necessary in view of the hypothesis that all binyanim are direct reflections of functional heads. A further result of the middle head µ is an explanation of the morphological similarity between middles and reflexives. Both, according to Doron, are formed by the head µ. The difference has to do with the place of attachment of this head. When µ is merged as a modifier of the root phrase or the head ι, the result is a middle verb, whose external argument is saturated. When it is projected as a sister of the root, the argument of the root remains free, and is bound by the verbal head that is merged with the predicate. This argument is bound by the verb, yielding a reflexive reading. This way, Doron syntactically captures Reinhart’s (1996) observation regarding the similarity between middle and reflexive verbs. Unlike the passive head π , w which only modifies external arguments, µ can modify either external or internal arguments, resulting in the formation of middles or of reflexives. 5.5.2. Results Doron’s system proposes a direct correspondence between syntax and morphology. Such an account, if it can be empirically maintained, has a great advantage from the acquisition standpoint: children can immediately “read” the structure of the verb off its morphological form. Although binyanim are equated with functional heads, Doron does not argue that the meaning of the root in a given binyan is predictable. A verb appearing in a given binyan contains a certain predictable meaning component (e.g. intensive), but we cannot yet predict its actual meaning. The system is thus not transparent (for the appearance of each root in a binyan), but merely systematic. Crucially, this systematicity is seen only in alternations. When a root appears

162

C HAPTER 5

in the simple binyan and in the intensive binyan, the intensive verb (e.g. yicer, manufacture) has an action component that the simple verb (e.g. yacar, produce) lacks.54 This enables Doron to capture the regularity that holds across the pattern system while not committing herself to a fully deterministic theory where the contribution of the pattern for each root is transparent. w Doron’s biggest contribution concerns such generalizations about relations between binyanim. The most important one, which has been already mentioned several times in this book, concerns the strict directionality of binyanim in marking argument structure alternations. That is, if a non-causative verb appears in binyan 1, the causative verb appears in binyan 5, and never the other way around: (41) a. b. c. d. e. f. g. h.

Inchoative 1 2 2 7 ∗ 5 ∗ 5 ∗ 5 ∗ 3

Causative 5 5 1 3 ∗ 1 ∗ 2 ∗ 2 ∗ 7

This directionality holds also for the transitive–reflexive alternation (the reflexive alternant is always in pattern 7, or in some rare cases in pattern 2, and the transitive alternant is in most cases in pattern 3, and, more rarely, in pattern 1 or 5). Passive forms also obey strict regularity, as noted above. Doron’s generalization exposes the intricate relations between the patterns and adds a great regularity to the pattern system. Under Doron’s system, this generalization is derived immediately from the structural make-up of Hebrew verbs and from the direct relation between morphological form (binyan) and syntactic structure. Recall that binyanim 3 and 5 reflect the presence of an intensive and a causative heads, respectively, while binyan 1 is the default form, where no functional head is present. Therefore, it is always the case that the verb in binyan 5 will be a causative one while a verb derived from the same root in binyan 1 will be non-causative. This is the way a the mapping from syntax to morphology goes. There is no way of getting a causative or an intensive form in binyan 1, because the relevant functional 54

For this particular pair, Doron points out subtle differences in interpretation related to this action component. Within the context “ha xelbon . . . nogdanim” (the protein . . . antibodies), yacar (create) entails that the protein may have participated in some indirect way in the creation of antibodies, while yicer (manufacture) entails that the protein was directly and actively involved in their creation.


IN

H EBREW

163

head is not present. In other words, Doron offers a direct, nearly one-to-one mapping from syntactic structures into morphological templates:55 No functional head present 1 ι head present 3 γ head present 5 ι plus µ present 7 (middle) µ present, attaching to root 7 (reflexive) or 2 (otherwise). ι plus π present 4 γ plus π present 6

(42)

In addition, Doron makes a number of finer-grained observations about the Hebrew verbal system. One such important observation concerns passive verbs in Hebrew: the external argument of passives must be an actor, even if that of the correlating active verb is not an actor. When a causer may be an external argument of an active verb, it may not appear in the by-phrase of the passive verb. (43) a.

b.

ha agronomit / eyxut ha qarqa the agronomist / quality the soil higdila et ha yevul increased OM the crop ‘The agronomist / the quality of the soil increased the crops’ Ha yevul hugdal The crop icreased—pass. al yedey ha agronomit / ??al yedey eyxut ha qara by the agronomist / ??by the quality of the soil

Doron assumes that the highest v node is missing in passive sentences. The relation usually denoted by v (regarding an external argument) is introduced by the passive morphology itself. She further assumes that the passive morpheme forces the role of an actor. For that reason, passive phrases with causers (nonagentive external arguments) turn out to be incoherent. Another fine-grained distinction concerns verbs that seem synonymous, such as heʔedim (redden, binyan 5) and hitʔadem (redden, binyan 7). It has been noted before (Borer 1991) that pattern 5 is ambiguous between a causative and an inchoative reading (just like redden r in English is). Doron notes that the form in binyan 7 corresponds to an intrinsic change caused by the person who reddens, while the form in binyan 5 is compatible with an interpretation of an 55

These mapping rules work, recall, only for those roots that alternate in several binyanim (less than half the roots in our corpus, if we omit Single Occurrence roots and passive only roots).

164

C HAPTER 5

external causer that causes the reddening. A particular context is required to highlight this subtle difference: (44) a. b.

Every time I look at you through these glasses. . . ata maʔ adim you redden ??ata mitʔ adem you redden

Doron suggests that this contrast is explained if we take all verbs in binyan 7 to contain a middle head µ (which, recall, is responsible also for the reflexive reading, as noted above), so that the causer and the theme are identified. In binyan 5, on the other hand, the functional head γ , w which introduces a causer. This causer is external to the basic event denoted by the root, and cannot be identified with the theme. To summarize: Doron offers the simplest, and therefore most theoretically desirable, mapping between syntax and morphology. This direct correlation between structure and form also entails the generalizations regarding the directionality in the marking of verb alternations, as well as derives some finergrained observations regarding the meaning of verbs in different binyanim. 5.5.3. Problems The basic tenet of Doron’s account, the direct relation between syntax and morphology, is also what makes her account problematic in several respects. Most problems result from two complementary aspects of the theory. On the one hand, the theory makes very strong predictions, which are not corroborated by the empirical facts. On the other hand, the theory proves too restrictive in its tools to accommodate the full spectrum of phenomena occurring in Hebrew. There are two ways to unpack Doron’s theory: strong and weak. On the strong reading, Doron argues that there is a tight correlation between binyanim and ‘intensive’ and ‘causative’ understood in their usual semantic sense. On the weak reading, Doron does not intend her ‘intensive’ and ‘causative’ to have a fully fledged semantic significance, but instead merely means by them a certain kind of syntactic structure (to be specified below). It appears that Doron herself intends the weak reading (the empirical inadequacy of the strong reading is very obvious indeed). Further, if the weak reading fails, so a fortiori does the strong one. Still, I find it useful to discuss both readings so as to get a fuller sense of the empirical data, in anticipation of my own discussion in Chapter Six.


IN

H EBREW

165

5.5.3.1. Doron’s theory: the strong reading Let us address the first problem, namely, the very strong predictions Doron’s theory makes regarding the correlation between syntax and binyan morphology, taking ‘causative’ and ‘intensive’ to have their usual meaning. In reality, the relation between the two is typically not so direct. As a consequence, Doron is often obliged to force the meaning of verbs so that they include such component as indicated by the binyan in which they appear. This is evident in particular if we examine pairs or triplets of verbs, derived from the same root, in patterns 1–3–5. The main prediction Doron makes is that for each triplet, the verb in binyan 1 will be unmarked with respect to its external argument, the verb in binyan 5 will be causative, and that in binyan 3—an actor. This is very difficult to argue for these cases, in particular for binyan 3: (45) a. b. c.

Root Binyan 1 √ √tʔ m taʔ am (match) kpp kafaf (bend) √

stq ˇ

Binyan 3 Binyan 5 teʔ em (coordinate) hitʔ im (match) kofef (bend) hixpif (subordinate) sataq ˇ (be silent) siteq ˇ (paralyze) hiˇstiq ˇ (silence)

taʔam (P1) is a stative verb (match, as in “The jacket matches the shirt”). teʔem (P3) means “coordinate”. hitʔim (P5) can be either a stative verb, synonymous with taʔam (P1), or a non-stative verb, meaning “match things together”. Now, it is not clear in what way the external argument of teʔem (P3) is an actor and that of hitʔim (P5) is not. Both verbs require an actor (on the non-stative reading of hitʕim). Both verbs are also causatives: there √is no way in which hitʔim is more causative than teʕem. With W the second root, kpp, a similar problem arises. The verb in binyan 1, kafaf, f is highly bookish, practically synonymous with the more colloquial verb in binyan 3, kofef (bend). In binyan 5, the root creates a verb with a specialized meaning, make someone or something subordinate to someone. This verb is causative indeed, but so is the verb kofef (bend—transitive) in pattern 3. There is no principled reason to assume that bend, w which is included in causative alternations in many languages, is intensive rather than causative in Hebrew, only by virtue of the fact that it appears in a binyan that has been dubbed “intensive”. Hiˇsˇtiq, (silence, P5) has, indeed, an external √ argument which is a cause. But if we look at the triplet of verbs derived from sˇtq, then sˇataq (be silent, P1) requires a volitional, sentient actor, while sˇiteq (paralyze, P3) clearly entails a pure causer, not an actor (a stroke or fear can paralyze). There is no way a in which w paralyzing is more of an action than silencing. The alternation Doron calls the “intensive alternation” in binyan 3 has another problem associated with it. Many of the verbs Doron uses to illustrate this alternation are highly bookish and marginal, and are not actually a part of the knowledge of native speakers. Thus, speakers come across forms such

166

C HAPTER 5

as siber ˇ (break, intensive of sˇavar, break) or riqed (dance, intensive of raqad r , dance) in highly specialized, bookish contexts (and many speakers do not know them at all). In other cases, it is the non-intensive form that is bookish, such as xavaq (hold, hug, P1), corresponding to colloquial xibeq (hug, P3). Forms such as xavaq do not belong, in fact, to the lexicon of today’s Modern Hebrew speakers, and it is very doubtful whether speakers have intuitions as to whether xibeq is more intensive than xavaq. This marginality of many of the verbs in P3 makes it difficult to evaluate their external arguments in relation to the external arguments of the corresponding verbs in P1. Other verbs that supposedly fall into this alternation seem perfectly synonymous. For example, yaʕac (P1) and yiʕec (P3) both mean “advise”. There are no extra requirements on the external argument of the verb in pattern 3. We noted above that the external argument of the verb in pattern 3 is not necessarily an actor. In fact, in some cases this is the other way around. Samax (rejoice) and sˇataq (be silent, not speak), both in pattern 1, require an animate external argument. The verb in pattern 3 that is derived from the same root may take a non-animate causer: simeax (make happy) and sˇiteq (paralyze). Only animate entities can be happy or not speak, but causes and stimuli may trigger happiness (as in ha xadaˇot ˇ simxu oto, the news made him happy) or make someone paralyzed (as in ha paxad shiteq oto, fear paralyzed him). To sum up for pattern 3: Doron’s claims, that verbs in P3 contrast with their counterparts in P1 and P5 in that they demand an actor, is valid for only a relatively small set of genuine examples: about 15 or so roots have a stative interpretation in P1 and a non-stative (possibly agentive) interpretation in P3. Examples are: (46)

Root √ ˇ √xˇsb √ydʕ √rzh √xsr yˇsb ˇ

P1 think know want lack sit

P3 calculate notify placate subtract colonize

It is striking, however, that practically all these roots are also assigned highly specialized meaning as an MCM (and that most of the roots alternating between these two patterns—168 in all—display MCM alone). In other words, if indeed there is a P1/P3 alternation of the stative/agentive kind, than we should also explain why it differs from other alternations such as those involving, for example, causation. I shall return to this in Chapter Six.56 56

It sometimes appears as if Doron is making an even stronger statement: that verbs in P3 are more intensive (in some sense) than those of patterns 1 or 5. This claim however is simply very


IN

H EBREW

167

5.5.3.2. Doron’s theory: the weak reading As regards causative verbs, Doron’s predictions (taken on the strong reading) are clearly not borne out. Of course, the simplest way to set out Doron’s predictions—that all causative verbs in Hebrew are in pattern 5—is obviously at odds with the data from my corpus study. There are Hebrew causatives in patterns 1, 3, and 5, with pattern 3 hosting the largest number of causative verbs. A more likely way to accommodate Doron’s account of Hebrew causativization, then, is a weaker one. This would be a prediction regarding the properties of the external argument of Hebrew causatives. The appearance of causatives in different binyanim would then specify different types of causation, in particular the relation between the external argument that brings about the causation and the caused event. These would be unmarked external arguments in pattern 1, actors (involving volition or action) in pattern 3, and pure causes in pattern 5. In other words, we would then predict pattern 5 to be the least specified for intentionality and action, and pattern 3—the most. This then seems to be Doron’s prediction—which, once again, is simply not borne out by the facts. In what follows, I concentrate on the empirical and theoretical failings of Doron’s w theory on its weak reading. I have analyzed Hebrew causatives in both binyanim 3 and 5 (binyan 1 can be ignored in this context, as Doron’s theory does not make any specific predictions about it). In both, the number of causatives whose external argument can be characterized as actor (in terms of intentionality and action) is limited. However, more are so marked, proportionally, in binyan 5 than in binyan 3, directly contradicting Doron’s prediction. Over a third of the causatives in binyan 5 are marked as actors, while only about a quarter of the causatives in binyan 3 are marked in this way.57 Among the many examples of causes in binyan 3, we may mention yiʔeˇ (make despair, almost invariably said of depressing

tenuous. At best, there are a handful of such roots alternating between P1 and P3, where most speakers would not be able to identify the intensive meaning, and indeed almost invariably only √ √ √ √ one form of the verb is in real use in Modern Hebrew. (Examples are sˇbr, rqd, xbq, qlp, and perhaps a few more—certainly not than a dozen in the most permissive reading, out of 893 tokens of P3 in the system as a whole!) I shall return in Chapter Six to mention possible sources for this highly marginal residue of the intensive in P3, but one reason may be mentioned immediately: as I note in Section 5.6, there is apparently some tendency, cross-linguistically, for speakers to associate forms that are semantically more marked with the more marked verbal phonological form. This then would be more than enough for accounting for the slight tendency of Hebrew speakers to associate the intensive action with the geminated verb form. 57 It is very difficult to determine a precise quantitative distribution, as Doron’s characterization of causes and actors, perhaps inevitably, remains imprecise. By my estimate, there are about 50 actors in binyan 5 (out of 159 causatives), and, similarly, about 50 actors in binyan 3 (out of 206 causatives).

168

C HAPTER 5

circumstances), ʕiyef (tire, almost invariably said of external conditions), sinver (to blind, said of strong light), while among the examples of actors in binyan 5, we may mention hetis (make-fly, typically a plane), hinˇim ˇ (make-breath, resuscitate), himlix (make-king, anoint as king). Even this tendency, however, is clearly an epiphenomenon, easily accountable in terms of the fine structure of Hebrew alternations (and not in terms of some inherent properties of any of the binyanim). Binyan 5 as causative often alternates with binyan 1, while binyan 3 as causative alternates with binyan 7. Now, binyan 1, unlike binyan 7, may host transitives. Thus, binyan 5 often hosts causative alternants to verbs that, in their non-causative forms, are already transitive, whereas binyan 3 hosts causative alternants to verbs that, in their non-causative forms, are intransitive. Naturally, verbs that take two transitive forms—non-causative and causative—are more often marked for an actor as their external argument (as the markedly transitive semantics tends to correlate with agentivity and volition: see also Arad (2002)). At any rate, regardless of our account of the tendency of distribution of causes and actors between the binyanim, it is clear that Doron’s prediction regarding causation is invalid. Pattern 2 poses another empirical problem. Since in Doron’s system all the binyanim are spell-outs of functional heads, and pattern 2 is argued to be a reflection of the middle morpheme, it turns out there is no passive forms for pattern 1. Indeed, Doron argues that verbs in pattern 1 do not have passive forms, only middle ones. In a few cases, the occurrence of roots in pattern 2 may indeed be ambiguous between a middle reading and a passive reading. For example, nidxaq, in pattern 2, can be read as the passive of daxaq (squeeze in, push through) in pattern 1 or as a middle-reflexive verb (push oneself through). But these cases are few and marginal in comparison with the hundreds of pairs of active and passive verbs in patterns 1 and 2: (47) a. b. c.

Root √ √ʔ kl √ktb smr ˇ

Active: P1

Passive: P2 neʔ exal (eat—passive) katav (write) nixtav (write—passive) samar ˇ (guard) niˇsmar ˇ (guard—passive) ʔ axal (eat)

According to Doron, these forms are middles. But there is no independent motivation to assume that these verbs are middles, except, once again, the theory-internal need to make verbs’ meanings conform to the inherent semantics of binyanim. Semantically and syntactically these verbs behave like all other passives. They depend semantically on their active verb, never allowing special meaning, and they undergo the same type of syntactic operation, with the object of the active verb appearing as their subject.


IN

H EBREW

169

A similar problem has to do with the following verbs in pattern 7, which was mentioned in Section 5.5.1. According to Doron, these verbs are w middles: (48) a. b. c. c.

hitxaret (regret) hitʕ aqeˇs (insist) hitmaceʔ (be well oriented) hiˇsˇtaʕ el (cough)

Doron argues that middles, unlike passives, may be formed even when no corresponding active verb exists—thus accounting for cases like hitxaret (regret). But in fact, she does not provide any semantic or syntactic argument as to why such a verb should be considered a middle verb at all. The only reason why hitxaret (regret) or hitʕaqeˇ (insist) are considered middle verbs is the morphological fact that they appear in binyan 7, which, in Doron’s system, entails that the verb contains the middle head. Note also that none of these verbs is considered as “middle” in English (There is generally some correspondence between true middles across languages). This leads inevitably to circularity. Verbs are considered to be “middle” or “intensive” only by virtue of having V certain morphology. In short, keeping the neat system of syntax–morphology correspondence very often forces us to make assumptions about the meaning of verbs that go against the intuitions of the speakers. So far, empirical shortcomings in Doron’s analysis. Moving now to the structure of her theory, we may note Doron’s direct correspondence of syntax and semantics. This, I argue, is undesirable: we saw that the same alternation can be realized in different ways. In alternations between transitive and reflexive verbs, the transitive alternant may occur in pattern 1, 3, or 5: (49)

Root √ √rxc √glx kwn

Transitive raxac gileax hexin

Reflexive hitraxec hitgaleax hitkonen

Patterns 1–7 3–7 5–7

Interpretation wash shave prepare

Wash, shave, and prepare are all causative, although they appear in different W patterns. Their corresponding verbs in pattern 7 are reflexive. The stative alternation (between a state and a change of state) has the stative alternant in pattern 1, and the non-stative alternant in either pattern 2 or 7: (50)

Root Stative Inchoative Patterns Interpretation √ yˇ sb ˇ yaˇsav ˇ hityaˇsev ˇ 1–7 sit (down) √ ʕ md ʕ amad neʕ emad 1–2 stand (up)

170

C HAPTER 5

Finally, recall the causative alternation: (51)

Root √ √qpʔ ˇ √sbr xmm √ mss

Inchoative qafaʔ niˇsbar ˇ hitxamem namas

Causative hipqiʔ savar ˇ ximem hemis

Patterns 1–5 2–1 7–3 2–5

Interpretation freeze break heat melt

In all these cases, the direct correlation Doron postulates between syntax and morphology is challenged: the same syntactic alternation is realized in different morphological ways. This is particularly evident in the case of the causative– inchoative alternation, where both the causative and the inchoative alternant may be expressed in different ways. Moreover, in some cases the same pattern (P1) may host both causative verbs for certain roots (ˇsavar, break, sadaq, crack), and inchoative verbs for other roots (qafaʔ, freeze—intr., nafal, fall). f We saw above that there are many verbs in pattern 5 that are not causatives. Now, we see that there are many causative verbs that appear in patterns 1 or 3. The only way to account for these data under Doron’s system is to assume that only verbs alternating between patterns 1 and 5 (e.g. freeze) are “true” instances of the causative alternation. Indeed, Doron assumes that sˇavar (break— transitive, binyan 1) and ximem (heat—transitive, binyan 3) are not causative at all, but rather an unmarked verb and an intensive one, respectively. Similarly, in this system, their intransitive counterparts are not inchoative, but middles (niˇsˇbar, break—intr., in P2 and hitxamem, heat—intr., in P7). Note, however, that as far as the syntax and semantics are concerned, these alternations behave similarly in all respects. They involve the same argument structures and the same event structure. These verbs are often grouped together as verbs of change of state, forming a semantic class (Levin 1993, 1999, Levin and Rappaport Hovav 1995). Arguing that break involves a different event type than freeze only because they appear in different verbal patterns, forces us into finer grained semantics that has no syntactic motivation. But this is only a part of the problem with Doron’s theory. Even if we force verbs’ meanings to accommodate the meanings imposed by the semantics Doron assigns to the binyanim, there remains another problem: the system is too restrictive. That is, the system cannot account for any phenomenon that ffalls outside the scope of voice or agency, the two dimensions that, according to Doron, are encoded in the Hebrew verb system. When we examine a large corpus of Hebrew roots, it turns out that there are other types of alternations, related not to voice and agency, but to event structure. One such alternation is between verbs of being in a state and verbs of falling into a state. In Doron’s


IN

H EBREW

171

theory, since all binyanim are reflections of agency or voice, there is no room left to accommodate such alternations: (52) a. b. c. d.

Root √ ˇ √yˇsb skb ˇ √ ˇ √stq ʔ hb

State yaˇsav ˇ (be sitting) saxav ˇ (be lying) sataq ˇ (be silent) ʔ ahav (love)

Fall into state hityaˇsev ˇ (sit down) niˇskav ˇ (lie down) hiˇstateq ˇ (fall silent) hitʔ ahev (fall in love)

Patterns 1–7 1–2 1–7 1–7

Doron’s system predicts that the verbs in pattern 7 above are middles or reflexives, but in reality most verbs in pattern 7 have nothing reflexive about them. To keep the system one could try, of course, to force the semantics of the binyan on the meaning of the verb. For example, one could argue that “sit down” actually means “seat oneself down” and “fall silent” means “make oneself silent”. But there is no independent semantic or syntactic motivation to do so. Furthermore, in the case of hitʔahev, ffall in love, such tinkering does not seem to work: there is no way this verb could be construed as “love oneself”. Doron’s system thus ignores the evident aspectual characteristics of this group, since it is not fine-grained enough to capture this alternation. Doron’s system forces certain semantic relations on verbs appearing in different binyanim. In practice, these relations often fail to be realized, while, on the other hand, there are other relations that do occur, but are not predicted by the system. Two additional phenomena that fall outside of the scope of voice and agency, and therefore are not accounted for under Doron’s theory, are MCM and synonyms. When looking at a small sample, these phenomena may seen marginal, but the corpus shows that they occur, when combined, in more than a fifth of the roots. In Doron’s system, it is not clear what is the way to account for cases of MCM (e.g. macaʔ, find, P1 and himciʔ, invent, P5). Perhaps Doron’s account would be to argue that they are cases of systematic alternations, in this case, the causative alternation, yet I illustrated above the problems associated with such forcing of verbs’ meaning into the alleged semantics of the binyanim. On the other hand, I argue that the multiple meanings assigned to roots in the environment of different binyanim is one of the most important properties of the Hebrew verbal system, alongside the obligatoriness of the binyan morphology for all verbs. From these two facts, most other aspects of the system (e.g. the fact that it contains both systematic and non-systematic alternations) are derived. A system that does not explicitly deal with MCM, or aims to force it into systematic alternations of argument structure, ignores a crucial part of the Hebrew verbal system. As for synonyms, Doron’s system predicts that all synonymous verbs should differ with respect to their external arguments. In fact, there are many pairs of practically synonymous verbs, as illustrated above. Hitgorer (live in, reside, P7) is not more “middle” or reflexive than gar (live in,

172

C HAPTER 5

P1). hitqaneʔ (envy, P7) and qineʔ (P3) have the same meaning—there is no way a in which w the verb in pattern 3 is “intensive” or that in pattern 7 is “middle”. The occurrence of such synonymous pairs poses a problem for the voice and agency based system proposed by Doron. The only way to accommodate these cases would be to argue that the variants in P3 take an actor external argument and those in P5—a cause. As noted above, this analysis might apply to a small subset of the cases, but it in no way accounts for the entire set of Hebrew verbs. Finally, the theory is too restrictive in that it only applies to a subset of Hebrew roots, those roots that appear in several patterns. Doron notes that single occurrence roots do not exhibit the same amount of systematicity. She accepts the fact that in cases of single occurrence the contribution of the binyan is arbitrary. She assumes that such verbs are coined with their binyan in the lexicon, rather than in the syntax. The considerations behind the coinage of roots with binyanim may be, in such cases, purely morpho-phonological (e.g. roots with more than three consonants may only go into binyan 3 or 7). This partition between single occurrence roots and all other roots ruins the neatness of correspondence between morphology and syntax. Note, further, that no grammatical considerations drive us to exclude single occurrence roots from a theoretical account. Of course, had single occurrence roots been numerically marginal, then a theory that excludes them would be acceptable. But, as the corpus indicates, these roots make 27% of all Hebrew roots. Setting apart such a large group of verbs as lexically coined creates a considerable divide within the system, motivated by no syntactic or semantic considerations. After all, single occurrence roots obey the same syntactic restrictions as all other roots. When they appear in patterns 2 and 7 they are never transitives, when they appear in patterns 1, 3, or 5 they may be transitives. If binyanim are taken as overt realizations of verbal morphemes, but not necessarily spell-outs of functional heads, there is no need to distinguish single occurrence roots from other roots. I sum up by recapitulating the valid observations I take over from Doron’s account, based on my corpus study, and emphasizing once again the points on which I differ from her (starting with the active patterns): w For pattern 1, it is true that it is not marked for any specific type of external argument. But this is not because it invariably signifies the “unmarked” external argument: to the contrary, pattern 1 denotes many different marked external arguments, in different contexts. (Causative, in some cases, with pattern 2 as non-causative; stative, in other cases, with pattern 5 as possible causative, etc.) For pattern 3, I agree that there is a certain tendency for verbs in P3 to take as an actor external argument where, sometimes, their counterparts in P1 do not do so. This, however, is a much less pervasive phenomenon than Doron suggests. Furthermore, the phenomenon of MCM is highly linked to most of those cases, setting them apart from other well-defined alternations. Indeed, P3


IN

H EBREW

173

most often participates in alternations other than those predicted by Doron: in particular, it hosts causative verbs whose non-causative counterpart is in P7. For pattern 5, I of course follow Doron in noting the fact that it hosts many causative verbs. And yet, it is not even the pattern of preference for causatives: most causatives are hosted in pattern 3 (as noted above). If anything, the causative appearance of P5 seems to be an epiphenomenon of another fact, namely, that P5 participates in causative alternations with several other patterns (P1, P2, P7, as well as P5 itself). Still, this is not the only function of pattern 5: it hosts many SO and MCM forms as well, where no causative meaning at all is associated with the verbs. In short: it is in variance with the facts to suggest that binyan 1 is systematically the “unmarked” pattern while binyan 3 is systematically marked for an actor external argument. Given this, it is no longer natural (even though it remains a possibility) to take binyan 5 as systematically marked for a cause external argument. For patterns 2 and 7, Doron’s analysis is harder to maintain given the corpus analysis. Doron predicts that all forms in those patterns have to be “middle”. Besides demanding of us to stretch verbs’ meaning, this fails to account for many phenomena of the two patterns: in particular, P2 being primarily the passive of P1, and P7 being very often the non-causative alternant of P3. We must note, in conclusion, that Doron’s account is far superior to all other attempts to make a direct correlation between binyanim and semantic or syntactic roles. The empirical failures of Doron’s model thus should serve not to discredit Doron (to whose analysis I remain in debt) but rather to discredit the very notion that Hebrew verbal morphology can, at all, be interpreted along the lines of a single role for a single binyan.

5.6. BINYANIM AND THE TYPOLOGY OF VERB ALTERNATIONS I argued that Doron’s theory, which associates binyanim with semantic and syntactic properties, is empirically flawed. Specifically, the theory predicts that two verbs derived from the same root and appearing in two different binyanim should have certain properties, but the prediction is often not borne out. But it could be the case that the research program is on the right track, and it is only the specific formulation that fails empirically. Maybe, if the right semantic characterization is selected, there are correlations between morphological marking and semantic content, reflected in the binyan system? In this section, I examine the corpus findings on the background of the rich typological work on verb alternations. In particular, I compare the morphology of Hebrew verb alternations with Haspelmath’s (1993) comparative study

174

C HAPTER 5

and with Jacobsen’s (1992) work on Japanese. I will show that the marking of Hebrew verbs does not overlap with the semantic classes identified by Haspelmath and Jacobsen. The marking of verb alternations is, to a large extent, up to the choice of each root, rather than a reflection of semantic properties. Haspelmath (1993) distinguishes, following Nedjalkov (1969), between a causative alternation and an anti-causative alternation. In the causative alternation, the non-causative verb appears morphologically “basic” or “simpler”, while the causative verb appears with an extra morpheme. In the anticausative alternation, the causative verb is morphologically simple, while the non-causative (anti-causative) verb takes extra morphology (“basic” applies here to morphological complexity: the alternant carrying more morphology (affixes) is considered derived while the alternant carrying less morphology is considered basic). The following examples from French illustrates the two phenomena: (53) a. b. c. d.

Non-causative fondre (melt, intr.) bouillir (boil, intr.) tomber (fall, intr.) passer (pass, intr.)

Causative faire fondre (melt, tr.) faire bouillir (boil, tr.) faire tomber (fall, tr.) faire passer (pass, tr.)

a. b. c. d.

Causative reveiller ´ (wake up, tr.) secher ´ (dry, tr.) briser (break, tr.) ouvrir (open, tr.)

Anti-causative se réveiller ´ (wake up, intr.) se sécher ´ (dry, intr.) se briser (break, intr.) s’ouvrir (open, intr.)

(54)

The verbs in (53) are examples of the causative alternation: the non-causative verb appears “on its own”, while the causative is marked by a causative auxiliary verb ((faire, make). The verbs in (54) undergo the anti-causative alternation: the causative verb is morphologically simple, while the non-causative requires extra morphology, in this case the pronominal clitic se. Haspelmath takes “verbs” as the basic elements in the lexicon. He furthermore assumes that verbs that contain more morphological material are derived from “simple” verbs. Such an assumption is problematic if we take roots r , rather than verbs, to be the basic lexical elements. All verbs are composite entities, and there is always the possibility that both the morphologically simple verb and the morphologically complex one are derived from a common root. This is manifest in Hebrew, where the verb-creating morphology, the binyan, is always distinguishable from the root. While all verbs, on my account, are complex, Haspelmath assumes that some verbal morphemes are more complex than others. Hebrew is among the languages included in his cross-linguistic study


IN

H EBREW

175

of the causative alternation. He takes binyan 5, hiCCiC, to be more complex than binyan 1, CaCaC, by virtue of its having a prefix. For the same reason, binyan 7, hitCaCCeC, is considered more complex than binyan 3, CiCCeC (Haspelmath’s intuitions are captured in Doron’s system, as presented above). Verb alternations in patterns 1 and 5 are considered by Haspelmath an instance V of the causative alternation, while verb alternations in patterns 3 and 7 are taken by him to be an instance of the anti-causative alternation. Haspelmath makes it very clear that his cross-linguistic survey can make generalizations only with regard a large group of verbs and across many languages, and cannot predict the marking of each verb in any given language. The generalizations he draws (incorporating insights from Jacobsen’s work on Japanese transitivity alternations) are the following. Events that are easier to conceptualize as occurring spontaneously, without an intervention of an external force, are more likely to pattern with the causative alternations. Among the events that belong to this group are freezing, melting, sinking, drying, and laughing. Events that are easier to conceptualize as occurring with the help or instigation of some outside force are more likely to pattern with the anti-causative alternation. Among these are splitting, breaking, closing, opening, washing, and gathering. Before we analyze the applicability of Haspelmath’s analysis to Hebrew, we should now enrich our evidence by bringing in another set of data, that of Jacobsen’s (1992) work. This is an in-depth examination of transitivity alternations in one language, Japanese. Alternations are marked in Japanese through the presence or absence of affixation on the verb stem (creating a causative from a non-causative or an anti-causative from a causative), or through two different affixations on the same stems. There are several causativizing and de-causativizing morphemes, so that transitivity alternations fall into 16 different classes according to the type and number of morphemes involved. In his analysis of the alternations, Jacobsen’s conclusion foreshadows Haspelmath’s: verbs that are easily conceived as occurring spontaneously are likely to participate in the causative alternations (i.e. the morphologically marked form is the transitive one). Examples of these verbs are given in (55) below: (55) a. b. c. d. e. f.

Intransitive aku (open—intr.) tizimu (shrink—ontr.) sizumu (sink—intr.) ukabu (float—intr.) tumu (pack—intr.) yurumu (loosen—intr.)

Transitive akeru (open—tr.) tizimeru (shrink—tr.) sizumeru (sink—tr.) ukaberu (float—tr.) tumeru (pack—tr.) yurumeru (loosen—tr.)

176

C HAPTER 5

On the other hand, verbs that are more easily conceived as occurring through the initiation of an external force pattern with the anti-causative alternation, that is, their more marked form is the intransitive: (56) a. b. c. d. e. f.

Transitive kiru (cut—tr.) muku (peel—tr.) yaburu (break—tr.) niru (boil—tr.) momu (wrinkle—tr.) oru (break—tr.)

Intransitive kireru (become cut) mukeru (peel—intr.) yabureru (break—intr.) nieru (boil—intr.) momeru (wrinkle—intr.) oreru (break—intr.)

Having surveyed the evidence, let us now consider the analysis, concentrating on Haspelmath’s account. Recall that Haspelmath does not predict the behavior of any given verb in any given language, but offers a general tendency. Specifically, he explains his approach by using four verbs along a continuum, from the most extremely motivated by outside force (wash) to the most extremely spontaneous (laugh): (57)

wash

close

melt

laugh

Haspelmath predicts that verbs whose meanings are closer to “wash” (e.g. decapitate) are most unlikely to participate in the causative alternation. Verbs whose meanings are closer to “laugh” are unlikely to participate in the antiw causative alternation. Verbs whose meanings are somewhere along the continuum may differ from one language to another. For example, Haspelmath predicts that melt should prefer the causative alternation, but an anti-causative expression is still possible. Similarly, close should prefer an anti-causative alternation, but a causative expression is possible. At the general, methodological level, it should be pointed out that while Haspelmath’s theory is compatible with a derivational account (where the morphologically complex verb is derived from the more simple one), it is equally compatible with a root-based account, where both verbs are derived from the same root. The morphological differences may then be accounted for in terms of markedness: all we assume is that across languages there is a tendency to associate the more semantically marked form with the more phonologically marked one. The source of such markedness, however, need not be that of richer morphology: it can be, simply, richer phonology in the spell-out of the very same morphological structure. In my view, the very universality of Haspelmath’s theory suggests such a cognitive, rather than a derivational account. Languages differ considerably in their patterns of morphological markings, while the association of sounds and meanings is always shaped by the same w


IN

H EBREW

177

cognitive tendency to associate the more marked with the more marked.58 In short, then, Haspelmath’s and Jacobsen’s observations, in themselves, do not necessarily point in the direction of any specific morpho-syntactic theory. Let us now consider the applicability of Haspelmath’s theory to Hebrew. It so happens that if we look at Haspelmath’s four paradigmatic verbs (wash, close, melt, laugh) in Hebrew, Haspelmath’s prediction is indeed borne out. W Wash is expressed through patterns 1 and 2 (which is not in the regular anticausative expression in patterns 3 and 7, but note that the intransitive form, in pattern 2, is more marked than the transitive one). Laugh, on the other hand, appears in patterns 1 and 5 (the causative alternation). Close is expressed as anti-causative, in patterns 1 and 2 (similarly to wash) and melt is marked both on its transitive and intransitive form: (58) a. b. c. d.

Root √ √rxc √sgr √mss cxq

Transitive raxac (wash—tr., P1) sagar (close—tr., P1) hemes (melt—tr., P5) hicxiq (laugh—tr., P5)

Intransitive nirxac (wash—pass., P2) nisgar (close—intr., P2) namas (melt—intr., P2) caxaq (laugh, P1)

Haspelmath’s study is not limited, of course, to these four verbs. His corpus includes 31 verbs across 21 languages. Let us first consider the statistical validity of Haspelmath’s sample for Hebrew by looking at his distributions of the causative/anti-causative alternation. Haspelmath’s sample of 31 verbs compares to a total, in my corpus, of 361 verbs that participate in some causative alternation. Naturally, 31 is a large enough selection out of 361 to ensure rough representation; it is also small enough to ensure that small sub-groups within the sample would give rise to exaggerated effects. So, in Haspelmath’s Hebrew sample, 66% of the verb pairs are anti-causative, 24% causative and 10% nondirectional; in the total corpus, 57% are anti-causative, 26% causative, and 17% non-directional.59 The fit is rather good, statistically, but it does point to the limitations of the sample: Haspelmath concentrates on some relatively “marked” verbs (he has to—those verbs should be represented in as many as possible of his 21 languages). He thus concentrates on the forms that are more prototypical, exaggerating the role of the anti-causative and significantly under-representing the role of non-directional alternations. 58

Indeed, Haspelmath’s most recent account (2003) is not framed in terms of derivationality at all. Instead, he compares forms in terms of frequency and argues that shorter forms tend to be associated with more frequently used verbs. Obviously this is compatible with any underlying morpho-syntactic model. 59 I follow, here, Haspelmath’s own analysis, in considering pairs of patterns as more or less marked. See also the following note.

178

C HAPTER 5

Indeed, the extent to which we can say that Haspelmath’s prediction is borne out in Hebrew is that, and that alone: five prototypically non-spontaneous events (out of his sample of 31) are marked, in Hebrew, by the (otherwise less common) causative alternation: “boil”, “freeze”, “go out/put out”, “sink”, “learn/teach”.60 This observation is interesting, but is clearly not sufficient, in itself, as an argument for derivational relations in Hebrew morpho-syntax. Let us now look at the behavior of a wider variety of Hebrew verbs. If it turns out that the verbs appearing in patterns 1 and 5 and verbs appearing in patterns 1 and 2 or 3 and 7 correspond with well-defined semantic properties, this will motivate the hypothesis that the prefixed patterns are more complex than the non-prefixed ones. If, on the other hand, the appearance in prefixed and unprefixed patterns crosses semantic boundaries, the hypothesis of the complexity of prefixed patterns will be significantly weakened. With this in mind, let us examine the Hebrew data. Haspelmath predicts that events that are likely to happen spontaneously should participate in the causative alternation. His examples for that are freeze, melt, sink, dry, and laugh. Indeed, with the exception of dry, all these verbs appear in Hebrew with their transitive alternant in binyan 5 and their intransitive alternant in binyan 1 (melt has its intransitive alternant in binyan 2, niCCaC, which is prefixed). In fact, most agentive verbs that have causative alternants w in Hebrew appear in patterns 1 and 5 (Arad 2002), including work, sign, laugh, move, park, drive, turn, and storm: (59) a. b. c. d. e. f. g. h.

Root √ √cxq √ʕ bd √xtm √xnh √zwz √pnh √nsʕ sʕ r

Intransitive caxaq (laugh) ʕ avad (work) xatam (sign) xanah (park) zaz (move) panah (turn) nasaʕ (drive) saʕ ar (storm)

Transitive hicxiq (laugh) heʕ evid (work) hextim (sign) hexnah (park) heziz (move) hifnah (turn) hisiʕ a (drive) hisʕ ir (storm)

It is important to note, though, that there are causative verbs in patterns 1 and 5 that are easily construed as initiated by an outside force, as the following examples indicate: 60

Note that the assignment of learn/teach to the causative alternation is questionable. The Hebrew forms, lamad (learn) and limed are, in my view, better considered equally marked: for example, neither of the patterns is prefixed. In other words, I believe that Hebrew has, in fact, considerably more than 17% of non-directional alternations).

R OOTS A CROSS P ATTERNS (60) a. b. c. d. e. f. g.

Root √ √rtx √ʕ md ˇ √yˇsb skb ˇ √ √npl √zrm pxd

Intransitive ratax (boil) ʕ amad (stand) yaˇsav ˇ (sit) saxav ˇ (lie) nafal (fall) zaram (flow) paxad (fear)

IN

H EBREW

179

Transitive hirtiax (boil) heʕ emid (stand) hoˇsiv ˇ (sit) hiˇskiv ˇ (lay) hipil (fall) hizrim (flow) hifxid (frighten)

Although there is a certain tendency for agentive verbs to undergo causative rather than anti-causative alternation, it is not the case that patterns 1 and 5 are associated with a particular semantic type of events. Let us now look at the anti-causative alternation, in patterns 3 and 7. Indeed, many events that are easily construed as occurring through the intervention of an outside force appear in the anti-causative alternation: (61) a. b. c. d. e. f. g. h. i.

Root √ √rsq √npc √bqʕ √qpl √qlp √tnp √sgr √ptx sbr ˇ

Intransitive hitraseq (crush) hitnapec (shatter) hitbaqeʕ a (split) hitqapel (fold) hitqalef (peel) hitanef (soil) nisgar (close) niftax (open) nisbar ˇ (break)

Transitive riseq (crush) nipec (shatter) biqeʕ a (split) qipel (fold) qilef (peel) tinef (soil) sagar (close) patax (open) savar ˇ (break)

On the other hand, there is a number of verbs appearing in patterns 3 and 7, presumably in the anti-causative alternation, that are most easily construed as occurring spontaneously. Note, in particular, dry and melt, w which Haspelmath predicts to belong to the causative alternation, as well as heat, chill, and collapse: (62) a. b. c. d. e. f. g. h.

Root √ √ybˇs √xmm √qrr √mwt √ʕ rʕ r √zʕ zʕ √mss pwr

Intransitive hityabeˇs (dry) hitxamem (heat) hitqarer (chill) hitmotet (collapse) hitʕ arʕ er (deteriorate) hizdaʕ azeʕ a (shock) hitmoses (melt) hitporer (crumble)

Transitive yibeˇs (dry) ximem (heat) qerer (chill) motet (collapse) ʕ irʕ er (deteriorate) ziʕ azeʕ a (shock) moses (melt) porer (crumble)

180

C HAPTER 5

Jacobsen (1992) presents a detailed study of Japanese verb stems, which makes the comparison between Hebrew and Japanese transitivity convenient. Even from comparing the data in (55) and (56) above, it becomes evident that the matching between the behavior of verbs in both languages is far from perfect. Consider the Hebrew correlates of these Japanese verbs: (63) a. b. c. d. e. f. g. h. i. j. k.

Root √ √ptx √kwc √sˇqʕ √cwp √ʔ rz √rpp √xtk √qlp ˇ √sbr rt √ x qmt

Intransitive niftax (open, P2) hitkavec (shrink, P7) sˇaqaʕ (sink, P1) caf (float, P1) neʔ eraz (pack, P2) hitrofef (loosen, P7) nextax (cut, P2) hitqalef (peel, P7) nisbar ˇ (break, P2) ratax (boil, P1) hitqamet (wrinkle, P7)

Transitive patax (open, P1) kivec (shrink, P3) hiˇsˇqiʕ a (sink, P5) hecif (float, P5) ʔ araz (pack, P1) rofef (loosen, P3) xatax (cut, P1) qilef (peel, P3) savar ˇ (break, P1) hirtiax (boil, P5) qimet (wrinkle, P3)

Certain verbs behave similarly in both languages. For example, in both Hebrew and Japanese open, sink, float participate in the causative alternation, and wrinkle, peel, and break participate in the anti-causative alternation. The first group of verbs, whose intransitive alternant is marked, overlaps almost completely between the two languages, with the exception of boil, which w is causative in Hebrew but anti-causative in Japanese. On the other hand, a number of verbs, e.g. open, shrink, pack, and loosen (as well as bend, shake, change, and continue, cf. Jacobsen’s 1992 appendix, group II) are anti-causative in Hebrew and causative in Japanese. The mismatch between Hebrew and Japanese does not weaken Jacobsen’s semantic characterization of Japanese transitivity alternations, nor do the Hebrew data contradict Haspelmath’s cross-linguistic tendencies. Haspelmath stresses that his findings only point out cross-linguistic tendencies, and languages may choose to characterize verbs in different ways.61 By choosing a continuum between two extreme cases, wash vs. laugh, Haspelmath can predict which verbs will tend to be realized as causatives across languages, and which verbs will tend to be realized as anti-causatives. The very validity of Haspelmath’s overall predictions, together with the mismatch between his generalizations and the Hebrew facts, serve to weaken, 61

For example, crumble, w which is anti-causative in Hebrew, German, Modern Greek, Hungarian, Armenian French, Turkish, Russian, and Georgian, is causative in Arabic, Finnish, and HindiUrdu.


IN

H EBREW

181

instead, the claim that prefixed binyanim—2, 5, and 7—are semantically and morphologically more complex than the non-prefixed binyanim, 1 and 3. There is no strict correlation between binyanim that are presumably more complex and their semantic complexity in terms of verb alternations. Spontaneous events occur in the anti-causative alternation, while events occurring through the intervention of an outside force occur in the causative alternation. There is no semantic transparency in the Hebrew binyan system and so, as I will argue in the next chapter, the selection of a binyan morpheme is, to some extent, up to the root. In the following chapter, then, our task is to account for the systematic component in the Hebrew verbal system, within the constraint that all the non-passive binyanim, 1, 2, 3, 5, and 7, are equally complex, semantically as well as syntactically.

CHAPTER 6 A THEORY OF HEBREW VERBAL MORPHO-SYNTAX 6.1. Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 6.2. A Theory of Hebrew verbal morpho-syntax . . . . . . . . . . . . . . . . . . . . . . 189 6.3. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235 In this chapter, I offer a comprehensive account of Hebrew verbal morphology. My main concern is with the following two phenomena:1 (i) The association of roots with binyanim, giving rise to two questions: 1. What specifications do verbal morphemes (binyanim) carry regarding the environments where they can be inserted? 2. What specifications do roots carry regarding the environments where they can be inserted? (ii) The relations between the morphological markings of verbs that are derived from a common root, again giving rise to two questions: 3. How to account for the freedom of roots—both in their association with binyanim and in acquiring interpretations? 4. What are the constraints on such morphological freedom of roots? 1

A word is called for regarding the scope of this chapter. My focus in this work is the relationship between syntax and morphology. Morphology, of course, interfaces not only with the syntactic component of grammar, but also with phonology. In this work I do not address some important questions which fall, I believe, under the scope of phonology (and its interaction with morphology). I do not discuss the correct description of prosodic templates, but take binyan forms to be the realization of verbal morphemes (see Bat El 2001, Ussishkin 2001). I do not deal here with the way the different components of the verb (that is, the exponents of the root, the verbal morpheme, voice, tense, and agreement) are phonologically combined into a string, or with the derivation of verbal templates (see Lowenstamm 2002). Finally, I do not offer an analysis of the phonological changes that occur in the verbal stem in different tenses, the relation between vowel melody and the expression of tense and voice, or the phonological expression of verbal inflection, and in particular person agreement.

182

A T HEORY

OF

H EBREW VERBAL M ORPHO-S YNTAX

183

To anticipate: my main findings are as follows. (1) The only piece of information that binyanim carry in an absolute sense (the same binyan always carrying the same information) has to do with transitivity properties. (2) Roots, analogously, can be inserted freely in all five active binyanim. (3) The underlying reason for the freedom of association of roots and binyanim is the many–many nature of Hebrew morphology, so that the same kind of information can often be encoded through multiple binyanim. (4) That freedom is finally constrained—aside from the transitivity considerations themselves—by the systematic distinction between wordformation from roots and word-formation from words (cases of which discussed in this chapter are passivization and reflexivization; w Chapter Seven proceeds to discuss this distinction in word-formation in general). We have already seen the data that go against a simple one-to-one correlation between binyanim and any specific information they may encode, in the discussion of Doron’s theory in Chapter Five. We have also seen several sources for regularity in the system. The goal of this chapter is to capture the regularities, while providing a principled account for why the system should indeed not display any one-to-one correlation of binyanim and verbs’ meaning. In what follows I shall first sum up, in Section 6.1.1, the sources of regularity in the Hebrew verbal system. Following that, Sections 6.1.2 and 6.1.3 put the many–many nature of Hebrew binyan morphology within the context of the overall many–many nature of morphology in general. This may serve as a starting point for showing why Hebrew verbal morphology should have the degree of freedom it has. Following the introduction in 6.1, Section 6.2 develops in detail my account, and Section 6.3 offers a summary of Hebrew verbal morphology, including the morphological properties of each binyan.

6.1. PRELIMINARIES 6.1.1. Regularities of Hebrew Verbal Morphology Before offering an account of Hebrew verbal morphology, let us first review the three principal sources of regularity in the Hebrew verbal system, as described in the previous chapter:

184

C HAPTER 6

1. Transitivity constraints on binyanim (and their use in the marking of alternations)

Of the five non-passive binyanim, two (binyan 2, niCCaC, and binyan 7, hitCaCCeC) are marked as non-transitive. This in itself is an important contribution towards morphological transparency. Verbs that are obligatorily transitives (e.g. causatives) will not appear in the non-transitive binyanim. The three remaining active binyanim, binyan 1 (CaCaC), 3 (CiCCeC), and 5 (hiCCiC), are not marked with respect to transitivity: they may host either transitive or nontransitive verbs. Given this, we would expect to find verb alternations where the transitive alternant is in pattern 3 (CiCCeC) and the intransitive one is in pattern 5 (hiCCiC) or vice versa. Such marking does not violate the transitivity constraints on binyanim. In practice, no transitivity alternations between patterns 3 and 5 are found.2 The greatest majority of transitivity alternations involve one of the non-transitive patterns, niCCaC or hitCaCCeC: Intransitive 2 1 2 7 7 5

Transitive 1 5 5 3 5 5

Transitivity alternations almost invariably associate the non-transitive variant T with pattern 2 or 7. The combination of patterns 1 and 5 is the only exception to this rule (as is the alternation where both alternants are marked by pattern 5).3 This fact also contributes to the morphological transparency of the system. We do not expect any transitivity alternations in patterns 3 and 5, 1 and 3, and, of course, 2 and 7.4 2

Transitivity alternations between patterns 1 and 3 are very rare, and their status is not clear. The T only two cases in our corpus are lamad (learn, P1) and limed (teach, P3), samax (rejoice, P1) and simeax (gladden, P3). Other cases of roots appearing in patterns 1 and 3 are MCM (compare sˇataq, be silent, P1, and sˇiteq, paralyze, P3) or synonyms (compare naˇsaq, kiss, P1, and niˇ iseq, kiss, P3). I return to this question, why Hebrew has certain alternations and not others, in Section 6.2.6.3 below. 3 If we regard pattern 1 as a default verbal morpheme, which is inserted whenever the root is not specified for another pattern, then this combination 1–5 can be regarded as a default. The small group of roots alternating in P5 is discussed later in the chapter. 4 Note that there are also no transitivity alternations between patterns 2 and 3, although there is no principled reason why this should be the case. There is a limited number of pairs of binyanim through which roots alternate, and the pair is not included in those.

A T HEORY

OF


185

2. Passive marking

The freedom that argument structure alternations have in their morphological marking is not extended to passive verbs. As shown in Chapter Five, the morphological form of the passive depends on that of the active. A root does not have an ability to select for a passive form independently of the active one. While active binyanim cannot be equated with semantic or syntactic classes, there are two binyanim (CuCCaC and huCCaC) that are uniquely associated with the passive forms of two active binyanim (CiCCeC and hiCCiC). This is an important source of transparency: a verb in one of the passive binyanim is immediately identified as a passive form, and, in addition, its active form can be predicted. 3. Directionality

The third factor that ensures morphological transparency in the Hebrew verbal system is the directionality of verb alternations (Doron 1999, 2003). Given a pair of verbs made from the same root in patterns 1 and 5, we cannot tell by their morphology whether this is an instance of an argument structure alternation, MCM or synonyms. But we do know that if this is an alternation, then the transitive alternant will appear in pattern 5, and the intransitive one—in pattern 1. The same holds for other such pairs: the transitive verb appears in pattern 3, the intransitive—in pattern 7. The stative verb appears in pattern 1, the change of state (inchoative) verb—in pattern 2, and the causative—in pattern 5. A verb in pattern 1 may be causative (and then its inchoative counterpart is in pattern 2) or inchoative (with a causative counterpart in pattern 5). But, crucially, there is not a single case where a causative verb appears in pattern 1, with an inchoative counterpart in pattern 5. 6.1.2. Morphology is Many–Many With all the regularities above, we still do not find that a single binyan stands, W in Hebrew, for a single type of information. But why should it? Quite simply, morphology in general has not a one-to-one, but a many–many character. It seems that there are principled reasons why this should be the case. In Chapter Two, we characterized the indirect relationship between syntax and morpho-phonology with the following principle: (1)

a. b.

The number of spell-out rules in a language’s grammar is equal to the number of its morpho-syntactic features. In this, and only this, resides the isomorphism between syntax and phonology.

186

C HAPTER 6

(1a) is a strong statement of parallelism between syntax and morphophonological form: the number of spell-out rules is defined solely by the grammar. (1b) is a strong statement of independence between phonology of syntax: not only does the syntax know nothing about its future phonological spell-out; the spell-out rules, too, are specified each separately (referring to syntactic environments), but know nothing further about the syntax. Principle (1) makes natural a fundamental feature of morphology in general, which is the focus of my account of the Hebrew verbal system: the many– w many nature of morphology. By “many–many” I mean the double ambiguity of morphological forms: r One to many: the same morpho-syntactic feature is spelled out by several

different phonological exponents.

r Many-to-one (Syncretism): several morpho-syntactic features are spelled

out by the same phonological exponent. This is in itself puzzling. In a “perfect” language, where morphology directly and transparently reflects the syntax, each morpho-syntactic feature—voice, tense, aspect, etc.—is realized by a separate exponent (I will term them a, b, c, etc.). These exponents are shown on each verb in the order in which they appear in the syntactic tree. Lower heads are realized by affixes that are closer to the root, higher heads —by affixes further apart from the root, in accordance with the Mirror Principle (Baker 1988). In this perfect language, all the grammatical properties associated with the verb are reflected immediately on its morphology: (2)

Root-averbalizing -bvoice -caspect -dtense -eagreement .

In practice, however, languages deviate considerably from the “perfect” model above. This has two main causes. First, there can be no set of readymade spell-out rules, spelling out (specifically to each language) the set of universal features. This is because the choice of syntactic features is not universal: languages select a subset of the feature pool made available by Universal Grammar (Chomsky 1998, Marantz 1999b). To mention a well-known example: Russian has a syntactic distinction between perfective and imperfective aspect of verbs, while English or Hebrew do not. Even if two languages have exactly the same set of morpho-syntactic features, they may vary as to whether they have separate syntactic heads representing each feature, or whether two heads are syntactically fused into one. Bobaljik and Thrá´ insson (1998) and Bobaljik (2001) argue that in English the Tense and Agreement heads are syntactically fused, while in Icelandic they exist as two separate heads (a fact they relate to overt verb movement to T). For our purposes, we may note that the combination

A T HEORY

OF


187

of the above would, indeed, make universal morphology much more difficult to maintain. But even relative to the set of syntactic features used by a given language, perfect morphology cannot be maintained. First, one-to-many spell-out results from the fact that many, if not most spell-out rules are context-sensitive. If spell-out rules are one-to-one (i.e. if the spell-out of a given syntactic feature has to be uniquely determined in the spell-out rule), then there is no way to capture the contextual allomorphy in language. Second, spell-out rules do not know anything about each other (1b), hence there is no way to avoid possible syncretism. In short, the many–many nature of morphology seems to follow from the mutual opacity of syntax and phonology as stated in principle (1) above. Let us now consider the consequences of the many–many nature of morphology for the regularity and irregularity of language in practice. To illustrate this, let us take the very familiar example of Latin nouns. These nouns receive case and number affixes, according to their conjugation class: (3)

Singular: master Nom dominus Gen domini Acc dominum Dat domino Abl domino Plural: Nom domini Gen dominorum Acc dominos Dat dominis Abl dominis

rumor fama famae famam famae famâ

work opus operis opus operi opere

night nox noctis noctem nocti nocte

famae famarum famas famis famis

opera operum opera operibus operibus

noctes noctium noctes noctibus noctibus

The one-to-many nature of Latin noun morphology is clear: the same feature, e.g. genitive singular, is spelled out as -i, -ae, or -is, depending on the noun. And although this case system is rich, some syncretism, that is, one-to-many, occurs: dative and ablative case suffixes are identical in many cases, nominative and accusative suffixes are identical in the neuter (opus) and in some cases in the plural (nox), and dative singular is identical to the nominative plural in fama.5 5

Some linguists distinguish between syncretism across paradigms, for example, the -s morpheme in English which marks both plural in nominal forms and singular third person in verbal forms, and syncretism within paradigms, as is the case in the Latin nominal paradigm above.

188

C HAPTER 6

6.1.3. Hebrew Morphology is Many–Many Let us now review Hebrew verb morphology with the discussion above in mind. We saw in Chapters Four and Five that the contribution of the binyan is not transparent. The fact that Hebrew has five different verbal morphemes, not necessarily corresponding with semantic or syntactic properties (so that a causative verb may appear in three different binyanim), contributes to the non-transparency of the Hebrew verbal system. We presented a central example for this in Chapter Four. Even in a very coarse-grained account of Hebrew morphology, we can see that the binyanim serve two functions, (i) purely as MCM and (ii) as marking semantic or syntactic properties. The same morphological device—binyanim—serves two roles and thus cannot mark a single type of information, creating the non-transparent nature of Hebrew morphology. The same non-transparency can be carried into the fine-grained account as well. As described repeatedly in Chapters Four and Five, the same binyan may host a causative verb or an inchoative verb. For example, binyan 1 hosts causative verbs whose non-causative alternant is in binyan 2. It also hosts noncausative verbs whose causative alternant is in binyan 5: (4) a. b.

Causative Non-causative savar ˇ (break, P1) niˇsbar ˇ (break, P2) hiqpiʔ (freeze, P5) qafaʔ (freeze, P1)

Syntactic cues, in particular argument structure, can reveal the properties of the verb, but given a verb in binyan 1, there is no way to tell just by its morphology whether it is causative or non-causative, or even transitive or non-transitive. w Note now that the one-to-many nature of the Hebrew binyan system does not, in itself, create a major irregularity. We could easily accommodate the fact that causatives appear in both binyan 1 and binyan 5, or that inchoatives appear both in binyan 2 and in binyan 7, and regard the system as perfect, modulo the variation in marking. What makes the system appear so irregular is the combination of the one-to-many with syncretism. It is the fact that the same binyan is used to mark both causatives and non-causatives which makes the description of Hebrew, at first glance, so difficult. Once the patterns of oneto-many and many-to-one are understood, the behavior of the Hebrew binyan system becomes straightforward. The fact that the same binyan hosts both causative and inchoative verbs is just one example of the syncretism of the Hebrew verb morphology. The fundamental point is that syncretism is pervasive: the same groups of binyanim are used, on the one hand, for marking syntactically meaningful relations such as passivization or reflexivization, and on the other hand, merely for creating

A T HEORY

OF


189

several different verbs (MCM and synonyms). Anticipating the discussion in Section 6.2, we can say that the same phonological strings are used both as exponents of v and as exponents of other verbal features above the v head. We noted that certain combinations of binyanim are unique to MCM and synonyms (e.g. 3 and 5, 2, and 7). But in many cases, given two verbs derived from the same root, verb morphology does not tell us whether this is an instance of an argument structure alternation or whether these are two separate verbs. For instance, two verbs in patterns 1 and 5 or 3 and 7, which share the same root, may be an instance of an argument structure alternation, MCM, or synonymous verbs: (5) a. b. c. d. e. (6) a. b. c. d. e.

Pattern 1 qafaʔ (freeze—intr.) nafal (fall) macaʔ (find) nagaʕ (touch) safad (make a funeral speech)

Pattern 5 hiqpiʔ (freeze—trans.) hipil (make fall) himciʔ (invent) higiʕ a (reach) hispid (make a funeral speech)

Pattern 7 hitxamem (heat—intr.) hitbasel ˇ (cook—intr.) hitʔ abed (commit suicide) hitgareˇs (get divorced) hitqaneʔ (envy)

Pattern 3 ximem (heat—tr.) biˇsel ˇ (cook—tr.) ʔ ibed (lose) gereˇsˇ (expel) qineʔ (envy)

Adding to the imperfection is the fact that binyan 2 (niCCaC) and binyan 7 (hitCaCCeC) may have in them either verbs appearing on their own or verbs that are systematically related to other verbs. Binyan 2 hosts passive forms whose active is in binyan 1, and binyan 7 hosts some passive forms of binyan w 3, as well as reflexive forms whose transitive counterpart is in binyan 1, 3, or 5. In Section 6.2.4 I will argue that this imperfection results from the fact r , while w that binyanim 1, 3, and 5 may only be inserted in the context of roots binyanim 2 and 7 may be inserted either in the context of a root or in the context of another binyan.

6.2. A THEORY OF HEBREW VERBAL MORPHO-SYNTAX The principles of my account are now clear: on the one hand, I contend that Hebrew verb morphology reflects morpho-syntactic features. On the other hand,

190

C HAPTER 6

this reflection is obscured by the many–many nature of morphology as such. All of the above, however, is not specific to Hebrew. What should make Hebrew verb morphology so apparently irregular? With Aronoff (1994), I assume that there is a principled reason why the W Hebrew verbal system is not semantically transparent. I ascribe the simultaneous regularity and irregularity in the system the combination of the following ffacts, uncovered throughout the book: (1) All Hebrew verbs must be in the form of a binyan (see Chapter Two). (2) Hebrew roots may acquire multiple meanings in the environments of certain binyanim (see Chapter Three). (3) The binyan system shows morphological syncretism. The first fact entails that all the regular argument structure alternations must be expressed through morphological alternations in the binyan system. The second entails that there should also be irregular alternations, that is, MCM. Within the same, very narrow confines of a root-and-pattern morphology—no W more than five binyanim—everything has to be expressed, both verb-formation itself as well as verb alternations. The syncretism in the binyan system is due to the fact that the same binyanim, or groups of binyanim, are used to mark both regular processes and irregular ones. With Doron (1999, 2003), I adopt the hypothesis that verbs are formed by W the syntax through the combination of roots and verbal features. I also share her view that morphology reflects or interprets syntactic structures. I differ from Doron on the following two points. First, my position is that all verbs are created by the syntax. This position is stronger than Doron’s, who assumes that only alternating verbs are created by the syntax, whereas non-alternating ones, such as those formed by Single Occurrence roots, are created in the lexicon. Second, I believe that Hebrew morphology is many–many. Doron’s view, on the other hand, takes binyan morphology as a one-to-one spell-out of syntactic functional heads, but as shown in Chapter Five, this is an unlikely principle and is indeed empirically false. With this in mind, then, let us now proceed to consider a positive account W of the regularities underlying the system. 6.2.1. The Basic Structure of the Hebrew Verb Recall the structure of the Hebrew verb, √ as argued from Chapter Two onwards: roots are inserted under the root ( ) node, the lowest node in the tree. The binyan morpheme (CVCVC, nVCCVC, CVCCVC, Hvccvc, or hitCVCCVC) is inserted under the node v. Vow V el melody, spelling out voice features, is

A T HEORY

OF


191

inserted under the voice head. Tense and agreement morphology is inserted under the T and Agr nodes.6 (7) Syntactic structure/Insertion7 AgrP/TP

Agr/TAgr /T suffixes

VoiceP

xexternal argument

Voice

Voicevowel melody yobject

vP v v Vbinyan morpheme

√root Root√CCC

I have been assuming so far that the binyan morpheme is inserted under the v node. In what follows I present a modification of this view: in some cases the binyan form of the verb is inserted under voice, rather than v. Crucially, the insertion of the binyan under v or under voice correlates with the syntactic structure of the verb in question. My claim is that once we make two fundamental distinctions, the behavior of the Hebrew verb system becomes straightforward. One distinction is syntactic, regarding two types of verbs produced by the syntax, while the other is morphological, regarding the insertion of binyan exponents: r Syntactic distinction: distinguish root-derived verbs from verb-derived

verbs.

r Morphological distinction: distinguish binyanim that may only be inserted

in the environment of a root from binyanim that may either be inserted in the environment of a root or in the environment of a verb. These distinctions are the most central observations of my theory: what looks like a uniform entity, “a verb”, may have two different structures, 6

Note that the realization of agreement features in Hebrew depends on tense. Agreement is suffixal in the past tense, prefixal (or circumfixal) in the future. I leave open the question whether tense and agreement morphology are realized separately, or whether the same morphology realizes both. 7 The phonological processes involved are, most likely, those existing in other languages, such as morphological merger (Embick and Noyer 2001). I leave this question open here.

192

C HAPTER 6

root-derived or verb-derived; what looks like a single exponent, binyan, exhibits in fact syncretism, being a spell-out of v or of voice. Alongside these distinctions, I make in this Chapter Three more major claims, which, together, fully derive the behavior of Hebrew verbs. Of these four claims, two are least controversial and I begin my account by setting them out: r Roots may select the binyan(im) in which they appear (6.2.2) rT Transitivity properties are specified for binyanim (6.2.3)

Following that, I put forward in Section 6.2.4 the two more controversial features of the theory, which I then motivate in detail in Sections 6.2.5 and 6.2.6: r In Section 6.2.5 I point out the distinction between root-derived and

verb-derived verbs in Hebrew. From this results the distinction between binyanim that may only be inserted under v and binyanim that may be inserted under v or under voice. This leads finally to my fourth claim: r In Section 6.2.6 I show how the exponent of v may be determined by a conjugation of binyanim. Once this claim is incorporated into the theory, the behavior of verb alternations ffalls into place as well. 6.2.2. The Selection of Binyanim by Roots Roots are inserted freely under the root node: for each root node, we may choose any one of the thousands of roots existing in Hebrew. This is part of the root being “lexical”: being inserted under the lowest node in the tree, there is no prior context that may require insertion of a particular item. This sets roots apart from all other categories which may be sensitive to upper or lower context, triggering allomorphy (Bobaljik 2000). Binyanim, or pattern morphemes, are inserted under v. For each v head there is a choice of only five exponents: CVCVC, nVCCVC, CVCCVC, hVCCVC, and hitCVCCVC. Unlike roots, v heads (or any other category heads) do have some lower context, that of the root. The root is the element that stands in a local configuration with the v head. Given a root R, which binyan should serve as the exponent of v? This would not even be a question, had each root appeared in all possible binyanim. Yet, as we saw in Chapter Five, the Hebrew verbal system has many gaps. What, then, determines the actual pattern that a verb might appear in? r , that element that stands in I argue that one determining factor is the root a local configuration with the verbal morpheme. The root receives specific

A T HEORY

OF


193

interpretations when embedded in the context of particular environments. I argued in Chapter Three that the binyan serves as such an environment where the root is assigned an interpretation. The root, the lowest element in the tree, thus requires a certain context where it is assigned an interpretation. In the Encyclopedia, each root is listed with the interpretations it receives in different contexts: (8)

Root √ ˇ √sbr ˇ √sbr sbr √ˇ sbr ˇ

(9)

v

n/v Interpretation savar ˇ break (v) maˇsber ˇ crisis (n) sever ˇ fraction (n) savir ˇ fragile (adj.) √ Roots are freely inserted under the root ( ) node. But not every root has an interpretation relative to every binyan. I argue that the root locally selects the binyan that is inserted in the verbal node above it:8

v

Environment CaCaC maCCeC CeCeC CaCiC

√root Insertion: √tyl

when combined with v, √tyl locally selects for binyan 3, CiCCeC, to be inserted.

Take, as an example, the case of Single Occurrence roots. Such roots are listed in the Encyclopedia with the interpretation it acquires in the environment of one specific binyan. If a different binyan is inserted under v above that root, the result is pronounceable, but there is no meaning assigned to the “verb” at LF, and so it is ruled out:9 8

There are two ways to capture the selection relation between roots and binyanim, which are empirically equal. One way is to state that the root comes with a “tag” on which the binyan √ (or binyanim) it takes are specified. For example, the root tyl, which appears only in binyan √ 3 (CiCCeC), is specified as +3. The root qˇsˇb, which appears only in binyan 5 (hiCCiC), is specified as +5. This way, the root determines the identity of the verbal morpheme it combines with. A different way to capture the relation between roots and binyanim is to say that roots do not have any specifications on them. All we have is the listings in the Encyclopedia of the environments in which meanings are available for roots. If the binyan inserted is such that the root has no interpretation in its context, the result is ruled out at LF, since the verb is uninterpretable. 9 Some combinations of roots with patterns are impossible on phonological grounds. For example, all roots containing more than three consonants may only combine with patterns 3 and 7, where an extra consonant position is available. But in those cases, too, the Encyclopedia does not contain any interpretation for the root in other verbal environments (e.g. 1, 5), so the same principle applies here.

194 (10)

C HAPTER 6 Root Verbal environment (Binyan) √ qrc CaCaC niCCaC CiCCeC hiCCiC hitCaCCeC

Verb Interpretation qarac wink niqrac ∗ ∗ qerec ∗ hiqric hitqarec ∗ √ Only in the environment of pattern 1 the root qrc results in an actual Hebrew verb. The fact that the binyan morpheme is selected by the root r entails three important properties of the Hebrew verbal system: r gaps r special (unpredictable) meaning r some freedom in morphological marking

These three properties, I argue, are typical precisely of rroot-derived verbs. Take the first property, gaps. The way roots select for their verbal morpheme is similar √ to the way in which English roots select for their adjectival suffixes: √ the root atroc takes the adjectival suffix -ious, creating atrocious, while clums selects for -y, as in clumsy. Each of these roots is not assigned any adjectival meaning in the environment of the other suffix, i.e. ∗ atrocy and ∗ clumsious. It is perfectly possible that roots select only for a subset of the possible verbal morphemes in a language, just like English roots select for one adjectival morpheme out of many. Therefore, gaps g are characteristic of root-derivation: when a root selects only a subset of the set of binyanim, the result is gaps. The second property of the Hebrew verbal system, special meaning of verbs, is also characteristic of root-derivation. I argued in Chapter Three that roots are assigned their specific interpretations rrelative to the context of a binyan or misqal iˇ (nominal pattern). This is precisely what we expect if roots locally select for their binyan form, where they are assigned a specific interpretation, which may be highly specialized and idiosyncratic. w Finally, since roots may select for their binyan, they have a freedom not only in the interpretation assigned to them relative to that binyan, but also with respect to the choice of binyan and the morphological form of the verb they create. For a Single Occurrence root, there are five different options for a pattern in which it may appear. Whenever a binyan is selected by a root, the phenomena discussed above must follow (with special meaning being, of course, optional, as illustrated in Chapter Three). However, as we already began to see in Chapters Four and Five, not all Hebrew verbs are characterized by these three properties. Crucially, I argue that certain Hebrew binyanim display syncretism: they can be inserted in

A T HEORY

OF


195

the environment of a root, or in the environment of another binyan. Insertion of a binyan in the environment of another binyan takes place in cases of verbderived verbs. In w what follows I will draw a clear distinction between cases of root-derivation, involving selection of a binyan by a root, and verb-derivation, involving binyan selection by another binyan. But first, let us examine the properties of the binyanim themselves. 6.2.3. How Roots Become Verbs: Transitivity and the Nature of v Morphemes 6.2.3.1. Transitivity restrictions on binyanim There are some absolute transitivity requirements on binyanim: binyanim 2 and 7 cannot host transitive verbs. Assuming, as we do, that the binyan is a spell-out of the verbal morpheme, the simplest way to account for this is to assume that the two binyanim 2 and 7 are marked by the diacritic [−tr.] or [(−ACC case)]. Below are the insertion requirements for each binyan: (11) a. b. c. d. e.

Pattern CVCVC (1) nVCCVC (2) CVCCVC (3) hVCCVC (5) hitCVCCVC (7)

Environment for insertion +v +v, −tr. (−ACC case) +v +v +v, −tr. (−ACC case)

Patterns 2 and 7 are more specified than patterns 1, 3, and 5: apart from requiring insertion in a verbal environment, they require also a structure where no direct object is present. These specifications ensure that in the environment of transitive and ditransitive structures, only patterns 1, 3, or 5 are inserted. It is intransitivity, rather than transitivity, which is obligatory, and therefore there is no problem for patterns 1, 3, or 5 to be inserted in the environment of an intransitive verb. Unaccusatives, unergatives or verbs with oblique complements can be marked by any of the non-passive patterns: 1, 2, 3, 5, or 7. The reflection of transitivity in Hebrew verb morphology is well motivated in terms of the role of transitivity in syntax. Chomsky (1995, 1998, and subsequent work) draws a distinction between the lexical verb V and the functional category he calls “little v”. Little v is the transitivity head, present in all transitive verbs, and is responsible for checking accusative case features and projecting an external argument. Unaccusatives and passives, on the other hand, lack little v altogether (they have only a “lexical” V). In line with the distributed morphology framework, I distinguish not between the lexical verb and the functional head v, but between roots (lexical elements) and all other

196

C HAPTER 6

functional elements. All verbalizing morphemes are “functional”. Importantly, I take (abstract) morphemes such as “v” to be bundles of features. There can therefore be, in principle, different verbal morphemes, realizing different bundles of features. The question that arises is what types of features are bundled under the category “v”. Obviously, all verbal morphemes share the “v” property, that is, they make the category-neutral root into a verb.10 In Hebrew we note at least two more properties of v morphemes. First, the exponents of verbal morphemes (binyanim) are sensitive to the transitivity of the verb (or to the checking of accusative case). Second, binyanim also express verbs of different semantic type. That is, often a root creates a causative verb in binyan 5, an inchoative (change of state) verb in binyan 2 and a stative verb in binyan 1: (12) a. b. c.

Root √ √ʕ md √ʕ md ʕ md

Binyan CaCaC (1) niCCaC (2) hiCCiC (5)

Verb ʕ amad (be standing) neʕ emad (stand up) heʕ emid (make stand up)

Let us see how this can be captured in our theory. 6.2.3.2. The nature of v morphemes While the Root Hypothesis itself is relatively new, the idea that verbs are not lexical atoms has received considerable attention in the last decade. Within the framework of Government and Binding, the Projection Principle ensures that all the information regarding the meaning of a verb is encoded in its lexical entry. In contrast, a number of linguists pointed to the role the syntax plays in determining the meaning (or range of meanings) assigned to verbs. Such approaches are often referred to as non-projectionist (that is, not based on projection from lexical entries) or configurational approaches to argument structure. Among the prominent advocates of this type of approach are Borer (1994, 1998, to appear) and van Hout (1996), who concentrate on the different semantic and aspectual interpretations acquired by verbs in different syntactic environments, Hale and Keyser (1998, 2002), who aim to characterize different types of argument structures in terms of structural configurations, and Marantz (1997, 1999a, 2001), whose root-based approach to verb-formation is adopted in this work. The Hebrew data reviewed in Chapter Five give further motivation to a configurational approach to argument structure. Clearly, the same root can create several different verbs. This is true both for two separate meanings 10

It is an open question whether v exists as an independent property, or whether it is derivative of being in a local configuration with other verbal properties such as voice or T.

A T HEORY

OF


197

assigned to the root (MCM) and to systematic argument structure alternations, such as the causative and stative alternation. This phenomenon is hard to explain if verbs are taken to be lexical atoms, but is immediately derived if we assume that verbs are formed through the combination of lexical elements (roots) with grammatical features (including category features). As I will argue in Section 6.2.5, argument structure alternations involve derivation of two verbs from the same root r (and not a derivation of one alternant from the other). Schematically, we can present this as follows: √ (13) a. √root + va Æ stative b. √root + vb Æ inchoative c. root + vc Æ causative How exactly does the formation of verbs out of roots take place? The root provides the lexical, idiosyncratic part of the verb. All functional information emenates from the elements that further combine with the root. There are two ways to capture this. One, following the work of Hale and Keyser (1998, 2002), is to put the burden on the syntax alone. The category “v” is a label, but each verbal structure—causative, stative, or inchoative—is associated with a different structural configuration.11 Alternatively, we could take the category “v” itself to be not a unified label, but a bundle of features, as noted above. Each verbal head “v” contains, besides verbalizing features that make the root into a verb, formal features (object case-checking or transitivity) and semantic content. This semantic content could be causative, inchoative, or stative. The configuration, in this case, is the same: root combining with a verbal head. It is the content of that head that gives the verb its specific interpretation. This is the approach adopted in Borer (1994, 1998), Arad (1999a, 1999b), Marantz (1999a):12 (14) a. b. c. 11

Root √ √ʕ md √ʕ md ʕ md

v vstative vinchoative vcausative

Binyan spell-out CaCaC niCCaC hiCCiC

Verb ʕ amad (stand) neʕ emad (stand up) heʕ emid (make stand up)

In fact, Hale and Keyser (1998, 2002) take the transitive alternant in the causative–inchoative alternation to be derived from the non-transitive one. The same configuration, with an extra specifier position for the causative verb, is involved in both alternants. 12 See Arad (1999b) for a detailed discussion of verbal heads. Specifically, it is argued that while all verbal heads share the +v property, i.e., making a root into a verb, they differ with respect to their semantic content and their transitivity properties. Two heads may share the same semantic content while differing in their transitivity properties (i.e. the verbal heads involved in active and passive—the event is the same, but the passive does not check accusative case), or share transitivity properties while having a different semantic flavor (e.g. little v and the applicative head, which may check structural case).

198

C HAPTER 6

Each of these heads also has, potentially, transitivity features (checking accusative case). Case checking is up to the verbal element, rather than the root. When a verbal morpheme combines with a root it introduces, therefore, the presence or absence of case-checking features.13 This study concentrates on the morphological marking of Hebrew verbs. I do not decide here between the two possible accounts sketched above, the configurational approach and the feature-based approach. A configurational approach has, in principle, a better claim for explanatory adequacy: configurations are generated by the syntax “for free”, while features have to be added to the grammatical mechanism. However, configurational approaches are not fine-grained enough to capture the richness of the phenomena, and so their application to Hebrew requires further research. At the moment, there is no theory that successfully accounts for the formation of a stative, an inchoative and a causative verb from the same root, using different configurations alone (Hale and Keyser, recall, derive the causative verb from the inchoative one, rather than from the root). In what follows I will use the feature-based approach, noting that features could be seen as convenient labels. Further research may (or may not) convince us to see in such labels a shorthand for configurations. Let us now examine how these verbal features are related to the binyanim. In Chapter Four, we saw that there is no direct mapping from verbal semantic properties to binyan forms: the same binyan (e.g. hiCCiC) may host verbs of different semantic types, and verbs of the same semantic type (e.g. causative) may appear in different binyanim (CaCaC, CiCCeC, hiCCiC). On the other hand, we noted that binyanim are directly related to the other set of verbal properties: transitivity, or, more precisely, lack thereof. Two binyanim, niCCaC and hitCaCCeC, are inherently specified as non-transitive: (15)

CaCaC, CiCCeC, hiCCiC Æ any semantic content of v. niCCaC, hitCaCCeC Æ any semantic content of v, [−transitive].14

We see the distinction in Hebrew between the transitivity property, on the one hand, and other verbal properties, on the other hand. Transitivity is marked via 13

There is a relation between semantic content and transitivity. For example, causative verbal heads also include transitivity features, while inchoative ones normally do not. 14 A caveat applies for “any semantic content of v” for binyan 3, CiCCeC: in practice, it has no inchoative verbs, with the exception of ʔeraʕ (happen). Note that transitivity constraints interact with the semantic content of v: no causative verbs (which are inherently transitive) are found in patterns 2 and 7. Binyan 2, niCCaC, is not found also with agentive verbs, although it is not clear to me, at that stage, whether this has anything to do with transitivity or with the fact that there are relatively few verbs in that binyan.

A T HEORY

OF


199

a diacritic on binyanim, directly; other verbal properties, specifically, vinchoative , vstative, and vcausative , are marked in Hebrew via conjugation classes. This will be explained in Section 6.2.6.15 6.2.4. The Nature of Binyanim: Conditions on Insertion and Conjugation Classes There is, of course, much more that a theory of Hebrew verbal morpho-syntax has to account for than just the arbitrary selection of binyanim by roots, and the diacritic [−tr.] of binyanim 2 and 7. Yet what remains of our theory can be summed up in two simple assumptions: conditions on insertion of binyanim and the notion of conjugation classes. Let us first look at the first assumption: (16) Conditions on insertion of binyanim: Some binyanim may be inserted only in the context of a root; some may be inserted either in the context of a root or in the context of another binyan:16 • Binyanim 1, 3, or 5 Æ may be inserted only in the context of a root. • Binyanim 2 or 7 Æ may be inserted either in the context of a root or in the context of another binyan (1, 3, or 5). The environments in which binyanim are inserted explain their properties: insertion in the environment of a root is associated with the properties discussed in Section 6.2.2 (gaps, special meanings, freedom in morphological marking) while insertion in the environment of another binyan is associated 15

A cautionary note is required for the discussion of semantic contents of verbs. Most theories have concentrated on those semantic types of verbs that are both well-defined and alternate with other types of verbs. This includes, most notably, causative and inchoative heads (Harley 1995, Hale and Keyser 1998). The empirical study presented in Chapter Five strongly call for three such semantic types in Hebrew: causative, inchoative and stative. This, of course, does not cover all the possible types of verbal phenomena that exist in Hebrew. For some verbs, for instance, we can postulate an agentive verbal morpheme. In some cases, such agentive verb alternates with a causative verb (e.g. work, run). But it is very difficult to identify any systematic alternations involving such properties, and my claim is that all systematic properties of Hebrew verbal morphology can be described with the three semantic types above. Further properties of Hebrew verbs seem to belong to Hebrew semantics rather than to Hebrew morphology. 16 If we take binyan 4 and 6 as independent forms, we can say that they may only be inserted in the environment of another binyan (3 and 5, respectively), and never in the environment of a root. However, I regard these forms not as independent binyanim, but as different vocalisms of the active binyan, determined by passive voice. See Section 6.2.6.

200

C HAPTER 6

with the opposite properties (no gaps, no special meanings, constrained morphological marking). It is the syncretism of binyanim 2 and 7 that is responsible for a substantial part of the apparent irregularity of the Hebrew verbal system. Consider now the second assumption I make: (17)

The exponent of v may be determined by a conjugation of two binyanim.

This principle enables us to explain the morphological marking of verb alternations in Hebrew. Recall that there is some dependency between the verbs derived from the same root. That is, if the causative form is in pattern 3, the inchoative form is in pattern 7, and not in pattern 1. I claim that there are binyan conjugations, that is, pairs or triplets of binyanim, to which certain roots belong, and which determine the spell-out of the verbs that these roots create. Merely to illustrate the idea of conjugations, consider the behavior of Latin nouns. Each noun falls into a nominal conjugation. It is this conjugation that determines what exponents of case and number it will take: w (18)

Root √ √reg (class III) √reg (class III) √puell (class I) puell (class I)

Environment ACC. sg. DAT. sg. ACC. sg. DAT. sg.

Form of morphological case -em -i -am -ae

To repeat, this is merely an illustration and I do not insist that the above is the correct analysis of the Latin facts. I emphasize the fundamental structure: the root √ falls into, or selects for √ a conjugation class (Latin has five such classes; reg selects class III, and puell —class I). The conjugation class then determines which phonological strings serve as the exponents of case and number, given the context of the noun. Analogously, I suggest that Hebrew roots fall into a conjugation class that determines which binyanim will serve as the exponents of v, given the context of the morpho-syntactic verb: stative, inchoative, or transitive. Some features of my proposal deviate from traditional accounts. Principle (16) suggests a sharp distinction between two ways in which binyanim are inserted, based on the distinction between root-derived and verb-derived verbs. The same morphological form of binyan serves, in fact, in two separate functions (this, indeed, is the major syncretism of the Hebrew verbal system). Further, and more specifically, my classification of binyanim in (16) above deviates from traditional approaches as well. I argue that binyanim 2 and 7 may be inserted both in the context of a root and in the context of another binyan. As a result, for some verbs those binyanim display gaps, special meanings, and free-

A T HEORY

OF


201

dom in morphological marking, while for other verbs they display the opposite. This characterization follows from my taking (unlike traditional accounts) not only passivization, but also reflexivization, to be a case of binyan insertion in the environment of another binyan. Finally, traditional Hebrew grammar does not consider binyanim in terms of conjugation classes (but rather, often captures the dependency in the morphological marking of verbs in terms of derivation). I proceed as follows. In Section 6.2.5 I make the case for principle (16), showing the evidence supporting it and the consequences we can draw from it. I survey and explain, in order, the behavior of Single Occurrence roots, MCM, synonyms, the causative and stative alternation (all cases of root derivation), followed by passives and reflexives (the cases of verb derivation). One complication is left in the Hebrew verbal system: the pattern of distribution of the causative and stative alternations by binyanim. By introducing principle (17) in Section 6.2.6 we show how this complicated pattern, too, falls out of our theory, with the apparent irregularity of the system being, once again, merely a case of syncretism. 6.2.5. The Distinction between Root-derived and Verb-derived Verbs I argue for a structural distinction between root-derived verbs and verb-derived verbs. Empirically, this translates into a parallelism in behavior within all the following groups: Single Occurrence roots, MCM, synonyms, the causative alternation and the stative alternation, on the one hand (all root-derived by my account), and passives and reflexives on the other hand (all verb-derived by my account). This may seem surprising, as the categories in the first group do not seem to form a natural class in any way. This, I argue, is typical of their being root r -derived: root derivation, by nature, may yield verbs belonging to different types. Some root-derived verbs may stand on their own (Single Occurrence), others may be related to other verbs (alternations, MCM); the types of interpretations assigned to root-derived verbs may also vary, from very far apart (MCM), through systematic relation (causative vs. inchoative) to synonymy. Let us survey these types in detail. 6.2.5.1. Root-derived verbs Single Occurrence roots

Single Occurrence roots are assigned an interpretation in the context of one binyan only. They are listed in the Encyclopedia with the meaning assigned to them in the environment of that particular binyan. The following cases are single occurrence roots in patterns 1, 2, 3, 5, and 7, respectively:

202

C HAPTER 6

(19) a. b. c. d. e.

Root √ √xdl √hnh √tyl ˇ √qsb sˇʕ l

Binyan CaCaC (1) niCCaC (2) CiCCeC (3) hiCCiC (5) hitCaCCeC (7)

Verb xadal nehenah tiyel hiqˇsiv ˇ hiˇsˇtaʕ el

Interpretation cease enjoy travel listen cough

Each of these roots is listed in the encyclopedia with the interpretation assigned to it in the environment of that particular pattern: (20)

Root √ √xdl √hnh √tyl √qˇsˇb sˇʕ l

ÅÆ ÅÆ ÅÆ ÅÆ ÅÆ

Interpretations available in the Encyclopedia17 “cease” in the environment of [CaCaC] “enjoy” in the environment of [niCCaC] “travel” in the environment of [CiCCeC] “listen” in the environment of [hiCCiC] “cough” in the environment of [hitCaCCeC]

When embedded in any other verbal environment, these roots yield a phonologically well-formed element which has no interpretation. Consider, for example, √ the root qˇsˇb: (21)


Verb Interpretation ∗ qaˇsav ∗ niqˇsav ∗ qiˇsev hiqˇsˇiv listen hitqaˇsev ∗

From a morpho-phonological point of view, there is no problem to associate almost any root with any binyan.18 In the Encyclopedia, however, there may not be any interpretations available for such associations. If another binyan is inserted under v, which w is not selected by the root, the result is uninterpretable at LF and is ruled out. The Encyclopedia itself is expandable: roots may acquire more interpretations, assigned to them in the environment of different contexts. Indeed, Hebrew 17

I focus here on the interpretations available to roots in the environment of verbal patterns. As noted in Chapter Five, many of the roots that create only one verb are assigned an interpretation √ √ √ √ √ also in some nominal environment. The roots xdl, hnh, tyl, qˇsˇb, and sˇʕ l create, respectively, xidalon (emptiness), hanaʔa (enjoyment), tayelet (promenade), qeˇev (attentiveness), and sˇaʕelet (whooping cough). 18 Caveats apply mainly to roots with more than three consonants, which can only be associated with patterns 3 and 7. Roots containing combinations of glides (y, w) and glottal stops (ʔ ) and fricative (h) also may have some difficulties in fitting into certain patterns.

A T HEORY

OF


203

roots often come √ to acquire another meaning in the environment of another pattern. The root sˇpk is a case in point: (22)

Binyan Verb Interpretation CaCaC (1) sˇafax spill niCCaC (2) niˇsˇpax spill—passive ∗ CiCCeC (3) sˇipex hiCCiC (5) hiˇsˇpix ejaculate hitCaCCeC (7) hiˇsˇtapex wax lyrical √ Originally, the root sˇpk was assigned an interpretation in binyan 1 (and its passive counterpart, binyan 2). Over the last few decades, the root acquired two other interpretations in the environment of patterns 5 and 7.19 Single Occurrence roots exhibit the three properties characteristic of roots derivation: the morphological freedom is apparent, since the choice of binyan in this case is up to the root; the root may select one of five possible forms where it is assigned a meaning. Gaps in the system are precisely those binyan forms in which the root is not assigned an interpretation. Since those roots create only one verb, we cannot compare their meanings to those of other verbs, as in the case of MCM, but note that the meaning is assigned to SO roots in the environment of their selected binyan is “special” in that it is specific to the combination of the root with that binyan. Multiple Contextualized Meaning (MCM)

It is now clear that roots that are assigned MCM differ from Single Occurrence roots only in the number of verbal environments in which they are assigned an interpretation. While Single Occurrence roots are assigned an interpretation in the environment of one pattern, MCM roots are assigned an interpretation √ in the context of two or three different binyanim. Consider the example of bxn, which creates two different perception verbs in patterns 3 and 5: w (23)

19

Binyan

Verb

Interpretation available in the Encyclopedia examine examine—pass

CaCaC niCCaC CiCCeC hiCCiC hitCaCCeC

baxan nivxan ∗ bixen hivxin discern hitbaxen ∗

Conversely, certain meanings that were available to a root at a particular stage of the language √ may be taken out of the Encyclopedia with time. For example, bzh, in Biblical Hebrew, had an interpretation available in the environment of binyan 1 (bazah, neglect). In Modern Hebrew, only the interpretation in the environment of binyan 3 (bizah, humiliate) is available.

204

C HAPTER 6

As illustrated in detail in Chapter Three, the two interpretations are separate, and not mutually derived. The root is listed in the Encyclopedia with the two interpretations available to it in two different verbal environments, CaCaC and hiCCiC: √ (24) bxn ÅÆ “examine” in the context of [CaCaC] ÅÆ “discern” in the context of [hiCCiC] Like Single Occurrence roots, MCM roots exhibit gaps: they are not assigned an interpretation in the environment of all five active binyanim. Forms such as ∗ bixen (binyan 3) and ∗ hitbaxen (binyan 7), while morphologically possible, have no interpretation available to them, and are ruled out at LF. MCM roots also have a choice regarding the patterns in which they appear: they may select any pair or triplet of the five active binyanim. As expected for root-derivation, the meanings assigned to one root in the environment of two different binyanim may be highly specialized and semantically far apart. Synonyms

Roots that create synonymous verbs in the environment of two patterns are similar to MCM roots; they merely happen to be much closer in their meanings. We noted in Chapter Five the fuzziness of the border separating MCM from synonyms. We can now see the principled reason why the border should be fuzzy: the two categories are distinct merely by a matter of degree. Structurally, MCM and synonyms represent exactly the same forms. And so, like MCM roots, roots that create synonymous verbs most commonly appear in two binyanim, leaving many gaps. Once again, then, I assume that the meanings of synonyms are listed in the Encyclopedia with their interpretations relative to their environments. Take the √ case of qnʔ : (25)


Verb Interpretation ∗ qanaʔ ∗ niqnaʔ qineʔ envy (colloquial) ∗ hiqniʔ hitqaneʔ envy (bookish)

The root is listed in the Encyclopedia with its two practically identical interpretations, assigned to it in the context of patterns 3 and 7. √ (26) qnʔ ÅÆ “envy” (colloquial) in the environment of [CiCCeC] ÅÆ “envy” (bookish) in the environment of [hitCaCCeC]

A T HEORY

OF


205

The only difference between the two verbs seems to be one of register: qine ʔ (pattern 3) is colloquial, while hitqane ʔ (pattern 7) is bookish. The Encyclopedia is the place where the grammar interfaces with extra-grammatical factors, such as language use. In the case of synonyms, it is almost always the case the one interpretation is colloquial while the other is bookish. The information regarding register is listed in the Encyclopedia, alongside the interpretation assigned to the root relative to each binyan context. The same binyan may serve as an environment to a bookish verb when selected by one root, and a colloquial verb when selected by another. This, again, is typical of roots: the meanings assigned to them in the context of binyanim are arbitrary to a substantial degree. This arbitrariness includes also the register to which the verb belongs. The causative alternation

The causative alternation differs from the cases surveyed above. It involves two verbs, closely related in meaning, of which one is causative and the other is non-causative (most often inchoative, that is, change of state). Still, we see here, essentially, the same structure as above (with a difference in morphological marking explained in Section 6.2.6). The claim that the causative alternation ffalls under root-derivation is not uncontroversial. Many theories of argument structure relate the two verbs participating in that alternation derivationally. Some derive the causative verb from the inchoative one (Harley 1995, Hale and Keyser 1998), while others derive the non-causative verb from the causative one (Levin and Rappaport Hovav 1995, Reinhart 1996). Indeed, in any theory that does not distinguish roots from verbs, such a derivation may appear natural. However, a closer examination of causative verbs in Hebrew shows they exhibit all the properties typical of root-derivation: gaps, special meaning and morphological freedom. Let us now review the evidence in favor of the claim that the non-causative and the causative verb are both derived from the root and not from one another. First of all, notice the gaps in the formation of causative and non-causative verbs. It is a well-known fact that not all non-causative verbs can have a causative counterpart in Hebrew. Now, there are also numerous causative verbs in Hebrew, which do not have any corresponding non-causative verbs: w (27) a. b. c. d. e.

hicmiʔ (make thirsty, P5) hicʕir (make young, P5) hilʔah (tire, P5) hifnim (internalize, P5) hisdir (coordinate, P5)

206

C HAPTER 6

Conversely, there are inchoative verbs that do not have any causative counterpart: (28) a. b. c. d. e.

hitʕalef (faint, P7) g ʕ (die out, P1) gava hofiʕa (appear, P5) ʔazal (run out, P1) nirqav (rot, P2)

Verbs such as hicmiʔ (make thirsty) appears, in our root corpus, as Single V Occurrence: it is a perfect causative verb, but it does not participate in the causative alternation.20 In other words, they exhibit gaps: a causative verb appearing without a non-causative counterpart, or vice versa. As we shall see, this sets causatives apart from passives and reflexives, which may never appear as a Single Occurrence, without a transitive counterpart. The very ability of causatives or inchoatives to stand on their own indicates that they are derived from a root: if they were derived from a non-causative verb, there would only be pairs of causative and non-causative verbs, similarly to actives and passives. The second source of evidence for causatives being derived from roots is the fact that they may acquire a meaning that is not present in the non-causative verbs. This meaning may be some idiomatic meaning that the non-causative verb lacks (29a–b), or any other specialized meaning different from that of the non-causative (29c–g): (29)

Causative heʕ if (make-fly/fire) herʔ a (show / herʔa lo, teach a lesson) √ c. heʕ evid (enslave/employ) √ʕ bd ʕ avad (work) d. t ʕ h taʕ ah (be mistaken) hitʕ ah (mislead) √ e. hitrim (fund-raise) √trm taram (donate) f. p ʔ r hitpaʔ er (boast) pe ʔ er (glorify) √ g. zkr zaxar (remember) hizkir (remind) √ In a causative environment, the root ʕ wp acquires an idiomatic meaning, fire, which w does not exist in the non-causative verb. herʔa (show, 29b) does not mean “cause to see”, and, in addition, has an idiomatic meaning of “teach a lesson”. Similarly, heʕevid (29c) does not mean “cause to work”, but rather has the specialized meaning “enslave”, and “hitrim” means “fund-raise” rather a. b.

Root Non-causative √ √ʕ wp ʕ af (fly) rʔ h raʔ ah (see)

A stative verb derived from the same root, camaʕ, be thirsty, existed in Biblical Hebrew, but Modern Hebrew does not have it. Either way, Biblical Hebrew never had an inchoative verb, whose meaning is become thirsty. w

20

A T HEORY

OF


207

than “make someone donate”. Again, this stands in opposition to passives and reflexives, where the passive or reflexive verbs may never have any meaning independent of that of the transitive verb (modulo the change in meaning entailed by the grammatical process of passivization or reflexivization). If causatives, like reflexives, were derived from a verb, we would have expected them to share all aspects of their meaning with that verb. Under this account, pairs of causative and non-causative verbs are structurally similar to MCM roots: both are instances of two verbs derived from the same root, spelled out as two different binyanim. The question arises regarding the differences between argument structure alternations and MCM. I argue that there are two such differences. First, in the causative alternation the root combines with two verbal morphemes that are semantically well-defined: causative and non-causative (the non-causative morpheme may be inchoative or unmarked). In MCM, on the other hand, the root combines with two verbal heads, but these heads do not differ with respect to their semantic properties in such a well-defined way. In some cases, the semantic characterization of the verbal morphemes combining with the root is difficult to pin down (consider batax, trust, and hivtiax, promise). In other cases, a root combines with two verbal morphemes of the same type, spelled out as two different binyanim, and creates two verbs with different meaning, but same semantic type. For example, √ sˇtq creates two causative verbs— —ˇsˇiteq (paralyze) in binyan 3 and hiˇsˇtiq (make silent) in binyan 5. The second difference between the causative alternation and MCM is related to the former one. MCM roots tend to acquire meanings that are often far from one another. Causative verbs, on the other hand, are often very close in meaning to their non-causative counterparts. For example, ximem (heat) and hitxamem (become hot) refer to two events that are almost identical, modulo the difference in argument structure. Special meaning, recall, is an option reserved for roots, but is never obligatory. Only less than 15% of the roots in our corpus have the ability to be assigned two interpretations that are semantically far apart, as in MCM. In Chapter Three, we saw that roots of certain semantic types are more likely to be assigned MCM. We contrasted this with roots that are Single Occurrence. Here we contrast MCM with those roots that are assigned two meanings in the environment of two verbal heads, causative and non-causative. It is up to the root—and the lexical class to which it belongs—whether it is assigned MCM or whether it can be construed with a causative and a noncausative verbal head. Within those roots that participate in the causative alternation, the degree of W special meaning varies according to the lexical class of the root. Activity roots √ √ ( ʕ bd, ʕ wp) are more likely to acquire special meaning when causativized (heʕevid, enslave, heʕiff, fire). On the other hand, roots that name properties

208

C HAPTER 6

(being hot) or resultant states (being broken) are often assigned very close meanings when combined with a causative and a non-causative verb. I assume that this fact is related to the semantic properties of the root. Properties and resultant states do not easily lend themselves to MCM (cf. Chapter Three), but, on the other hand, are easily interpreted as events that come about (in the environment of a non-causative verb) or as events that are brought about (in the environment of a causative verb): (30) a. b. c.

Root √ √xmm √npl sbr ˇ

Non-causative hitxamem (become hot, P7) nafal (fall, P1) nisbar ˇ (break, intr., P2)

Causative ximem (heat, P3) hipil (make fall, P5) savar ˇ (break, trans,. P1)

Interestingly, even in these cases, one of the alternants has an additional meaning that is not included in the other alternant. For example, hitxamem (become hot) also means get angry, but ximem (heat) does not have the meaning of make angry. Conversely, the causative verb hipil (make fall) also has the meaning of abort, but the non-causative verb, nafal (fall) does not have a parallel noncausative meaning. Finally, niˇsˇbar (break—intr.) means being fed up when w used with a dative subject, but no parallel causative meanings is implied in the causative form sˇavar. The border between MCM and the causative alternation is thus often as fuzzy as the border between MCM and synonymy. In all these cases, a root acquires a special meaning in the environment of a verbal head. In the case of the causative alternation, an extra requirement, in the form of the semantic type of the verbal head, restricts the interpretations it may receive. The third piece of evidence for the root-derivation of causative verbs is the relative freedom in their morphological marking. The morphological form of the passive verb depends on that of the active. This is not the case with causative verbs. Below I list the different ways in which a pair of a causative and a non-causative verbs may be marked: (31) a. b. c. d. e. f.

Patterns (causative) CaCaC (1) hiCCiC (5) hiCCiC (5) CiCCeC (3) hiCCiC (5) hiCCiC (5)

Pattern (non-causative) niCCaC (2) CaCaC (1) niCCaC (2) hitCaCCeC (7) hitCaCCeC (7) hiCCiC (5)

As is evident from (31), given a non-causative verb in pattern 2 or 7, there is no way to tell what its causative form might be. A verb in pattern 7 might have

A T HEORY

OF


209

its causative alternant in pattern 3 or 5.21 A verb in pattern 2 might have its causative form in pattern 1 or 5. A verb in pattern 1 may either be inchoative, whose causative counterpart is in pattern 5, or a causative, whose non-causative w counterpart is in pattern 2. Finally, a small group of roots forms both causative and non-causative verbs in pattern 5. The independence in morphological marking is not expected if we assume that the causative verb is derived from its non-causative alternant. If, on the other hand, both the causative and the nonr , then this behavior is natural: causative verb are derived from a common root the root is free to acquire two different interpretations in the environment of a causative and of a non-causative verb: (32)

a.

b.

Vcausative

Vcausative

√root

Vnon-causative

Vnon-causative

√root

Morphological spell-out:

Morphological spell-out:

CaCaC, CiCCeC, hiCCiC

CaCaC, niCCaC, hiCCiC, hitCaCCeC

The purpose of the discussion above is to motivate the claim that the causative alternation involves two root-derived verbs. In Section 6.2.6 I will address the way this alternation is morphologically marked. For the time being, note the two factors that establish an order in the marking of the causative alternation. First, within each pair, the same pattern is associated with the causative verb or with the non-causative verb. (In Chapter Five we called this property directionality.) Furthermore, with the exception of the pair 1–5 (and the quantitatively marginal group alternating in binyan 5–5), the intransitive alternant is marked by patterns 2 or 7, the two patterns marked as non-transitive. This property is significant: although there is no problem for intransitive verbs to appear in patterns 1, 3, or 5, when the intransitive verb is part of an argument structure alternation, it strongly tends to appear in one of the patterns that are specified as non-transitive. The stative alternation

The second major alternation marked by the Hebrew verbal system is between stative verbs and non-stative ones, which includes verbs of falling into a state, 21

From a quantitative standpoint, the number of causative verbs in pattern 3 which have an inchoative in pattern 7 is much larger than those in pattern 5 (104 vs. 11—cf. Chapter Five), but this fact is significant: passives do not have even eleven non-predictable forms.

210

C HAPTER 6

or verbs of change of state: (33) a. b. c. d. e. f. g.

Root √ ˇ √skb ˇ √yˇsb ʕ md √ √ʔ hb √dlq zkr √

sˇʕ n

Stative verb saxav ˇ (be lying, P1) yaˇsav ˇ (be sitting, P1) ʕ amad (be standing, P1) ʔ ahav (love, P1) dalaq (be on, P1) zaxar (remember, P1) niˇsʕ an (be leaning, P2)

Fall into a state (inchoative) niˇskav ˇ (lie down, P2) hityaˇsev ˇ (sit down, P7) neʕ emad (stand up, P2) hitʔ ahev (fall in love, P7). nidlaq (turn on P2) nizkar (come to remember, P2) niˇsʕ an (come to lean on. . . , P2)

This alternation is more well-behaved morphologically, mainly because all the stative verbs that take part in this alternation are in pattern 1 (with the exception of niˇʕan, 33g, which marks both the stative and the non-stative verb).22 At first glance, it seems that the verbs participating in this alternation do not acquire any special meaning. This in itself would not be surprising, because, as in the case of the causative alternation, roots that allow this alternation tend to be roots that name states, and these are harder to construe with special meanings. However, one member of the pair of verbs participating in this alternation is often associated with a meaning that does not exist in the other member of the pair. The stative verb sˇaxav (lie, P1), for instance, has an additional meaning, sleep with someone, w which the inchoative verb, niˇsˇkav (lie down, P2), does not have. Conversely, the inchoative verb nidlaq (turn on, P2) has an additional meaning of being turned on, which w the stative verb, dalaq (burn, be on, P1) does not have. Furthermore, there are many gaps in the formation of the stative alternation: the presence of one member of the pair does not predict the presence of the other. Many stative verbs have no non-stative counterpart: (34) a. b. c. d. e. f. g.

22

sanaʔ (hate, P1) sˇaxax (forget, P1) ʔahad (favor, P1) sagad (adore, P1) baz (despise, P1) baxal (abhor, P1) paxad (fear, P1)

Biblical Hebrew had more stative verbs that Modern Hebrew, such as yavaˇ (be dry), kavad (be heavy), xazaq (be strong).

A T HEORY

OF


211

There is no verb in Hebrew whose meaning is come to forget or come to favor, etc. Conversely, there are verbs of falling into a state that do not have a stative counterpart: (35) a. b. c. d.

nivhal (take fright, P2) nigʕal (get disgusted, P2) hitragez (get angry, P7) niffʕam (become affected, P2)

These verbs have no stative correlates whose meaning is “be afraid”, “be in a state of disgust” or “be angry” (a stative verb, ragaz, be angry, in binyan 1, existed in Biblical Hebrew). If one of the two verbs were derived from the other, we would expect there to be no “singletons” without a stative or a non-stative counterpart. The fact that such verbs are found corroborates our assumption that these verbs are derived from the root. Finally, the alternation also exhibits a degree of morphological freedom. Although the majority of stative verbs appear in pattern 1, their non-stative counterpart may appear either in pattern 2 or in pattern 7. This freedom suggests that, as in the causative alternation, both the stative verb and the non-stative one are derived from the common root r , rather than from one another. For the sake of presentation I discuss here the stative alternation separately from the causative alternation, but often the same root creates a stative, a nonstative and a causative verb, as the following examples illustrate: (36) √ Stative non-stative a. skb ˇ saxav ˇ (be lying, P1) niˇskav ˇ (lie down, P2) √ b. dlq dalaq (be on, P1) nidlaq (turn on, intr., P2) √ c. zkr zaxar (remember, nizkar (come to P1) remember, P2) √ d. yˇsb ˇ yaˇsav ˇ (be seated, hityaˇsev ˇ (sit down, P1) P7)

causative hiˇskiv ˇ (lay down, P5) hidliq (turn on, tr., P5) hizkir (remind, 5) hoˇsiv ˇ (seat, P5)

In such cases, the same root combines with v morphemes of three different types, inchoative, stative, or causative, yielding three verbs with related meanings but different event structures: (37)

vstative

vinchoative vinchoative

√root

vstative

vcausative √root

vcausative

√root

212

C HAPTER 6

6.2.5.2. Verb-derived verbs So far, we analyzed cases of root-derivation, where binyanim are inserted in the context of roots. The following two cases, passives and reflexives, involve derivation of verbs from other verbs, where binyanim are inserted in the context of other binyanim.

Passives

There is an asymmetry between active verbs and their passive counterparts. First, there are many active verbs that lack passive counterparts, as noted in Chapter Five, but there is never a passive verb standing on its own, without an active counterpart. Furthermore, the semantics of the passive verb may never deviate from that of the active. That is, passives never acquire a special meaning which is unavailable for the active verb. The opposite case is widely attested: a narrowing of meaning in the passive verb relative to its active counterpart, or an idiomatic meaning in active verbs that is lost under passivization.23 (In this respect, passives differ from the causative alternation, where either alternant may acquire a specialized meaning that does not exist w with the other.) Finally, there is an asymmetry between passives and actives with regard to their morphological form. Most active verbs (with the exception of reflexives, which will be discussed in what follows) have some freedom in their choice of binyan morpheme. I argued that this freedom is given by the fact that the root may select any of the possible binyan forms. Passives, on the other hand, do not have this freedom. As noted in Chapter Five, the morphological form of the passive depends on, and is predictable from that of the active. The passive pattern is determined not by the root, but by the active pattern: (38) a. b. c.

Active Passive CaCaC Æ niCCaC CiCCeC Æ CuCCaC24 hiCCiC Æ huCCaC

All the properties of passives—no gaps (no existence independently of an active verb), no special meaning and no freedom of morphological marking—stem

23

An example in English is the idiom kick the bucket, which w does not retain its idiomatic meaning when passivized: the bucket was kicked. w 24 With a handful of exceptions, which were noted in Chapter Five. W

A T HEORY

OF


213

from their being derived from verbs rather than from roots. Whatever process is involved in the formation of passives, this process takes as its input not the root, but the active verb. Following recent work by Marantz (1997, 2000) and Pylkkä¨ nen (2002), I assume that passives are formed through a modification of an active vP. Verb V phrases combine with a Voice head (Kratzer 1996), the head that determines active or passive voice, as well as the projection of the external argument. Passivization involves a voice head that does not project an external argument (as a result, accusative case-checking features of the verb are lost, otherwise, there would be two cases and only one argument). I do not take a position as to whether the process takes place after merging the external argument (following w which it is suppressed) or whether the external argument is not projected in w the first place (and only its existence is implied). What is important for our analysis is that passivization takes place after the root has selected for an active pattern that is inserted under v. After the root has merged with a verbal morpheme, spelled out in the form of an active binyan (P1, P3, or P5, because passives are formed only from transitive verbs), the voice head combines with the vP: (39)

Insertion

Syntax VoiceP

Voice merges with VP vP

Voice

v

Object v

√root

Root locally selects verbal morpheme: CVCVC, CVCCVC, hVCCVC

When the root selects CVCCVC or hVCCVC as the active verbal morphemes, insertion of vowels under voice spells out passive or active voice (see Chapter Two). The root fills in the consonant slots, voice fills in the vowel slots, and the result is a continuous, pronounceable string, representing √ a Hebrew verb with voice specifications. Let us take as an example the root xbr and binyan 3, CVCCVC:

214

C HAPTER 6

(40)

Structure VoiceP

Insertion

vP

Voice

i/e (active), u/a (passive) v

Object

v

√root

CVCCVC, √xbr

Depending on the value of voice, the vowels {i, e} or {u, a} are inserted, and the root forms the active verb xiber (combine-active) or the passive verb xubar (combine—passive). The case of pattern 1, CaCaC, is different. The passive of binyan 1 is in binyan 2, niCCaC. While the passives of patterns 3 and 5 involve the same CV template, and differ from the active only in their internal vowels, this case involves two different patterns. Furthermore, pattern 2 is not uniquely used as the passive form of pattern 1, but also marks root-derived verbs: single occurrence roots, MCM, synonyms, and inchoative verbs alternating with causative ones. Binyan 2, like binyan 7 (see discussion of reflexives in what follows), thus exhibits syncretism: in some cases it is inserted in the environment of roots, but in others it is inserted under voice, in the environment of binyan 1: (41)

structure VoiceP

Insertion

vP

Voice

active: no change. Passive: insert niCCaC v

Object

v

√root

CVCVC, √smr

The root selects for binyan 1, CVCVC, in the environment of v, creating the active verb sˇamar (guard). I assume that if voice features are [+active], no change in pattern morphology takes place, and the vowels {a, a}, that is, the spell-out of active voice relative to the context of CVCVC, are inserted under voice. The external argument is projected in the specifier of voice, and the active

A T HEORY

OF


215

verb is formed. If, however, voice features are [−active], then the following spell-out rule applies: (42) Voice [−active] Å Æ niCCaC in the environment of [CaCaC]25 When voice is [+active], it has both an external argument and case-checking features; when it is [−active], no external argument is projected (and suppression of case-checking features follows, otherwise there are two cases and only one argument). It is the morpho-syntactic features of voice that force the insertion of passive morphology, but the specific exponent of voice is determined by the exponent of the verbal morpheme that stands in a local configuration to that voice head, CVCVC. Just like a root can determine the exponent inserted for v, the binyan form inserted under v can determine the exponent inserted for the next head above it, voice. The specific strategy taken for the morphological marking of passives is less important than the very variety of strategies: once again, we see many–many morphology. The morphology of Hebrew passive is one-to-many in that the same feature of [−active] voice is spelled out by three separate phonological exponents and, indeed, by two quite different spell-out strategies. The morphology is also many-to-one, in that binyan 2 serves as the exponent both of MCM and other forms of verb-formation from roots, as well as of passive itself. The contribution of this to the apparent irregularity of the Hebrew verbal system is obvious. Reflexives

All theories of reflexives note the fact that the reflexive verb has a corresponding transitive verb. This is true regardless of the morphological means used by the language to mark reflexives: pronominal clitics (French and other Romance languages), different verbal templates (Hebrew), or null marking (English). In fact, practically all theories of reflexives assume that the reflexive is derivationally related to its corresponding transitive verb.26 The nature of this derivation differs from one theory to another, depending on its basic tenets and background. 25

I assume that the CVCVC pattern is deleted and rewritten as nVCCVC, following which the voice vowels specified for this context are inserted. Alternatively, it could be that the pattern with voice vowels, niCCaC, replaces CVCVC under voice. I leave this question open here, as it does not bear on any crucial aspect of my analysis. 26 The only exception I am aware of is Doron (2003), who assumes that reflexives are formed through the head µ. This head may modify either the root (in the case of reflexives in P2) or the agency head ι (in reflexives appearing in P7), but not a transitive verb. This is in opposition to the Hebrew facts, as illustrated here.

216

C HAPTER 6

Kayne (1975), working within a theory of transformations, is interested in the derivation of the surface structure of se clitics in French. Marantz (1984) sees reflexives as created through an affix bearing specific features ([−logical subject], [−transitive]) in the lexicon, which affects the syntactic structure of the verb. Reinhart (1996), who is interested in the structure of the lexicon, considers the derivation of reflexives as a lexical operation (which she calls reduction), affecting the lexical entries of transitive verbs.27 If we look at the set of Hebrew reflexive verbs, we find precisely those properties that are characteristic of verb-formation from verbs. First, no gaps: we never find reflexive verbs without transitive counterparts. (In this reflexives differ from causatives or statives, which may exist on their own without a corresponding non-causative verb). Furthermore, reflexives do not acquire special meanings that are not present in the transitive verb: raxac r (wash—tr.) and hitraxec (wash—refl.) both refer to a washing event.28 Finally, the morphological marking of this alternation is regular, much more regular than that of the causative or stative alternations. Consider the following examples: (43) a. b. c.

Root √ √rxc √glx lbˇs

Transitive raxac (wash, P1) gileax (shave, P3) hilbiˇs (dress, P5)

Reflexive hitraxec (wash oneself, P7) hitgaleax (shave oneslef, P7) hitlabeˇs (dress oneself, P7)

An interesting pattern is revealed: while the transitive verb may appear in any of the patterns that may host transitives (1, 3, 5), the reflexive verb appears in pattern 7. That is, the transitive verb has some choice as for its marking, but not the reflexive verb.29 These three facts combined—no gaps (no reflexive without a transitive), no special meaning and no choice in morphological marking—indicate that the reflexive verb is derived from the transitive verb. The transitive verb itself 27

By assuming that reflexivization affects (“reduces”) the internal argument of the transitive verb, Reinhart (1996) is able to explain the similarity in the marking of reflexives and unaccusatives in many languages: unaccusatives are also derived from transitive verbs, but in this case their external argument is reduced. The morphological effects of reduction, whether of internal or external argument, is the same. 28 The transitive verb, which is derived from the root, may take up an idiomatic meaning, e.g. rraxac be niqyon kapav (wash one’s hand off something, i.e., not take responsibility). 29 The group of alternating verbs in pattern 3 is the biggest: 74. There are eight verbs in pattern 1 and six in pattern 5 (see Chapter Five). Regarding the marking of reflexives, Doron (2003) mentions two examples of reflexive verbs in pattern 2: nidxaq, squeeze oneself in, nidxaf, f push oneself in. Note, though, that the verbs in pattern 2 are ambiguous between a reflexive reading and a passive reading (i.e., being squeezed in by someone else). This passive reading does not exist in pattern 7.

A T HEORY

OF


217

is derived from a root, as is evident by its relative freedom in morphological marking and its ability to take up a special meaning. The structure of reflexives is subject to some debate. One view, advocated by Kayne (1975) and Marantz (1984) and based on the behavior of the Romance clitic se (in particular, its auxiliary selection patterns and its incompatibility with passivization), argues for an unaccusative analysis of reflexives.30 A different view, held by Reinhart (1996) and Reinhart and Siloni (2000, 2003), rejects the unaccusative analysis of reflexives. Showing that reflexives fail certain unaccusativity tests such as partitive cliticization and inverted word order, these authors argue that reflexives do have an external argument. I do not take here a stand regarding the question whether reflexives actually lack an external argument. Specifically, I am agnostic as to whether the formation of reflexive verbs takes place after all the arguments of the transitive verb (including the external one) have been merged, or whether the process takes place before the merging of the external argument. What is crucial for us is that reflexivization takes place in the syntax, at a stage where the binyan morpheme of the transitive verb has already been decided by the root: (44)

v

object

v

v

√root

selects for verbal morpheme: CaCaC,

The root stands in a local configuration with the verbal morpheme. It can therefore select for any of the (possibly) transitive patterns to be inserted there: 1, 3, or 5. Whatever merges with that root, or affects it, has, as its input, the verb, not the root. Suppose that the element above v that is responsible for reflexivity is the voice head. As in passives, voice affects the projection of the direct object, resulting in an intransitive verb. We need a rule that makes sure that hitCaCCeC exponent is inserted under voice if the structure is reflexive: (45)

30

Voice [−transitive] ÅÆ hitCaCCeC, in v [CaCaC, CiCCeC, or hiCCiC]31

While Marantz (1984), as well as Grimshaw (1990), argue for a lexical absorption of the external argument in reflexives, Kayne (1984), followed by Pesetsky (1995) and Sportiche (1998), assume that the external argument is present in the syntax through the clitic se. 31 As with the passive of binyan 1, I am agnostic as to whether the overwriting of the template takes place before the vowel melody is inserted or afterwards.

218

C HAPTER 6

I term voice as [−transitive] to abstract away from the question regarding the structure of reflexives. The similarities between reflexives and passives is obvious: both are derived from transitive verbs and involve de-transitivization. But while it is widely agreed that passives do not project an external argument, the status of the external argument in reflexives is less agreed upon.32 What is crucial for my point is, once again, the role of syncretism in the system. Pattern 7, hitCaCCeC, may either be inserted in the context of a root, which selects for it, or in the context of another binyan, spelling out the synw tactic process of reflexivization. When the binyan is selected by a root, the result is an intransitive verb that happens to appear in binyan hitCaCCeC and takes on a special meaning in the environment of that binyan. This applies to Single Occurrence roots, as well as to roots assigned MCM, roots that create synonyms or roots participating in an argument structure alternation, creating a non-causative root in hitCaCCeC, with a causative (and/or stative) counterpart appearing in another pattern. The transitivity restriction on hitCaCCeC applies also to root-derived verbs, of course. In this case, we have root-derived verbs that happen to be intransitives. When the hitCaCCeC morpheme is inserted in the environment of one of the transitive binyanim, 1, 3, or 5, following reflexivization, the result is a reflexive verb, which is systematically related to a transitive verb. The fact that two binyanim, niCCaC and hitCaCCeC, may be selected either by roots or by other binyanim, contributes greatly to the apparent irregularity of the Hebrew verbal system. The greatest irregularity, however, is presented by the conjugation classes used for marking root alternations, which I shall now address. 6.2.6. Motivating Conjugation Classes 6.2.6.1. Introducing conjugation classes What is left to explain now is the way in which the causative and the stative alternations are morphologically marked. My assumption has been that each root is listed in the Encyclopedia with the binyanim relative to which it is assigned an interpretation. In principle, then, we could do without conjugation 32

From the point of view of Hebrew verb morphology, we must distinguish passives and reflexives syntactically, and not group both of them as [−transitive]. This is because in the context of binyan 1 (e.g. raxac r , wash) passive is spelled out as binyan 2 (nirxac, wash-passive) and the reflexive counterpart appears in binyan 7 (hitraxec, wash-reflexive). A possible distinction between passives and reflexives is that in passives de-transitivization takes place after the external argument has been introduced (and then suppressed), while reflexives lack an external argument from the start (and therefore have both agent and patient functions identified). I leave this issue for future research.

A T HEORY

OF


219

classes and just have the distribution of, say, causative and non-causative verbs listed, case by case for roots: (46) a.

b.

c.

Root Encyclopedia √ qpl ÅÆ fold f (trans.) in the environment of CVCCVC ÅÆ fold f (intransitive) in the environment of hitCVCCVC √ bhr ÅÆ clear (trans.) in the environment of hiCCVC ÅÆclear (intransitive) in the environment hitCVCCVC √ ʕ bd ÅÆ enslave in the environment of hiCCVC ÅÆ work in the environment of CVCVC

The root may select patterns 1, 3, or 5 for its causative verb, and pattern 1, 2, 5, or 7 for its non-causative verbs, which could be represented in various ways, a e.g. by further diacritics on binyanim. Such an approach, however, does not capture some of the most salient facts about the causative and stative alternations. Despite the relative freedom in morphological marking, there are certain pairs of patterns that do not exist. For example, there is no non-causative verb in pattern 2 whose causative counterpart is in pattern 3. That is, there seems to be a limited number of possible pattern combinations. For the causative alternation, these are 1 and 5, 1 and 2, 2 and 5, 3 and 7, 5 and 7, and 5 and 5. Evidently, the marking of the non-causative verb does not determine that of the causative one, or vice versa. But it is also not the case that any pair of patterns may mark a pair of a causative and a non-causative verb. The most straightforward way to capture the dependency between the pairs of binyanim would be postulating insertion rules similar to those we postulated for passives and reflexives, stating: “insert binyan 5 in the environment of binyan 1 when the verb is causative”. Note, however, that unlike the case of reflexives, we would need six such rules, for each existing pair of binyanim: 1–2, 1–5, 2–5, 3–7, 5–7, and 5–5. But the most important drawback of this approach is that it is empirically inadequate: as shown above, pairs of causative and non-causative verbs behave as if they are both derived from a common root and not from one another. Doron (1999, 2003) offers a solution that immediately and elegantly captures these facts with no further stipulations. She takes binyan 1 to be the spell-out of an unmarked v and binyan 5 to be the spell-out of the causative head (γ ). Therefore, by definition, all causative verbs must appear in binyan 5 (there is no other way to get a causative verb in Hebrew) and all non-causative ones appear in binyan 1. However, as noted in Chapter Five, Doron’s theory also runs into many

220

C HAPTER 6

empirical problems precisely because it is so restrictive. Doron only regards those verbs in pattern 5 as causatives. Verbs in patterns 1 or 3, despite being causatives by any semantic or syntactic criterion, are not considered causatives because they do not appear in the “causative” binyan. Doron’s theory explains the directionality between patterns 1 and 5, but fails to account for the other five possible ways in which the same alternation is marked. What we need, then, is a theory that simultaneously captures two facts: first, that the verbs participating in the causative alternation are root-derived, and, second, that there are six different pairs of binyanim that may mark this alternation. In other words, it appears that alternations are defined not in terms of single binyanim, but in terms of the combination of two or three binyanim. This immediately accounts for the most salient facts: the restricted number of possible pattern combinations, as well as the directionality of binyanim in alternations. I therefore suggest that the pairs of patterns that mark roots participating in the causative alternation are best described as conjugations. Each root is specified for one√of these conjugations. Let us consider the root ptx, which is specified for what we shall now tentatively call conjugation 1. It creates patax (open—causative) in pattern 1 and niftax (open—inchoative) in pattern 2. In other words, we define (47)

√

Conjugation 1 = {inch./P2, caus./P1}

The root qpʔ creates qafaʔ (freeze—inchoative) in pattern 1, and hiqpiʔ (freeze—causative) in pattern 5. This may be termed a second conjugation: Conjugation 2 = {non-caus./P1, caus./P5} √ The root mss creates namas (melt—inchoative) in pattern 2 and hemes (melt—causative) in pattern 5:33 (48)

(49)

Conjugation 3 = {inch./P2, caus./P5}34

(50)


√ The root xmm is specified for conjugation 4. It creates hitxamem (heatinhcoative) in pattern 7 and ximem (heat—causative) in pattern 3: √ Note the phonological peculiarities of roots with two identical consonants, such as mss (according to McCarthy’s 1979, 1981 analysis, such roots have only two consonants, and the last consonant is an instance of spreading). 34 Occasionally the combination of {inchoative/P2, causative/P5} goes along with a third member, {stative/P1}. Instead of seeing here a three-term conjugation I interpret such cases as a combination of two conjugations, 2 and 3, so that P2 is marked as inchoative (based on conjugation 3), P5 as causative (based on both conjugations), while P1, which based on conjugation 2 alone is merely marked as non-causative, gets to be more specifically marked in the Encyclopedia, as stative, to mark it off from P2. 33

A T HEORY

OF


221

√

bhr creates hitbaher (become clear, inch.) in pattern 7 and hivhir (make clear, caus.) in pattern 5: (51)


√ Finally, ʔ dm creates both a causative and an inchoative verb (heʔedim, redden/be red) in pattern 5: (52)

Conjugation 6 = {inch./P5, caus./P5}35

This may serve to introduce the tool of binyan conjugation classes. In what follows, I first clarify how this morpho-syntactic tool is understood, in Section 6.2.6.2. Following that, there are two parts of the argument left to be made. First, I need to provide an account explaining why Hebrew morphology marks the alternations it does, in the particular binyanim it uses. I derive this in Section 6.2.6.3. Finally, in Section 6.2.6.4, I elaborate the detailed account of the morphology of the alternations in action. 6.2.6.2. Clarifying conjugation classes In Section 6.2.4, we compared such conjugation classes with the familiar example of morphological case. Limited as it is, we may further pursue the analogy to explain the nature of my suggestion. Thus, there is no inherent morphosyntactic content in the combinations {1, 2} or {2, 5}. Rather, these are the overt spell-out of classificatory groups into which roots fall. A Latin noun belonging to the second conjugation receives -i suffix for genitive singular and -is suffix for dative plural, while a noun belonging to the third conjugation

35

The above list may perhaps be extended to include two further conjugations. One is that of {stative/P1, inchoative/P7}, shown in the two examples 33b, d. Since these are the only two examples in Hebrew of this alternation, I prefer not to postulate an entire conjugation for these alone. Further, one may consider the conjugation {stative/P1, agentive/P3}, which w we have seen already as the empirical kernel to Doron’s claim that P3 denotes an ‘intensive’ head. I leave aside this possibility, because this alternation would then stand out from all other alternations in that nearly all of its tokens are in fact cases of MCM. (Besides being less clearly defined in terms of argument structure, and being quantitatively marginal). My suggestion concerning this group of verbs is simple (I will not pursue it here in detail, as such considerations go beyond the scope of this book): Hebrew had historically (perhaps at the biblical stage), a P1/P3 alternation, which is no longer part of modern Hebrew except in some frozen and residual cases. The interest of this suggestion is in that it reminds us of what Hebrew morphology really consists of: it consists not of seven binyanim, but of a structure of MCM, Single Occurrence and alternations. The very same binyanim may come to encode different morphologies at different stages of the language—in this case, I suggest, we may see an erosion of an alternation taking place.

222

C HAPTER 6

receives -is suffix for genitive singular and -ibus for dative plural. Similarly, a root belonging to the first verbal conjugation class in Hebrew selects for binyan 1 when w combining with a causative v head and for binyan 2 when combining with a non-causative v head. A root belonging to the second conjugation class selects binyan 5 in the environment of a causative v head and binyan 1 in the environment of a non-causative v head. The ensuing syncretism is obvious; I argue that to look for “the meaning of binyan 5” is rather like looking for “the meaning of the Latin nominal suffix -is”. The analogy with morphological case stops there. The following point is a central clarification concerning the nature of binyan insertion in Hebrew. While I take conjugation classes to be the morphological tool for determining the choice of binyan to be inserted, it is a single, particular binyan that is inserted in the immediate environment of the root. The phonological spellout rules for Hebrew define the spell-out of v (sometimes) via conjugation classes, but there is no rule of the form “insert conjugation class 1” (in which one, in turn, would insert an actual binyan). There seems to be an obvious reason for this: the root only becomes pronounceable or meaningful in the environment of a binyan. The abstract √ entity “conjugation class 1” does nothing, in itself, to make the root qpʔ pronounceable or meaningful; it becomes a pronounceable and meaningful string once a specific binyan is inserted. And since, as will be pointed out in the following chapter, the first category head above the root constitutes a phase where the verb is sent to LF and PF, a pronounceable, meaningful string is in fact required r at the level of this head. Thus, the conjugation class is a morphological device but does not serve, as such, as spell-out of v: the only spell-out of v are the binyanim themselves. All verb-formation from roots ought, therefore, to be structurally similar: in Single Occurrence verbs, in MCM, in synonyms as well as in the causative and stative alternation, the binyan is always the spell-out of a v head. In alternations, however, the root receives its interpretation not once, relative to the conjugation class as a whole, but twice, relative to each of the binyanim in that conjugation. We therefore predict a continuum between MCM and alternations: MCM are cases of root-derived verbs that are assigned two interpretations separately, relative to two binyanim. Alternations are root-derived verbs that are interpreted relative to a pair of binyanim, or a conjugation, which correlates with a different event structure of the v morpheme. Because both alternations and MCM are root-derived, both may acquire special meanings. But alternations, unlike MCM, are constrained by the semantic types of v involved in them, which, in combination with the lexical content of the root, limit somewhat the degree of freedom in the meaning assigned to the verb.

A T HEORY

OF


223

The crucial point is that both alternations and MCM exhibit special meaning, and the difference between them is only a matter of degree: (53) a. b.

MCM baxan (examine, P1) qalat (absorb, P1)

a. b.

ʕ af (fly, P1) heʕ if (make fly/fire, P5) ʕ avad (work/make a practical heʕ evid (enslave P5)

c.

joke, P1) saxav ˇ (lie/sleep with, P1)

(54)

hivxin (notice, P5) hiqlit (record, P5)

Alternations

niˇskav ˇ (lie down, P2)

Consider the difference between the alternations and MCM above. The semantic type of v, as well as its morphological spell-out (in the form of conjugations), is well-defined in alternations, but not in MCM. The theory predicts that there are two ways in which the same root may appear in both patterns 1 and 5: either by choosing the two binyanim separately, or by being assigned them through the morphological device of conjugation class 2. In the first case, no special argument structure alternation is predicted while in the second the argument structure alternation is required. But this is the only requirement imposed by being assigned to a conjugation class: nothing is predicted as to the interpretations assigned to the two verbs in the alternants. These interpretations might be far apart (in some cases, almost as far as in MCM) or very close (as in synonyms, modulo the difference in event structure). This dual way for a root to obtain meanings, in both binyanim 1 and 5, is another example of syncretism in the Hebrew verbal system. It is not the case, however, that all roots are equally amenable to MCM, where the two verbs are semantically far apart. We saw in Chapter Three that a w certain important root class, in particular, does not tend to display such MCM: roots naming a property or a state.36 This group is exactly where the Hebrew event-structure alternants apply most naturally: √ (55) xmm: ximem (heat, P3, causative) hitxamem (P7, become √ hot, inchoative) qffʔ : hiqpiʔ (freeze, P5, causative) qafaʔ (freeze, P1, inchoative) Roots naming a property or a state quite naturally lend themselves to the interpretation of being in a state (stative), coming into a state (inchoative), or 36

For suggestions concerning why certain semantic classes may or may not undergo MCM see Chapter Three, where this difficult problem is discussed mainly in the simpler, nominal context.

224

C HAPTER 6

bringing about a state (causative). Most alternations in Hebrew are of this root class. This semantic property, then, easily accounts for the tendency of alternations to be, on the whole, closer to synonyms (modulo the differences in event structure, imposed by the v morpheme) than to MCM. Perhaps the best way to show how assignment to binyanim via MCM and assignment to binyanim via a conjugation class are parallel, is to note that the two do not exclude each other. Just as a root is assigned an interpretation relative to two binyanim (in MCM), it may also be assigned an interpretation relative r to two conjugation classes. These are the roots that are listed in our corpus under both Alternation and MCM. The prediction is that in such cases there will be four different patterns in which the root appears, two associated with each conjugation, and that the same patterns of conjugations will be observed √ in such cases. This is indeed the case. The root ptx, which was discussed above, created both patax (open—causative) in pattern 1 and piteax (develop— causative) in pattern 3. Each of these verbs has an inchoative counterpart in patterns 2 and 7, respectively: √ (56) ptx P1 P2 P3 P7 patax niftax piteax hitpateax open— open— develop— develop— causative inchoative causative inchoative A root may thus be specified for two different conjugations—in this case, conjugation 1 (patterns 1–2) and conjugation 4 (patterns 3–7). The strict √ directionality is kept within each conjugation. That is, if a root appears, like ptx, in patterns 1, 2, 3, and 7, we know that patterns 1 and 2, and 3 and 7 will mark an argument structure alternation, and not patterns 1 and 3 or 2 and 7. 6.2.6.3. Why does Hebrew have the conjugation classes it has? The conjugation classes (as well as the passive and reflexive pattern) visible in Hebrew can, surprisingly, largely be derived from a small number of principles. These classes are not arbitrary stipulations but derive from the central properties of Hebrew morphology. The first general principle is that Hebrew binyanim are marked only for two very abstract properties: transitivity, and insertability above or below v. Specifically, Hebrew binyanim may be marked for (−transitive) and/or for (−above v)37 . T Transitivity and insertability below v are the “unmarked” options. 37

It might be argued that binyanim 4 and 6 are in fact marked as (+above v), but in line with the argument in Chapter Two above these are better considered not as independent binyanim at all but as voice exponents for the true binyanim 3 and 5.

A T HEORY

OF


225

In other words, we have the following grid: (−transitive)

(± transitive)

(−insertable above v) (±insertable above v)

Note now the following: Hebrew (in a pattern present in many other languages) has only Voice between v and T. It thus follows that binyanim inserted above v must involve some voice alternation (which changes transitivity). Specifically to Hebrew, all such voice alternations involve in practice de-transitivization.38 An immediate result of that is that one of the slots above— (−transitive) together with (−insertable above v)—is in fact uneasy: Hebrew tends to mark verbs that are de-transitivized (passives and reflexives) specifically by inserting a binyan above v. Of course, it remains as a logical possibility to allow a binyan which is marked specifically for non-transitive, root-derived verbs, but we can also see the logic behind Hebrew’s decision to avoid such forms. At the same time, we can also see the logic behind Hebrew’s decision to avoid forms that are marked as both (±transitive) and (±insertable above v). Simply, such a form would not be marked at all, and it does appear that Hebrew binyanim must have some information, however abstract, encoded in them. In other words, we find that there are exactly two types of binyanim allowed in Hebrew, with two and three exponents, respectively: (−transitive)

(±transitive)

(−insertable above v)

Not allowed

Binyanim 1, 3, 5

(±insertable above v)

Binyanim 2, 7

Not allowed

At this point, I add two systematic tendencies of conjugation classes. The first is the strong tendency to avoid conjugating a binyan with itself. Of the five possible combinations of self-conjugation, only one is fulfilled, and that with a small group of verbs (about a dozen)—namely, conjugation 6 from Section 6.2.6.1, P5/P5. The motivation for this may well be purely at the interface of language with communication: without such a principle, Hebrew syncretism would give rise to mass homophony. 38

This results from my analysis above, according to which Hebrew causatives are root-derived and not verb-derived. It may well be that languages with analytical causatives (causatives involving an independent causative verb or causative morpheme), such as Turkish or Japanese, also allow alternations above v were transitivity is added.

226

C HAPTER 6

The second principle is motivated not by considerations of communication, but by those of phonology. Namely, there is a very strong tendency to have the patterns in each conjugation as either both geminated (binyanim 3–7, C1 VC2 C2 VC3 ) or both non-geminated (binyanim 1–2–5, C1 VC2 VC3 , nVC1 C2 VC3 , hVC1 C2 VC3 ). This does not reveal anything about the “meaning” of any of the binyanim: it is merely a matter of phonological expedience. The system simply prefers to avoid “crossing the lines” from the geminates to non-geminates or vice versa. It is perhaps natural that, for verbs that are classed together in a single conjugation class, Hebrew may choose to mark this cluster by the phonological similarity. Whichever way we account for this principle, the observation remains as a fundamental restriction on the forms available for Hebrew conjugations.39 As noted above, this principle is a matter not of strict rule, but of tendency: the main exception is that of conjugation 5 (binyanim 5–7), which has relatively few verbs (once again, about a dozen). In other words, out of the six binyan pairs that cross the gemination border (obtained by multiplying (P1, P2, P5)∗ (P3, P7)), only one is actualized as a Hebrew alternation. This compares directly with the single self-conjugating binyan, among five hypothetical possibilities. It now follows that Hebrew may have, in principle, the following four natural types of binyan pairs marking alternations: (57)

1ÅÆ2 1ÅÆ5 2ÅÆ5 3ÅÆ7

(both non-geminated) (both non-geminated) (both non-geminated) (both geminated)40

Remarkably, all of these four types are actualized in Hebrew, each with numerous verbs appearing in it. Three of these pairs take binyanim from two different slots in the table above: for instance, P1 is (±transitive), (−insertable below v), P2 is (−transitive), (±insertable below v). It is thus natural that the system marks this contrast, and makes the P1 alternant transitive, and the P2 alternant—intransitive (in this case: P1 is causative while P2 is inchoative). A similar relation holds for the pairs 5 and 2, as well as 3 and 7. The pair 1 and 5 does not have any obvious transitive contrast, and indeed it is used for marking an alternation where both verbs may be transitive (but with P5 always causative). Indeed, as pointed out in Arad (2002), causativization of transitive verbs (such as ʔaxal, eat, or ʕavar, pass), invariably appears in this conjugation—for reasons which have now been 39

I owe this crucial observation to Ur Shlonsky. Once again, I leave aside conjugation 6, where both alternants appear in binyan 5—while allowed, this strategy clearly is not favored in Hebrew.

40

A T HEORY

OF


227

made obvious. This further accounts for the observation that alternating agent and action verbs strongly tend to take this conjugation. To review the ground covered so far: we have seen why Hebrew ought to have the four conjugations 1 Å Æ 2, 1 Å Æ 5, 2 Å Æ 5, 3 Å Æ 7, and why each of these should have its specific character. This accounts for the conjugations 1–4 from Section 6.2.6.1. Two conjugations, left aside above, are clearly less central: 5 Å Æ 7 (which w violates the tendency not to allow conjugations to cross the gemination border), and 5 Å Æ 5 (which w violates the tendency not to allow a conjugation of a binyan with itself). These are conjugations 5 and 6. Of course, the derivation offered here specifically asks us to consider conjugations 5 and 6 as exceptional. I consider such exceptions to be beyond the scope of this study. To complete the account, then, I proceed to provide an explicit set of spell-out rules for Hebrew verbal morphology, based on our conclusions so far. 6.2.6.4. How Hebrew roots make verbs: the spell-out rules We may sum up the spell-out rules for v in its various forms in Hebrew as follows. First, roots directly select for binyanim or they may also select for conjugation classes. To capture this, suppose that there are 11 diacritics on roots. Each diacritic represents the root’s selection for a certain binyan or conjugation class.41 Crucially, such diacritics are not exclusive: roots may have several diacritics on them, as they may choose more than one binyan or conjugation class. To simplify the presentation, I number the diacritics as follows: α, β, γ , δ, or ε correspond to direct selection of the five binyanim, 1, 2, 3, 5, or 7. A, B, , , E, and Z correspond to selection of the six conjugation classes 1, 2, 3, 4, 5, or 6, respectively. Thus, if a root has the diacritic α, this means that it may select binyan 1 directly (and may form there a case of Single Occurrence, MCM, or synonym). If a root has the diacritic A, it belongs to conjugation class 1. When it meets with the morpho-syntactic head vinch it takes binyan 2 and when it meets with the head vcaus it takes binyan 1.42 w 41

Of course, while my account is in terms of diacritics on roots, it may equivalently be presented in terms of meanings assigned in the Encyclopedia: the two presentations, as already noted above, are mere notational variants. 42 The types of v available in language are presented here in terms off features - vinch , vcaus , vstat . As noted in Section 6.2.3.2 above, the same morphological system can also be derived via a configurational account of event structure (such as in Harley (1995), Hale and Keyser (1998)). In such a case, our spell-out rules would take the form of ‘in the configurational environment X, given the diacritic Y, the root selects binyan Z’. (The spell-out rule will have to display context-sensitive allomorphy which, however, given the syntactic nature of the configurations, is well-established cross-linguistically: see Bobalijk 2000). Note however that Harley, as well as Hale and Keyser, see event-structure alternations as derivationally related (and their analysis of the

228

C HAPTER 6

Note at this point that the choice of diacritics is not completely arbitrary: selection of diacritics by roots follows the simple rule that roots prefer not to select two diacritics whose exponents overlap. While this avoidance of√overlap is a marked preference of the system, it is not total. Thus, for instance, mlʔ in pattern 3 seems to have two separate meanings: once as the causative alternant of the same root in pattern 7 (“make-full”), and once again as another verb created by MCM from the same root (“fulfill, occupy”). In other words, here is a case of a root choosing√ two overlapping diacritics (in the terms above—γ and ). Similar cases are rʔ h in Pattern 5 (being the causative of see, as part of conjugation class 2, with the B diacritic,√as well display w when selecting the δ diacritic for binyan 5 independently) and btʔ in pattern 7 (meaning the inchoative of reveal with the diacritic for conjugation class 4, and express oneself w when directly selecting for binyan 7 through the ε diacritic). Hebrew has perhaps a few dozen cases of this kind. In other words, the phenomenon of diacritic overlap is marginal, yet not ungrammatical. This is reminiscent of another case: the avoidance of self-conjugating patterns. As we saw above, there is only one such case of self-conjugation, the relatively small group of verbs in pattern 5 where one of the alternants is inchoative and the other— causative. Just because this preference is a matter of tendency, rather than of a structural principle, we have concluded above that this is likely to represent the considerations of expediency in the structure of the Encyclopedia, and not any rules of Hebrew grammar. By analogy, the same seems to hold for diacritics overlap: they represent a choice—at the level of the association of sounds with meanings—not to assign too many overlapping meanings to a root in the same context. We can say more than this: there seems to be at least the following preference: other things being equal, when faced with the choice of which meaning to assign to a root in several contexts—an alternation or an MCM—the Encyclopedia often prefers the alternation. That is: when the meanings are available, Hebrew would prefer to assign a diacritic such as and not a diacritic such as (γ , ε) so that many roots that are assigned, for instance, to binyanim 3 and 7 do so via an alternation and not via MCM. Indeed, the number of verb-pairs (derived from the same root) in patterns 3 and 7 that are an instance of an alternation is significantly larger than the number of verbs that are MCM (210 alternations as opposed to 58 cases of MCM). This may be motivated in terms of informativity: an alternation is more informative than MCM, and it might be different configurations is motivated by the need to provide an account of that derivation). Since there is evidence that, in Hebrew at least, event-structure alternations result from independent root-derivations and are not derivationally related, we lose much of the motivation to represent event-structure alternations in terms of configurations.

A T HEORY

OF


229

that the Encyclopedia aims for optimal informativity. Alternatively, this may be considered in terms of language acquisition. (In this case we may say that the learner’s first assumption is typically that the same root, in different binyanim such as 3 and 5, represents an alternation, and only later she may come to learn them as special meanings.)43 The combination of the two preferences—to avoid overlap and to choose alternations, when available, over MCM—together with the presence of allowed alternations only in six conjugations (as derived in the previous section), allows us now to derive an important (if limited) tendency of the system. I mentioned in the previous chapter the tendency of Hebrew to display a certain complementary distribution between MCM and alternations: certain combinations of binyanim typically host alternations, while others typically host MCM. There is some overlap between the two, but also a tendency to have them marked separately. In particular: the combinations P1/P3, P2/P7, and P3/P5 are exclusively reserved for MCM or synonyms. These three combinations all cross the geminationborder (P1, P2, and P5 are non-geminated, P3 and P7 are geminated), and all fail to cross the transitivity border (P1, P3, and P5 are all ±(transitive), while P2 and P7 are—(transitive)). Naturally, none of these combinations mark alternations. Now, given the two principles above we can predict that MCM would tend to be marked in the binyan combinations that may not mark alternations (otherwise there would be a preference to have alternations and no MCM marked there). On the other hand, it is an immediate consequence of the same two principles above that, in those binyanim where alternations are marked, MCM would be less common, resulting in the tendency for complementary distribution of alternations and MCM.44 43

We can put these two preferences as follows: The non-overlap preference: ∼ ((α&A)V (α&B)V (β&A)V (β&)V (γ & )V (δ&B)V (δ&)V (δ&E)V (δ&Z) V (ε& )V (ε&E)) The preference of alternation over MCM: (A, B, , , E, Z) > (α, β, γ , δ, ε)

Once again (as the notation might be misleading in this respect), it has to be insisted upon that those rules are preferences and not strict constraints. 44 Once again, note that this is a mere tendency, and that there are in fact many cases of MCM √ within binyan combinations that primarily mark alternations: 7 in P1/P2 (e.g. mnʕ ) as against √ √ 21 alternations, 72 in P1/P5 (e.g. dxq) as against 129 alternations, 8 in P2/P5 (e.g. ngˇs), ˇ √ as against 39 alternations, 58 in P3/P7 (e.g. xˇsb) ˇ as against 210 alternations. On the other √ hand, there are 46 cases of MCM marked in P5/P7 (e.g. zmn), as against only 27 alternations: interestingly, this is also an alternation that crosses the gemination border.

230

C HAPTER 6

We may now proceed to set out the spell-out rules themselves. As v may come in Hebrew as one of four syntactic heads (unmarked, stative, inchoative, causative), each with its own spell-out rules, we now require four rules. In each of the rules, the left-hand column is the diacritic on the root, and the right-hand column is the pattern selected.

45

(58)

vunmarked :45 αÆ1 βÆ2 γ Æ3 δÆ5 εÆ7

(59)

vstat :46 αÆ1 Æ1 EÆ1

(60)

vinch :47 αÆ1 βÆ2 δÆ5 εÆ7 AÆ2 BÆ1 Æ2 Æ7 EÆ7 ZÆ5

Unmarked v may be a verbal morpheme of any type. Crucially, it does not participate in argument structure alternations, and there are no semantic constraints on its type. 46 In this and the following spell-out rules I take vinch , vcaus, or vstat in Single Occurrence or MCM contexts as a spell-out of a diacritic selecting for a single binyan. An alternative account would be to have vinch , vcaus, and vstat always resulting from the spell-out of a conjugation class, with Single Occurrence or MCM cases explained as cases of gaps in a conjugation class. This is a notational variation which I do not pursue here. 47 As we noted in Chapter Five, although the majority of roots combine with an inchoative (change of state) v morpheme when participating in the causative alternation, there is a class of roots that create a non-causative verb which is not inchoative: ʕavad (work, P1), xatam (sign, P1), rac (run, P1).

A T HEORY (61)

OF


231

vcaus : γ Æ 348 εÆ5 AÆ1 BÆ2 Æ5 Æ3 EÆ5 ZÆ5

Let us now illustrate how these diacritics work. (58) specifies the spell-out of (unmarked) vv—that is, any verbal head, with no semantic specifications. This v can be spelled out as binyan 1, 2, 3, 5, or 7. The actual binyan is determined by the diacritic on the root: the diacritics α, β, γ , δ and ε stand for the five possible binyanim in which Single Occurrence verbs, synonyms and MCM appear. Single Occurrence roots are classified for one binyan only: (62) a. b. c. d. e.

Root √ √ʔ hdα √hnhβ √tylγ √qˇsˇbδ lbtε

Binyan CaCaC niCCaC CiCCeC hiCCiC hitCaCCeC

Verb ʔ ahad (favor) nehenah (enjoy) tiyel (travel) hiqˇsˇiv (listen) hitlabet (dither)

MCM roots (63) and synonyms (64) may be classified into more categories (typically, however, not more than two or three): (63) a. b. c. d. e.

48

Root √ bxnα, δ √ √

grˇsγ , ε btxα, γ , δ

√ √

ʔ klα, γ

mcʔ α,γ ,ε

Verbs baxan (examine, P1) gereˇs (expel, P3) batax (trust, P1) ʔ axal (eat, P1) macaʔ (find, P1)

hivxin (discern, P5) hitgareˇs (divorce, P7) biteax hivtiax (insure, P3) (promise, P5) ʔ ikel (consume, P3) himciʔ hitmaceʔ (invent, P5) (be savvy, P7)

Although causatives may appear in patterns 1, 3, or 5, there are no Single Occurrence causatives in pattern 1. That is, there are no roots that select for binyan 1 via the α diacritic and which create causative verbs. All causative verbs in binyan 1, and almost all causative verbs in binyan 3, alternate with inchoative verbs and are formed via conjugations classes.

232

C HAPTER 6 f. g. h. i.

(64) a. b. c. d. e. f. g. h. i.

√

ʕ lmβ,ε neʕ elam

√ √ √

qdˇsγ ,δ

mkrα,ε prʕ δ,ε

hitʕ alem (disappear, P2) (ignore, P7) qideˇs hiqdiˇs (sanctify, P3) (dedicate, P5) maxar (sell, P1) hitmaker (get addicted, P7) hifriʕ a hitpareʕ a (disturb, P5) (behave wildly, P7)

Root √ √lxmα,β √qccα,γ √nqˇsα,δ √gwrα,ε √rqbβ,δ √zqqβ,ε √xsrγ ,δ √nmnmγ ,ε qrxδ,ε

Verbs laxam (fight, P1) qacac (chop, P1) naqaˇs (knock, P1) gar (live, P1) nirqav (rot, P2) nizqaq (need, P2) xiser (subtract, P3) nimnem (snooze, P3) hiqriax (grow bald, P5)

nilxam (fight, P2) qicec (chop, P3) hiqiˇs (knock, P3) hitgorer (reside, P7) hirqiv (rot, P5) hizdaqeq (need, P7) hexsir (subtract, P5) hitnamnem (snooze, P7) hitqareax (grow bald, P7)

Unlike conjugations, which are restricted to specific pattern combinations, roots that directly select for binyanim exhibit almost all possible combinations of diacritics: (1, 3), (2, 7), (3, 5), (1, 3, 5). This is expected, since diacritics do not exclude one another. Consider, next, roots whose verbal morphemes may have a specific semantic value: stative, inchoative, and causative. In some cases, a root creates only a stative verb, while in others it alternates with an inchoative and/or a causative reading. In the first case, a root is specified for binyan 1, bearing the α diacritic. The verb it creates appears in binyan 1 and happens to be stative: (65) a. b.

Root Verb √ sgd sagad (adore) α √ ʔ hdα ʔ ahad (favor)

Inchoative verbs may appear in patterns 1, 2, 5, or 7. Selection for one of these binyanim may either be direct, in which case the root is marked with an α, β, δ, or ε diacritic, or through a conjugation class, in which case the root falls into the A, B, , , E, or Z conjugations. Note that in both cases the actual spell-out of the binyan is the same: one of the four options, 1, 2, 5, or 7. Conjugations, unlike direct selection of a root for a binyan, give us information regarding the morphological form of other verbs

A T HEORY

OF


233

derived by this root. (66) includes roots that create an inchoative verb but do not alternate: (66)

Root √ a. √gwʕ α b. √rqbβ c. √ypʕ δ d. ʕ lpε

Verb ggavaʕ (die out, P1) nirqav (rot, P2) hofiʕa (appear, P5) hitʕalef (faint, P7)

(67) lists the cases of roots that combine with an inchoative v as part of an alternation, via conjugation class: (67) a.

Root √ ptxA

b. c.

√ √nplB mss

d. e. f.

√ √ √

xmm bhrE sˇxr Z

Verb niftax (open— inchoative, P2) nafal (fall, P1) namas (melt— inchoative, P2) hitxamem (heat— inchoative, P7) hitbaher (clear— inhcoative, P7) hisxir ˇ (blacken— inchoative, P5)

patax (open— causative, P1) hipil (fell, drop, P5) hemes (melt— causative, P5) ximem (heat— causative, P3) hivhir (clear— causative, P5) hiˇsxir ˇ (blacken— causative, P5)

Further, causative verbs in Hebrew may appear in patterns 1, 3, or 5. As with inchoative verbs, the choice of the binyan may be done directly by the root (if it only creates a non-alternating causative verb) or by a conjugation class to which the root is assigned: w (68) a. b. d.

Root √ √ʔ mllγ √cmʔ δ ptxA

e. f.

√ √nplB mss

g. h. i.

√ √ √

xmm bhrE sˇxrZ

Verb ʔ imlel (make miserable, P3) hicmiʔ (make thirsty, P5) patax (open— causative, P1) hipil (fell, drop, P5) hemes (melt— causative, P5) ximem (heat— causative, P3) hivhir (clear— causative, P5) hisxir ˇ (blacken— causative, P5)

niftax (open— inchoative, P2) nafal (fall, P1) namas (melt— inchoative, P2) hitxamem (heat— inchoative, P7) hitbaher (clearinhcoative, P7) hiˇsxir ˇ (blacken— inchoative, P5)

234

C HAPTER 6

Just as binyan diacritics do not exclude one another, conjugation classes do not exclude each other either. A few verbs belong to two conjugations, creating two causative and two inchoative verbs, while others are marked via a diacritic for a certain binyan and via a conjugation for an alternation in two other binyanim. (69) a.

b. c. d. e. f. g. h. i.

Root √ ptxA, √ √ √ √

pcc ,δ bˇsl ˇ ,δ sˇtqE,γ

ʕ lm,ε

√

xzq ,δ

√

ʔ bdB,γ

√ √

lkd ,α ptr ,α,γ

Verbs patax (open—caus., P1) piteax (develop—caus., P3) pocec (explode—caus., P3) hifcic (bombard, P5) hitbasel ˇ (cook—inch., P7) hivˇsˇil (ripen, P5) sataq ˇ (be silent, P1) siteq ˇ (paralyze, P3) neʕ elam (disappear, P2) hitʕ alem (ignore, P7) hitxazeq (become strong, P7) hexziq (hold, P5) ʕ avad (work, P1) ʔ ibed (process, P3) hitlaked (become cohesive, P7) laxad (ensnare, P1) hitpater (resign, P7) patar (exempt, P1) niftar (pass away, P2)

niftax (open—inch., P2) hitpateax (develop—inch. P7) hitpocec (explode— inch., P7) biˇsel ˇ (cook—caus., P3) hiˇstateq ˇ (fall silent, P7) hiˇstiq ˇ (silence, P5) heʔ elim (makedisappear, P5) xizeq (strengthen, P3) heʔ evid (enslave, employ, P5) liked (make cohesive, P3) piter (fire, P3)

We saw above with the diacritics α − ε (selecting directly a binyan) that roots often select a single diacritic and in general do not select more than three. Naturally, bringing the diacritics A–Z in (selecting via conjugation classes) does not alter that rule. We see that roots, in general, take between one and three diacritics. In other words, Hebrew verbal morpho-syntax can be characterized as follows: each root is associated with a selection, among eleven, of one to three diacritics, a choice which then determines its full morphological behavior in the binyanim system. I conclude with the following three observations. First, we may now see why Hebrew has no other alternations except for event structure alternations. This is because Hebrew alternations are not at all relationships between different lexical categories as such. (In such a case, why shouldn’t we have alternations between, say, “doing X”, “doing X with the aid of an artifact”?) Instead, Hebrew alternations are the epiphenomenon of

A T HEORY

OF


235

several different syntactic heads merging with the root. Since Hebrew has not only v, but also vinch , vstat , and vcaus , it has to have four separate spell-out rules for the syntactic head merging with the root; and to the extent that diacritics systematically demand different spell-outs with different heads, alternations ensue49 . This is the only route for alternations to show up in Hebrew. And so, we predict—what is otherwise an unexplained generalization—that no lexical relationship in Hebrew could be systematically morphologically marked, except for that between different syntactic heads (namely, except for that of event structure alternations). Second, note that spell-out rules specify no “otherwise” lines. In other words, when a spell-out is not defined for a certain diacritic, the verb is simply not w allowed to form with that combination of root and head. This is understood in light of the basic result from Chapter Two, that no Hebrew verb may be formed outside the binyan system. This rules out the possibility of a zero form assigned by an otherwise line in the spell-out rule. When a root is not specified for any binyan or binyan conjugation, it simply fails to form a verb. Third, and finally, note that the rules above, together with the rules for formation of passives and reflexives, sum up the morpho-syntax of the Hebrew verbal system. The system is thus seen to be, ultimately, fully regular. The seemingly great irregularity in the system is simply a function of its widespread syncretism. I return to discuss this syncretism in detail in the summary. 6.3. SUMMARY Based on an extensive corpus study, I argued throughout this book that the Hebrew verbal system has a degree of regularity and a degree of irregularity in it. We noted that the mixed regularity and irregularity is due to the presence of MCM and regular alternations, but we did not explain why each of these verb types behaves the way it does. Furthermore, there was no clear prediction regarding where and how much irregularity can exist. In this chapter, I offered a theory that derives the regular and irregular aspects of the system and predicts precisely where irregularity might be found. The theory is based on two distinctions, one syntactic and the other morphological. Syntactically, I suggested distinguishing between: r Root-derived verbs and verb-derived verbs.

49

The same may be restated under a configurational account, pointing out in this case that since Hebrew morpho-syntax is sensitive to four types of configurations, there have to be four spell-out rules for the syntactic head merging with the root, with the epiphenomenon of alternations.

236

C HAPTER 6

Morphologically, I draw a distinction between: r Binyanim that may only be inserted in the local environment of a root

(CaCaC, CiCCeC, hiCCiC) and binyanim that may be inserted in the environment of a root or in the environment of another verb (niCCaC, hitCaCCeC). My empirical study shows that irregularity—gaps, special meaning, and morphological freedom—is strictly confined to root-derived verbs: Single Occurrence, MCM, synonyms, and the causative and stative alternation. Verb-derived verbs—reflexives and passives—are always regular: syntactically, semantically, and morphologically. It is important to note that the distinction between rootderivation and verb-derivation, which underlies the theory proposed, is structural. A structural boundary, defined by the v head that merges with the root, draws the line between regularity and irregularity. This gives further support to the Root Hypothesis, as well as to our working hypothesis regarding the syntactic nature of word-formation: the root/non-root distinction successfully distinguishes the two groups of verbs. When combined with the hypothesis that roots select for their binyan forms locally, it derives the differences between root-derived and verb-derived verbs: Group 1—root-derived verbs: r May exist on their own, either causative or non-causative may exist. r May acquire special meaning (SO, MCM, idioms in causatives, Synonyms

specialized)

r Some freedom in morphological marking (MCM, SO, causatives, Syn-

onyms) Group 2—verb-derived verbs: r Must have a transitive counterpart. Asymmetry: transitives do not need to

have passives or reflexives.

r No special meaning for passive/reflexive that is lacking in the active. r No morphological freedom: morphology is up to the base verb, not up to

the root. While regularity and irregularity are associated with clearly demarcated syntactic domains, the morphological expressions of the two are mixed together, resulting in the apparent irregularity of the Hebrew verbal system. And they are mixed together for a reason. We may see here the operation of two more general principles: first, syntactic features, such as voice, must be spelled out. And because voice exponents are defined relative to the environment of binyanim, this in turn makes binyan morphology obligatory for all verbs: everything must be expressed within the confines or five basic binyanim (see Chapters Two and Four). Second, within this system, syncretism has major consequences.

A T HEORY

OF


237

Obviously, the fewer forms one has at one’s disposal, the more widespread syncretism is. The Hebrew verbal system, operating with five forms, must therefore display rampant syncretism. We may sum up this syncretism as follows: r P1 is the most heavily syncretized: it serves as the spell-out of v itself,

giving rise to MCM, synonyms, and Single Occurrence forms. It also serves in most conjugation classes as the exponent of any of the three types of v heads: stative, inchoative, or causative. In other words, P1 serves in all root-derived Hebrew forms and for this reason it is a completely unmarked form—in some sense, as many linguists suggest, this is the default binyan.50 r P2 has a more nuanced syncretism; besides serving as the exponent of v (hosting MCM, synonyms, and Single Occurrence roots), it also participates in some (but fewer) conjugation classes. Within those conjugation classes, unlike P1, it can only serve as the exponent of an inchoative v. At the same time, it can serve as the exponent of a verb-derived form: passivization from binyan 1. This syncretism creates a richer structure than that of binyan 1: binyan 2 cannot express all root-derived forms and for this reason it feels to have some kind of “content”. At the same time, the syncretism goes even deeper in this case: it is in fact a syncretism between two completely different forms, root-derived and verb-derived. r P3 as well as P5 display least syncretism (and so give rise to the greatest temptation to characterize them as bearing some inherent “meaning”). As P1 and P2, they serve as exponents of v with all the special meanings that entails, besides serving in conjugation classes as causative only, in the case of P3, or as both causative and inchoative, in the case of P5. (It is puzzling that, given all this, P5 is still so often taken to be the “causative”; one assumes this is primarily because this pattern serves, in one important conjugation class, as the causative alternant to P1 which, in turn, is so often taken as the prime form). r P7 displays an especially elegant pattern of syncretism: it may be directly inserted as the exponent of v; through a conjugation class, as the exponent of an inchoative v; and finally as a verb-derived verb, as the reflexivization of P1, P3, or P5 (and, for a handful of roots, also as the passive of P3). This binyan serves in each of the three forms through which a binyan may 50

Hagit Borer (p.c.) suggests that binyan 1 is, in fact, not a binyan, but a default verbal form. She points out that it practically does not host newly coined verbs (see also Bolozky 1978b), it has no canonical argument structure configuration associated with it and it is the only binyan that has irregularity in its vowel melody, which depends on the root (e.g. yiˇmor i , he will guard, vs. yirkav, he will ride).

238

C HAPTER 6 be inserted: freely (where it may function in SO, MCM, or synonyms in a completely free manner), as dictated by a conjugation class (where it has to be inchoative) and as verb-derived (where it has to be reflexive or passive).51

The theory proposed here takes morphology to be a reflection of syntax, but not a direct reflection: the many–many nature of morphology precludes any such simple and direct reflection. Therefore, we are not bound to assume that each binyan necessarily reflects a specific functional head or syntactic structure. The morphological marking of verbs is subject to coarse-grained syntactic constraints of transitivity, but crucially, a root may select for, or be specified for a particular pattern. The fact that roots may locally select for the pattern they are associated with explains not only the degree of freedom in marking verbs, but also, more crucially, the lack of such freedom in passive and reflexive verbs. Other theories of the Hebrew verbal system (Aronoff 1994, Doron 1999, 2003) do not account for the difference between the great regularity of passives and reflexives, as opposed to the degree of freedom that exists in non-passive verbs. My explanation of this phenomenon relies strongly on the Root Hypothesis. In particular, I assume that active verbs (with the exception of reflexives) are created from roots, while passive and reflexive verbs are formed from transitive VPs. Without a root (as in Aronoff’s system), there is no obvious explanation for the behavior of passives and reflexives, as opposed to the behavior of other verbs. This difference follows directly from my theory. Without the Late Insertion hypothesis, it is not possible to explain the syncretism in the Hebrew verbal system, in particular, the fact that patterns 2 and 7 may be inserted under v or under voice. A further advantage of the theory is that it explains why phonological factors often play a role in the association of roots with patterns. Insertion of phonological material takes place post-syntactically. It is, in essence, a morphophonological process. It is not surprising that roots may select for those patterns with which they fit together phonologically. Recall that Doron (1999, 2003) draws a distinction between those cases where the binyan supposedly spells-out a functional head, and those cases where the association of the root with the binyan is “purely phonological”. No such artificial divide is required: phonology plays a role in the association of roots with patterns because the association is done in the mapping from syntax to phonology. The combination of the Root Hypothesis, syntactic word-formation, and Late Insertion proves both strong enough to capture the regular aspects of the 51

This elegant pattern is precisely identical with that of the reflexive se in Romance languages, a parallelism that is pursued in Arad and Shlonsky (2001).

A T HEORY

OF


239

Hebrew verbal system and flexible enough to account for the degree of freedom in morphological marking. Finally, the distinction between root-derivation and verb-derivation subsumes the distinction between “lexical” and “syntactic” verb-formation in Hebrew. Instead of two places for word-formation, the lexicon, and the syntax, all derivation takes place in the syntax. The correct distinction is not in terms of location—lexicon vs. syntax—but in terms of structural position: derivation from a root (lexical atom) or from an existing verb (whose semantic and phonological properties are already incarnated). What Doron and Aronoff call “lexical” verb-formation is, in fact, derivation of verbs from roots. What they term “syntactic” (passive) is derivation of verbs from existing verbs. The structural distinction between root-derivation and verb-derivation proved to play a crucial role in explaining both the morphological properties (gaps, freedom in marking) and the semantic properties (special meaning and lack thereof) of different groups of Hebrew verbs. In the following chapter, we shall see that this distinction goes even deeper, and plays a role in all aspects of word-formation. Through a case study of root-derived and wordderived (denominal) verbs, I will show how this structural distinction accounts for important semantic and morphological differences between root-derived and noun-derived verbs. This, in turn, will provide strong motivation for two hypotheses entertained throughout the book: word-formation in the syntax and h the Root Hypothesis.

CHAPTER 7 ROOTS IN WORD-FORMATION: THE ROOT HYPOTHESIS REVISITED 7.1. Roots and Word-Formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241 7.2. Root-Derived Verbs and Noun-Derived Verbs . . . . . . . . . . . . . . . . . . . . 245 7.3. In the Absence of Morphology: The Semantic Properties of Denominals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254 7.4. The Remaining Piece: Verb-Derived Nouns . . . . . . . . . . . . . . . . . . . . . . 261 7.5. Back to the Root: The Phonological Properties of Denominals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265 7.6. Roots: Between the Universal and the Language Specific . . . . . . . . . . 271 This chapter brings to completion the argument made in this book. It provides new evidence that motivates both the Root Hypothesis and the structural distinction between root-derived elements and non-root-derived elements. Generalizing the distinction made in Chapter Six, I examine verb-formation from nouns and adjectives, a very productive process in Hebrew. Specifically, I compare verb-formation from roots with verb-formation from existing words (in this case, nouns and adjectives). The results are striking: roots—and only roots— may be assigned multiple interpretations in different environments. When the basis for the derivation is a word, this word forces its semantic and phonological properties on any element derived from it. To account for this difference, a generalized constraint on the interpretation of roots is postulated: the first nominal or verbal head that merges with the root serves as the immediate environment for determining its interpretation. In Hebrew, this head is spelled out as the nominal or verbal pattern with which the root combines. Any further derivation takes as its input not the root itself, but an element whose semantic and phonological properties have been cashed out. Word-derived words thus have access only to the words they are derived from, not to the root. This work studies the Hebrew root, but at the same time the Root Hypothesis itself is universal. This allows us to make further cross-linguistic predictions.

240

R OOTS

IN

WORD-FORMATION

241

In particular, although English roots do not have the property of being assigned multiple meanings relative to different contexts, it is shown here that the distinction between word-formation from roots and word-formation from words applies equally to English. To illustrate this, I examine English zero-related pairs, which exhibit the same contrast between word-formation from roots and from words. This further motivates my initial claim, that in all languages the root is distinct from “words”, even if this distinction is not always morphologically manifested. The argument is built as follows. Section 7.1 places our discussion in the context of the debate on the status of word-formation, and succinctly recapitulates the relevant Hebrew data. In Section 7.2, the central claim is made: roots, but not words, may acquire multiple interpretations. The domain for assignment of interpretation is shown to be locally constrained by the first category head with which the root is merged. Sections 7.3 and 7.4 further extend and develop the claim made in Section 7.2, generalizing it to all root-derived elements. Section 7.3 extends the analysis to cases where no morphological cues are available, especially English zero-related pairs. Another group of wordderived elements, verb-derived nouns, is explored in Section 7.4. As the theory predicts, these nouns are dependent in their interpretation on the verbs from which they are derived. Having established the semantic differences between w root-derived and noun-derived Hebrew verbs, Section 7.5 points out important phonological differences between the two types of verbs. Going through Bat El’s argument against the root, I show how her findings actually support the distinction between root-derived and noun-derived verbs in Hebrew, which, in turn, further supports the existence of the root. Finally, Section 7.6 locates the discussion of denominals in a cross-linguistic context, reducing “lexical” differences among languages to specific choices made by languages with respect to their basic inventory and the possible combinations they employ in building verbs. But before we examine word-formation from roots and non-roots, a brief introduction to the relationship between syntax and word-formation is called for. 7.1. ROOTS AND WORD-FORMATION In this book, I adopted the hypothesis of syntactic word-formation. This view is, of course, controversial: word-formation is, more often than not, taken to be a component distinct from syntactic computation. Indeed, one of the main reasons for the distinction between word-formation and all other computation is what is known as the “double nature” of word-formation. While some aspects of w

242

C HAPTER 7

word-formation are morphologically productive and semantically transparent, others exhibit paradigmatic gaps and non-compositional meaning. This distinguishes word-formation from syntactic computation, which is always productive and compositional. To account for this “double nature”, many theories postulate two places for word-formation: one for the regular, productive processes, another for the non-productive ones. Whether they take word-formation to occur at two grammatical components (lexical vs. syntactic) or at two different levels of the same component (derivational vs. inflectional, level one vs. level two), all theories share the following intuition: “lower” (lexical, derivational, level one) wordformation is associated with idiosyncrasies (gaps, non-compositional meanings), while “higher” (syntactic, inflectional, level two) word-formation is productive and regular. Recent developments challenge the two-place approach for word-formation both empirically and theoretically. First, many processes that fell, traditionally, under word-formation, were shown to be syntactic in nature. To mention three influential cases: Baker’s (1988) work on incorporation as head movement, Pesetsky’s (1995) work on zero morphemes, and Hale and Keyser’s (1998, 2002) on the syntactic nature of “lexical items”. Furthermore, within the framework of distributed morphology (see especially Marantz 1997), it has been pointed out that, both empirically and theoretically, there is no need for distinction between “word-formation” and other types of computation: a single engine performs all computation, whether of small (words) or large elements (phrases and sentences). The data seem, then, somewhat contradictory: on the one hand, there is a clear intuition, based on gaps and productivity, in favor of distinguishing two types of word-formation. On the other hand, theoretical and empirical considerations motivate a single engine for all computation. Seeking to reconcile the single engine hypothesis with the evidence in favor of two places for word-formation, Marantz (2000) proposes to reconstruct the two places for building words within the syntax. The crucial distinction holds between creating words from roots r (atomic elements, devoid of all functional material) and creating words from existing words, that is, from roots that are already merged with some wordcreating head.1 A category head x may thus merge either with a root, or with an existing word (a noun, a verb or an adjective):

1

Note that the divide between roots and non-roots does not correspond precisely to levels of locality postulated in other theories. Specifically, traditional derivational affixes attach to both root-derived and word-derived elements (e.g. curi-ous, heal-th vs. realiz-ation, realizational). See Marantz (2000) for discussion of productivity facts and identification of English roots.

R OOTS (1)

(a)

IN

(b)

x √

WORD-FORMATION x n, v, a…

x √

word-formation from roots

243

n,v, a…

word-formation from words

The root stands in a local configuration with the head bearing a category feature, n, v, or a (1a). When another head merges with this noun or verb, as in (1b), this head can only “see” the head below it, not the root. Marantz further argues that the contrasts that have been attributed to the distinction between lexical (derivational) vs. syntactic (inflectional) word-formation follow, in fact, from the distinction between roots and non-roots. “Lower” word-formation from roots may exhibit idiosyncrasies (this is the domain where the specific meaning of the root is fixed) and be non-productive (roots may select for their category √ forming heads arbitrarily, e.g. clums takes -y as its adjectival head while √ malic takes -ous). In word-formation from words, on the other hand, an affix merges with a word whose interpretation is already fixed in a given environment. It is therefore semantically transparent and productive. The same properties that characterize English root-derived words, gaps and non-productivity, were shown, in Chapter Six, to characterize root-derived Hebrew verbs. This uniform behavior supports, in turn, the universality of the Root Hypothesis. The universality of the Root Hypothesis is further motivated by the following claims: (1) Roots may be assigned a variety of interpretations in different morphophonological environments. This property is language specific, occurring in some languages but not in others. (2) The ability to be assigned multiple interpretations is strictly reserved for roots. Word-formation from words retains the semantic and phonology of the base word. This locality constraint is universal. The first claim was defended in detail in Chapter Three. The second claim is the center of this chapter. A prediction suggests itself immediately: root-derived and word-derived elements in Hebrew will differ in their range of interpretations. The latter will necessarily be tightly related in their meaning (in a sense to be made precise) to the word from which they are derived. This prediction is indeed confirmed. Noun-derived verbs are shown here to depend in their interpretation on the noun

244

C HAPTER 7

from which they are derived, while root-derived verbs may take on multiple, semantically varied interpretations. Before we get to the contrast between roots and words, let us briefly recapitulate the most important point made in chapter three regarding the interpretations available for roots in Hebrew. It is a characteristic property of Hebrew word-formation that a single root may appear in several nominal and verbal patterns, acquiring a different meaning in each of these patterns. In Chapter Three, we termed this phenomenon Multiple Contextualized Meaning (MCM), that is, multiple meanings assigned to a root in specific environments or contexts. These contexts, in Hebrew, were shown to be the nominal and verbal heads (miˇsˇqalim and binyanim) with which the root combines. Some examples of the wide range of interpretations that a single root acquires across verbal and nominal environments in Hebrew are given below: (2) a. b. c. d. e. f. (3) a. b. c. d. e. f. g. h. i. (4) a. b. c. d. e. f. g.

sˇmn CeCeC (n) CaCCeCet (n) CuCaC (n) CaCeC (adj.) hiCCiC (v) CiCCeC (n) √ xˇsb ˇ CaCaC (v) CiCCeC (v) hiCCiC (v) hitCaCCeC (v) maCCeC (n) maCCaCa (n) CCiCut (n) CiCCon (n) taCCiC (n) √ qlt CaCaC (v) hiCCiC (v) miCCaC (n) maCCeC (n) taCCiC (n) CaCCeCet (n) CeCeC (n)

sˇemen sˇamenet sˇuman sˇamen hiˇmin iˇ sˇimen

(oil, grease) (cream) (fat) (fat) (grow fat/fatten) (grease)

xaˇav xiˇev hexˇiv x xˇ hitxaˇev maxˇev x maxˇava x xaˇivut ˇ xeˇsˇbon taxˇiv x xˇ

“to think” “to calculate” “to consider” “to be considerate” “a computer/calculator” “a thought” “importance” “arithmetic/bill” “calculus”

qalat hiqlit miqlat maqlet taqlit qaletet qelet

(to absorb, receive) (to record) (a shelter) (a receiver) (a record) (a cassette) (input)

R OOTS

IN

WORD-FORMATION

245

For the following discussion, let us keep in mind the wide range of unrelated interpretations that may be assigned to roots in different environments. Such a degree of variation in meaning cannot be reduced to a derivational relation between two words, but is easily explained if we take all the words in each group to be derived from a common root.

7.2. ROOT-DERIVED VERBS AND NOUN-DERIVED VERBS While most Hebrew verbs are derived from roots, there is a significant group of verbs that are derived not from roots, but from existing nouns or adjectives. Noun-derived verbs are similar to root-derived verbs in many respects. They carry the same verbal morphology of tense and agreement, and, most importantly, they must appear in the form of a binyan (see Chapter Two). There is one important difference between root-derived and noun-derived Hebrew verbs: the option to take on multiple interpretations in different environments is strictly reserved for root-derived verbs. Noun-derived verbs must share an interpretation with the noun from which they are derived. In other words, roots, but not nouns, may be assigned MCM when appearing in different patterns. √ To illustrate this point, consider the interpretations assigned to the root sgr in various verbal and nominal environments: √ (5) sgr a. CaCaC (v) sagar (v, close) b. hiCCiC (v) hisgir (v, extradite) c. hitCaCCeC (v) histager (v, cocoon oneself) d. CeCeC (n) seger (n, closure) e. CoCCayim (n) sograyim (n, parentheses) f. miCCeCet (n) misgeret (n, frame) From the noun misgeret (frame, 5f) a new Hebrew verb, misger (to frame) is formed: (6) a. b.

miCCeCet misgeret (a frame) CiCCeC misger (to frame)

Consider the relation between the noun misgeret (frame) and the verb derived from it. The verb bears a morpho-phonological similarity to the noun: it contains not only the root consonants, sgr, but also the prefix m-, which is carried over from the nominal pattern (miCCeCet) into the verbal form. The presence of this prefix overtly indicates the nominal origin of this verb. The vowels of the verb, on the other hand, are those typical of the verbal pattern CiCCeC, not of the nominal pattern in which the noun appears (see discussion in

246

C HAPTER 7

Chapter Two).2 I assume that the formation√of the verb misger (to frame) is as follows. First, the consonants of the root sgr are combined with the nouncreating morpheme, phonologically spelled out as miCCeCet (7a). The noun misgeret (frame) is then embedded under a v head (7b):3 (7)

a.

Root-derived noun: Nmisgeret

√sgr

NmiCCeCet b.

/misgeret/

Noun-derived verb: Vmisger

VCiCCeC

Nmisgeret

Nmisgeret

√sgr

/misger/

But noun-derived verbs differ from root-derived verbs not only morphologically, but also semantically. It may seem natural or even trivial that the verb made from the noun frame means to frame. But this, note, is a crucial property of noun-derived verbs. While root-derived verbs may pick up numerous interpretations in different environments, noun-derived verbs are tied to the meaning of the com√noun from which they are derived. To illustrate this point, √ pare the root sgr in (5) with the noun misgeret (frame) in (6). The root sgr is assigned numerous interpretations in√different environments, but when the basis for the derivation is not the root sgr itself but a noun derived from it (misgeret), that noun seems to force its meaning on any element further √ derived from it. Although the verb misger contains the consonants of the root sgr, it 2

In the active, the vowel melody, i/e, does not differ from that of the nominal pattern, but in the passive, the verb acquires the vowel melody u/a, typical of passives: musgar (was framed). The truncation of the nominal suffix -et, as opposed to the prefix m- that is carried into the verb, is yet to be explained. If the formation of verbs out of nouns involves stem modification, as argued by Bat El (1994), then perhaps m- is taken to be part of the stem, and therefore has to be syllabified, w while -et is a separate affix which could be truncated by Stray Erasure. Cf. Bat El (ibid.) for other cases of truncated final syllables. 3 The phonological realization of this process involves melodic overwriting, that is, matching the vowels of the stem with those of the third pattern, CiCCeC. See Bat El (1994) and discussion in Section 7.5 of this chapter.

R OOTS

IN

WORD-FORMATION

247

cannot have access to the underspecified core meaning of the root, or to any of the interpretations assigned to that root in different environments. Something √ seems to interfere between the verb misger and the root sgr. This interfering element, I argue, is the noun misgeret. Why should the presence of the nominal projection interfere between the root below it and the verb above it, not allowing the verb any access to the root? Recall that roots are assigned their interpretation locally, that is, when they stand in local relation with a certain head that serves as an environment for fixing their interpretation. Following Marantz (2000), I postulate the following locality constraint on the interpretation of roots: (8)

Locality constraint on the interpretation of roots: roots are assigned an interpretation in the environment of the first category-assigning head with which they are merged. Once this interpretation is assigned, it is carried along throughout the derivation.

The generalization above seems to be correct, but it would be theoretically desirable to derive it from an independently needed principle. The claim is that the first category head that merges with the root defines a closed domain for interpretation. Cyclicity and closed domains play an important role in most areas of generative grammar: phonology, morphology, and syntax. The locality constraint above could thus be formulated in a number of ways: it could be stated structurally, but in the lexical component, as a condition governing complex words; it could be formulated as a purely lexical rule, postulating different levels in the lexicon. However, there are some advantages to postulating this locality condition as a syntactic condition, in accordance with the single-engine hypothesis adopted here, where all computation is done by the computational h system. Such an account requires no further assumptions: the locality constraint on roots follows immediately from the independent syntactic notion of cyclicity. I will discuss the advantages of such an approach later in this section, as well as in Section 7.5. Let us now consider how this is done. If we follow the line of argumentation suggested in Marantz (2000), then the first category head merging with the root defines a phase (cf. Chomsky 1999), that is, a stage in the derivation where the element built by the computational system is spelled out both semantically and phonologically.4 The locality constraint in (8) then falls immediately out of the definition of the phase. Once the root has merged with the first category head, the product of 4

The assumption that the first category head merging with the root defines a phase is not trivial. In Chomsky (1999) phases are defined by specific heads (C, v, possibly D) and have an effect on movement. I explore here the possibility that any head that creates a semantic or phonological domain defines as a phase. It is possible that the phases that are relevant for movement are those distinguished as “strong phases” (cf. Chomsky 1999).

248

C HAPTER 7

the computation is sent off to the interface levels. The interpretation of the output (noun, verb, or adjective) in that environment is then fixed both semantically and phonologically. Whatever comes next in the course of the derivation will not merge directly with the root: it will combine with an element whose features have already been shipped to and interpreted at LF, being assigned there an interpretation in their specific context. Since the phase is a closed domain, any material above it cannot have any access to what is inside, and as a result, further derivational elements cannot alter the interpretation of a “word”, nor can it have any access to the root itself. The only alterations may be those modifications forced by the additional heads. For example, adding a v head to the noun misgeret (frame) changes its grammatical category, yielding a verb. Since the verb misger (to frame) is created from a noun, it cannot “look back” into the closed domain defined by the nominal head. Crucially, the verb cannot access the underspecified core meaning of √ the root, and take on any of the numerous interpretations assigned to the root √sgr (cf. 5 above). Instead, it is tied to the particular instantiation of the root sgr in the nominal environment of miCCeCet. That is, to the noun misgeret (frame) from which it is derived: (9)

Vmisger

V

V merges with the noun misgerett and has no access to √sgr

Nmisgeret

first category head N defines a phase. Interpretation is assigned to the noun at LF

Nmisgeret

√sgr

The locality constraint in (8), as well as the phase hypothesis, are strongly supported by the Hebrew data. In all cases of noun-derived verbs attested, the verbs are tied to their base nouns and have no access to the root. The nominal prefix m- is not unique to the pattern miCCeCet. It is carried into the verb in other cases of denominal verbs. Note the tight semantic relation between the noun and the verb derived from it:5 5

In most cases, those roots creating the base nouns are assigned multiple interpretations across √ patterns, none of which are available to the denominal verb. For example, xzr is assigned the following interpretations: √ (i) xzr CaCaC xazar (to return/repeat) CiCCeC xizer (to court) CCaCa xazara (rehearsal/return) maCCoC maxzor (cycle)Æ mixzer (CiCCeC, recycle)

R OOTS (10)

Root a. b. c. d. e. f. g.

√ √xzr sxr

IN

WORD-FORMATION

Nominal Noun Pattern maCCoC maxzor (cycle) miCCaC misxar (commerce)

249 Verb

mixzer (recycle) misxer (commercialize) √ xsb ˇ maCCeC maxsev ˇ (computer) mixˇsev (computerize) √ spr miCCaC mispar (number) misper (number) √ skn ˇ maCCoC ma skon ˇ (pawn) miˇ sken ˇ (pawn) √ 6 ysd miCCaC mosad (institute) mised (institute) √ dwr maCCoC mador (compartment) mider (compartmentalize)7

Another nominal √ prefix that often appears in denominal verbs is t-. Consider, first, the root xzq, which acquires multiple meanings in nominal and verbal environments: √ (11) xzq a. CaCaC xazaq (adj., strong) b. CiCCeC xizeq (v, to strengthen) c. hiCCiC hexziq (v, to hold) d. CiCCa xezqa (n, in math: power) e. CCaCa xazaqa (n, a hold, custody) f. CoCeC xozeq (n, strength) g. taCCuCa taxzuqa t (n, maintenance) h. CiCCeC tixzeq (v, to maintain, from taxzuqa, maintanance) √ One of the nouns derived from xzq, taxzuqa (maintenance), further creates the verb, tixzeq (maintain). This verb retains the nominal prefix √ t- and is tied semantically to the noun from which it is derived: while the root xzq creates words with varied meanings, the noun, taxzuqa (maintenance), can only form a verb whose √ meaning is maintain. Access to the various meanings associated with the root xzq is barred once the noun-creating head has merged with the root and fixed its specific interpretation in that environment. The nominal prefix t- is very often carried into the denominals verb, as the following examples illustrate: 6

Roots with initial glide y often exhibit phonological peculiarities when appearing in patterns; see the discussion in Section 7.5. 7 Note the phonological peculiarities of roots with middle glides.

250

C HAPTER 7

(12)

Root a. b. c. d. e. f. g.

√ qˇsˇr

Nominal pattern tiCCoCet

Noun

tiqˇsoret ˇ (communication) √ √ʕ sh taCCiCiya taʕ asiya (industry) gbr tiCCoCet tigboret (reinforcement) √ rwm taCCuCa truma (donation) √ qcb taCCiC taqciv (budget) √ pqd taCCiC tafqid (function) √ rgl taCCiC targil (exercise)

Verb tiqˇser (communicate) tiʕ es (industrialize) tigber (reinforce) taram (donate)8 tiqcev (budget) tifqed (to function) tirgel (exercise)

Note the variety of patterns with the t- prefix: taCCiC, tiCCoCet, taCCuCa, taCCiCiya. As in the case of misgeret (frame, miCCeCet) above, the prefix tis taken into the verbal form, while suffixes are erased (this indicates, perhaps, that prefixes and suffixes have a different status in nominal patterns: prefixes make part of the stem, while suffixes do not). In the case of t-prefixed nouns, too, most roots are assigned multiple interpretations, √ to which the denominal forms have no access. One representative example, qcb, illustrates this: √ (13) qcb a. CaCaC qacav (v, to ration) b. hiCCiC hiqciv (v, to allot) c. CaCCaC qacav (n, butcher) d. CiCCa qicba (n, pension) e. CeCeC qecev (n, beat, rhythm) f. miCCaC miqcav (n, tempo) g. taCCiC taqciv (budget) h. tiqcev (to budget, CiCCeC, derived from taqciv, budget) In some cases, nominal suffixes, too, are carried into denominal verbs. Consider √ the root xˇsb, ˇ and the noun xeˇsˇbon (bill, calculus) derived from it: √ (14) xˇsb ˇ a. CaCaC xaˇav (think) b. CiCCeC xiˇev (calculate) c. hiCCiC hexˇiv x xˇ (consider) d. hitCaCCeC hitxaˇev (be considerate) e. taCCiC taxˇiv x xˇ (calculus) f. maCCaCa maxˇava x (thought) 8

Note the phonological peculiarities of roots with middle glides in the nominal pattern.

R OOTS g. h. i. j.

maCCeC CiCCeC CiCCon hitCaCCeC

IN

WORD-FORMATION

maxˇev x mixˇev x xeˇsˇbon hitxaˇsˇben

251

(computer) (computerize, from maxˇev x , computer) (account, arithmetic, calculation, bill) (settle accounts with someone, from xeˇsˇbon, account)9

√ The root xˇsb ˇ forms various nouns and verbs. But the verbs mixˇev x (14h, note the nominal prefix m) and hitxaˇsˇben (14j, note the nominal suffix, -n), derived, respectively, from the nouns maxˇev x (14g), and xeˇsˇbon (14i), depend in their interpretation on their base noun. This is the only case I am aware of, of a nominal -on suffix entering a verb. The suffix -an, which w is very common in adjectives, appears on many deadjectival verbs as evidence for their origin (and, since the vowel is replaced by that of the verbal pattern, appears similar to -on): √ (15) a. qamcan (miser, from √ qmc) Æ hitqamcen (act miserly) b. sˇaxcan (arrogant, from sˇxc) Æ hiˇsˇ taxcen (act arrogantly) √ c. baxyan (complainer, from√ bkh) Æ hitbaxyen (complain) d. xucpan (impudent, from xcp) Æ hitxacpen (act impudently) Hebrew denominal verbs were shown here to differ from root-derived verbs with respect to their variety of interpretation. While the constraint on the interpretation of roots could be formulated as a lexical rule of a structural rule applying specifically to words, I suggested that it is best captured in syntactic terms, and argued that the first category head merging with the root creates a phase. This explains the semantic dependency between denominal verbs and their base nouns. It also explains the presence of the nominal prefixes in denominal verbs 9

A question arises regarding the semantic relation between two verbs derived from the same noun. Some Hebrew speakers also accept the verb xiˇsˇben (to keep accounts, make calculations), derived from the noun xeˇsˇbon (account, calculation). The crucial facts for the argument made √ here are two. First, recall that the root xˇsb ˇ has an underspecified meaning, which gets incarnated in different nominal and verbal environments. The noun xeˇsˇbon (account) is such an incarnation. Both verbs derived from xeˇsˇbon, xiˇsˇben (keep accounts), and hitxaˇsˇben (settle accounts with someone), have access to, and contain in them, the meaning of the noun from which they √ are derived ((xeˇsbon ˇ , account), namely, of the particular instantiation of the root xˇsb ˇ in the environment of CiCCon. Second, and most importantly, neither of these noun-derived verbs √ have any access to the other instantiations of the root r xˇsb ˇ in other environments—maxˇava x (thought), taxˇiv x (calculus), maxˇev xˇ x (computer), etc. These are precisely the predictions made here regarding such cases. Note that we do not predict that two verbs derived from the same noun should be identical—only that they will both include the meaning component of their base noun (A similar case in English involves two verbs derived from the noun seed: seed a field—a location verb, and seed a watermelon—verb of removing. While both are related to the noun, their meanings are not identical).

252

C HAPTER 7

and their absence in root-derived verbs: denominal verbs only have access to the phonological output of the phase, which includes the nominal prefix. Note, furthermore, that the same structural domain—the first category head merging with the root—constrains both phonological and semantic interpretation. Why should these semantic and phonological domains coincide? Under a syntactic approach to word-formation, this convergence is straightforward. The phase is the point of semantic and phonological spell-out. Everything that merges above that point takes as its input the semantic and phonological incarnation of that element. Note that under the phase-based theory, the opposite case is also predicted: not only should denominal verbs resemble their base nouns, but also, root-derived verbs may have phonological peculiarities, similarly to their semantic peculiarities. Such phonological peculiarities should not occur with denominal verbs, which have no access to the root consonants. Anticipating the discussion in Section 7.5, we can say that the phonological peculiarities associated with root-derived verbs never occur when the basis for the derivation is an existing word. So far we were concerned with noun-derived verbs, but the locality constraint on the interpretation of roots applies equally to all non-root-derived elements. We therefore expect to the same contrast in noun-derived nouns and root-derived nouns in Hebrew. This is indeed the case, as illustrated in (16) and (17): √ (16) yld a. CiCoC yilod (newborn) b. CCuCa yeluda (birth rate) c. CCeCa leda (childbirth; initial root consonant y drops: ∗ yleda) d. CaCiC yalid (a native) e. CeCeC yeled (child) √ The root yld forms many different nouns. Any heads that further combine with the noun yeled (child, 16e), made from this root, take as their input the semantics of that noun, and not of the root: √ (17) a. yld + CeCeC yeled (n, child, boy)10 b. yeled + ut yaldut (n, childhood) c. yeled + ut + i yalduti (adj., childish, pertaining to childhood) d. yeled + ut + i + ut yaldutiyut (n, childishness) 10

Nouns such as yeled belong to the segolite group, and are phonologically special. Their singular forms bear non-final stress, which is unusual for Hebrew nouns. This stress is changed into final one in all suffixed forms. Except in the plural form ((yeladim), such nouns also exhibit vowel contraction in all suffixed forms (17b–f), which corresponds to the change in stress from initial ( (yeled ) to final.

R OOTS

IN

WORD-FORMATION

253

yeled + on yaldon (n, little child) yeled + a yalda (n, girl) yeled + hitCaCCeC hityaled (v, act like a child, be childish)

e. f. g.

Evidently, nominal and adjectival heads combining with the noun may alter some aspects of it. They may make an abstract noun out of a concrete noun (17b) or out of an adjective (17d), create a property from the noun (as does the adjectival head in 17c), form a diminutive (17e), or change gender (18f). But crucially, all these heads√operate on the basis of the noun yeled (child), and not on the basis of the root yld. They √ therefore retain the meaning of “child” and do not have access to the root yld or to any of the interpretations of that root in other environments; it is impossible to add any of these suffixes to the noun child and get, for instance, birthrate (16b).11 Finally, the locality constraint on root interpretation works trivially with borrowed nouns or verbs. Consider Hebrew verbs that are derived from foreign words: (18) a. b.

Foreign noun telephone click

Borrowed Hebrew Borrowed verb noun telefon tilfen, CiCCeC (to telephone) qliq hiqliq, hiCCiC (to click)

Borrowed verbs or nouns are semantically tied to the foreign word, and very rarely change their basic meaning in the act of borrowing.12 I take this as evidence that what is borrowed is not a root, but a noun or a verb, that is, a unit whose lexical and semantic properties have already been sent off to PF and LF w (more on this in Section 7.5). The contribution of Hebrew to the theory of word-formation is twofold. First, because Hebrew roots create several nouns and verbs in different morphophonological environments, Hebrew enables us to appreciate fully the striking contrast between the variety of interpretations associated with word-formation 11

Certain Hebrew suffixes may alter the meaning of the Hebrew words in the following way: mapa (table cloth) vs. mapit (serviette), beyca (egg) vs. beycit (ovum), mexona (machine) vs. mexonit (car), and miˇsˇpaxa (family) vs. miˇspaxt ˇ on (day care), see Bat El (1997). While true (note, though, that with the -it suffix the words formed may be seen as diminutives that have acquired a specialized meaning), this does not contradict the main claim made here, that further derivation of words does not have any access to the root r , nor to any of the interpretations assigned to that root in other environments. 12 Some cases of semantic change in borrowing exist. For example, the German word for mobile phone, handi, from English “handy”, or the Hebrew word buk, model’s portfolio, from English “book”. This suggests that such changes involve narrowing of meaning or specialization, where the borrowed word is used in a specialized sense.

254

C HAPTER 7

from roots, as opposed to the strict semantic dependency forced on wordformation from words. This contrast would not have been noticed if we only looked at English-type languages, where roots are not assigned multiple interpretations. Second, because the root and the exponent of the word-creating morpheme are distinct in Hebrew, morphological evidence for the nominal origin of noun-derived verbs is often available. This provides an overt manifestation for the derivational relation between certain nouns and verbs. But while the assignment of multiple meanings to roots is language specific, the distinction between root-derived and word-derived elements is universal. We therefore expect that even in languages where no morphological cues are available, we will find the same semantic contrast between word-formation from words and word-formation from roots. This is indeed the case, as will be illustrated for English zero-related pairs in the next section.

7.3. IN THE ABSENCE OF MORPHOLOGY: THE SEMANTIC PROPERTIES OF DENOMINALS Morphological cues are not always available to determine whether a verb is derived from a noun. Consider the following Hebrew pairs: (19) a. b. c. d. e. f.

Noun cevaʕ kis miˇxa i sid zefet galʕin

(paint/color) (pocket) (paste) (whitewash) (tar) (a pit)

Verb cavaʕ kiyes maˇax siyed zipet gilʕen

(to paint/color) (to pick-pocket) (paste) (to whitewash) (cover with tar) (to pit)

Two possibilities suggest themselves: either both the noun and the verb are derived from the same root, or the noun is derived from the root and the verb is derived from the noun. The case of (19) above is of course not specific to Hebrew, but rather, the general case. This is precisely the situation in English, where pairs of nouns and verbs are often zero-related—hammer, tape, kiss, w button, anchor, etc. (see Clark and Clark 1979), with no morphology indicating whether the noun and the verb are derived from a common root or whether one w member of the pair is derived from the other. Based on the universality of the constraint on the interpretation of roots, I argue that in the absence of morphological cues, semantic cues can distinguish between root-derived and noun-derived verbs. Following Kiparsky (1982a), I suggest that English verbs that are zero-related to nouns are not a uniform group. Rather, they may be either root-derived or noun-derived. The crucial

R OOTS

IN

WORD-FORMATION

255

observation made by Kiparsky is that English verbs that have zero-related nouns differ with respect to the relation they bear to these nouns. Specifically, such verbs fall into two groups, as illustrated below: (20) a. b. c. d.

I paddled the canoe with a copy of the New York Times. String him up with a rope! She anchored the ship with a rock. He hammered the nail with a rock. (Kiparsky’s 1982a example 14)

(21) a. b. c. d.

∗

She taped the picture to the wall with pushpins. They chained the prisoner with a rope. ∗ Jim buttoned up his pants with a zipper. ∗ Screw the fixture to the wall with nails! (Kiparky’s 1982a example 16) ∗

A clear distinction is drawn between the two types of verbs above. The verbs in (21) entail the existence of the corresponding noun—there is no way to tape, chain or button without using tape, a chain or a button. In (20), on the other hand, the meaning of the verb does not entail that of the noun. As Kiparsky notes, to tape roughly means “apply tape”. To hammer, by contrast, does not mean “strike with a hammer”, but rather, “to strike with a flat surface of a solid object”. One can hammer not only with a hammer, but also with a rock or a shoe. But one cannot tape using anything other than tape. The difference between hammer type and tape type verbs is also apparent in the range of interpretations associated with the first group, but not with the second. Thus, there is hammer out an argument, hammer out a revised version of the paper, and hammer away. These interpretations share the manner component of the verb hammer—they r refer to fashioning, shaping, or resolving issues—but they are not semantically tied to the instrument “hammer”. Such an interpretation is very rarely associated with tape type verbs, where the verb invariably means “perform an action using the entity”—tape, button, bicycle, etc.13 13

Two notable exceptions to this are chain up, which w does not involve a chain, and ladle out money, w which does not involve a ladle. It could be that noun-derived verbs may, in the course of time, develop meanings that are independent from that of the noun. Note that, historically, tape type verbs may turn into hammer type verbs. One such example is the French verb sucrer. Historically (according to the Petit Robert Dictionary) it is derived from the noun “sucre”, sugar. In present day French, though, it is being used as English “sweeten”, and does not imply necessarily the use of sugar: one can “sucrer” with honey, saccharine, etc. Perhaps there is an asymmetry between tape type verbs and hammer type verbs, in that hammer type verbs should not evolve into a verb whose action must imply using a hammer. If this asymmetry turns out to be correct, it may be the result of a possibility to reanalyze a noun as a root, thus using sucrer with no reference to sugar.

256

C HAPTER 7

Following Kiparsky, I assume that in (20) the zero-related nouns and verbs are independently derived from a common root, while in (21), the noun is derived from the root and the verb is further derived from the noun.14 Consider hammer-type verbs, where both the noun and the verb are derived from the same root: (22)

a.

b.

V

N

√hammer

V

√hammer

N

√ The root hammer, from which both the noun and the verb are derived, has a core meaning of a certain manner—shaping, r forming, etc. This core meaning is manifest in both the noun and the verb derived from the root. The default interpretation of roots in verbal√environment is that of an action or an event. In a verbal environment, the root hammer is interpreted as an action performed in a certain manner (the precise manner, hammering, is given by the properties specific to the root, cf. Levin 1999). In a nominal √ environment, the default interpretation of roots is that of an entity. The root hammer, when embedded in a nominal environment, it is interpreted as an entity, or an instrument used in that particular manner (the specific entity, a hammer, is determined by the root). It is expected that the noun hammer and the verb hammer will have close meanings—both are derived from the same root. But because they are both derived from a root r , none entails the existence of the other.15 Now consider the tape type group, where the verb is derived from the noun: (23)

a.

N

N

V

b.

√tape

V

N

N 14

√tape

I differ from Kiparsky in assuming that the derivation in question takes place not in the lexicon, but in the syntax. Instead of ordered levels in the lexicon, there is structural hierarchy in the syntax, defined by the first category head that merges with the root. 15 The diagnostics and intuitions are much sharper with concrete nouns. Thus, manner-instrument verbs (tape, button, bicycle), location verbs (sugar, salt, frame), and verbs of removing (stalk, pit) clearly indicate the relation between the verb and the noun. In cases such as love, hate, or r request (historically, noun-derived) and exit, produce, or release r (historically verb-derived), the distinction is less clear.

R OOTS

IN

WORD-FORMATION

257

In this case, the noun denotes an entity, and the corresponding verb refers to some activity, which necessarily involves that entity. The specific type of activity will depend on the particular noun involved—tape, chain, or button, as well as on the specific syntactic structure of the verb (e.g. bicycle is a motion verb, tape—location verb).16 There is an implicit assumption in Kiparsky (1982a) that the meaning of the noun enters into the meaning of the verb when the verb is noun-derived. The reason why this should be the case is not discussed. Let me make this claim explicit. Word-derived words take as their input something whose meaning has already been fixed (cf. 8). Therefore, the meaning of the noun must be carried into the meaning of the noun-derived verb. A word of explanation is required for the step we just made. In Section 7.3, I argued (based on morphological evidence) that a verb that is structurally derived from a noun also depends on it semantically. This analysis was now extended to cases where no morphological cues exist. The basis for this extension is the assumption that when a verb semantically depends on a noun, it must be structurally derived from it. In other words, the entailment works both ways: a derivational relation between a noun and a verb entails a semantic relation between them, and a semantic relation between a noun and a verb entails a derivational relation between them. This follows from the theoretical assumptions made so far, as well as the following hypothesis: (24) Each specific interpretation of the root is only available in a specific (syntactic and phonological) environment. (24) is an extension of an assumption present in practically any theory: semantic polysemy aside, the relation between sound and meaning is unique. If the sound /dog/ refers to a dog, then the sound /kæt/ does not refer to it. (24) refers to roots r , taking into account the multiple environments where√roots may be assigned an interpretation. In English, this means that the root hammer is assigned the meaning of the instrument hammer only in a nominal environment, not in a verbal one, and the interpretation of an activity (in a certain manner) only in a verbal environment. In Hebrew, where a single root may form numerous nouns and verbs, (24) claims that each nominal or verbal environment is associated √ with one specific interpretation. For example, in (4) above the root qlt is assigned the interpretation shelter only in the nominal environment miCCaC, and receiver r r—only in the nominal environment of maCCeC. √ Let us see how (24) applies to denominal verbs. Recall that the root tape does not equal the noun tape: the root is only interpreted as an entity (tape) w when 16

Note that the difference between hammer type and tape type verbs is based on synchronic evidence, concerning speakers’ intuitions, and not on historical and etymological evidence.

258

C HAPTER 7

√ put in a nominal environment. There is no way to get directly from the root tape to the verb to tape, w which entails the meaning of the noun tape, without going first through a nominal projection. Since roots acquire their interpretations as nouns only when embedded in nominal environment, whenever a verb entails the meaning of a noun it cannot be derived directly from the root. There must be a nominal projection intervening. This is particularly evident in Hebrew, as seen √ in (5)–(6) above. The root sgr may be assigned many different interpretations; the interpretationn frame is only available √ in the nominal environment miCCeCet. There is no way to get from the root sgr to the verb misger (frame) without going first through the noun misgeret (frame), as is evident by the nominal prefix it carries. The verb must be derived from that particular noun—not from any other noun that the root creates (e.g. seger, closure, sograyim, parentheses). In contrast, root-derived verbs will never entail the meaning of their corresponding (root-derived) noun—because the interpretation of the noun is only available at the nominal projection, which is lacking in that case. Given the assumptions made here, we would expect to find not only semantic, but also phonological evidence that should distinguish between root-derived and noun-derived English verbs. Such evidence does exist, and it concerns stress assignment in zero-related pairs. There are several correlations that hold between stress assignment and semantic relation in zero-related pairs. First, as has been observed initially by Hayes (1981), as well as by Kiparsky (1982a), Raffelsiefen (1983), and Myers (1984), when zero-related pairs share the same stress pattern, there tends to be a tight semantic relation between the noun and the verb (25). When stress assignment is different in the verbs and the nouns, the semantic relation between the two is in most cases quite tenuous (26): (25) d´´ısciplineN,V , có´ ntacttN,V , dó´ cumenttN,V , expé´ rimenttN,V , bá´ lanceN,V , có´ mmenttN,V , hé´ ralddN,V . (Myers 1984) (26) ré´ corddN /recó´ rdV , sú´ bjecttN /subjé´ ctV , o´ bjecttN /objé´ ctV , pró´ jecttN / projé´ ctV , ré´ belN /rebé´ lV , digé´ sttN /d´´ıgestV , có´ nducttN /cond´ ductV , d´ a´ bstracttN /abstrá´ ctV , có´ mbineN /comb´ıneV , e´ xploittN /expló´ itV , pró´ duceN /prod´ duceV , dé´ fecttN /def´ fectV , có´ ntesttN /conté´ stV , có´ nvicttN / f´ conv´´ıctV , có´ ntracttN /contrá´ ctV . (Myers 1984, Raffelsiefen 1993) Under the assumptions made here, the convergence of stress difference and freer semantics vs. similar stress and tight semantic relation is hardly surprising. It is immediately derived if we assume that the difference between the two groups lies in the relation between the verb and the noun. In (25), the verbs are derived from the nouns. They therefore share the typical nominal stress in English (recall they take as their input both the semantics and the phonology of their base noun) as well as the interpretation of the noun. In (26), on the other hand,

R OOTS

IN

WORD-FORMATION

259

both the noun and the verb are derived from a common root. They are assigned typical nominal and verbal stress, respectively, and are assigned meanings that are sometimes far apart. Thus, the relation between a combine (a machine for harvesting grain) and the verb combine, between the noun defect and the verb to defect, or between the noun subject and the verb to subject is quite free. This is opposed to zero-related pairs that share their stress pattern—discipline, experiment, comment, etc.—where the semantic relation between the two is clear. Further, phonological evidence distinguishing English zero-related pairs emenates from the triplets of zero-related forms. As noted by Kiparsky (1982a) and Myers (1984), many nouns that have zero-related verbal forms may further create zero-derived verbs: (27) a. b. c. d. e. f.

perm´´ıtV —pé´ rmittN aff´ fıxV —á´ ffixN proté´ stV —pró´ testtN digé´ stV —d´´ıgesttN compó´ undV —có´ mpounddN contrá´ ctV —có´ ntracttN

Æ Æ Æ Æ Æ Æ

pé´ rmitV a´ ffixV pró´ testV d´´ıgestV có´ mpoundV có´ ntractV

In such cases, there is a striking difference between the two verbs in each triplet. The verb that does not share the stress pattern with the noun also has a different interpretation, while the verb that shares the nominal stress is also semantically dependent on that noun. I take this as evidence that the verbs with the nominal stress pattern, such as affix ´ and pé´ rmit, are derived from their corresponding noun, while verbs such as aff´x f and perm´t ´ are root-derived. So far we were concerned with a particular group of English zero-related pairs, the group Kiparsky calls noun-instrument pairs. But the same semantic criteria hold for other lexical groups. I will illustrate this with two additional such groups, location and locatum verbs, and show that the same semantic contrast appears within each group. Consider first location verbs: (28) a. b. c. d. e.

They housed their guests in their barn/their castle. John shelved the books on the mantelpiece. ∗ John boxed the apples in his bag. ∗ John bottled the vodka in the jar. ∗ John corralled the horses into the stable.

The verbs in (28) are all location verbs (naming the location into which the object goes, cf. Levin 1993), yet they differ with respect to the semantic relation the verb bears to the corresponding noun. One can shelve not only on a shelf, but also on the windowsill or the mantelpiece. To box o , on the other hand, means “put in a box”—and not in any other location (bag, bottle, etc.). This difference

260

C HAPTER 7

is immediately explained if we assume that (28a and b) are root-derived, hence the verb (e.g. house) does not necessarily entail the meaning of the noun, while (28c and d) are noun-derived, hence to jar or to box must mean put in a jar—or a box—respectively. Locatum verbs, that is, verbs naming the entity that is being located or applied, show the same distinction: (29) a. b. c. d.

She dusted the cake with sugar. She powdered her face with crushed chalk. ∗ She sugared her tea with jam. ∗ She starched the collar with chemicals.

Again, the contrasts in (29) are immediately derived if we assume that dust and powder (29a and b) are root-derived, defined independently of the noun: one can powder using crushed chalk. Similarly, the verbs in (29c and d) are derived from their corresponding noun: to starch entails “apply starch”, to sugar—apply r sugar (same for salt, pepper, flour).17 We may now return to the Hebrew verbs in (19) and show that they behave like their English counterparts. Consider the pair paint (19a) and whitewash (19d): (30) a.

b.

et ha kirot be Dani cavaʕ Dani painted OM the walls with Dani painted the walls with varnish. ∗ Dani siyed et ha kirot be Dani whitewashed OM the walls with Dani whitewashed the walls with varnish.

laqa varnish laqa varnish.

cavaʕ only specifies “add color”: the material used is not part of the denotation of the verb. Like English paint, the verb may include a manner component. That is, it may be used for application of material with some instrument (brush, crayon) or in a particular manner (i.e. splashing material all over the surface is not painting). The manner component, however, is optional, and in its absence the verb may take a non-agentive external argument (as in “The blood that spilled all over the place colored John’s shirt deep red”). The related noun, cevaʕ, can be whatever is prototypically used for coloring: liquid paint, a crayon, etc. The noun and the verb are related in their meaning, which is expected, given that 17

For lack of space I limit the discussion here to location and locatum verbs. Note, interestingly, that verbs of removing (Levin 1993), which are very similar to location verbs, seem to be all noun-derived. They include in their meaning the entity that is removed—a shell, pits or a stalk: (i)

a. b. c.

She shelled the lobster/the peas/the nuts (∗ the banana) She pitted the orange (∗ the melon) She stalked the artichokes

R OOTS

IN

WORD-FORMATION

261

√ they are derived from the same root r , cvʕ , but neither of these semantically entails the other. In contrast, siyed (whitewash) roughly means “cover with whitewash”. The noun sid refers to a specific material, whitewash. In this case, w the verb entails the noun, and is derived from it: one cannot whitewash using any material other than whitewash (cf. 30b). By applying the same criteria, we can distinguish the root-derived verbs in (19b and c) from the noun-derived verbs in (19e and f). Nouns and verbs derived from the same root often have more specialized meanings: the noun kis (19b) denotes a container in a general sense (not only a pocket in one’s clothes) while the verb kiyes has the specialized meaning, to pick-pocket (One can pick-pocket a wallet, a hand-bag, a pouch, etc.). The noun miˇxa can be any material with a paste texture, and the verb specifies a manner ( (paste ), but not the material applied. This stands in contrast to the nounderived verbs in (19e and f), where the meaning of the verb depends on that of the noun. One may not use the verb zipet (cover with tar) if the material applied is not zefet (tar), and one may only use gilʕen (pit) if the entity removed is a pit. To conclude: starting with the observation that zero-related nouns and verbs ffall into two groups, based on their relation to their corresponding noun, I argued that this semantic relation directly results from the difference in the derivation of the verb in each of these groups. In one group the verb and the noun are derived from the same root, in the other—the noun is derived from the verb. Both English and Hebrew distinguish between word-formation from roots and word-formation from words, but while in Hebrew nominal morphology often appears on noun-derived verbs, in English the morpho-phonological evidence is less direct. Crucially, the semantic effect is similar. In both languages nounderived verbs are tied in their interpretation to their base noun.

7.4. THE REMAINING PIECE: VERB-DERIVED NOUNS The discussion so far mentioned three groups: root-derived verbs, root-derived nouns, and noun-derived verbs. A fourth group suggests itself: verb-derived nouns. The question arises whether such nouns differ from root-derived nouns in the same way that noun-derived verbs differ from root-derived verbs. Nominal forms have received much attention in recent years, and a considerable amount of work has been dedicated to their typology (for some recent references see Engelhardt 2000, Alexiadou 2002, and references therein). The existence of verb-derived nouns is hardly disputed. Gerunds (destroying, growing, or reading r ) are universally taken to contain a VP layer in them, and event nominalizations (assignment, examination) are also assumed by many to contain a

262

C HAPTER 7

VP.18 As expected, verb-derived nouns depend in their meaning on the verbs they are derived from (e.g. examination and examining both share the meaning of examine). A detailed study of nominalizations is beyond the scope of this chapter (see references above, especially Marantz 1997 and Alexiadou 2002, for evidence distinguishing root nominalizations and verb nominalizations). I therefore concentrate here on two specific, less studied, groups of verb-derived nouns: English zero-related pairs and their Hebrew counterparts, where verbal morphology is overt. As expected, both types of verb-derived nouns depend in their interpretation on their base verb, as predicted by the locality constraint on the interpretation of roots. Consider, first, the following English pairs: (31) a. b. c. d.

Noun kiss, cough roast walk, jump slap, kick

Verb kiss, cough roast walk, jump slap, kick

At first glance, these pairs seem identical to those discussed in Section 7.4, such as tape or button. But the relation between the noun and the verb is different here. It is the noun that seems to be semantically dependent on the verb, in the following way: a kiss is an outcome of a kissing event—and owes its existence to the activity of kissing. A roast is the result of roasting something. A walk is a (temporally) bounded “piece” of walking activity. While noun-derived verbs include in them the entity specified by the noun, verb-derived nouns entail the activity denoted by the verb. There is no way to bring about a cough or a jump without the activity of coughing or jumping, just like one cannot use the verb tape if the material applied is not tape. The interpretation of the noun-derived as a result of the activity or as a temporally bounded activity seems to depend on the lexical class to which the verb belongs (cf. Harley 1999): (32) a. b. c. d.

18 19

Lexical class Noun interpretation change of state (roast, crack) result of change of state (a roast, a crack) activity (wipe) the smallest “piece” of activity (a wipe) semelfactive (kick, cough) the smallest “piece” of activity (a kick)19 motion verb (walk, jump) motion bounded in time/ quantity (a walk)

Cf. Borer (1990), Alexiadou (2002), and Engelhardt (2000). See Levin (1999) for the similarity between activities and semelfactives.

R OOTS

IN

WORD-FORMATION

263

If the verb is an action verb, the noun denotes a bounded piece of the event. This boundedness could either be arbitrary (e.g. a walk is any “a piece of walking” bounded in time, not necessarily with a specific destination), or it could be taken as a single event (semelfactive) in a series of activities (a kick is a single occurrence, the verb kick could be a single kick or a series of kicking), or a result r that is part of the meaning of the verb (the verb roast r could be interpreted as either an activity or an accomplishment, the noun roast r encodes the accomplishment—something that has been roasted). Noun-derived verbs of the “noun-instrument” class (e.g. pitchfork) were shown to have corresponding pairs where both the noun and the verb are rootderived (e.g. hammer). The question arises whether the verbs in (31) also have such corresponding pairs of the same lexical class (e.g. motion verbs, change of state verbs). While this matter cannot be pursued in detail here, I believe that such cases indeed exist. Consider the verb run and the noun a run. Unlike a walk or a jump, a run has a meaning independent of the motion verb run. That is, a run does not require a running event in the same way a walk requires a walking event. Similarly, a break does not depend semantically on the change of state verb break. Hayes (1981), Kiparsky (1982a), Myers (1984), and Raffelsiefen (1993) point out other cases where the semantics of the verb is quite independent of its zero-related noun. Remarkably, in all these cases there is a strong correlation between the semantic and phonological properties of the noun and the verb. Nouns that are assigned typical nominal stress are also semantically independent of their corresponding verb: (33) recó´ rdV /ré´ cord dN , perm´´ıtV /pé´ rm``ıttN , convé´ rtV /có´ nverttN , e´ xploittN / expló´ itV , pró´ duceN /prod´ duceV , dé´ fecttN /def´ fectV , có´ ntesttN /conté´ stV , f´ có´ nvicttN /conv´´ıctV , có´ ntracttN /contrá´ ctV The pairs in (33) are derived from a common root—which explains their relatively independent meanings. In other pairs, the nouns share the verbal stress pattern, indicating their verb-derived status, and are also semantically dependent on the verb: (34) consé´ nttN,V , rebú´ fffN,V , def´ feattN,V , divórceN,V , attá´ ckkN,V , debá´ teN,V , f´ dispá´ irrN,V , suppó´ rttN,V , demá´ nddN,V , escápeN,V , rewá´ rd dN,V , ref´ form mN,V , distréssN,V To conclude now the discussion of English zero-related pairs, note that a closer examination of this group forces a finer-grained analysis. These pairs are often treated uniformly due to their morphological properties.20 However, members 20

See, notably, Kiparsky (1982a), Myers (1984), and Raffelsiefen (1993) for a different view. Interestingly, the authors that advocated a difference in derivation were also those concerned with the phonological properties of this group. I believe that this is not a coincidence, as the

264

C HAPTER 7

of this set were shown here to fall into three sub-groups. In one group both the noun and the verb are derived from the same root (hammer, run). In another group, the noun is derived from the root and the verb is derived from the noun (tape); finally, there is a third group where the verb is derived from the root, and the noun is derived from the verb (kiss). This claim is at odds with some current work, which takes all zero-related pairs to be noun-derived (most notably, see Hale and Keyser 1998), but I believe that the distinction between these three classes is well grounded, given the semantic and morpho-phonological evidence brought up here. Hebrew provides further support for the existence of verb-derived nouns. Each Hebrew verbal pattern (with the exception of the two passive patterns) has a nominal pattern related to it: (35) a. b. c. d. e.

Verbal pattern CaCaC niCCaC CiCCeC hiCCiC hitCaCCeC

Related nominal pattern CCiCa hiCCaCut CiCCuC haCCaCa hitCaCCCut

These nominal patterns are related to their verbal counterparts both morphologically and syntactically: they retain the syllable structure and some of the prefixes of the verbal pattern, as well as the argument structure of the verb (Engelhardt 2000). Interestingly, many Hebrew nouns of the type kick, walk, etc. appear in verb-related nominal patterns: (36) a. b. c. d.

Root √ √str √bʕ t √hlk xbq

Verb satar (slap) baʕat (kick) halax (walk) xibeq (hug)

Related noun stira (a slap) beʕita (a kick) halixa (a walk) xibuq (a hug)

The nouns in (36) bear overt morphological relation to the correlating verb. As expected, they are also tied in their interpretation to that verb. The evidence for this semantic relation is in fact even stronger. Note the relation between two verbs derived from the same root through MCM, and their corresponding derived nouns: (37) a. b.

Root Verb Derived nominal √ btx biteax (insure) bituax (insurance) √ btx hivtiax (promise) havtaxa (a promise)

correspondence between sound and meaning in this case tends to coincide at the same structural level.

R OOTS (38) a. b. √

IN

WORD-FORMATION

265

Root Verb Derived nominal √ bˇ sl ˇ biˇ s ˇ el (cook) biˇsˇul (cookery, cooking) √ bˇsl ˇ hivˇ vsil (mature) havˇ vˇ vsala (maturing) vˇ

The root bˇsl ˇ acquires two verbal interpretations in the context of two verbal patterns, cook and ripen. The noun biˇsˇul, cookery, is semantically tied to the verb biˇsˇel (cook) from which it is derived. Crucially, the noun has access neither √ to the root bˇsl ˇ nor to any of the interpretations assigned to it in other contexts. Biˇsˇul (cookery) can never mean havˇ vsala (ripening), just as bituax (insurance) vˇ in (37a) cannot take on the meaning of promise, assigned to the root in another verbal context (37b). The relation between verbal and derived nominal patterns is not completely regular. In particular, nouns may appear in these patterns even when there is no corresponding verb (in which case they are derived from a root) or if the corresponding verb appears in another pattern.21 But this irregularity does not weaken the central generalization made here: if a noun appears in a verb-derived nominal pattern does have a corresponding verb, then it must be derived from that verb. In other words, while there may be some gaps in the morphological system (non-existing verbs or nouns), if a verb has a related nominal form, then this form is semantically and morphologically dependent on it.

7.5. BACK TO THE ROOT: THE PHONOLOGICAL PROPERTIES OF DENOMINALS Throughout the book we accumulated evidence in favor of the Hebrew root. Let us now consider the phonological argument against g the root, as pointed out most notably in Bat El (1994), as well as in Ussishkin (1999). The reason I bring up this topic at this point is that the phonological argument against the consonantal root is drawn entirely from denominal and borrowed verbs. The argument against the root runs as follows: there is sound phonological evidence that denominals and borrowed verbs are formed not from a consonantal root, but from an existing word. If we generalize this analysis to all Hebrew 21

Irregularities fall into three types. First, a verb-derived noun may appear not in the relevant, verb-derived pattern, but in another pattern, which is not specifically verb-derived: qacar, to √ harvest, from qcr, does not have ∗ qcira, but qacir (a harvest). In other cases, the noun-derived verb appears in a nominal pattern derived not from the matching verbal pattern, but from another pattern: sˇiʕul, a cough, appears in the nominal pattern derived from the (non-existing) verb ∗ sˇiʕel, w while the verb, hiˇta ˇ ʕel, appears in another pattern. Finally, a noun may appear in one of √ the verb-derived nominal pattern even when no corresponding verb exists. Thus, the root sˇpt, which creates the verb safat w ˇ (to judge) and its derived noun sfita ˇ (judging), forms also sˇiput (judgement), which does not have a corresponding verb, ∗ sˇipet.

266

C HAPTER 7

verbs, that is, if we assume that all Hebrew verbs are derived from bases, stems or lexemes, rather than from roots, we get a unified account of the Hebrew verbal system, with the notion of the consonantal root made redundant. As we shall see, the phonological evidence regarding denominals is indeed convincing. In fact, given what we know about denominal verbs, this is precisely what is predicted. Denominals and borrowed verbs were shown to differ from root-derived verbs with respect to their semantic interpretation. If, as I suggested, the nominal head that merges with the root creates a phase, then we should expect denominals to differ from root-derived verbs not only semantically, but also phonologically. Just as a denominal verb takes as its semantic input the noun from which it is derived, it also takes that noun as its phonological input, and therefore we expect that different phonological mechanisms should be involved in the formation of noun-derived and root-derived verbs. My argument here is twofold. First, I examine the phonological arguments against the consonantal root. These arguments, I claim, are in fact arguments supporting the claim that denominal verbs are formed on the basis of existing words rather than roots, and not arguments against the consonantal root itself. I then show that root-derived and noun-derived Hebrew verbs differ not only semantically, but also phonologically. This further motivates the structural distinction between root-derived and word-derived verbs. But before we turn the argument upside down, let us review Bat El (1994), where four sources of evidence in favor of the word-based approach for denominals are presented. First, Bat El notes that denominal verbs carry affixes that are typical of nouns, such as m-, t- (this has also been shown here in Section 7.2), showing that the basis for the derivation is the actual noun rather than the root. The second argument has to do with the way in which consonant clusters in foreign words are transferred into the Hebrew verb: cluster transfer in borrowed verbs preserves the original consonant cluster of the word from which they are derived. The arrangement of the consonants within the Hebrew verbal pattern differs according to the phonological form of the base, as the following examples show: (39) a. b. c.

Base transfer (transfer) streptiz (striptease) sinxroni (synchronic)

Derived verb trinsfer (to transfer) striptez (to perform a striptease) sinxren (syncronize)

Consonant clusters in the base word are kept together in the verbal form. Note that this is not due to phonological restrictions: Hebrew phonology would allow forms such as “tirnsfer”, “stirptez”, or “snixren”, where the cluster is broken.

R OOTS

IN

WORD-FORMATION

267

As Bat El points out, accounting for these patterns of consonant transfer is very difficult if we assume that those verbs are made of consonantal roots, but is immediately explained if we take these verbs to be derived from the borrowed noun. Furthermore, Bat El shows that borrowed and denominal verbs strongly tend to appear in the verbal pattern in which the phonological structure of the base word is best preserved. For example, when the base word has the vowel i in it, the verb will appear in the fifth pattern, which has the vowel melody i/i in the active voice: (40) a. b. c.

Base qliq (a click) fliq (a slap) sˇpritz (a splash)

Derived verb hiqliq (to click) hifliq (to slap) hiˇsˇpritz (to splash)

Finally, Bat El (1994) and Ussishkin (1999) show that when the base noun is monosyllabic, thus requiring some modification to make the base fit into the prosodic structure of the verbal pattern, the form of the denominal verb is predictable from the form of the nominal base. Bases containing the vowel o such as xoq (law) and qod (code) give rise to verbs with the less frequent vowel melody {o, e} (41a and b), while other short bases give rise to verbs containing the more common vowel melody, {i, e} (41c–e): (41) a. b. c. d. e.

Base xoq (law) qod (code) dam (blood) faks (fax) cˇ at (chat)

Verb xoqeq (legislate) qoded (codify) dimem (bleed) fikses (fax) cˇ itet (chat)

Based on this evidence, Bat El argues that borrowed and denominal verbs are formed by a process called Stem Modification (modifying the stem so as to fit the syllabic structure of the binyan), followed by melodic overwriting (replacing the vowels of the base word by those vowels typical of the verbal pattern), rather than by root-to-template association, as originally assumed in McCarthy (1981). Having established the Stem Modification analysis for borrowed and denominal verbs, Bat El then extends this analysis to all Hebrew verbs. It is at this point that I depart from Bat El’s assumptions. I argue that it would be wrong to generalize from word-derived verbs to all Hebrew verbs, precisely because of the property that characterizes these verbs, their being made from existing words. Category heads serve as the immediate environment for fixing

268

C HAPTER 7

the meaning of a root, where its semantic and phonological features are shipped off to the interface levels. Shipping off phonological features to PF means that the unit has to be phonologically incarnated. Similarly, it has no access to the phonological representation of the root, only to the actual phonological output at the phase level, √ that is, the noun or verb. Consider, for example, the representation of the root sgr (cf. (5)–(6) above), in the nominal environment miCCeCet: (42)

LF: interpretation: “frame” N

NmiCCeCet

√sgr

PF: pronunciation: /misgeret/

Because, after it merges with a category head, the derivation no longer has √ access to the phonological representation of the root sgr, any word-derived verb must be phonologically derived by manipulation on existing phonological structures. When a verb is formed from a foreign word, such as telefon, or from a root-derived noun, such as misgeret (frame), the interpretive component has access only to the meaning of the noun, and the phonological component has access only to the phonology of the noun. We W thus expect borrowed and denominal verbs to be formed by Stem Modification of the noun rather than by root-to-template association. This is the only choice the grammar has: it cannot extract the consonants of the root, to which it has no access; all it has is the existing stem. All the phonological properties of denominals—preservation of clusters, preservation of the base word, nominal morphology carried into the verb (resulting, according to Bat El, from the requirement that all the consonants of the base be syllabified)—are precisely what we expect, given their being derived from existing words. And this is also where these verbs differ from root-derived verbs. If the argument made here is on the right track, then we expect that rootderived words will have not only semantic idiosyncrasies, but also phonological idiosyncrasies. Furthermore, we expect that word-derived words will not exhibit such phonological idiosyncrasies. This turns out to be the case in Hebrew. Note the following phonological idiosyncrasies in certain Hebrew verbs. Initial n assimilates before a stop (43a), initial y assimilates before c (43b) middle glides are dropped, giving rise to a contracted form (43c and d), and two final identical consonants yield a change in vowels of some patterns (43e) and a contracted form in others (43f):

R OOTS (43) a. b. c. d. e. f.

Root22 √ √ncl √ycb √qwm √qwm √sbb sbb

IN

Pattern hiCCiC hiCCiC CaCaC hiCCiC CiCCeC hiCCiC

WORD-FORMATION

269

Verb hicil (save; ∗ hincil, cf. yinacel) hiciv (to position; ∗ hiyciv, cf. yicev) qam (rise; ∗ qawam) heqim (raise; ∗ hiqwim) sovev (turn around, tr., ∗ sibev)23 hesev (turn towards, tr., ∗ hisbiv)

The verbs exhibiting idiosyncrasies are those verbs that are traditionally taken to be formed from consonantal roots. On the other hand, there are Hebrew verbs which do not exhibit such idiosyncrasies. Crucially, those verbs are indepenw dently argued to be created from existing words. Examples are given in (44). If the base word contains an n, this consonant does not assimilate (44a and b). Medial glides are preserved (44c-d), even if not pronounced in the base word, and identical consonants do not give rise to a vowel change if they are part of the base from which the verb is derived (44e):

22

I refer here to “consonants” as to consonants of the (hypothetical) root, but note that our argument does not presuppose the existence of the root. For ease of presentation, example (43) lists the (hypothetical form of the) root, alongside the attested form of the verb (for assimilated forms, we represent other forms of the same verb in parentheses, where the assimilated consonant appears). The evidence upon which I rely here is not that of (43) alone, but of (43) and (44), taken together. My observation is not simply that the (hypothetical forms of) roots differ from the attested forms of verbs from which they are associated. Instead, it is that certain verbs show certain quirkiness in the phonology of their attested forms, which other verbs do not display. It is then striking that verbs of the first kind are those that, on other grounds, are likely to be root-derived, while those of the second kind are, on other grounds, likely to be word-derived. 23 There are certain Hebrew word-derived verbs with such vowel melody—xoqeq — (legislate) and koded (codify). Crucially, these verbs are derived respectively from the nouns xoq (law) and qod (code). The form of verbs such as xoqeq and qoded arises not from an idiosyncrasy of the root, but from the phonology of the word from which they are derived. As observed by Bat El (1994), word-derived words strive to preserve the phonological structure of their base word. This is also predicted by the theory advocated here, where the noun-derived verb has as its input the phonological form of that noun. Of course, in order for forms such as xoqeq to exist, Hebrew phonology must allow such variation in vowels. But it is important that only verbs derived from monosyllabic words containing o may have this form (note that other verbs derived from monosyllabic bases, such as qav, line, always have their vowels modified, as in the verb qivqev, draw a line). A further observation is that forms containing o vowels often have free variants with i. This is correct both for root-derived verbs (cf. sivev, colloquial variation of sovev, turn, and kided, free variation of koded, codify). I believe such modification takes place post-syntactically, at the level of PF, readjusting the output of phonology.

270

C HAPTER 7

(44) a.

b. c. d. e.

Base neged (opposite)

Pattern hiCCiC

Verb hingid

(to put in opposition; ∗ higid) necax (eternity) hiCCiC hinciac (make eternal; ∗ hiciax) tiq (file) CiCCeC tiyeq (to file; ∗ toqeq) dox (report; acronym) CiCCeC diveax (to report; ∗ doxeax) √ rax (soft, from rkk) CiCCeC rikex (to soften, ∗ roxex)

It is striking that phonological processes such as assimilation occur in verbs that are assumed to be root-derived (43a and b) and not in those that are independently argued to be word-derived (44a and b). If we retain our assumption, that the verbs in (43) are root-derived and those in (44) are word-derived, this ffact is immediately and elegantly explained. The derivation in (44a) has no √ access to the phonology of the root ngd, only to the word neged, a particular phonological incarnation of the root. This incarnation serves as the base for any further derivation, as in (44a), and no phonological alternations of that word (such as assimilating the sound n) occur. These phonological distinctions do not always divide root-derived and wordderived verbs neatly. For example, some root-derived verbs also keep their middle glides. In the third pattern, some root-derived verbs have two forms, one that drops the middle glide and one that retains it.24 In addition, certain phonological changes, such as metathesis of t with s, sˇ, and c in the hitCaCCeC pattern and changes in the final syllable when the final consonant is h occur both in root-derived and in word-derived verbs.25 But the crucial point here is that no word-derived verb shows the mutations typical of roots, such as those in (43). Root-derived verbs may yield phonological idiosyncrasies, just like they can yield semantic and lexical idiosyncrasies (i.e. specialized meaning in certain contexts), but word-derived verbs can only alter the lexical and phonological output of an already existing word, and therefore do not exhibit such idiosyncrasies. We are now in a position to examine the implications of the phonological data presented here for the theory of the root. In addition to the initial √ For example, the root qwm has, according to the dictionary, two forms in the pattern CiCCeC: qiyem and qomem. The first form retains the medial glide while in the second the final consonant is doubled to make up for the medial glide. 25 Both metathesis and final h involve changes at the edge of the stem, and are therefore perhaps not related to the association of the root with the pattern, but to the actual phonological output. 24

R OOTS

IN

WORD-FORMATION

271

observation of special meaning assigned to roots, we saw that root-derived words may also exhibit phonological idiosyncrasies. Word-derived words, by contrast, preserve the semantics and the phonological form of their base word. This phonological difference further motivates the initial distinction between root-derived and word-derived words. It also provides a strong argument supporting the existence of the consonantal root in Hebrew, by showing that one must not generalize from the behavior of denominal and borrowed verbs to all other Hebrew verbs. But these phonological data also support the theoretical idea explored here, concerning the single engine hypothesis, or syntactic word-formation, and the cyclic (phase-based) nature of that engine. Note that the first category head merging with the root defines a closed domain both semantically and phonologically: semantic and phonological peculiarities occur below, but not above that domain. Why should properties of sound and meaning converge at the same structural level? If word-formation is indeed performed by the syntax, and the output of the syntax may be spelled out at different points, or phases, then the convergence of the boundary for sound and meaning at the same structural level follows immediately. If the first category head that merges with the root creates a phase, then the properties of the syntactic construct are spelled out both semantically and phonologically. In this case, we expect the first category head merging with the root to be a domain for both special semantics and special phonology. While it is perfectly possible to account for these facts under a theory that assumes word-formation in the lexicon (e.g. by postulating morphological conditions on the structure of complex words), the coincidence of semantic and phonological domains is immediately and elegantly explained in a theory that assumes a cyclic, syntactic engine for word-formation.

7.6. ROOTS: BETWEEN THE UNIVERSAL AND THE LANGUAGE SPECIFIC Both English and Hebrew were shown here to be subject to the same universal locality constraint on the interpretation of roots. But the two languages also differ substantially. Hebrew roots may be assigned numerous interpretations in different morpho-phonological environments, while English roots lack that property (with the exception of Latinate bound roots combined with prefixes, as noted above). Let us concentrate for the moment on Hebrew-type languages. Given the locality constraint on the interpretation of roots, it is predicted that all languages where a single root may be assigned multiple interpretations will exhibit a sharp contrast between the variety of meanings assigned to the root as opposed to the strict dependency in meaning between word-derived elements

272

C HAPTER 7

and their base word. This is indeed the case, as will be illustrated here on Russian and Georgian. Consider Georgian first:26 √ (45) Root: cx a. acxobs (to bake, v) b. namcxvari (cake; literally, participle of bake) c. sicxe (fever, n) d. cxeli (hot, adj.) e. acxelebs (to heat, v, from cxeli, hot) √ (45a–d) are derived from the root cx. (45e), the verb acxelebs, is derived from the adjective cxeli (note which is carried into √ the adjectival suffix -el w the verb). While the root cx may be assigned several interpretations, the adjective-derived verb must depend in its meaning on that adjective. The same phenomenon exists in Russian, where a large number of roots may acquire radically different interpretations in the √environment of different prefixes. Consider one typical example, of the root kaz: √ (46) kaz a. skazat’ (say) b. r raskazat’ (tell) c. otkazat’sja (refuse) d. dokazat’ (prove) e. prikazat’ (order) f. okazat’sja (turn out) g. pokazat’ (show) The root may acquire many interpretations in combinations with different prefixes, but once its meaning has been assigned, it is retained throughout all further derivation:

26

(47) a. b.

prikazat’ prikaz

(v, to order) (n, an order, command).

(48) a. b. c. d. e.

pokazat’ pokazatel pokaz pokazuxa pokazatelniy

(v, to show, indicate) (n, an indicator, index) (n, a showing, demonstration) (n, a show) (adj., significant, revealing).

(49) a. b. c.

rrasskazat’ r rasskaz r rasskazcik

(to tell, narrate) (a story) (narrator)

Georgian examples are from Lé´ a Nash (p.c.).

R OOTS

IN

WORD-FORMATION

273

√ Once the root kaz has merged with the prefix pri or po, it no longer has any access to the meanings assigned to it in the environment of other prefixes, and all the words further derived from it also share the specific meaning of the root in the context of that prefix. The difference between English-type languages and Hebrew-type languages is not often noted. When variation between languages is discussed, most linguists concede that variation is located in the lexicon and in morphology (a view that, within generative grammar, originates with Borer 1984). What this is usually taken to mean is that, roughly, different languages have different words. Of course, the initial inventory of roots and features differs from one language to another. Languages may have a subset of the features made available by UG, and this subset is not identical in all languages (e.g. Russian, but not English, has a Prepositional case feature and grammatical gender). The inventory of roots is not uniform either. In sheer numbers, some languages, like English, have a large number of roots, while others, like Hebrew, have a relatively small one. Furthermore, roots themselves are not uniform: different languages cut the conceptual pie in different ways, specifying different signs, or roots. To take an example from kinship terms: Arabic has two signs, one for a maternal uncle, xaal, and another for a paternal uncle, ʕamm, w while in English there is only one sign, uncle. But all these variations do not cover the difference between English and Hebrew with respect to word-formation. In this chapter, I suggest an additional source for language variation: variation lies not only in the initial lexical inventory, but also in the number of interpretations assigned to each root in different morpho-syntactic and morphophonological environments. Hebrew makes available several interpretations for the same root, across different environments, while English normally allows at most one interpretation of the root for each category head, n, v, or adj. In Section 7.3, we noted yet another source of variation: the degree of freedom languages have in deciding what kind of verb is formed from a particular root. Even if two languages have similar roots, there could still be some difference in the verbs each language creates. For example, while English pocket is a verb of location, putting something in your pocket, its Hebrew counterpart, kiyes (to pickpocket), is a verb of removing. I assume that this difference is related to the type of (abstract) preposition present in such structures. Finally, it was shown that languages have the option either to derive both a noun and a verb from the root, or to derive first a noun, and from it a verb. In both English and Hebrew, within the same lexical class (e.g. instrumentmanner verbs), the verb may either be root-derived (hammer), or noun-derived ( (pitchfork ). Spelling out the set of possible choices available to languages yields a finergrained formulation of the widely held view, ascribing language variation to

274

C HAPTER 7

morphology. But the most important conclusion that arises from the Hebrew and English data is the universality of the lexical atom, the root. Independently of their morphological properties, all languages have roots that are distinct from the noun or verb-creating heads, and in all languages roots are subject to locality constraints on assignment of interpretation. No matter what their initial cards are or what moves they make, all languages play the same game—one in which the cards are roots, and the rules are locality constraints. w

REFERENCES Alexiadou, Artemis. 2002. Functional Structure in Nominals: Nominalizations and Ergativity, Mouton de Gruyter, Berlin. Anderson, Stephen. 1992. A-morphous Morphology, Cambridge University Press, Cambridge. Arad, Maya. 1998. VP structure and the syntax-lexicon interface. Doctoral dissertation, University College London (published by MIT Working Papers, MITOPL 16, 1999). Arad, Maya. 1999a. On the contents of v: evidence from the Hebrew binyan system. Paper presented at the Second Conference on the Syntax of Semitic languages, University of Illinois, Urbana-Champaign. Arad, Maya. 1999b. ‘On “little v”’, in Karlos Arregi, Benjamin Bruening, Cornelia Krause and Vivian Lin (eds.), MIT Working Papers in Linguistics 33, Papers on Morphology and Syntax, Cycle 1, pp. 1–25. Arad, Maya. 2002. ‘Hebrew lexical causatives’, in Jamal Ouhalla and Ur Shlonsky (eds.), Themes in Arabic and Hebrew Syntax, Kluwer, Dordrecht, pp. 241–266. Arad, Maya. 2003. ‘Locality constraints on the interpretation of roots: the case of Hebrew denominal verbs’, Natural Language and Linguistic Theory 21: 737–778. Arad, Maya and Ur Shlonsky. 2001. ‘Locality domains for interpretation: affixation above and below vP’. Paper presented at the ApplFest, MIT. Aronoff, Mark. 1976. Word-Formation r in Generative Grammar, MIT Press, Cambridge, Massachusetts. Aronoff, Mark. 1994. Morphology by Itself, f MIT Press, Cambridge, Massachusetts. Baker, Mark. 1988. Incorporation, University of Chicago Press, Chicago, Illinois. Bat El, Outi. 1994. ‘Stem modification and cluster transfer in Modern Hebrew’, Natural Language and Linguistic Theory 12: 571–596. Bat El, Outi. 1997. ‘On the visibility of word-internal morphological features’, Linguistics 35: 289–316. Bat El, Outi. 2001. In search of the roots of the root. Paper presented at the Root and Template Morphology Workshop, University of Southern California.

275

276

R EFERENCES

Beard, Robert. 1966. The affixation of adjectives in contemporary literary serbo-croatian. Doctoral dissertation, University of Michigan, Ann Arbor. Beard, Robert. 1995. Lexeme-Morpheme Based Morphology: A General Theory of Inflection and Word-Formation r , State University of New York Press, Albany. Berman, Ruth. 1975. ‘Rishumam ha-leksikali sel ˇ pealim, soraˇ ˇ sim ˇ u-vinyanim’ (Lexical entries for verbs, roots and patterns), in Rephael Nir and Ben Zion Fischler (eds.), Kilshon Amo, Council on the Teaching of Hebrew, Jerusalem. Berman, Ruth. 1978. Modern Hebrew Structure, T Tel Aviv University Press, Tel Aviv, Israel. Bobaljik, Jonathan. 2000. ‘The ins and outs of contextual allomorphy’, in University of Maryland Working Papers, vol. 10, pp. 35–71. W Bobaljik, Jonathan. 2001. ‘The implications of rich agreement: why morphology doesn’t drive syntax’, in Proceedings of WCCFL 20, pp. 82–95. Bobaljik, Jonathan and Hö¨ skuldur Thráinsson. ´ 1998. ‘Two heads aren’t always better than one’, Syntax 1: 37–71. Bolozky, Shmuel. 1978a. ‘Some aspects of modern Hebrew phonology’, in Berman (1978). Bolozky, Shmuel. 1978b. ‘Word-formation strategies in the Hebrew verbal system: denominative verbs’, Afroasiatic Linguistics 5.3: 2–26. Borer, Hagit. 1984. Parametric r Syntax, Foris, Dordrecht. Borer, Hagit. 1990. The syntax and morphology of derived nominals. Unpublished manuscript, University of Massachusetts, Amherst. Borer, Hagit. 1991. ‘The causative–inchoative alternation: a case study in parallel morphology’, The Linguistic Review 8: 119–158. Borer, Hagit. 1994. The projection of arguments, in E. Benedicto and J. Runner (eds.), Functional Projections, UMass Occasional Papers, pp. 19–47. Borer, Hagit. 1998. ‘Passive without theta grids’, in S.G. Lapointe, D.K. Brentari and P.M. Farrell (eds.), Morphological Interfaces, CSLI Publications, Stanford, CA, pp. 60–99. Borer, Hagit. 2005. Determining Structures, Oxford University Press, Oxford. Brown, Cecil H. 1999. Lexical Acculturation in Native American Languages, Oxford University Press, Oxford. Chomsky, Noam. 1970. ‘Remarks on nominalization’, in Roderick Jacobs and Peter Rosenbaum (eds.), Readings in English Transformational Grammar, Ginn, Massachusetts, pp. 184–221. Chomsky, Noam. 1995. The Minimalist Program, MIT Press, Cambridge, Massachusetts.

R EFERENCES

277

Chomsky, Noam. 1998. ‘Minimalist inquiries: the framework’, MIT Occasional Papers in Linguistics 15, MIT, Cambridge, Massachusetts. Chomsky, Noam. 1999. ‘Derivation by phase’, MIT Occasional Papers in Linguistics 17, MIT, Cambridge, Massachusetts. Clark, Eve and Herbert Clark. 1979. ‘When nouns surface as verbs’, Language 55: 767–811. Di Sciullo, Anna Maria and Edwin Williams. 1987. On the Definition of Word, MIT Press, Cambridge. Doron, Edit. 1999. The semantics of semitic templates. Unpublished manuscript, Hebrew University, Jerusalem. Doron, Edit. 2003. ‘Agency and voice: the semantics of semitic templates’, Natural Language Semantics 11: 1–67. Embick, David. 1998. ‘Voice systems and the syntax/morphology interface’, MIT Working Papers in Linguistics 32, Papers from the UPenn/MIT Roundtable on Argument Structure and Aspect, MIT, Cambridge, Massachusetts, pp. 73–118. Embick, David. 2000. ‘Features, syntax, and categories in the Latin perfect’, Linguistic Inquiry 31.2: 185–229. Embick, David and Rolf Noyer. 2001. ‘Movement operations after syntax’, Linguistic Inquiry 32.4: 555–595. Engelhardt, Miriam. 2000. ‘The projection of argument-taking nominals’, Natural Language and Linguistic Theory 18: 41–88. Even Shoshan, Avraham. 1978. Ha milon ha ivri he xadaˇ (the new Hebrew dictionary), Kiryat Sefer Press, Israel. Fodor, Jerry A. 1970. ‘Three reasons for not deriving ‘kill’ from ‘cause to die”, Linguistic Inquiry 1: 429–438. Fodor, Jerry A. 1994. The Elm and the Expert: Mentalese and Its Semantics, The 1993 Jean Nicod Lectures, MIT Press, Cambridge, Massachusetts. Fodor, Jerry A. 1998. Concepts: Where Cognitive Science Went Wrong, The 1996 John Locke Lectures, Oxford University Press, Oxford. Fodor, Jerry A. and Ernie Lepore. 1998. ‘The emptiness of the lexicon, critical reflections on James Pustejovsky’s The Generative Lexicon’, Linguistic Inquiry 29(2): 269–288 Gesenius. 1910. Gesenius’ Hebrew Grammar, edited by E. Kautzsch, revised by A.E. Cowley, Clarendon Press, Oxford. Glinert, Lewis. 1989. The Grammar of Modern Hebrew, Cambridge University Press, Cambridge.

278

R EFERENCES

Goshen-Gottshtein, Moshe, Zeev Livne and Shlomo Shapan (1967). Hadikduk haivri haˇimuˇ ˇ ˇ i, Schocken Press, Jerusalem. Graf, Dafna and Adam Ussishkin. 2002. Loanword phonology in Modern Hebrew. Unpublished manuscript, Heinrich-Heine-Universitä¨ t Dusseldorf ¨ and University of Arizona. Grimshaw, Jane. 1990. Argument Structure. MIT Press, Cambridge, Massachusetts. Hale, Ken and Samuel J. Keyser. 1998. ‘The basic elements of argument structure’, in Heidi Harley (ed.), MIT Working Papers in Linguistics 32, Papers from the UPenn/MIT Roundtable on Argument Structure and Aspect, MIT, Cambridge, Massachusetts, pp. 73–118. Hale, Ken and Samuel J. Keyser. 2002. Prolegomana to a Theory of Argument Structure, MIT Press, Cambridge, Massachusetts. Halle, Morris. 2000. ‘Distributed morphology: impoverishment and fission’, in Jacqueline Lecarme, Jean Lowenstamm and Ur Shlonsky (eds.), Research in Afroasiatic Grammar: Papers From the Third Conference on Afroasiatic Languages, John Benjamins, Amsterdam, P pp. 125–149. Halle, Morris and Alec Marantz. 1993. ‘Distributed morphology and the pieces of inflection’, in Ken Hale and Samuel. J. Keyser (eds.), The View from Building 20, MIT Press Cambridge, Massachusetts, pp. 111–176. Harley, Heidi. 1995. Subjects, events and licensing. Doctoral dissertation, MIT. Harley, Heidi. 1999. ‘Denominal verbs and aktionsart’, in Liina Pylkkä¨ nen, Angeliek van Hout and Heidi Harley (eds.), MIT Working Papers in Linguistics 35, Papers from the Upenn/MIT Roundtable on the Lexicon, MIT, Cambridge, Massachusetts, pp. 73–85. Harley, Heidi and Rolf Noyer. 1998. ‘Licensing in the non-lexicalist lexicon: nominalizations, vocabulary items and the encyclopedia’, in H. Harley (ed.), MITWPL 32, Papers from the MIT/Upenn Round Table on argument structure and aspect, pp. 119–137. Haspelmath, Martin. 1993. ‘More on the typology of inchoative/causative verb alternations’, in Bernard Comrie and Maria Polinsky (eds.), Causatives and Transitivity, John Benjamins, Amsterdam, pp. 87–120. Haspelmath, Martin. 2003. Against iconicity and markedness. Paper presented at the Linguistics Colloquium, Stanford University. Hayes, Bruce. 1981. A metrical theory of stress rules. Unpublished Ph.D. dissertation, MIT, distributed by Indiana University Linguistics Club. van Hout, Angeliek. 1996. Event semantics of verb frame alternations. Doctoral dissertation, Tilburg University. Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar, MIT Press, Cambridge, Massachusetts. Jackendoff, Ray. 1990. Semantic Structures, MIT Press, Cambridge, Massachusetts.

R EFERENCES

279

Jacobsen, Wesley. 1992. The Transitive Structure of Events in Japanese, K Kurosio Publishers, Tokyo. Kayne, Richard. 1975. French r Syntax, MIT Press, Cambridge, Massachusetts. Kiparsky, Paul. 1982a. ‘Word-formation and the Lexicon’, in Fred Ingeman (ed.), Proceedings of the Mid-America Linguistics Conference, University of Kansas, pp. 3–29. Kiparsky, Paul. 1982b. ‘Lexical morphology and phonology’, in Linguistics in the Morning Calm, Hanshin Publishing Company, Seoul, Korea. Kratzer, Angelika. 1996. ‘Severing the external argument from its verb’, in Johan Rooryck and Laurie Zaring (eds.), Phrase Structure and the Lexicon, Dordrecht, Kluwer, pp. 109– 137. Kripke, Saul. 1972. Naming and Necessity, Harvard University Press, Cambridge Massachusetts. Lakoff, George. 1970. Irregularity in Syntax, Holt, Rinehart and Winston, New York. Levin, Beth. 1993. English Verb Classes and Alternations, The University of Chicago Press, Chicago, Illinois. Levin, Beth. 1999. ‘Objecthood: an event structure perspective’, in Proceedings of CLS 35, vol. 1, The Main Session, Chicago Linguistic Society, University of Chicago, Chicago, Illinois, pp. 223–247. Levin, Beth and Malka Rappaport Hovav. 1995. Unaccusativity at the Syntax–Lexical Semantics Interface, MIT Press, Cambridge, Massachusetts. Lieber, Rochelle. 1980. On the organization of the lexicon. Doctoral dissertation, MIT. Lowenstamm, Jean. 2002. Templates and verbal derivation. Unpublished manuscript, University of Paris VII. McCarthy, John. 1979. Formal problems in semitic phonology and morphology. Doctoral dissertation, MIT. McCarthy, John. 1981. ‘A prosodic theory of nonconcatenative morphology’, Linguistic Inquiry 12: 373–418. Marantz, Alec. 1984. On the Nature of Grammatical Relations, MIT Press, Cambridge, Massachusetts. Marantz, Alec. 1991. ‘Case and licensing’, in Proceedings of ESCOL, pp. 234–253. Marantz, Alec. 1995. A late note on late insertion. Unpublished manuscript, MIT. Marantz, Alec. 1996. ‘Cat’ as a phrasal idiom’. Unpublished manuscript, MIT. Marantz, Alec. 1997. ‘No escape from syntax: don’t try morphological analysis in the privacy of your own lexicon’, in Alexia Dimitriadis et al. (eds.), UPenn Working Papers in Lin-

280

R EFERENCES

guistics 4.2, Proceedings of the 21st Annual Penn Linguistics Colloquium, Philadelphia, pp. 201–225. Marantz, Alec. 1999a. Case and derivation: properties of little v. Unpublished manuscript, MIT. Marantz, Alec. 1999b. Creating words above and below little v. Unpublished manuscript, MIT. Marantz, Alec. 2000. Roots: the universality of root and pattern morphology. Paper presented at the Conference on Afro-Asiatic languages, University of Paris VII. Marantz, Alec. 2001. Words. Unpublished manuscript, MIT. Myers, Scott. 1984. ‘Zero derivation and inflection’, MIT Working Papers in Linguistics 7, MIT Workshop in Morphology, Cambridge, Massachusetts, pp. 53–69. W Nedjalkov, Vladimir. 1969. ‘Nekatorye verojatnostnye universalii v glagolom slavoobrazovanii’, in I. Vardul (ed.), JJazykovye universalii I lingvistiˇceskaja ˇ slovoonrazovanii, Nauka, Moscow, pp. 106–114. Nir, Rephael. 1976. ‘ˇsˇkifut lesonit ˇ u-psixoleˇsonit ˇ ba-ivrit ha-xadaˇsa’ ˇ (Linguistic and psycholinguistic transparency in Modern Hebrew), in Rephael Nir and Ben Zion Fischler (eds.), Kilshon Amo, Council on the Teaching of Hebrew, Jerusalem. Noyer, Rolf. 1997. F Features, Positions and Affixes in Autonomous Morphological Structure, Garland Publishers, New York. Oltra Massuet, Isabel. 1999. On the notion of theme vowel: a new approach to catalan verbal morphology. Master’s thesis, MIT. Ornan, Uzi. 1971. ‘Binyanim uvsisim, netiyot ugzarot’, Hauniversita 16: 15–22. Pesetsky, David. 1995. Zero Syntax, MIT Press, Cambridge, Massachusetts. Pinker, Steven. 1994. ‘How could a child use verb syntax to learn verb semantics’, in L. Gleitman and B. Landau (eds.), Lexical Acquisition, Special Issue of Lingua, MIT Press, Cambridge, MA, pp. 377–410. Postal, Paul. 1969. ‘Anaphoric islands’, in R.I. Binnick, et. al. (eds.), P Papers from the Fifth Regional Meeting Chicago Linguistic Society, Department of Linguistics, University of Chicago, Chicago, Illinois, pp. 209–239. Postal, Paul. 1970. ‘On the surface verb remind’, Linguistic Inquiry 1: 37–120. Postal, Paul. 1971. Cross-Over Phenomena, Holt, Rinehart and Winston, New York. Prince, Alan. 1975. The phonology and morphology of Tiberian Hebrew. Doctoral dissertation, MIT. Prince, Alan and Paul Smolensky. 1993. Optimality Theory: Constraint Interaction in Generative Grammar, Rutgers University Center for Cognitive Science, Technical Report 2.

R EFERENCES

281

Pustejovsky, James. 1995. The Generative Lexicon, MIT Press, Cambridge, Massachusetts. Pylkkä¨ nen, Liina. 2002. Introducing arguments. Doctoral dissertation, MIT. Raffelsiefen, Renate. 1993. ‘Relating words, a model of base recognition’, Linguistics Analysis 23: 3–159. Ravid, Dorit. 1990. ‘Internal structure constraints on new-word-formation devices in Modern Hebrew’, F Folia Linguistica 24(3–4), 289–347. Reinhart, Tanya. 1996. ‘Syntactic effects of lexical operations: reflexives and unaccusatives’, Occasional Paper, University of Utrecht. Reinhart, Tanya and Tal Siloni. 2000. Against an unaccusative analysis of reflexives. Unpublished manuscript, Tel Aviv University. Reinhart, Tanya and Tal Siloni. 2003. ‘Against the unaccusative analysis of reflexives’, To appear in A. Alexiadou, E. Anagnostopoulou and M. Everaert (eds.), The Unaccusativity Puzzle: Studies in the Syntax–Lexicon Interface, Oxford University Press, Oxford. Rosen, Haim. 1962. A T Textbook of Israeli Hebrew, Chicago University Press, Chicago. Sadka, Yitzchak. 1975. Hadikduk Hamaasi, Kiryat sefer, Jerusalem. Schwarzwald, Ora. 1974. ‘ˇsˇoraˇsim, ˇ bsisim umivne hamorfemot’, Leˇonenu 35: 131–136. Schwarzwald (Rodrigue), Ora. 1998. ‘Inflection and derivation in Hebrew linear word formation’, Folia Linguistica 32(3–4), 265–287. F Shlonsky, Ur. 1997. Word Order and Clause Structure in Hebrew and Arabic, Oxford University Press, Oxford. Sportiche, Dominique. 1998. P Partitions and Atoms of Clause Structure: Subjects, Agreement, Case and Clitics, Routledge, London. Ungebaum, Boris. 1967. Russian Grammar, Oxford University Press, Oxford. Ussishkin, Adam. 1999. ‘The inadequacy of the consonantal root: Modern Hebrew denominal verbs and output–output correspondence’, Phonology 16.3: 401–442. Ussishkin, Adam. 2001. Fixed prosody and the epiphenomenon of root-and-pattern morphology. Paper presented at the Workshop on Root and Template Morphology, University of Southern California. Waltke, Bruce K. and Patrick O’Connor. 1990. Biblical Hebrew Syntax, Eisenbrauns, Winona W Lake, Indiana.

INDEX

Akkadian, 33 Alexiadou, Artemis, 261-262 Alternations, 18-19, 111, 114, 116, 123, 126-127, 132-144, 168, 173-174, 185, 197, 228-229, 234 see also argument structure, causative alternation, stative alternation Anderson, Stephen, 8-9 Anti-Causative Alternation, 174-175, 177, 179-180 Arabic, 5, 35 Arad, Maya, 9, 75, 129, 136, 139, 155, 168, 178, 197, 226, 238 Argument Structure, 113, see also Alternations Aronoff, Mark, 1, 11, 13, 19, 35-39, 120-121, 149-153, 156-157, 190, 238-239 Baker, Mark, 1, 186, 242 Bat El, Uti, 1, 13, 15, 22-23, 35-36, 47-49, 51, 76, 95, 182, 246, 253, 265, 267-269 Beard, Robert, 9 Berman, Ruth, 17, 61, 119-120 Binyanim, see verbal patterns, see also under individual Binyanim Binyan 1, 28, 30, 43, 46, 67, 69-70, 80-92, 96-97, 99, 102-103, 109-114, 123-149, 151-155, 157-160, 162-163, 165-173, 175, 177-181, 185, 188-189, 194-204, 206, 208-214, 216-220, 223-224, 226-227, 229-233, 237 Binyan 2, 28-29, 43, 80-82, 84-87, 89, 109-110, 112-114, 123-126, 128-141, 143-149, 151-155, 160-163, 168-173, 177-178, 180-181, 185, 188-189, 195-204, 206, 208-215, 220, 223-224, 226-233, 237

Binyan 3, 28-30, 43, 46, 67-69, 80-90, 92, 96, 99, 108-115, 123-129, 131-137, 139, 141-142, 144-149, 152-155, 157-160, 162-163, 165-170, 172-173, 175, 177, 179-181, 185, 189, 193, 195, 198-205, 208-209, 212-214, 216-221, 224, 226-233, 237 Binyan 4, 28, 30, 43, 108, 120, 125-127, 131-134, 137, 139, 141, 152, 155, 163, 185, 212-213 Binyan 5, 28-29, 43, 67-70, 80-91, 93, 96-99, 108, 111-115, 120, 123-126, 128-149, 152-155, 157-159, 162, 163, 165-173, 175, 177-181, 185, 188-189, 195-199, 201-206, 208-209, 211-214, 216-221, 223, 225-233, 237 Binyan 6, 28-30, 43, 108, 120, 125-126, 131-134, 137-140, 152, 155, 163, 185, 212-213 Binyan 7, 28-30, 43, 46, 67, 80-90, 108-109, 111-114, 120, 123-129, 131-137, 139-143, 145-149, 152-155, 160-163, 168-173, 175, 177, 179-181, 185, 188-189, 193, 195, 198-206, 208-211, 214, 216, 218-219, 221, 224, 226-233, 237 Bobaljik, Jonathan, 9-10, 15, 26, 40, 45, 186, 192, 227 “Body-Part” Alternation, 137 Bolozky, Shmuel, 28, 30, 61, 237 Borer, Hagit, 154, 163, 196-197, 237, 273 Brown, Cecil, 102-104 Catalan, 47 Causative, 19, 110-111, 113-114, 132, 135-136, 168, 174-175, 188, 199, 205-209, 233-234

283

284

I NDEX

Causative Alternation, 174-175, 177, 179-180, 205-209 Chomsky, Noam, 22, 32, 41, 186, 195, 247 Clark, Eve, 254 Clark, Herbert, 254 Classical Greek, 33 Conjugation Classes, 199-201, 218-235, 237-238 Consonantal Roots, 14, 25, 27, 31, 42-43, 45 Decomposition, 2-7 Denominal Verbs see under noun-derived verbs, see also word formation from words Diacritics on Roots, 227-228, 231, 234 Directionality, 185 Di Sciullo, Anna- Maria, 8 Distributed Morphology, 1, 8-12, 24-26, 35, 40, 52 Doron, Edit, 20, 67, 110, 114, 120-121, 130, 140, 149, 157-173, 185, 190, 215-216, 219-221, 238-239 Embick, David, 8-10, 15, 26, 32, 40, 42, 191 Encyclopedia, 9, 77, 193, 201-205, 218-219, 228-229 Engelhardt, Miriam, 75, 261, 264 English, 5, 11, 26, 35, 57, 101, 241 Even Shoshan, Avraham, 34, 60, 70, 121 Exponents, 9-10, 46, 48, 49 Features, Morpho-syntactic, 9-10, 14, 40-41 Features, Morpho-phonological, 41, 43-44 Fodor, Jerry, 3-6 French, 174 Georgian, 272 Gesenius, 60, 119 Glinert, Lewis, 60, 119 Goshen-Gottshtein, Moshe, 119 Graf, Dafna, 50 Grimshaw, Jane, 217 Hale, Ken, 154, 196-199, 205, 227, 242, 264 Halle, Morris, 8-10, 25-26, 41 Harley, Heidi, 10, 199, 205, 227, 262 Haspelmath, Martin, 20, 121, 173-181 Hayes, Bruce, 258, 263

Van Hout, Angeliek, 196 Inchoative, 111, 113, 199, 233-234 Irregular Alternations, 140, 142 Jackendoff, Ray, 2-4, 6 Jacobsen, Wesely, 174-176, 180 Japanese, 175-176, 180 Kayne, Richard, 216-217 Keyser, Jay, 154, 196-199, 205, 227, 242, 264 Kiparsky, Paul, 8, 254-259, 263 Kratzer, Angelica, 40, 155, 213 Kripke, Saul, 4 Late Insertion, 8-9, 24, 26, 52, 238 Latin, 151, 187, 221 Lepore, Ernie, 3 Levin, Beth, 55-56, 59, 79, 98, 104, 113, 122, 142, 170, 205, 256, 259-260 Lexeme-Based Lexicon, 94-95 Lexicalist Hypothesis, 25 Lieber, Rochelle, 8 Little v, 195 Locality Constraint on the Interpretation of Roots, 247-248, 271 Lowenstamm, Jean, 182 Manner of Motion, 75, see also under Roots, semantic types of Many-Many Morphology, 185-189 Many-to-One Morphology, 186 Marantz, Alec, 8-11, 25-26, 32, 42, 57, 97-98, 155, 186, 196-197, 213, 216-217, 242-243, 247, 262 McCarthy, John, 13, 30, 43, 61, 99, 122, 151, 220, 267 MCM, 17-19, 56-57, 59, 62-63, 65-66, 68-75, 78-83, 85, 87, 90, 91-94, 101-105, 107, 115-116, 122-127, 130, 133, 142, 144-149, 171-172, 188-189, 203-204, 207-208, 222-224, 228-229, 231, 236-237, 244-245 Melodic Overwriting, 47 Miˇsˇqalim, see under nominal patterns, see also non-miˇsˇqalic nouns Morphemes, 10 Myers, Scott, 258-259, 263

I NDEX Nash, Lé´ a, 272 Nedjalkov, Vladimir, 174 Nir, Rephael, 61 Non-Miˇsˇqalic Nouns, 34 Non-Transitive Patterns, 184 Nominalization, 57, 261 Nominal Patterns, 25, 31-35, 42-43, 48, 71-72, 78, 244-245, 249-250 Non-Transitive Verbs, 109 Noun/Verb Asymmetry, 15, 31-44 Noun-Derived Nouns, 252 Noun-Derived Verbs, 13-14, 22, 245-261, 266-267 Noyer, Rolf, 8, 10, 32, 40, 42, 191 O’connor, Patrick, 119 Oltra Massuet, Isabel, 47 One-to-Many Morphology, 186 Ornan, Uzi, 120 Passive, 18-19, 30, 126-127, 131-132, 185, 195, 212-215 Pattern morphology, 13-15, 17-19, 21, see also nominal patterns, verbal patterns Pesetsky, David, 217, 242 Phase, 247-248, 251-252, 266, 268, 271 Pinker, Steven, 56 Postal, Paul, 2 Prince, Alan, 50, 152 Pustejovsky, James, 2-4, 6 Pylkkä¨ nen, Liina, 40, 136, 213 Raffelsiefen, Renate, 258, 263 Rappaport Hovav, Malka, 113, 170, 205 Ravid, Dorit, 61, 100 Reflexive, 110-111, 113, 133, 136, 215 Reinhart, Tanya, 67, 113, 161, 205, 216-217 Roots, semantics contents of, 53-105., see also MCM Roots, semantic types of: Activity, 88, 129 Adding, 85-86 Allotting, 87 Configuration, 82-83 Cutting, 87 Damaging, 86 Defining, 87 Directed Motion, 85

285

Dividing, 87 Entity, 89 Getting, 86 Holding, 86 Letting Go, 86 Manner of Motion, 89, 129 Manner of Speaking, 90 Mental Activity, 80-82 Motion from Source, 84 Motion, 83-84 Needing, 86 Perception r , 80-82 Piercing, 86 P Possession , 86 Property, 88 Psychological State, 89 Putting Together, 87 Receiving, 86 Relation, 82-83 Satisfying, 86 Substance, 89 Subtracting, 85-86 Temporal Concepts, 87 T Transfer r , 83-84 Roots in word formation, see Root Derived Verbs, Word formation from roots V Root Derived Verbs, see also word formation from roots, 14, 20-22, 191, 194, 201, 205, 235-236, 240, 245-261, 270 Root Hypothesis, 6, 8, 10-17, 23, 26, 34-35, 44, 52-53, 94-96, 196, 236, 238, 240, 243 Rosen, Haim, 119 Russian, 26-27, 36-39, 156, 272 Sadka, Yitzchak, 119 Schwarzwald, Ora, 61, 120 Separation, 8-9 Shlonsky, Ur, 51, 72, 108, 238 Siloni, Tal, 217 Single Occurrence Roots, 18-19, 123, 126, 128-130, 172, 201-203, 222, 231, 237 Smolensky, Paul, 50 Spell-Out, 44, 46, 227, 229-235 Sportiche, Dominique, 217 State Roots, 88 Stative, 111, 113, 133, 199, 209-212 Stative Alternation, 209-212

286

I NDEX

Stem Modification, 47-48, 50-51 Sumerian, 33 Syllabic Roots, 34-35, 42 Syncretism, 21, 187, 188, 190, 194, 200-201, 218, 222, 235-237 Synonyms, 19, 123, 146-149, 189, 204-205, 222, 229, 231, 236-237 Syntactic Word-Formation, 8, 52, 238 Templates, 29, 42 T Thrá´ insson, Hoskuldur, 186 Transitivity, 192, 195-198, 224-226 T Transitivity Alternations, 184 T Transitivity Constraints, 184 T Truncation, 11 Ungebaum, Boris, 36 Ussishkin, Adam, 36, 50, 61, 110, 114, 182, 265, 267

183-184, 189-190, 192-196, 198-203, 207, 218, 222, 224, 229, 231, 237-238, 244 Verb-Derived Nouns, 252, 261-262 V Verb-Derived Verbs, 20-21, 191, 201, 216, V 235-236 Vocabulary Insertion, 45-46 V Voice, 30, 43-44, 46, 49, 190-191, 213, V 214-215, 217, 236 Waltke, Bruce, 119 W Williams, Edwin, 8 W Word Formation, 1, 14-15, 17, 55, 154, 242, W 244 see also Syntactic Word-Formation, Word Formation from Roots W Word Formation from Roots, 11, 54, 183, W 240-243, 253-254, 268, 271 Word Formation from Words, 183, 240-243, W 254, 257, 268, 270-271 Yiddish, 35 Y

Verbal Patterns, 25, 28-31, 35-36, 42-44, 46, V 48, 60, 77, 97, 106-110, 150-152, 154, 156, 162, 164, 168-169, 171-173,

Zero-Related Pairs, 7, 26, 254-256, 259, 262-263

Roots and Patterns: Hebrew Morpho-syntax

Issues in Zulu Verbal Morphosyntax

Roots

Roots

Roots

Roots

Hebrew life and times

Roots and Wings

Multiculturalism: Roots and Realities

Yupik Roots and Affixes

Roots and Wings

Polynomials, roots, and interlacing

Patterns

Patterns

העברית שלנו והעברית הקדומה (Contemporary Hebrew and Ancient Hebrew)

העברית שלנו והעברית הקדומה (Contemporary Hebrew and Ancient Hebrew)

Dead Roots

Grass Roots

Twisted Roots

Hebrew Literature

Ancient Aramaic and Hebrew Letters

Hebrew Humour and Other Essays

Tilings and Patterns

Pagan Patterns and Designs

An Index to Aquila: Greek-Hebrew, Hebrew-Greek, Latin-Hebrew, with the Syriac and Armenian Evidence

Dead Roots

Macroeconomic Patterns and Stories

Patterns, Programming and Everything

Mystic Roots

Nerve Roots

Twisted Roots

Roots and Patterns: Hebrew Morpho-syntax

Issues in Zulu Verbal Morphosyntax

Roots

Roots

Roots

Roots

Hebrew life and times

Roots and Wings

Multiculturalism: Roots and Realities

Yupik Roots and Affixes

Roots and Wings

Polynomials, roots, and interlacing

Patterns

Patterns

העברית שלנו והעברית הקדומה (Contemporary Hebrew and Ancient Hebrew)

העברית שלנו והעברית הקדומה (Contemporary Hebrew and Ancient Hebrew)

Dead Roots

Grass Roots

Twisted Roots

Hebrew Literature

Ancient Aramaic and Hebrew Letters

Hebrew Humour and Other Essays

Tilings and Patterns

Pagan Patterns and Designs

An Index to Aquila: Greek-Hebrew, Hebrew-Greek, Latin-Hebrew, with the Syriac and Armenian Evidence

Dead Roots

Macroeconomic Patterns and Stories

Patterns, Programming and Everything

Mystic Roots

Nerve Roots

Twisted Roots

Recommend Documents