Key Features and Parameters in Arabic Grammar

Key Features and Parameters in Arabic Grammar

Linguistik Aktuell/Linguistics Today (LA) Linguistik Aktuell!Linguistics Today (LA) provides a platform for original monograph studies into synchronic and diachronic linguistics. Studies in LA confront empirical and theoretical problems as these are currently discussed in syntax, semantics, morphology, phonology. and systematic pragmatics with the aim to establish robust empirical generalizations within a universalistic perspective. For an overview of all books published in this series, please see

http:!!benjam ins.com!cataloglla

General Editors VVerneri\braharn University ofVienna I

Elly van Gelderen Arizona State University

University of Munich

Advisory Editorial Board Josef Bayer University of Konstanz

Christer Platzack

Cedric Boeckx I CREA/Universitat Autonoma de Barcelona

Ian Roberts Cambridge University

Guglielmo Cinque University of Venice

Lisa deMena Travis McGill University

Liliane Haegem.an University of Ghent

StenVikner University of Aarhus

Hubert Haider University of Salzburg

C. Jan-Wouter Zwart University of Groningen

University of Lund

Terje Lohndal University of Maryland

Volume 182 Key Features and Parameters in Arabic Grammar

by Abdelk.ader Fassi Fehri

Key Features and Parameters in Arabic Grammar Abdelkader Fassi Fehri KAICAL, Ryad & Mohammed V University, Rabat

John Benjamins Publishing Company Amsterdam I Philadelphia

The paper used In this publication meets the minimum requirements of the American National Standard for Information Sciences - Permanence of Paper for Printed Library Materials, ANSI z39.48-1984.

Library of Congress Cataloging-in-Publication Data

Fassi Fehri, Abdelkader. Key features and parameters in Arabic grammar I Abdelkader Fassi Fehri. p. em. (Linguistik Aktuell/Linguistics Toda~ ISSN 0166-o829; v.l.!l2)

Includes bibliographical references and index.. 1. Arabic language--Grammar. I. Title. PJ6w6.F348

2012

492.7'5--de23 ISBN 978

90 272 5565 5 (Hb; alk. paper)

ISBN 978

90 272 7496 o (Eb)

© 2012- John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher.

John Benjamins Publishing Co.· P.O. Box 36224 • 1020 ME Amsterdam ·The Netherlands John Benjamins North America· P.O. Box 27519 ·Philadelphia PA 19118-0519 • usA

For Heroes ofArab Revolutions

Table of contents

Foreword

XV

Provenance of chapters PART 1.

XIX

Temporality, aspect, voice, and event structure

1

CHAPTER 1

Tense/Aspect interaction and variation 1. Past, Perfect, Perfective 5 1.1 The Past/Perfect ambiguity 5 1.2 Temporal and modal qad 7 1.3 One or two projections ofT 9 1.4 Perfective 9 2. Present, Imperfect, Imperfective 10 3· Imperfect and SOT 13 4- Perfectivity 16 4-1 ST as Perfective 16 PT as Imperfective? 20 4-2 4-3 From Tense to Aspect 22 The Tense/ Aspect language typology revisited 4-4 5· Conclusion 25

3

24

CHAPTER 2

Transitivity, causativity, and verbal plurality Issues 28 1.1 Problem 1: Semitic morpho-syntax 28 1.2 Problem 2: Transitivity theory 29 2. Number Theory 31 2.1 Ingredients of Num theory 31 2.2 Verbal plurality and distributed Num 32 2.3 Distributed plurality 33 2.4 Causative complexity; verbalization, and distributivity 33 2.5 Two sources of transitivity 35 2.6 Parallel plural morphology 36 Summary 40 2.7 3· Cross-linguistic evidence 41 1.

27

vm Key Features and Parameters in Arabic Grammar Causatives, transitives, and event quantification 41 3.1.1 Causativization and transitivization 42 3.1.2 Multiple behaviour 43 3.1.3 Event quantification 44 3.2 Moravcsik:'s resistant cases 45 Conceptual motivations and competing analyses 46 4.1 Little v: Verbalizer or transitivizer? 46 42 Aspect 47 43 Voice 50 43.1 Anti-transitive reflexives 51 43.2 Reflexive causatives .52 43·3 Agentive and "expositive" causatives 53 43·4 "Requestive" causatives 53 43·5 Ergative Num and intensive forms 54 44 Further empirical motivations 55 44.1 Ergative and unergative Num in event plurality and transitivity 55 44.2 Adicity, (in)transitive alternations, and multiple uses Num theory and Num heights 57 5.1 Sg and Pl Merge 57 5.2 Language variation 57 Summary and conclusion 58 3.1

4·

5·

6.

55

CHAPTER 3

Synthetic/analytic asymmetries in voice and temporal patterns 1. Analysis, voice, and temporality 65 1.1 The problem 65 Nominal auxiliaries 66 1.2 1.3 S/0 Agr split and auxiliary selection 67 Temp auxiliaries 69 1.4 1.5 Voice 71 1.5.1 Arabic and anaphoric Agr 71 1.5.2 Latin and split Agr 72 1.5.3 Modern Greek 74 Albanian 75 1.5.4 1.5.5 Moroccan Arabic 76 2. Formal complexity and categorization 79 2.1 Further analytic and synthetic questions 8o 2.1.1 Pass and additional complexity 8o

61

Table of contents

Two finite Agrs So 2.1.3 Ancient Greek as fully synthetic 81 2.2 Reanalysis as the source of analytic pass or perfect 2.3 A splitting analysis (of Temp and Agr categories) Peculiarities and structural heights 85 3.1 Imperfective passive 85 3.2 Verbal and adjectival voices 86 3·3 Multiple functions across heights 89 Summary and conclusion 91 2.1.2

3·

4·

82 83

4 Arabic Perfect and temporal adverbs 1. Salient properties of the Arabic TR system 94 1.1 Polyfunctionality ofTI Asp forms 94 1.2 The PresPerf split: Synthesis and analysis 95 1.3 The Past split: Simple Past Pfv and complex Past Impfv 2. The Perfect/Past ambiguity 99 2.1 Aspects and Tenses 100 2.2 Positional "deictic" adverbs 103 2.3 Perf and modal qad 104 2.4 Adverbs and simple vs. complex tenses 106 2.5 Durational adverbs 107 3· Temporal adverbs and kinds of Perfect 110 3.1 Positional mundu uo 3.1.1 Imperfective tenses 110 3.1.2 Perfective tenses 111 3.1.3 PresPerf tense 112 3.2 Durational mundu 113 3·3 Perf of Res and Post-state 113 4· Summary and conclusion 115 4.1 T/Asp morphology 115 42 T/Asp adverbs or particles 116 43 Conclusion n6 CHAPTER

PART 11.

93

96

DP, np, bareness, and count/mass structures

5 The grammar of count and mass 1. Toward a wider count grammar 1.1 Ways of 'numeralizing' CHAPTER

121 122 124

IX

x

Key Features and Parameters in Arabic Grammar

Partitive numeral 124 Predicative numeral 125 1.1.3 Numeral verbs 125 1.1.4 Numeral adverbs 125 1.2 Ways of count quantifying 125 1.2.1 Count quantifier bi4r in a construct state 125 1.2.2 Predicative count quantifier 126 The count/measure ambiguity of kam 126 1.2.3 1.2.4 More partitives 127 1.3 Events selecting count nouns 127 General nouns 128 2.1 General atomicity 129 2.2 Singulative atomicity 130 General nouns and general Number 130 2.3 2.4 Counting and numeralizing 132 The general noun is not plural 132 2.5 2.6 The general noun is not mass 134 The general noun is not a group 134 2.7 Collective varieties 134 3.1 What 'counts' in the grammar of collectives 134 Syntactic groups 136 3.2 Masses 138 41 Mass as atomless 138 42 Mass is cumulative, and non-divisive 140 43 Mass as distinct from plural 141 44 Plural of mass is productive 142 Count/mass architecture, features, and functional categories 144 5-1 Count and mass syntax 146 5.2 Interpreting plural and singular inflections 147 Summary and conclusion 151 1.1.1 1.1.2

2.

3·

4

5·

6.

CHAPTER 6

Synthesis in Arabic DPs 1. Setting the stage for synthesis 154 1.1 Synthetic ingredients 154 1.2 Synthetic 'articles' 156 1.3 Quantifiers 158 1.4 Numerals 158 1.5 Demonstratives 159 1.6 Indefinites in Arabic dialects

153

160

Table of contents 2.

3·

4

DP architecture 162 2.1 Adjective modification 163 2.1.1 The NA Order 163 2.1.2 MIO 164 2.1.3 Adjectival classes and non-intersectives DA extensions 167 2.1.4 2.2 Mirror image alternations 168 Synthetic indefinites 169 2.3 Core properties and types of synthetic possession 171 3.1 How real is Definiteness spreading? 172 3.2 Possessor placement 174 Two classes of synthetic possessives 175 3·3 Characterizing the variation 176

165

CHAPTER7

Bare, generic, mass, and referential DPs 1. Overt D contrasts and genericity 181 1.1 Definite, generic, and mass 181 1.2 Predicates and anaphors 183 1.2.1 Kind Level Predicates 183 Stage Level Predicates 183 1.2.2 1.2.3 Individual Level Predicates 184 Kind anaphora 185 1.2.4 1.3 Plural as 'plural of the singular' 186 Scope 186 1.3.1 1.3.2 Opacity 186 Telicity 187 1.3.3 2. Arabic BNs are indefinites 187 2.1 Arabic BNs as indefinites? 188 2.2 GenP and N-to -Gen 190 2.3 Arabic/Romance distinctions and the Numeral Parameter 191 2.3.1 Varieties of bareness 193 2.3.2 The Numeral Parameter 196 3· Further discussion 197 3.1 Gen contexts 197 Modification and D-binding 198 3.2 Modalized contexts 198 3·3 3·4 Definite and indefinite generics 200 4 BNsandPNs 200

179

XI

xn Key Features and Parameters in Arabic Grammar 5· 6.

Mass specification Conclusion 204

CHAPTER

202

8

Determination parameters in the Arabic and Semitic diglossia 1. Indefinites 207 1.1 N moves over indefinite quantifier 208 1.2 N movement to indefinite D 210 1.3 Adjective movement and movement over adjectives 211 1.4 Semitic overt indetermination 212 2. Definites 213 3· Double (in)determination in Semitic 216 3.1 Definites and indefinites are not complementary 218 3.2 Searching for minimal (in)definite pairs 220 3·3 PN s are not indefinite 221 4· Bare determination 223 4-1 Individuating and non-individuating vocatives 223 4-2 'Unique' superlatives 226 5· Account and summary 228 5-1 The computational history of determination 228 5.2 Two features in computation 230 PART III.

205

Clausal structure, silent pronouns, and Agree

CHAPTER 9 Time/space anchors, logophors, finiteness, and (un)interpretability of inflection CP anchoring, double access tenses, and logophors 236 1. 1.1 Person double access 237 1.2 Time double access 238 Arabic as a DAR language 238 1.3 Double access and Mood 240 1.4 Root and logophoric Cs 242 1.5 2. Finiteness 243 2.1 Subject properties 244 2.1.1 Cases of subjects 245 2.1.2 Positions of subjects 246 Subject agreement 246 2.1.3 Expletive subjects 246 2.1.4 2.2 Truncated structures 247 2.2.1 Raising 247

235

Table of contents xm

Auxiliary (complex tense) structures 2.2.3 ECM 249 2.2.4 Control 249 The structure of Tense 250 3.1 Tense on T and Person 250 Synthetic and analytic temporality 251 3.2 3·3 V movement 253 Summary and conclusion 256 2.2.2

3·

4-

248

CHAPTER 10

Arabic silent pronouns, person, and voice 1. The referential/non-referential correlation 258 2. Impersonals/indefinites 261 2.1 Arabic 261 Comparison with French, Finnish, Irish, and Italian 2.2 2.3 The human feature 266 3· Referential pro 268 3.1 A topic approach 268 A Probe-Goal implementation 271 3.2 4- 'Passive, 'impersonal: 'indefinite' 273 5· Expletives and EPP 2']6 5.1 Arabic expletives 277 5.2 Pronouns and EPP 281 5·3 Finnish 283 6. Some consequences 284 6.1 Variation around Person 284 6.2 Topicality and Person 286 6.3 A new approach to Voice 288

257

264

CHAPTER 11

Plural verbs and Agree Introduction 291 Nominal Number 292 2.1 Where is number? 295 Non-human plurality 299 2.2 2.3 Lexical collectives 299 2.4 Syntactic collectives 301 2.5 Plurals of plurals and similar matters 305 3· Verb plurality 309 3.1 Pluractional morphology 311 3.2 Collective and distributive plural 313 1. 2.

291

XIV

Key Features and Parameters in Arabic Grammar

Semantic Pl in SVO 314 Kinds of plural agreement: Collective and non-collective Reciprocity 318 4.1 Lexical reciprocals and symmetric events 318 4-2 Morphological reciprocals 320 4-3 Syntactic reciprocals 324 Summary and conclusion 326 3·3 3·4

4·

5·

317

References

329

Index

349

Foreword This book is based on a collection of recently published contributions by the author in the last decade (suitably revised and annotated with additional new references), complemented with new unpublished work (Chapters 5 & 9), or to appear (Chapter 11). The whole material has been appropriately restructured in three parts. Part I deals with essential issues in temporality, aspectuality; and actionality. Part II examines various aspects of nominal architecture, including (but not limited to) distinctions in count/mass expressions, bareness, (in)definiteness, or other determination parameters involving N-to-D Move or Agree. Part III is dedicated to investigating time/space anchoring in clauses (or CPs), anaphors, pronominals, and the interpretability of inflectional features. Although there is now a rather flourishing and valuable modern linguistic literature on the various themes and issues of Arabic grammar, the key features of the analyses proposed, as well as the nature of the parameters involved are, to my knowledge, both new and appealing. First, the profound study of plurality (and pluractionality) of verbs in Arabic (Chapters 2 & 11), including the interpretability of Number on the verbal category, is rather unprecedented in the literature. Second, the discovery that Arabic bare nouns behave like true indefinites in English or Romance (Chapter 7), coupled with the discovery that Arabic postnominal adjectives are in a strict mirror image order of their Germanic counterparts (in analytic contexts) provide even more important new keys of the Arabic (or Semitic) nominal architecture than the traditionally studied construct state (reanalyzed here in terms of macro-parametric synthesis; Chapter 6). Third, the traditional count/mass dichotomy is re-examined in light of the role played by general, collective, or singulative DPs, and replaced by a more adequate binary feature system ([±atomic] and[± singulative]; Chapter 5). Fourth, the synthetic nature of temporal and voice categories (Chapter 3) and determiner categories (Chapter 6) is given a formal characterization. Finally. a Past/Perfect/Perfective ambiguity is established and solved in the contexts of (potentially ambiguous) collocating temporal adverbs, or the role played by CP in temporal anchoring, DAR (double access reading), or SOT (sequence of tense), Arabic being argued to be a DAR language. These key features and issues in the grammar of Arabic are treated in light of the most recent developments of generative theory, building on comparative and parametric approaches to language variation, minimalist design of gram-

XVI

Key Features and Parameters in Arabic Grammar

mar, and computational efficiency (cf. in particular Chomsky (1995, 2008), and Kayne (2000, 2005)). The semantic/formal nature of parameters is amply investigated, chiefly in connection with the Longobardi (1994, 2001)/Chierchia (1998) debate, as well as the more general debate on whether there are macro-parameters (Baker (2008) or parameter schemata (Longobardi (2005), Guardiano & Longobardi (2003)) that cannot be fully reduced to micro-parametric variation (as in Kayne (2005)). The cartographic methodology is also explored (cf. Rizzi (1997) and Cinque (1999) in particular), and typological results discussed and integrated (e.g. Greenberg (1972, 1991), Aikhenvald (2001), WALS (Haspelmath et al. 2005), etc.). One peculiarity and persistent state of the art in approaching the structure of Arabic is that three rather separate methodological and conceptual trends coexist, although they hardly converge, or communicate the results of their investigations: (a) a remote Arabist tradition, which is rich and valuable with regard to traditional grammatical thought (see e.g. Sibawayhi (8th century), Astarabaadii (12th century)), but cannot replace (or be transposed as such) in modern analysis or theory; (b) a western Orientalist (and/or semiticist) tradition, which makes use of distinct notions and terms, although it builds (critically) on the Arabist lines of thought and data (see e.g. Brockelman (1913), Wright (1974), Blachere & Gaudefroy-Demombynes (1939), Blau (1972), or more recently Retso (1997), Owens (2000), Versteegh (2001), Badawi, Carter & Gully (2004), or Ryding (2005)), and (c) modern linguistic theory applied to Arabic (for very recent applications, see e.g. Bahloul (2008), Bardeas (2009), Aoun, Benmamoun & Choueri (2010), etc.). The texts of the Encyclopedia of Arabic Languages and Linguistics edited by Kees Versteegh and others is an illustration of this state of affairs, since they gather together these three noncommunicating traditions. The present book attempts to bridge gaps between these traditions, taking into account important results reached in the various frameworks. The language described is basically Standard Arabic, although dialect (or colloquial) varieties are also brought in, in addition to Hebrew, for the sake of describing and identifying micro-variation. For more general comparison (or macro-variation) various other languages are used (typically Germanic and Romance). Standard Arabic is the language of more than 350 million speakers around the globe (including 22 Arab states in which it is the main or unique official language), more than 60 million internet users, and more than 30 million social network users. It is also the media language of the most influential TV channels in the Arab World, the most read language of the newspapers or magazines, etc. Huge electronic corpora in Standard Arabic are then now available to check grammatical judgements, a tool which developed so rapidly and quantitatively only in the last decade, and now provides more modern and reliable data.

Foreword xvu By putting these new contributions together (with new considerations of

theory and data), the book is expected to have a wide and substantial audience among students, researchers, and teachers of Arabic, Semitic, comparative, typological. or theoretical linguistics. Following the path of its predecessor, Issues in the structure of Arabic Clauses and Words, published by Kluwer Academic Publishers in 1993, it is hoped that it will become one of the popular references in the field.

Provenance of chapters Chapter 1, Tense/aspect interaction and variation, appeared in The Synta.:"<: of Time, Jacqueline Gueron & Jacqueline Lecame (eds), 235-257, 2004. Cambridge MA: The MIT Press. Chapter 2, Transitivity, causativity, and verbal plurality, appeared in Research in Afroasiatic Grammar 5, Jacqueline Lecarme, Jean Lowenstamm & Ur Schlonsky (eds), 131-185,2003. Amsterdam: John Benjamins. Chapter 3, Synthetic/analytic asymmetries in Voice and Temporal Patterns, appeared inA~mmetry in Grammar 2, Anna Maria Di Sciullo (ed.), 95-128. 2003. Amsterdam: John Benjamins. Chapter 4, Arabic Perfect and temporal adverbs, appeared in Perfoct Explomtions, Artemis Alexiadou, Monika Rathert & Arnim von Stechow (eds). 2003. Berlin: Mouton de Gruyter. Chapter 6, Synthesis in Arabic DPs, appeared in Relative Clauses and Genitive Constructions in Semitic. Journal of Semitic Studies Supplement 25, Jan Retso & Janet Watson (eds), 153-180, 2009. Oxford: OUP. Chapter 7, Bare, generic, mass, and referential Arabic DPs, appeared in Phrasal and Clausal Architecture. In Honor of]oseph Emonds, Simin Karimi, Vida Samian & Wendy Wilkins (eds), 40-65, 2007. Amsterdam: John Benjamins.

Chapter 8, Determination parameters in the Arabic and Semitic diglossia, appeared in Proceedings of NACAL 35, Charles Haberl (ed.), 149-182, 2007. Newcastle: Cambridge Scholars. Chapter 10, Arabic silent pronouns, person, and voice, appeared in Brill's Annual ofAfroasiatic Languages and Linguistics 1: 1-38, 2009. Chapter 11, Plural verbs and Agree, is to appear in Nominal and Verbal plurality, Patricia Cabredo-Hofherr & Brenda Laca (eds). Berlin: Walter de Gruyter.

xx Key Features and Parameters in Arabic Grammar

Thanks to the publishers for having granted me permission to reproduce this material. Chapter 5, The grammar of count and mass, appears here for the first time, and is based on two invited talks: Kinds of Number (and Gender). International Workshop on Nominal systems across languages. Barcelona: UAB, 2009; & Ways of counting and numbering. Workshop on nominal and verbal Plumlity, 2009. Paris: CNRS & Univ. of Paris VIII. Chapter 9, Time/space anchors, logophors, finiteness, and (un)interpretability of inflection, appears here for the first time, and is based on two talks. Finite time inflection and double access construal. Invited talk. Postgraduate Conference at the Univ. ofNewcastle, 2007. & Talk delivered at NACAL 36, 2008. Chicago.

PART I

Temporality, aspect, voice, and event structure

CHAPTER 1

Tense/Aspect interaction and variation*

Temporal and aspectual relations and differences in their grammaticalization across languages can be traced back to detectable morpho-syntactic properties. Past. Perfect, and Perfective (and their negative/neutral counterparts) will be taken as the core grammatical components building the temporal/aspectual systems. In the context of Semitic, there is an old-recent debate on whether the expression of grammatical Tense is lacking, which would then make Semitic an "aspect language': 1 There is also a debate on whether among "Tense languages~ only languages like Slavic or Greek are "truly'' aspectual, in the sense that Aspect is grammatically expressed only in those languages. 2 The ambiguity of morphological expression of some temporal or aspectual categories in some languages, in addition to its total absence in others (in tenseless and/or aspectless languages), is sufficient to stress that the descriptive program of the temporal/aspectual variation is basically morphological (or morpho-syntactic), and that semantically temporal or aspectual cross-linguistic generalities have to be built in general syntax. 3 Furthermore, in addition to Past, Perfect, and Perfective interactions, there is also the contribution of actional (Dowty/Vendler) classes of predicates (the so-called

*

This Chapter is based on Fassi Fehri (2000/2004).

The idea has been initiated by the German philologists Caspari (1848/1859) and Reckendorf (1895), the former being later translated (and modified) in Wright (1858/1971). Later developments along a similar line include a huge list of references, among which see Meillet (1910) and Cohen (1924). More recently, Cohen (1989) has defended the "pure aspect" view of Semitic verbal morphology. On the opposite side, Kurylowicz (1972 & 1973) in particular denies the existence of aspect (or perfective) as a grammaticalized category in Semitic (typically Arabic and Hebrew). 1.

1.

On a partial review of references on the matter, see e.g. Szemerenyi (1965).

3· It is an empirical question whether the lack of some temporal words is correlated with the lack of their meaning altogether, or whether its absence can be compensated through default (morphosyntactic) correlations.

4

Key Features and Parameters in Arabic Grammar

"lexical aspect"), and their properties (such as telicity), in determining temporal interpretation. 4 Arabic is typically characterized by the ambiguous use of the same finite verbal form for Past or Perfect (or non-Past, Imperfect) to express Anteriority (or non-Anteriority) of Reference Time (= RT) with respect to either Utterance Time (UT) or Event Time(= ET). Perfectivity (or imperfectivity), on the other hand, is not morphologically expressed, being either correlated with Past or Perfect T (or non-Past, Imperfect). The latter interacts with telicity (or more gererally actionality), but does not depend on it. For concreteness sake, I assume a neo -Reichenbachian model of tenses, conceived as expressing relations between times, with two syntactic TP projections headed by predicative Ts, to account for Perfect tenses. Each T defines a temporal ordering relation between two temporal arguments: Tl orders UT with respect toRT (and/or ET), and T2 orders RT and ET. Tl is usually qualified as "deictic" or "absolute" (Past, Present, or Future), and T2 as "relative" (Perfect/Imperfect or Anterior/non-Anterior). 5 As for Perfectivity (= Pfv), it is conceived as an expression of completeness, boundedness or culmination of events or situations (which cannot be further extended). In contrast, Imperfective events do allow completion or addition of an end. Following Giorgi & Pianesi (2001; = GP henceforth), I will take the semantic correlate of Perfectivity to be terminativity. Put in a tree, the core configurational structure ofT and Asp is roughly as follows: (1)

T1 (±Past)

~

T2 (± Perf/ Ant)

~

Asp (± Pfv/Term)

~

VP (±Telic)

Within this background, I provide a new description of the essential ingredients of the Arabic aspectuo-temporal system, within a cross-linguistic perspective. The latter focuses on the following questions:

4· Note that Mood and Modality, which also interact with temporal interpretation, will not be dealt with here (see e.g. Fassi Fehri (1990, 1993, 1996) on this interaction in Arabic, as well as Chapter4). 5· On this model, see Zagona (1990), Hornstein (1990), Fassi Fehri (1990), Giorgi & Pianesi (1991), Stowell (1993), and Gueron (1995), among others.

Chapter l. Tense/Aspect interaction and variation

a. b.

the Past/Perfect ambiguity; the dependency of Perfective on Past or Perfect, and the dependency of the Imperfective on the Present; c. the existence of a dependent Imperfect in Arabic, with past interpretation arising from contextual distribution, typically in sequence of tense (SOT); d. the role played by synthesis and analysis in determining complementary values of aspectuo-temporal chains.6

To the extent that Asp can be separated from T, the picture that emerges is that Arabic is more of a "tense language': than an "aspect language". 7 This typology, once redefined, can be used to differentiate languages in which (a covert) Perfective is a feature associated with an expressed Past T from languages in which (a covert) Past is inferred from an expressed Perfective (depending on whether the association is oriented from up to bottom, or from bottom up). A third type briefly discussed is Slavic, where both T and Asp are morphologically expressed.

1.

Past, Perfect, Perfective

Verbal morphology exhibits two contrasting finite forms characterized by Person placement (as a suffix for Past, and prefix for non-Past), internal vocalic changes of the verb stem, and (suffixed) Mood marking in the present (and its absence in the Past). These forms contribute to mark Mood, Tense, and Aspect interactions. For convenience sake, I limit the discussion to the latter two categories, and designate the two contrasting forms as suffixed Tense (= ST), and prefixed Tense (= PT), taking into account only Person Placement. 1.1

The Past/Perfect ambiguity

ST expresses Past in neutral (non-dependent, non-embedded) context, as evidenced by its co-occurrence with appropriate deictic adverbs: 8

6. On a detailed description of synthetic/analytic variation, see Fassi Fehri (2001a, and Chapter 3). On temporal chains, see Gueron & Hoekstra (1988) and Fassi Fehri (1990). 7. This view is very dose to that of Kurylowicz (ibid), and differs radically from the aspectualist view defended in many works on Arabic and Chamito-Semitic, as in e.g. Cohen ( 1989). 8. According to Wright (1898), III, (1 ), ST here is equivalent to English Past, Greek Aorist, or German Imperfect. But as will be shown, this description is incorrect.

5

6

Key Features and Parameters in Arabic Grammar

(2)

a.

katab-a r-risaa.la.t-a

zamsi

(*gad-an)

wrote-3 the-letter-Ace yesterday ("tomorrow) He wrote the letter yesterday.

b.

jaraa zamsi

(*gad-an)

ran yesterday (*tomorrow) He ran yesterday. ST expresses also Perfect (= Present Perfect) in neutral (non-dependent, nonembedded) context. The adverbial test also applies here, similar to what happens in English: 9 (3)

a.

katab-a r-risaa.la.t-a l:.zaan-a, qablu (*gad-an) wrote-3 the-letter-ac now before (*tomorrow) He has written the letter now, before (*tomorrow).

b.

jaraa l:.zaan-a, qablu (*gad-an) ran now before (*tomorrow) He has run now, before (*tomorrow).

(4)

John has written the letter (just) now, before (*tomorrow).

In both (2) and (3), ST serves to locate ET prior toRT, or liT, without relying on contextual information. In the Semitist literature, ST has often been designated as Perfect, "Accompli" (a French term), or Perfective, and treated as Aspect. However, ST in this context can be reasonably construed as denoting ordering temporal relations, or time reference (whether "absolute" or "relative").10 Given its collocation with time reference adverbials, it is hard to see how ST can be conceived as marking primarily Aspect (perfectivity or non-durativity), given that the latter is not sensitive to such adverbial contrasts (typically past and future). 11

9· Wright (ibid) equates the Arabic Perfect with that of English, German, or Greek. This is also incorrect, as we will see. to. Higginbotham (2000b), among others, regards "speech time" or UT as a special case ofRT.

n. The ambiguity of ST in (2) and (3) has a synthetic analogue in Portuguese (i), and an analytic counterpart in French "passe compose"' (ii): (i)

Comi as quatro. I ate/have eaten at four.

(ii)

a. b.

Jean a mange la pomme hier. Jean a mange la pomme en ce moment.

See GP (1997) for an analysis of the Portuguese form as T2 (which parallels the Latin Perfect form), and for a treatment based on T2 of the French "passe compose~ see Gueron (1995).

Chapter l. Tense/Aspect interaction and variation

ST can also be the embedded member of other perfect tenses (i.e. Past Perfect or Future Perfect), typically when a copular auxiliary is overtly realized. These uses confirm the T nature of the Perfect ST: (5)

kaan-a (qad) katab-a r-risaalat-a lammaa daxal-tu had already wrote-3 the-letter-Ace when entered-! He had (already) written the letter when I entered.

(6)

sa-yakuunu (qad) kata.b-a r-risaa.la.t-a gadan FUT-is already wrote-3 the-letter-Ace tomorrow He will have (already) written the letter.

In (5), the writing event (which occurs in the past) is prior/anterior to the entering (which occurs also in the past). This indicates that two "shifting" tenses are involved in the matrix clause (a past of the past, or a past perfect). The two anteriority relations do not follow if the thematic verb is interpreted as expressing Aspect only (i.e. perfective). They do not obtain in e.g. (7), where the thematic verb, although read as Past perfective, is not able to express the same temporal orderings: (7)

katab-a r-risaalat-a lammaa daxal-tu wrote the-letter-Ace when entered-! He wrote the letter when I entered.

In (7), both verbs can be construed as Past, but the entering is understood as prior/ anterior to the writing. As for (6), although the action occurs in the future (which is posterior toUT), the time adverbial is interpreted as a bound prior to which the completion of the writing must occur (hence the anterior future interpretation, a complex time reference).

1.2

Temporal and modal qad

Let the two temporal interpretations be attributed to Tl (for Past) and T2 (for Perfect). If we state that ST expresses either T1 or T2, then a significant progress is made in identifying the nature of the ST alternations. The ambiguous use of ST is not to be qualified as "temporal" vs. "aspectual~ In both cases, temporal Anteriority is associated with ST. The variation arises only from the nature of the time arguments that ST orders (UT and RT, or RT and ET). Other temporal components in the clause (such as adverb level type) or the context may play a coercive role, which leads to finer time reference, or disambiguation.

7

8

Key Features and Parameters in Arabic Grammar

The particle qad is typically acknowledged in traditional literature as favoring the Perfect reading. It is usually interpreted as meaning "already" (i.e. precedence as in (5) and (6)), or "just" (immediate precedence), as in (8): (8)

qad Zataa just came He has just come.

But note that in (8), qad is ambiguous in a way that it cannot be in (5). It can mean "indeed~ or "in fact~ to stress the factual certainty of the event, and the sentence is translated as "He has come indeed" or "He did come". In the latter case, it serves a modal function. Interestingly enough, qad in this interpretation can co-occur with past time adverbials. It is then interpreted only as a modal and it looses the possibility of being temporal, as the interpretation of (9) illustrates: (9)

qad zataa zamsi indeed came yesterday He did come yesterday (*he has just come yesterday).

In these contexts, qad appears to be higher than T1, and inserted under Mod (Modal). This behavior is confirmed by its possible occurrence in front of the auxiliary of a complex tense: (10)

qad kaana y-u~llii indeed was 3- pray He was indeed praying.

If all perfects exhibit a structure with two (hierarchized) TPs, where a high auxiliary supports Tl. and a lower thematic verb supports T2, then the duality of ST in Arabic may find its source in the emptiness of the present copula in Arabic. The double tensed (auxiliary) structure, with an empty auxiliary in Tl, and the thematic verb in T2, is diagrammed in (11): T1

(11)

~

0

T2

~

kataba

VP

Such an analysis of the Present Perfect is in consonance with other perfects being doubly headed structures. The overt periphrastic character of Past or Future

Chapter l. Tense/Aspect interaction and variation

Perfects and Passive Pluperfects provide clear support for the double T analysis, given that both Ts are temporally specified. 12 1.3

One or two projections ofT

As seen in the examples above, Arabic exhibits an important peculiar property: it makes no morphologica.l distinction between Agrl and Agr2, and no morphological distinction between T1 and T2. 13 Since both the auxiliary V and the thematic V carry the same finite morphology, based on Person placement essentially, but also on Agr (Number and Gender), the morphology cannot be taken as discriminatory as far as the Tlff2 distinction is concerned. 14 Furthemore, although the syntheticity of the Present Perfect makes it identical in form to the Past, the interpretations of the two tenses are difierent. In order to distinguish Perfect from Past, we propose that the latter projects only one T, whereas the Perfect projects two Ts. This morpho syntactic difference is supposed to account for temporal divergences and time adverb co-occurrences.15 1.4

Perfective

Up till now, I have established the (primary) temporal nature of ST. But I have not dealt with its aspectual character, namely when it expresses perfectivity, and how the two facets of this nature can be connected. These questions will be examined

For more on the interpretation of qad, see Dahl & Talmoudi (1979). The Arabic verbal copula, when unspedfied temporally, surfaces as null. The pronominal copula "huwa" (he) can appear as an overt expression of Agr in verbless sentences, when needed, indicating the existence of a dear-cut distinction with the verbal copula. The visibility of the latter has nothing to do with Agr specification, but only with temporal specification (see Fassi Fehri (1990, 1991/1993)). Since Present Perfect is synthetic, it is reasonable to think that T2 merges with T 1, when T 1 is null In analytic Perfects, however, T 1 (and Agr or phi features) must be visible (cf. Fassi Fehri (2001) for detail, as well as Chapter 3). 12.

Agr1 and Agr2 have been used in the pre-minimalist representations as projections of the phi features of subject and object, respectively. In current analyses, they can be dispensed with, and replaced by nominal features on vN and T, along the lines proposed in Chomsky (1995).

13.

Note that the two Agr must be anaphoric, due to conditions on chains, which preclude the repetition of identical heads in the same chain, as argued in Fassi Fehri (1990 & 1996).

14.

I see no reason to favour a sole T2 interpretation of the form over a complex T one. The situation of T here is exactly parallel to that of time adverbs or particles like qad, which can have di1ferent interpretations depending on the level of structure at which they are inserted However, the question arises more seriously with the ambiguity of analytic present perfects, typically when interpreted as Past. See e.g. Gueron (1995) and GP (1997) for plausible though different analyses.

15.

9

10

Key Features and Parameters in Arabic Grammar

in the next sections by using various tests. For the moment, I will only mention that ST as a simple Past can be interpreted only perfectively, and so is ST as Perfect. This statement is true regardless of the actional class of the verb. For example, the accomplishment (2a) cannot be uttered if the writing of the letter is not already finished, and the letter completed. Likewise, the activity (2b) is terminated, and cannot be further extended. Furthermore, the perfectivity is also true of the Perfect in (3).

Present, Imperfect, Imperfective

2.

In Western traditional grammars, PT is usually named "Imperfect': and commonly thought of as expressing "no idea of time"; it denotes "an act which takes place at all time~ 16 However, contrary to this view, PT, like ST, can express Tense. It is associated with Present in neutral (non-dependent, non-embedded) contexts, regardless of whether the verbal predicate is stative, or non -stative: (12)

y-abnii d-daar-a l-1aan-a (*1amsi) 3-build the-house-Ace now (*yesterday) He is building the house now.

(13)

y-ajrii l-1aan-a (*1amsi) 3-run now (*yesterday) He is running now.

(14)

y-arrifu l-jawaab-a l-1aan-a (*wnsi) 3-know the-answer-Ace now (*yesterday) He knows the answer now.

The construction (12) is an accomplishment, (13) an activity, and (14) a state. Constructions (12) and (13) mean "He is involved presently in the act of building or running': Furthermore, there is no culmination of that act, which indicates that the Present is also imperfective. As for (14), it means roughly that "a certain state holds of the subject at the present moment': The Present interpretation of (12) and (13) cannot be obtained for English; "John runs" or "John eats" are normally construed as habitual or reports. English can form an actual present only from statives ("knows': "loves': etc.). This difterence in the interaction between actional properties of eventualities and the availability of actual Present reading recalls the well-known difference between English, on the one hand, and Romance and

Cf. Wright (ibid, p. 18). For the Arabic grammatical tradition, on the contrary, PT is basically non-Past, being Present or Future (see e.g. Sibawayhi (8th century)).

16.

Chapter l. Tense/Aspect interaction and variation

other Germanic languages, on the other hand. In the latter languages, accomplishments and activities do form actual Present, but in English, they do not, unless progressive morphology is used. Arabic is on a par with French or German, rather than English, in this respect The interaction mentioned might serve as a base for postulating that only languages which allow Present interpretation of nonstatives (typically accomplishments and activities) have Present T. Moreover, it is the Present meaning which requires the latter eventualities to be imperfective (i.e. with no telos or culmination reached)P GP (1991) suggest that the variation between English and Romance/Germanic can be implemented through a difference in X -bar structure height (X0 for English, but X1 for Romance/Germanic). They argue that simultaneous reading for nonstatives require duration (or more precisely imperfectivity). It is the languages exhibiting this interaction that have Present Tin their morphological inventory, while English has Present as lexical, and only progressive -ing is morphological. 18 The lexical/morphological contrast can be observed or tested with complements of perceptual reports: (15)

a. b.

John saw Mary eat an apple John saw Mary eating an apple.

In the (a) construction, where the verb is bare, only a perfective reading is possible, whereas (b) is interpreted only as a progressive. In Arabic, the situation is similar to that of English (15b) in that the complement sentence in (16), which contains a (morphologically null) Present, is interpreted only as simultaneous imperfective: (16)

razay-tu r-rajul-a y-azkulu t-tuffaat~at-a saw- I the man 3-eat the-apple-Ace I saw the man eating the apple (*eat the apple).

The impossibility of perfective interpretation with the Present of these classes (cf GP (1991, 1995), Gueron (1995)) is attributed by GP (1997) to the Punctuality Constraint, as stated in (i), and the interpretive principle (ii), once the notion of punctuality is appropriately defined:

17.

(i) (ii)

A closed event cannot be simultaneous with a punctual event. (p. 163) The anchoring event is punctual. (p. 160)

The speech event is one case of anchoring events. Punctual has to be durational with activities and accomplishments to make Present possible with these classes. GP follow Kamp in defining an event as punctual "iff it is not temporally partitioned by other events" (p. 159). In other words, Present can be an interval, not necessarily a moment, as in Gueron (ibid). 18.

See Gueron (ibid) for a similar view.

11

11

Key Features and Parameters in Arabic Grammar

Note that in Romance/Germanic, infinitive verbs are used in such contexts. and the interpretation is ambiguous between (lSa) and (15b). 19 The fact that PT cannot be ambiguous in (16) suggests that it behaves (with regard to relevant properties) like a progressive in English (although it has no progressive morphology as far as we can tell). Arabic also differs from Romance/Germanic in another important respect. Factual present of achievements is possible in Arabic, but not in those languages: (17)

y-ablugu r-rajul-u l-qimma.t-a l-'laan-a 3-reach the-man-NOM the-top-Ace now The man is reaching the top now.

The construction (17) is interpreted as imperfective, like a progressive form in Romance or Germanic; but unlike the present form in the latter languages, it has no performative (or perfective) interpretation: (18)

John reaches the top.

(19)

Gianni trova un libra.

Thus the Arabic Present is strongly associated with imperfectivity, compared to that of Romance/Germanic. As a corollary, it is insensitive to actionality. 20 In addition to being Present, PT is also interpreted as Imperfect (associated with non-anteriority), typically when used with past auxiliary in complex tenses,

19.

This is e.g. the case in Italian (i), from GP (1997,p. 165), which is ambiguous:

Ho vista Gianni mangiare una mela I saw Gianni eat-INF an apple. For the English progressive, see Higginbotham (2004). (i)

10. Jacqueline Gueron (p.c.) pointed out to me that achievements in the Present are normally interpreted as perform.ative in a number oflanguages. On the other hand, GP (1997) claim that achievements are inherently perfective in all languages, and hence can form no actual Present. Contrastively, Arabic achievements can form actual Present, but it is normally ST which is used for performatives. This difference in behavior is correlated with (in)sensitivity of the Present to actionality. It also indicates that Arabic Present is correlated with Imperfective, but Romance/Germanic Present inherits imperfectivity only through (a particular) actionality. The existence of Present progressive achievements even in Romance/Germanic (GP (ibid), pp. 172 & 177) appears to be truly problematic for the hypothesis of inherent perfectivity. To solve this problem, GP resort to a notion of"continuity': which is associated (according to them) with the progressive, and is not incompatible with perfectivity (imperfectivity not being a prerequisite of continuity). A more convincing analysis. it seems to me, has to abandon the inherent perfectivity hypothesis, and postulate that the progressive is a case of a morphologically marked imperfective, while other cases of imperfectives in Romance/Germanic are only induced from actionality (in interaction with Present T).

Chapter l. Tense/Aspect interaction and variation

to express roughly what Greek or Latin Imperfect expresses, or English past progressive: (20)

kaana y-aktubu r-risaalat-a was 3-write the-letter-Ace He was writing the letter.

The same is true with Future or Habitual Imperfect: (21)

yakuunu y-aktubu r-rasaaZil-a

fli l-raJiyy-i

is 3-write the-letters-Ace in the-evening He will be (is usually) writing letters in the evening. Observe that Imperfects here are periphrastic, like their Perfect counterparts. Other properties of PT as Imperfect will become clear when dealing with dependent tenses. PT is often misinterpreted as (primarily) Imperfective (cf. Comrie (1976)). Wright (ibiti) takes it to express only duration. Here, I maintain the Imperfect/Imperfective distinction, not only for the sake of making precise the right notion involved, but also for knowing whether PT is the neuter-negative (non-Past/Imperfect) counterpart of the positive Past/Perfect. or whether it grammaticalizes imperfectivity, as against perfectivity. a position that I will reject. Recall that Perfect denotes anteriority, an ordering before relation.21 Imperfect is non-ante1"im·, usually simultaneous, expressing a within relation. 22 I can then conclude that the Arabic Present is morphologically (or morphosyntactically) expressed, although it interacts interestingly with Imperfective. PT also expresses Imperfect, which is distinct from Imperfective. This morphology turns out then to be one of "tense" (Present or Imperfect), rather than "aspect': Imperfective can be associated with these tenses, and we will return to further clarifications on this interaction, after examining other specific contexts in which PT is used as Imperfect.

3·

Imperfect and SOT

In addition to its use as Present or non-Anterior/Simultaneous in complex tenses, PT has other specific uses, which can be contextually dependent on a Past tense. It does not form a complex tense with the latter, but has roughly the interpretation which a French "Imparfait" or Italian Imperfect has, i.e. the interpretation of

21. I leave aside here the notion of the "relevance" of the reference or evaluation time (or simultaneity meaning), which can be used e.g. to distinguish pluperfect from past of the past. 11.

I leave aside the after/posterior relation for the moment.

13

14

Key Features and Parameters in Arabic Grammar

Past hnperfect. That interpretation can be determined by anaphoricity with an immediately dominating Past tense, or it can be contextually conditioned by other temporal elements. The following construction with an adjunct clause is an example of an hnperfect use: (22)

jalasa y-a.Srabu l-xamra sat 3-drink the-wine-Aec He sat, drinking wine.

In the adjunct clause, only the PT can be used, and it expresses "Present-underPast reading': The use of the Perfect here is impossible, thus indicating that it cannot be referentially dependent or anaphoricP Likewise, in complement clauses, the simultaneity with the past is solely acceptable if the hnperfect is used (i.e. no SOT effect is possible, just like the situation in Japanese or Hebrew):24 (23)

qaal-a l-ii 1inna-hu y-aktub-u r-risaalat-a said-3 to-me that-him 3-write the-letter-Ace He said to me that he was writing the letter.

The use of the Past/Perfect does not yield that interpretation: (24)

qaa.l-a l-ii zinna-hu katab-a. r-risaalat-a. said-3 to-me that-him wrote-3 the-letter-Ace He said to me that he wrote the letter.

In (24), there is no overlapping of the writing and the saying, as in (23). The writing is rather anterior to the saying (with a shifted past reading). This situation is in contrast with the SOT behavior in complement clauses in English, where the embedded verb can be interpreted as anaphoric (or simultaneous, with the meaning "Mary is ill"):25

23.

For anaphoricity of Present/Imperfect, see Fassi Felui ( 1990 & 1993).

24. Cf. Ogihara (1995) and Abush (1997), among others. Cf. also Higginbotham's (2000a) analysis of SOT, based on anaphoridty.

25.

The Arabic stative counterpart to (25) read as anaphoric is (i), where no copula surlaces: (i)

qaal-a 1-ii

1inna-hu mariicf-u11

said-3 to-me that-him sick-NoM He said to me that he is sick. When the past copula is used, no "present-under-past" reading is possible: (ii)

qaal-a I-Ii

1inna-hu kaan-a mariitf-an

said-3 to-me that-him was-3 sick-Ace He said to me that he (was) was sick.

Chapter l. Tense/Aspect interaction and variation

(25)

John said Mary was ill.

GP (1995) compare languages which possess Imperfect (like Italian) with languages that do not (e.g. German or English). They observe that Italian Imperfect is a dependent anaphoric tense, which denotes Present-under-Past tense, the Past being provided normally by the matrix verb. Moreover, the (simple) Past cannot be used in this language as dependent, and in embedded contexts, only Imperfect can be used with a SOT effect. 26 In contrast, languages like German use Past (ambiguously) as dependent or non-dependent, the latter being neutral with respect to perfectivity. It can be read either way, as simultaneous (imperfective) or non -simultaneous (perfective): (26)

Hans sagte, da.B Marie einen Apfel aB a. b.

Hans said that Mary was eating an apple. Hans said that Mary ate an apple.

In (a), the simultaneous reading is available, and the Past is interpreted as Imperfective. In the (b) reading, the Past is Perfective and the simultaneous interpretation is excluded. Thus (simple) Past is necessarily Perfective in Italian, but not so in German. The conclusion then is that if a language has an Imperfect Past, as opposed to a simple Past, the latter cannot be used as dependent in embedded contexts. On the other hand, the Imperfect is dependent and used to denote simultaneity. In languages in which the Past is unspecified, it can be used as dependent/ simultaneous. Arabic is close to Italian in this respect. Past (which is also Perfect) is perfective and non-dependent, contrary to Imperfect, as illustrated above. Furthermore, Past Imperfect interpretation can be provided by the context: (27)

li-maa t-aqtutuluuna zanbiyaaza

l-laah-i

min qabl-u

why you-kill-PL.M. prophets-Ace Allah-GEN before Why were you killing the prop bets of Allah before? (28)

baynamaa n-nabiyy-u

y-u~aUii

ziti

zaqbala

ruqba.t-u

while the-prophet prays then appeared Oqbah-NOM While the prophet was praying, Oqbah showed up. In these examples, the subordinate clause acts as a temporal adverbial forcing the shifting of the Imperfect in the matrix clause to a past interpretation (through a "hidden" past auxiliary, just like what happens in embedded dependent contexts). Such sorts of examples have been used thoroughly by the "aspectualists" to show that PT can express all tenses, from which they conclude that it is devoid ofTense.

16.

Compare their examples (17) and (2l),p. 349.

15

16

Key Features and Parameters in Arabic Grammar

But these distributions can be taken as evidence to the contrary, namely that PT is an Imperfect Tense (denoting simultaneity, possibly with Past). If the contrary was true, i.e. if PT were "purely aspectual then it is hard to see how the Presentunder-Past interpretation can ariseP

Perfectivity

4· 4.1

ST as Perfective

We have seen that Arabic Past and Perfect are not anaphoric, and do not allow simultaneous imperfective reading in SOT (cf. (24) above). Consequently, Past and Perfect ST are associated (or correlated) with Perfective. The correlation between Perfective and Past or Perfect obtains normally in languages that distinguish Imperfect from simple Past (i.e. Italian, and somewhat differently Arabic, as opposed to German). Furthermore, like Perfect, Perfective is not compatible with simultaneous interpretation. Present and Imperfect, however, are usually associated with Imperfective (ct~ the interpretations of e.g. (16), (22), and (23)). But unlike what happens in Russian (or Slavic), Arabic Perfectivity is not marked by two distinct verbal conjugations (or two distinct lexemes), opposed in the same tense, as illustrated by e.g. the following Russian minimal pair, contrasting the imperfective and the perfective in the preterite: (29)

a. b.

ya pisal ya napisal

This pair is rendered only approximatively by English I was writing and I have written or French Jecrivais and Jai ecrit. The latter pairs, contrary to the Russian ones, are rather oppositions of Perfect/Imperfect (Anterior/Simultaneous), although Aspect may be associated with them. Likewise, Perfectivity in Arabic is a feature of a specific Tense, rather than a grammaticalized (derivational) morpheme (or a lexicalized opposition) differentiating the same tense. Arabic has only one Past/Perfect synthetic form, which is also perfective/terminative (see (2) and (3) above). It contrasts minimally with the Imperfect, which is either a complex tense (with an expressed past auxiliary, as in (20)), or occurs in an appropriate context (with a "hidden" past interpretation), as in e.g. (23), (27), or (28). But as clarified above, Arabic Imperfect is not necessarily a Past (contrary to what is found with Italian Imperfect, or French "Imparfait"). Only the analytic

17. See e.g. Hisle (1990), Zaborski (2002), and in particular Fassi Fehri (1990, 1993) for various kinds of arguments leading to this conclusion.

Chapter l. Tense/Aspect interaction and variation

Past Imperfect is. Moreover, Imperfect is (normally) imperfective and (roughly) equivalent to both Italian Imperfect or English Past Progressive. Consider e.g. (2b) expressing a past activity (repeated here as (30a) for convenience), and its analytic Past Imperfect counterpart, given in (30b ): (30)

a.

jaraa r-rajul-u ran the-man-NOM The man ran.

b.

kaana r-rajul-u

y-a;m

was the-man-NOM 3-run The man was running. c.

[ ... ] wa maa zaala y-ajrii and still 3-run and he is still running.

The event in (30a), which is completed or terminated, cannot be further extended, contrary to that in (30b ), which is non-terminated and can be extended. Hence the sequence in (30c) can be used after (30b ), but it leads to ungrammaticality after (30a). In fact, the combination of (30a) and (30c) is acceptable only if two events are involved, one terminated to make the first clause true, and one non-terminated and ongoing at liT, to make the second sequence possible. 28• 29 The Perfective nature of ST is corroborated by its use in so-called "performative" sentences, as instantiated by the following examples: (31)

jur-tu hungered- I I am hungry (literally: I hungered).

GP (2001), who analyze quite similar contrasts in Italian, reach a similar conclusion. Italian simple Past and Perfect are perfective, and they contrast with Past Imperfect, which is imperfective. The former tenses cannot be extended (unless the interpretation is different, as indicated). See e.g. the contrasts in their (22) and (23). 18.

Simple ST being perfective, it cannot be equated with German Imperfect or even English Past (as claimed by Wright (II I, 1). Comrie (1976) correctly analyzes Wright's example which appears to be non-terninative (= (i) here) as Past perfective: 29.

(i)

jalas-uu ralaa 1-baab- i sat-3.PL on the-door-GEN They sat down at/on the door.

The natural interpretation of (i) is that the event cannot be further extended here (with the proviso made above with respect to (30a) ).

17

18

Key Features and Parameters in Arabic Grammar (32)

bir-tu. sold-I I sell (literally: I have sold).

Sentences like these (with appropriate contexts and presuppositions) express merely reaching a terminating point (the state of "being hungry': the decision of "selling"), rather than any past action. In English or French, they are normally translatable by a Present (which should be construed as terminative). In Arabic, the use ofST focuses on their perfectivity, although reference time (a sort of resultative Perfect very close in these cases to a stative present) may be playing a less direct (or "metaphoric") role. Observe that the perfective/terminative nature of ST vs. the imperfective/non-terminative nature of PT should come as a natural consequence of the Western philologist or Semitist view, which takes the nature of the opposition to be aspectual (whatever the terms used: Perfect/Imperfect, Imperfective/Perfective, or Accomplished/Unaccom plished). 30•31 But note that Kurylowicz (1972) denies the existence of verbal aspect as a (discrete) grammatical category in Semitic, and he takes the perfective/ imperfective meaning to be a tertiary function (in his terms) of the ST/PT opposition, its primary function being preterite/non-preterite (or past/present), and its secondary function anterior/simultaneous (or perfectum/infectum; p. 83). Since Tense and Aspect are non -separable from a semantic point of view, he claims that Aspect is "... a non-distinctive feature of tense forms in every language. The present tense referring to the moment of speaking is by itself imperfective or linear, the preterite or the future are by themselves perfective or punctual" (p. 79). He then observes that it would be an elementaty error to treat equally the tertiary temporal/aspectual system of Western Semitic (cf e.g. Akkadian) and the binary system found in Arabic and Hebrew. In the latter case, Aspect must be subordinate to Anteriority (a tense notion), the primary content of the sole opposition (p. 80). Arabic Perfect and Imperfect function primarily as Preterite/ Present, when related to the moment of speaking, and secondarily as Anterior/ non-Anterior when related to another tense. Thus Kurylowicz's view represents an undeniable significant progress in the adequate description of the Arabic and

Even Wright (1971, III, p. 21 ), who awkwardly attributes to ST the meaning of German Imperfect, correctly states later on that" to express the imperfect of Greek or Latin languages, kaana is frequently prefixed to the imperfect':

30.

Recall that Present is not normally used in performatives in Arabic, due to its general imperfective nature, which then makes it a grammatically mlllked form for Aspect (the value of which depends on that ofT), compared to Romance or Germanic, in which Aspect is only inherited from Actionality, and is somehow more "lexical': as already explained above.

31.

Chapter l. Tense/Aspect interaction and variation 19

Hebrew system, and it comes close to ours (and in part to that of traditional Arabic grammarians) when it denies the grammaticalization of Aspect as the primary content of the verbal inflectional opposition. It favours Tense instead, and organizes the aspectuo-temporal system around three "functions: with their dependencies and correlations. 32 But divergences between Kurylowicz's system and ours, although less crucial than those with the "aspectualists~ are still of important significance for more precise description, as well as for conceptual foundations and empirical consequences. 33 Likewise, Comrie (1976, pp. 78-82) also defines a mixed (or combined) Tense/ Aspect interpretation of Arabic verbal inflection, but does not take issue with the predominance of Tense or Aspect, nor with the question of how the system is descriptively organized. Comrie correctly observes that when the rest of the sentence contains no overt specification of time reference (e.g. no temporal adverbs), ST (named Perfective) has perfective and past meanings (as in (i) in Footnote 29, taken from Wright (ibid)), while PT (named Imperfective) is interpreted with imperfective and present meaning (as in (33), also taken from Wright (ibid)): (33)

llaa.hu.

y-arlamu bi-maa

t-armal-uu.na.

Allah-u 3-know about-what 2-do-PL.M. Allah knows what you are doing. The contrast between these two constructions can be used to favor neither the sole tense nor the sole aspect distinction. The system appears to be mixed. First, the use of PT as future does not necessarily imply imperfective meaning, but could well

The primacy of Tense in his view is assessed at different places, despite the fact that he some times talks about a system with no Tense and no Aspect, only a binary oppositioiL Anteriority (or time reference) is temporal, and it is needed for Kurylowicz (1973, p. 114) to assess that "... whereas the existence of tense does not entail that of aspect, the latter presupposes that of tense~ the dependency of aspect vis-a- vis tense in Arabic, the inherent nature of aspect as non -distinctive feature of Tense in every language, and the systematic imperfectivity of present and perfectivity of past and future. All these statements are almost true (as rough descriptions), but regrettably questionable, when faced with precision, as will become dear.

32.

Clearly, Kurylowicz's (ibid) claims either make wrong or no predictions about variation in the aspectual/temporal expression in languages, and how it is organized 0 n the other hand, the general "primacy" of tense over aspect that he adopts is not empirically motivated It is lan-

33·

guage specific, as explained below. Furthermore, even if it turned out to be true of Arabic, no motivation for such orientation is provided Similar criticisms carry over to Zaborski's (2002) work, who adopts Kurylowicz's views entirely.

20

Key Features and Parameters in Arabic Grammar

correspond to a Slavonic perfective, as illustrated by the following sentence (found also in Wright (ibid)): (34)

fa-llaah-u y-a~kum-u bayna-hum yawm-a l-qiyyaamat-i and-Allah 3-judge between-them day-Ace the-resurrection-GEN And God will judge them on the day of resurrection.

Given that the day of resurrection takes place some time in the future, RT is future. Butthe day of resurrection functions as a telos of the event, which suggests thatthe sentence is terminative (and/or "perfective"). PT then ends up being both terminative or non-terminative in our terms, and hence the ST/PT opposition cannot be purely aspective. Similarly, ST is not always denoting past time reference, but could express future RT, as in the often quoted sentence (35): (35)

z-ajiizu-ka zicjaa ~marr-a. l-busr-u I-come-you when redden-ed the-unripe date I will come to you when the unripe date ripens.

Thus although the embedded verb carries ST, the ripening is understood as occurring in the future. What ST then denotes is "relative" (anterior) RT, since the ripening is prior to my coming. Comrie then establishes the combined nature of the opposition, but does not go any further. However, his contribution is a step forward, when compared to the "pure aspect" view (initiated by Caspari & Reckendorf; cf. Footnote 1). 4.2

PT as Imperfective?

I have established earlier that ST (being either Past or Perfect) is correlatively Perfective, and hence terminative. The question now is whether PT (being Imperfect and Imperfective) is necessarily non-terminative, or it can be also terminative (or "perfective"). Consider first PT in analytic Past Imperfect/Imperfective. The latter is typically non-terminative, regardless of the actional properties of predicates. This is true of accomplishments as well as achievements: (36)

ka.ana 1'-rajul-u y-~ilu zilaa l-qimma.t-i lammaa. was the-man-NOM 3-arrive to the-top-NOM when zatlaq-a ~aaruux-an launched-3 missile-Ace The man was reaching the top when he launched a missile.

Chapter l. Tense/Aspect interaction and variation (37)

ka.ana 1'-rajul-u was

y-aqtaru l-maa.sajat-a

the-man-NOM 3-cross

rann-a

lammaa

the-distance-Ace when

l-ii

appeared-3 to-me The man was crossing the distance when he appeared to me. In both cases, the action in the matrix clause is not terminated, whereas that in the embedded one is. The analytic Past Imperfective behaves then like the English or Italian progressive with regard to its interpretation. In contrast, PT in simple Imperfect/Imperfective interpretation is not so committed to non-terminativity, but can be terminative as well. This value depends on the actional properties of the predicate (which include telicity and durativity). but also on the Tense of the Imperfect. The telic/atelic distinction is clearly observable in opposing (38) to (39), all of which are finished or terminated events: (38)

a.

qata.ra r-rajul-u

l-ma.saafat-a.

.fii saara.t-in (*saarat-an)

crossed the-man-NOM the-distance-Ace in hour-GEN (*hour-Ace) The man crossed the distance in an hour (*for an hour).

b.

wa.fala r-rajul-u

zilaa. l-qimmat-i

.fii saarat-in (""saa.rat-an)

arrived the-man-NOM to the-top-GEN in hour-GEN (*hour-Ace) The man reached the top in an hour (*for an hour). (39)

a.

jaraa sa.arat-an ("fii saarat-in) ran-3 hour-Ace (*in hour-GEN) He ran for an hour (*in an hour).

b.

sarlb-a

l-xamr-a

saarat-an ("fii saarat-in)

drank-3 the-wine-Ace hour-Ace (*in hour-GEN) He drank wine for an hour (*in an hour). The first set of clauses is telic, in the sense that a certain goal or endpoint, the telos, has been reached, or that the event culminates at the telos. The distance has to be entirely covered, the top reached, etc. The ongoing of the event cannot be continued any further, e.g. by adding a clause like "and he is still .. .':The second set of clauses is atelic, in that although the event is finished, it does not culminate at any endpoint or goal. It is conceivable that it might be continued. In- X vs.for-X adverbials yield felicitous testing results for identifying tetic and atelic sentences, respectively. With simple Imperfect clauses, however, telicity cannot be (directly) tested by using the in-XIfor-X adverbial opposition. In clauses with Present tense interpretation (normally ambiguous between actual (or progressive) and habitual

21

:u Key Features and Parameters in Arabic Grammar readings), the use of the adverbial removes the actual reading, making available only the habitual interpretation: (40)

a.

y-aqtaru r-rajul-u

l-masaafat-a

fii saarat-in

3-cross the-man-NOM the-distance-Ace in hour-GEN The man crosses the distance in an hour. b.

y-~ilu

r-rajul-u lilaa l-qimmat-i fii saarat-in 3-arrive the-man-NOM to the-top-GEN in hour-GEN The man reaches the top in an hour.

With these telic predicates, the use of in-X adverbs is only possible when the interpretation is habitual or non-actual. Actual Present is not compatible with those adverbs. But note now that with Imperfect Future, the in-X test is completely felicitous: (41)

y-aqtaru r-rajul-u

l-masaafat-a

gad-an

fii saarat-in

3-cross the-man-NOM the-distance-Ace tomorrow in hour-GEN The man will cross the distance tomorrow in an hour. The same result obtains with achievements. The common behaviour of Future and Past/Perfect vis-a-vis the in-X test, as opposed to that of the Present, indicates that the difference is due to the value of Perfectivity; associated with Tense: Past/Perfect and Future are perfective, whereas Present is imperfective. With actual tenses, the in-X/for-X opposition then applies not to test telicity (as commonly thought), but to test a combination ofperfectivity and telicity. 34 4·3

From Tense to Aspect

Granted that the PT/ST opposition expresses a mixed combination ofT and Asp, and having indicated that Asp is merely associated with T (rather than being the value of the mentioned opposition), I would like to corroborate this view by showing how the system is oriented, from top (Tense) to bottom (Aspect), rather

34· Various authors arrive through various routes to a similar conclusion (see e.g. Bertinetto (2001), and Iatridou, Anagnostopoulou & Izvorski (2001)). GP (2001) claim that the telic/ atelic distinction does not apply to imperfective predicates, and that the perfective/imperfective distinction is privative, in the sense that only perlectives are associated with the (positive aspectual) value of terminativity, whereas imperfectives are left unspecified. My analysis of the Arabic Imperfect suggests that the postulated privative nature is not established, since only the dearly imperlective factual Present cannot be terminative, as far as I can tell, even when telic. As for the simple Imperfect interpreted contextually as Past, it has been shown that it can be interpreted as simultaneous, and hence imperlective.

Chapter l. Tense/Aspect interaction and variation 23

than the other way round. The opposite orientation appears to be true in e.g. Western African languages, as will be explained. In Arabic analytic aspect/tenses, both Perfective and Imperfective can be dominated by either Past or non-Past of the auxiliary kaana "be" (see e.g. the (5)/(6) contrast for Perfective, and (20)/(21) for Imperfective). Hence no Aspect/ Tense correlation is predictible. In the synthetic aspect/tenses, however, simple Past is always perfective, Future can be perfective, and Present is imperfective (as shown in e.g. (38), (41), and (12) respectively). The imperfective, on the other hand, can be associated with Imperfect Past, as in (37), Present, as in (12), or nonbounded Future, as in (42): (42)

sa-y-ajrii FUT-3-run

He will run. Contrary to (41) above, (42) is neither terminative nor culminative (or telic). It is imperfective. These contrasts suggest that no Aspect can determine (a specified) Tense, since perfective can be Past or Future, and imperfective can be associated with Present, Past, or Future. But a (specified) Tense can determine Aspect (simple Past is perfective, actual Present is imperfective, etc.). The conclusion then is that in Arabic Aspect correlates with Tense, but not vice versa. Assuming that these categories project separately, as indicated above, the system can be said to be oriented from top to bottom. Comrie (1976, pp. 82-84) observes that in a number of Western African languages, including Yoruba and Igbo, there are no specific markers of Tense, although there are markers of Aspect. Thus non-stative verbs have the prefix marker n in Yoruba and na in Igbo if they have imperfective meaning, whereas perfectives have no morphology (statives have only imperfective meaning, with no marker). But sentences with these forms are non-ambiguous with regard to time reference (in the absence of any contextual indication): perfectives are interpreted as referring to Past, and imperfectives to Present: (43)

a. b.

Yoruba: 6wa Igbo: 6 byara. He came.

(44)

a. b.

Yoruba: 6 n sise Igbo: 6 na am 6ru He is working; he works.

Thus in these languages, Aspect is marked overtly, and time reference is a "secondary consequence" of the aspectual mar king (except in contexts where overt time reference adverbs like "yesterday" are used, in which case the imperfective

24 Key Features and Parameters in Arabic Grammar

refers to the past). Clearly then, although these sentences are morphologically "tenseless~ Tense is not absent from their interpretation. It is presumably projected as null Tense, and interpreted by association with imperfectivity or actionality (i.e. ingredients from the bottom of the structure). Contrastively. Arabic can be thought of as morphologically "aspectless~ but Asp values are associated with specific values of Ts. 4·4 The Tense/ Aspect language typology revisited I have observed that there are two types oflanguages which regard to the lack of either Tense or Aspect marking: the Arabic type (call it A-type), and the Western African type (call it W-type). A third type like Russian, (or more generally Slavic), appears to mark both categories (call itS-type). But as amply explained in the article, the first two types do express both categories of time reference (or tense) and aspect, like the third type, except that they differ in how they do it morphologically. In the A-type, the aspectual value is (normally) associated with specific time reference marked on the verb, and in the W-type, time reference is associated with aspect marking. Leaving aside other factors that play an important role like markedness (in the past/non-pastor the perfective/imperfective distinctions), finiteness, and actionality (vs. perfectivity), a rough picture of the typology can be given at this point, assuming that all languages instantiate the same syntactic partitioning (as exhibited in (1) above), and only morphology makes differences, in addition to morphosyntactic derivations. In the A-type, T has a specified value (Past or Perf), and the value of the Perfv feature is inherited by T. Assuming that V moves to T, then the displacement is motivated by V checking of its features against those of "rich" T (i.e. through a sort of attraction). In theW-type, Perfv (induced from actionality) is only compatible with Past. If T is null, then it is V which carries the relevant value of features which percolate to "poor" T, to make it interpretable (assuming that there are universal default Asp, Actionality, and T correlations (as proposed e.g. by Kurylowicz (1972)). The Imperfective in the W-type, which is morphologically marked, can be associated with Present through the same mechanism of movement, except that other overt expression of time reference (like past time adverbs) may prevent interpreting imperfective as Present (i.e. blocks aspect values from percolating up, by blocking movement). The Imperfective in the A-type appears to exhibit a similar default behaviour. However, Asp has no morphological mark. Since PT can be both imperfective and perfective, it must be that it is Tense, in fact Present T, which imposes imperfectivity, rather than the other way round. Language type distinctions can be based on the grammaticallymarked members of the pairs (Perfv, Perf, Past), or their semantically marked positive counterparts (when Actionality operates). Many details of the analysis

Chapter l. Tense/Aspect interaction and variation

remain to be worked out, but it is clear that if the T/Asp language typology is to be properly defined, the A-type would qualify as aT language, while theW-type would represent an Asp language, the S-type being marked for both T and Asp. 35

5·

Conclusion

In this chapter, I have examined how Aspect and time reference (or T) correlate in languages, and typically in Arabic. I have rejected the quite spread view among Western philologists that Arabic is an "aspect" language in the sense that the PT/ST opposition does not express reference time. The result turned out to be the contrary, in that the opposition has been shown to have no primary aspect content, compared toW-type or S-type languages. A more principled description of how the system of aspectuo-temporal relations and correlations is organized has been provided, based on the morpho syntactic typology proposed. Many questions obviously remain, but they would presumably be addressed more adequately if the "tense" dominant view of Arabic aspectuo-temporal morphology is adopted, being clearly understood that no language seems to lack the essential semantic notions associated with T and Asp. 36

35· In work in progress, I develop an analysis of Tense/Aspect marking distributions which is parallel to Number/Classifier marking distributions. See Fassi Fehri (2011). 36. In this description, I have neglected the active participle because it does not play a central role in the temporal system, being carrying no Tense morphology. See e.g. Fassi Fehri (1993), Hisle (1990), and Shlonsky (1997) on a quite convergent view of this matter.

15

CHAPTER 2

Transitivity, causativity, and verbal plurality* Two essential problems are addressed in this chapter. The fist one is desc1·iptive; it consists in the appropriate characterization of the variable behavior of Arabic morpho-syntactic conjugation forms (or Semitic binyanim). The second one is theoretical; it concerns the appropriate theory of transitivity or causativity (or more generally complexity oflexico-syntactic diathesis). I argue that the two problems can be dealt with adequately only through the introduction of Numbe1· (Num) as a feature category of verbs, and the configurations in which it is found, in parallel to nominal Num. Various singular (= Sg) or plural (= Pl) Num configurations would then lead to various syntactic distributions and alternations. Furthermore, the collective/distributive dichotomy of Num is shown to play an important role in determining configurational and semantico-aspectual differences associated with complex predicates. Arabic (as well as Semitic) provides a clear morphological evidence for general nominal/verbal Num marking. The morphology used in nominal forms to express Pl is essentially the same as that used in verbal forms. If Num is generalized (or orthogonal) to categories, then Num theory can serve as the basis for providing a unifying treatment of (a) the 'intensive' (or repetitive) and the 'causative' (or transitive) alternations (of e.g. the second form; FII), and (b) the verbalization (or 'inchoativization') and the transitivization (or causativization) properties (of e.g. FIV). Further properties of other forms (e.g. Fill, FVI, and FX) are analyzed along similar lines. The general Num theory of event/argument complexities receives strong support from alternative uses and complex polysemies of causative reduplicating morphemes across various languages. Significant consequences and predictions oflanguage variation are then naturally accounted for.

*

Part of this work has been presented at the UQAM Asymmetry Project, Montreal December 1999, the 5th Conference of Afro-asiatic languages, Univ. of Paris VII, June 2000, and during my visit to the Department of Linguistics at MIT in September 2000. I would like to thank the audiences there, and especially Anna Maria Di Sciullo, Ken Hale, Alec Marantz, Noam Chomsky, Richie Kayne, Marie-Therese VInet, and Driss Seghrouchni for helpul remarks and discussions. An early and extended version of this work appeared in Fassi Fehri (200 1a).

:18

Key Features and Parameters in Arabic Grammar

Num theory is implemented in the Minimalist framework (as defined in e.g. Chomsky (1995 & 1998)), coupled with Distributed Morphology (cf. Halle & Marantz (1993), and subsequent work). It builds on significant ideas advanced in the literature with regard to verbal Num marking and reduplication uses (typically Kulikov (1993 & 1999), Greenberg (1991), Dolinina (1999), Moravcsik (1978a &b), and the many references cited there), as well as ideas proposed in the context of VP shell structure (Larson 1988), L-syntax composition (Hale & Keyser (1993, 1998 & 1999), little v (Chomsky, ibid), and recent work on word and category formation (Marantz (1997 & 2001), and other work along similar lines). As far as implementation is concerned, it is proposed that Num is a (potential) functional projection of verbs, and that its realization involves either a single Merge (collective), or a plural Merge (distributive).

1.

1.1

Issues Problem 1: Semitic morpho-syntax

The correspondence between morphologically complex Arabic forms (or Semitic binyanim) and their syntactico-semantic properties, alternations, and uses has until now resisted any systematic and/or unifying treatment How can we candle two of the traditionally acknowledged properties of e.g. so-called Form II(= FII) and Form IV(= FIV), stated in (a) and (b)? a.

First, FIV (which involves glottal prefixation) in (1) and FII (which involves second consonant gemination) in (3) appear as the transitivized forms of Form I (= FI) in (2) and (4), respectively: (1)

1-axraja

r-rajul-u

l-walad-a

caus-wentout the-man-NOM the-child-Ace The man made the child go out (2)

xaraja

l-walad-u

went.out the-child-NOM The child went out. (3)

ra,4qama

l-'lamr-a

made-important the-matter-Ace He made (took) the matter (as) important. (4)

ra4uma

l-'lamr-u

became-important the-matter-NOM The matter became important.

Chapter 2. Transitivity, causativity, and verbal plurality

b.

Second, FII and FIV are used in (5) and (6), without inducing any adicity increase: (5)

a.

kassara

l-walad-u

l-kazs-a

broke.intens. the-child-NOM the-glass-Ace The child broke the glass into pieces. b.

jawwala

r-rajul-u

took.walks.intens. the-man- NOM The man took a lot of walks. (6)

a.

kasara l-walad-u

l-kazs-a

broke the-child-NOM the-glass-Ace The child broke the glass. b.

jaala

r-rajul-u

took.walks the-man-NOM The man took walks. (7)

a.

z-aftata

r-rajul-u

come.escaped the-man-NOM The man came to escape.

b. falata

r-rajul-u

escaped the-man- NOM The man escaped. In this use, morphology is rather associated with 'intensive' or 'repetitive' meanings, as in (5), or 'inchoative~ as in (7), i.e. with interpretations which are traditionally thought of as aspectual in nature. Moreover, the transitive (a) and the aspectual (b) uses (or alternations) are normally taken to be non-related, and hence no unifying treatment of them has been proposed. 1 1.2

Problem 2: Transitivity theory

What is the source of transitivity? Does it arise from a VP shell structure (Larson 1988), the projection of little v (Chomsky 1995), or the projection of Voice (Kratzer 1996)? Chomsky (1995, 1998), relying on work by Larson (ibid) and Hale & Keyser (1993; passim) proposed that transitivity arises through a v functional head, which is external to the projection of VP. and which: (a) licenses a second Merge, introduces an external argument (in its Spec), and (b) forms a Case (or

t. In addition to these two characteristics, variable meanings and uses of these forms are documented (for details on the matter, see e.g. Fassi Fehri (1996a & 2001b)).

29

30

Key Features and Parameters in Arabic Grammar

Agree) relation with an object.2 Further work has shown that (c) v has a verbalizing property(combineswitha root R to form a verb) and (d) vhasvarious semantic flavors (agentive, causative, process, stative). 3 Consider the (b) and (d) properties. Unergative vhas the (b) property only if it is transitive, but not if it is intransitive. As for (d), it is not a (semantic) definition of v, since nothing unifies these various semantic flavors. Then what about (a) and (c)? If vPs and their nP nominalizing counterparts have the same internal functional structure, then (c) cannot be of strictly categorical (morpho-syntactic) nature. It has to be replaced by some 'lexico-aspectual' type feature (let it be ev for eventuality), to be able to account for the parallel functional structure of nominalizations, as in (8): 4 (8)

ntiqaad-u r-rajul-i l-masruur-a criticizing-NOM the man-gen the project-Ace The man's criticizing the project.

Assume that ev licenses an argument (arg) in its Spec. Then arg is "internal" in unaccusatives, and "external" in transitives and unergatives. The internal or external positioning of arg is semantically motivated (ct~ e.g. Kratzer (1996)). Ev can be merged with an internal argument (let it be evl), and assigns it an internal theta role. In the case of transitives, a second ev (let it be ev2) is merged with the external argument, and assigns it an external theta role. If so, then transitivity is derived through the occurrence of two evs, one of each licensing one of the two arguments. In other terms, the bivalent (or complex) argument structure of transitives is (bidirectionally) correlated with the bivalent/complex structure of evs. The bidirectionally ofcomplexity between args/Specs and evs/Heads is questionable, however. For

2. These two properties form the so-called Burzio's (1986) generalization. Chomsky (1995, pp. 315-6) claims that "... if a verb has several internal arguments, then we have to postulate a Larsonian shell [... ],where vis a light verb to which V overtly raiseS:: He then adds that "If intransitive (unergative) verbs are hidden transitives, as Hale & Keyser(= HK) suggest, then only unaccusatives lacking agents would be simple VP structures. [ ... ] The external role is a property of the 1'-VP configuration, and a specifier bearing this role is therefore a necessary part of the configuration; a transitive verb assigns an external theta-role by definition': But note that HK's analysis is located at the L-syntax, a level which deals with lexical complexity and diathesis, but not necessarily with Case.

3· See in particular Marantz (1997, 1999, 2001), Alexiadou (2000), Harley (1999), Arad (1999), and Embick (2000), among others. 4· I take a neutral position with respect to the category label of nominalizations. Another option is an early introduction of v, which is later nominalized (a position adopted e.g. by Marantz ( 1997), and criticized by Alexiadou (2000) ).

Chapter 2. Transitivity, causativity, and verbal plurality

example, unergatives might be taken to be headed by dyadic evs, although they merge only with one ev-Spec. 5 Likewise, anti-causative predicates might be analyzed as merging with only an internal Spec, although they are headed by dyadic evs. If true, then the projection of an extra syntactic functional layer (via ev2) does not force transitivity. Furthermore, if v has the category forming property (c), then it is not clear how (a) and the latter property combine. In other words, we are in need of a connection between the category forming property and the complexity property. I take these two problems (the lack of direct correlation between ev and arg complexity and the lack of correlation between category and complexity forming) to undermine seriously the external v theory of transitivity. 6

2.

2.1

Number Theory Ingredients of Num theory

Any theory of transitivity must be able to take into account the following properties: (i) the complex nature of arg/Spec structure (in transitives), (ii) the (potential) complex nature of evs/Hs, (iii) possible relations between complex Specs and complex Heads, with no necessary bidirectionality, and (iv) relations between the use of ev/v as category forming, and its use as complex predicate forming. I claim that complexity relations described in (i)-(iv) can be captured by Number Theory(= Num T). Num Tis based on the following hypotheses: (a) Num (with a Pl value) is the source of complexity; (b) Num is a property ofH/ev or Spec/arg (or both, hence "distributed plurality, with potential multiple Specs); (c) Pl Num when applied to H or Spec can be realized as single or multiple Hs or Specs, depending on whether it is collective or distributive; (d) various cases of complexity are instances of various Num configurations; (e) Voice may affect complexity, but is not the crucial factor in determining it.

5· I interpret Kratzer's proposal in this sense, since the source of e-theta role for the external arg necessarily involves internal binding into the lower V (via what she calls "event identification"). 6. The objections raised against (the external) little vas a source of transitivity carry over to the VP shell theory (proposed by Larson (1988) and the VoiceP (advanced by Kratzer (1996, 2004) ), if they are construed as structural sources of transitives. Voice (applied to L-syntax, to determine the "lexical" internal/external diathesis, or to S-syntax in Hale & Keyser's (1993 & 1997) sense can be viewed as the source of hierarchical organization of arguments (= HOA), which refiects argument complexity only indirectly, at different heights of functional structure (cf. infra subsection 4.3 for discussion).

31

32

Key Features and Parameters in Arabic Grammar

2.2

Verbal plurality and distributed Num

Arabic FII is acknowledged to have the characteristics of event plurality. Medieval Arabic grammarians associate it with the essential meanings of taktiir "multiplicity" and mubaalagah "exaggeration (i.e. augmentative). Western philologists translate these two meanings as "intensive" and "extensive': respectively. Greenberg (1991) has shown that this form (as well as the Semitic piel) has the basic characteristics of verbal plurality, as originally stated by e.g. Swadesh (1946). The latter essentially consist in: 7 (a) reduplication (although partial as in second consonant gemination), (b) temporal1-epetition (or repetitive action), (c) plura.l action on/by many. and (d) natural ergativity (i.e. action on many with transitives, and action by many with intransitives).The two first characteristics are easily observed in the so-called intensive or repetitive FII, illustrated in (5) above. Action on plural objects is illustrated by the following constrasts: (9)

a.

jarra.l;la.

l-junuud-a

wounded.intens the- soldiers- Ace He wounded many (of) the soldiers. b.

jarral;la r-rajul-a wounded.intens the-man- Ace He inflicted many wounds (to) the man.

The difference between the two constructions is that (a) has, in addition to the repetitive/intensive reading, an interpretation which (b) does not have, namely the interpretation of distributed action on many objects. This interpretation of the so-called intensive is sufficiently documented for Standard Arabic, Arabic dialects, and other Afro-Asiatic languages (as reported e.g. in Greenberg ibid).8 Let us take the property (b) to be an instance of Pl Num on verbs (or evs, following in essentials the Arabic tradition and Greenberg), marked here by reduplication. The question then is how this property relates to property (c), i.e. to

7·

Greenberg(1991) takes (c) and (d), which I have given as distinct, to be the same property.

Also, I do not discuss here another characteristic of plurality, namely "spatial dispersion': 8. Action by many (or plurality of subjects) is not so documented. Greenberg (ibid) provides, after Caspari/Wright (1858/1971 ), the following contrast:

(i)

baraka

1-jamal-u

kneeled the-camel-NoM The camel kneeled. (ii)

barraka

n-naram-u

kneeled. intens the-drove-NOM The (whole) drove (of camels) kneeled.

Chapter 2. Transitivity, causativity, and verbal plurality 33

plurality of action on (or by) many? A natural way to relate the two properties is to take advantage of their configurational positioning: temporal repetition arises from Num operating on ev/H, and action on/by many from Num operating on args/Specs. In other words, Pl Num applies to the H/Spec domain, and can be a property of either of H/ev or Spec/arg (or both). I will use the expression dL¢ributed Number to designate the placement ofNum in the second instance, and nondistributed in the first instance. 2.3

Distributed plurality

Greenberg takes properties (b) and (c) (as well as the ergative character of verbal Num, stated in (d)) to follow from the distributive nature of Num, marked prominently by gemination. "[If ... ] the expression of distributive plurality in the verb has as its original sound symbolic meaning "temporal repetition~ then the transition to the meaning "plural object" is not so difficult': He then adds that if one does something repeatedly " it is highly likely to be action on separate objects, hence to involve a distributive plural': (p. 584). 9 I will argue, however, that the cases described (informally) by Greenberg cannot be taken as prototypes of (strong) distributives. To do so, I keep as separate for the moment the distributed Num question and the natural ergativity question. In repetitive events, I take H Pl to be non-distributive, in the sense that the Pl ev is realized by a single H. I also take actions on many or by many to be instances of weak distributivity applied to Specs (or args), since I assume that Pl interpretation can obtain in a Sg Spec. In this view, the distributed Pl in H or Spec in the cases examined above is not taken as (strongly) distributive. If so, I will have to look for cases where strong distributivity applies to H and/or Spec. 2.4

Causative complexity, verbalization, and distributivity

As shown above, the introduction of glottal prefixation (in FIV) may or may not result in transitvization. This variation depends on the nature of the root R to which this form applies. If R is non-verbal, then the form "verbalizes" it (without

9· Greenberg then goes on explaining why (d) holds: "'f one or more agents act on a plurality of things, the action on plural objects is almost certain to be viewed distributively as separate action on a number of distinct objects. On the other hand, separate agents can only carry out the same act on a particular plurality of objects by a concerted action, thus collectively. For an intransitive verb, there is no object toward which action is being directed, each subject will in the usual case be acting independently" (pp. 583-4). It is worth noting that Greenberg keeps the causative meaning separate from the intensive, hence postulating no connection between the two.

34 Key Features and Parameters in Arabic Grammar

adicity increase), but if R is already "verbal': then the outcome is transitivized. The question then is how these two properties can be related, although they have not been so treated. The verbalizing property of the glottal prefix [1] is supported by its ability to create "verbs" from non-verbal Rs, thus forming denominal intransitives, as in (10): (10)

1-anbata l-f:w.ql-u inch-planted the-field-NOM The field became with plants (started having plants).

Likewise, the same prefix verbalizes Rs which can be thought as adjectival, or prepositional, as illustrated by (11) and (12), respectively: (11)

1-axbata-t l-mar1at-u inch-toughed-f the-woman-NOM The woman became tough.

(12)

1-a.rbar-a r-rajul-u inch-spring-ed the-man-NOM The man came to be in spring.

The transitivizing property is illustrated by (1) above, compared to (2). In this case, the outcome is necessarily transitive. Why is that so, and how is transitivization forced? It is reasonable to think that transitives like (1) have both complex heads and complex Specs. The complex His morphologically realized as [1- V], where [1] is taken as another discrete instance of ev (or V/v). In other words, [1] can be taken as reduplicating ev/V, and hence pluralizing it. Spec complexity is instantiated by the existence of two args, an indication of another form of "reduplication" or plurality. Suppose then that the complexity arises from the reduplicating plurality in both cases. Since the complex His not composed of identical evs, and the complex Spec is not either, I take plurality to be of "distributive" character in both cases (with distributive complex Hs and distributive complex Specs). Thus the application of glottal prefixation to already formed verbs results in creating a reduplicated verb configuration, which is read as distributive (by virtue of having two discrete morphemes for the two verbs, although they form a single complex H). Likewise, the realization of discrete Specs in transitives is an indication of the distributive character of Spec Num. If this is true, then [1] prefixation is read off as reduplication of ev/Y, although it is not a straightforward mark of reduplication. However, despite the fact that [1] prefixation results in plurality, it cannot be interpreted as "intensive': e.g. as denoting temporal repetition. This follows automatically if [1] is taken to be (necessarily) interpreted as distributive, due to its "discrete" nature. In contrast, gemination morphology of FII is not so interpreted. FII is normally

Chapter 2. Transitivity, causativity, and verbal plurality

interpreted as "intensive" because reduplication reads naturally as collective. The reason that (Sb) has no extra argument, compared to (6b), can be attributed to the fact that reduplication marks collective plurality of the event. Collective plurality of evs is interpreted as complex H (read collectively. hence "intensitivityj, without necessarily inducing complexity of Spec. The core meanings of FIV and FII, acknowledged in the traditional literature, are then derived, namely that FIV has a transitivizing property, and FII an "intensive" property (see Fassi Fehri (1996a & 2001b) for detail, as well as Larcher (1996) for FIV). But as also acknowledged in the literature, matters are more complex, because FII can also be a transitivizer. This is the subject of the next subsection. 2.5

Two sources of transitivity

There are cases in which FII results in transitivization when applied to verbal Rs, as exemplified by (3) above, compared to ( 4). Likewise, there is an alternate to FII in (Sb ), which can be transitive, as (13) illustrates: (13)

jawwala

r-rajul-a

took.walks.intens the-man-Aec He made the man take walks.

If reduplication is a mark of verbal plurality. how can it be the source of both the intensive (Sb) and the causative (13)? To show how the readings can be related, I need to establish: (a) how FII can behave like a transitive FIV, and (b) how FII can be intensive, whereas FIV cannot. FII is a plural form. When Pl is read collectively on H, intensive results. If Pl on FII is read distributively (on Spec), just like what happens with FIV, then transitivity is yielded in the same way. In other words, Pl can be read ambiguously with FII. The distributive/collective ambiguity of Pl in FII recalls the ambiguity found in the interpretation of universal nominal quantification described in Fassi Fehri (1999); see also Chapter 11 of this volume. There, it is shown that under specific structural conditions the universal quantifier kull can be ambiguously interpreted, as illustrated in the following examples: (14)

zakala kull-u

r-rijaal-i

dajaajat-an

ate

all-NOM the-men-GEN chicken-Ace All the men ate a chicken.

(15)

zakala kull-u

rajul-ayn-i

dajaajat-an

ate all-NOM man-dual-GEN chicken-Ace Every two men ate a chicken. The quantifier in (15) can be strongly distributive in the sense that there should be more than one chicken, and each two men ate one. It is rather collective in (14), in

35

36

Key Features and Parameters in Arabic Grammar

that there could be only one chicken eaten by all the men (or only weakly distributive, in the sense that the existence of more than one chicken is not required by the interpretation). The transitivizing property of FIT is hardly questionable. Arabic dialects have simplified their conjugation form system by doing without FIV, hence transferring its functions to FII, which has become the unique form for expressing both intensive and causative meanings. The ambiguity of FII and its transitivizing property in Standard Arabic come then as no surprise. 10 The question to be addressed now is why FIV cannot be intensive. In order to answer this question, we need first to examine more closely some details of the verbal morphology of plurality. 2.6

Parallel plural morphology

Up till now, I have established a parallel between nominal and verbal quantification for the distributive/collective distinction. It would be a welcome step to reinforce this parallelism by examining other features/properties of plurality. In fact, other parallels can be, and have been (at least partly) acknowledged in the literature. I will point to some (presumably) non-accidental similarities here, leaving aside the details for further research. Typically, glottal prefixation and gemination (whether consonantal and vocalic) play a role in nominal and adjectival plurality, which is parallel to its role in plural verbs. Consider first glottal prefixation. There is a class of so-called broken plurals which denotes small sets (from three to ten normally), named jumuur 1-qillah "paucity plurals" by traditional grammarians, and which has the essential common property of being headed by the glottal prefix: (16)

a.

~ayn "eye" --71a~n; 4i1b

b.

yawm "day" --71ayyam; zaman "time" --71azmaan; ~inab "raisin" --71amaab; tiql "heaviness" --71a!qaal; burj "tower" --7?abraaj; ~unuq "neck" --71a~aq

c.

~imaad "support" --71a~midah; ta~aam "food" --71at~imah; ragnf"flat loaf of bread, roll" --71argifah; burj "tower" --7 ?abrijah; ~amuud "column" --71a~idah

"wolf" --71a4_1ub;

to. In Fassi Fehri (1996b & 200 1b), it is argued that transitive FII selects a stative (adjective or noun) as an internal complement, whereas FIV selects a non-stative, and that FII induces a change of state of the complement, but FIV does not. When FII selects non -statives, it is interpreted as "coercive': Le. implying the use of energy or force by the causer, against the will of the internal agent. Fwthermore, the subject of the internal verb controls the action with FIV, but not with FII.

Chapter 2. Transitivity, causativity, and verbal plurality 37

I take the prefix [1] in (16) to be the essential morphological segment for marking plurality. Vowels of the singular are not transferred to the plural in any significant sense. For example, vocalic change is not regularly "apophonic': 11 Whenever vocality is relevant, it has to do with lenghtening/geminating the vowel. It is then clear that the unifying feature of these forms is glottal prefixation, a morphology which parallels that of the "verbal" plural causative found in FIV. Consider now gemination and/or lengthening. Consonantal gemination is involved in the so-called "forms of exaggeration'' ($iyag al-mubaalagah) by Arabic grammarians. The latter designate the subject of the event which repeats "exaggeratedly" the action, as in the following examples: (17)

a.

?akkaal "a big eater': sarraab "a big drinker': k.addaab "a big liar': nawwam "a big sleeper': sikkiir "a big drinker (of alcohol)".

b

sakuur "very gratefur, waduud "very friendl~ kasuul "very lazy, jahuul "very ignorant'; waluud "giving birth to many children, fertile~ saruub "big drinker'; 1akuul "big eater'; samiis "big listener': qadiir "very apt'; saliim "big knower': sariib "big drink.e~

The translation with a quantity adjective (or adverb) is a mere approximation. I take the gemination to express an "intensitivity~ applied to an adjectival subject. 12 Vowel gemination (or lengthening) appears also as an essential building segment of a class of broken plurals: (18)

a.

qi1b "wolf"~ 4i1aab; jabal "mountain"~ jibaal; rum~"lance" ~ rimaal}.; rajul ~ rijaal

b.

qayd "constraint" ~ quyuud; jisr "bridge" ~ jusuur; namir "tiger" ~ numuur; burj "tower" ~ buruuj

These plurals are taken to be less marked, compared to paucity plurals. They are interpreted as "plurals of abundance" when contrasted with the latter, otherwise

n. For various relevant versions of apophony paths and their motivation, see Kurylowicz (1973), and Guerssel & Lowenstamm (1996), among others.

u. It is worth noting that these subject forms cannot be "'derived" from the verb, in the traditional sense of derivation, since the corresponding intensive verbs do not exist to my knowledge (cf e.g. *sakkara r-rajul-u"'the man drank a lot': *?akkala r-rajul-u "'the man ate a lot"). Cf. Fassi Fehri (2001a) for detail as well as analysis of forms making use of vocalic gemination, and yielding eventually doublets or triplets.

38

Key Features and Parameters in Arabic Grammar they are just plurals, with no mere specification. They can also be contrasted with "plurals of plurals" (munta.haa 1-jumuur), exemplified in (19): 13 (19)

a.

qawl "saying'--+ ?aqwaal--+ ?aqawiil

b.

jamal "camel" --+ jimaal --+ jamaa?il

c.

kuttaab "Koranic school" --+ kataatiib; jumhuur "public" --+ jamaahiir; diiwaan "cabinet" --+ dawaawiin

Nominal vowel gemination has a verbal counterpart in Fill. The latter, according to traditional grammarians, express the meaning of"participation" (mu5aarakah): (20)

a.

maaJaa-hu walked.PART.-him He walked with him (or: he shared walking with him).

b.

saa.baqa-hu raced.PART-him He raced him.

c.

saaraba-hu drank.PART.-him He drank with him.

What "participation" means in this case is the fact that both participants are interpreted as actors of the action. But although the event is the same (e.g. "the race"), it has plural participants. The plurality of participants (the "nominal" part of the verb meaning) appears then to be distributed on Pl Specs. FVI, the reciprocal form, plays the role of unifying the two participants: (21)

t-asaa.baqa zayd-un

wa.-ramr-un

rec-raced Zayd-NOM and-Amr-NOM Zayd and Amr raced each other. But consider the following pair: (22)

saabaqa zayd-un

ramr-an

raced Zayd-NOM Amr-Acc Zayd raced Amr. (23)

saabaqa ramr-un

Zayd-an

raced Amr-NOM Zayd-Acc Zayd raced Amr.

13.

These plurals involve double gemination of the vowel, or a double pluralization process

(with various forms, arguably related to a much simpler plural form). But the important idea is that these forms are semantically "intensive" plurals, hence involving "double" plurality.

Chapter 2. Transitivity, causativity, and verbal plurality 39 Although the interpretation of both (22) and (23) appears to involve reciprocal action, this is not to be confused with the reciprocal. In Fill, the subject is more active than the object, and the two sentences are not equivalent. In the reciprocal FVI in (21), both participants are understood as "subjects" of the "same" event. I take these subjects to have "fissioned" into subject and object positions with Fill, and licensed in Pl Spec positions. Vowel gemination, read as a functional Pl head, is satisfied through Pl Specs in this case. If so, then the question is why this Pl cannot be read as intensive, and why Fill is interpreted as only distributive. 1his situation recalls that found with FIV. Interestingly enough, FIV and Fill can be, and in fact have been, related morphologically. McCarthy and Prince (1990), adopting Levy's (1971) view, propose that the two forms be related through a local metathesis of the open vowel in the first syllable. The relatedness view can also be implemented morpho-syntactically. If the two forms are both marked as Pl. realized by a long vowel [aa], then the latter fissions in FIV (in the sense of Fassi Fehri ( 1996b ); see also Chapter 12), i.e. it has two segment heads, instead of one. In other words, FIV has more structure than Fill, with one extra head up. The two segments of the Pl head in FIV then license "discrete" evs, whereas the Pl head in Fill licenses only one head, hence the "participation" meaning. FIV has no participation meaning, because the two participants are args of different evs. Put simply, Fill has only a distributive Spec, but no distributive H, but FIV has both. Then what about difierences between FII and Fill? Recall that FII can be intensive or transitive, its transitivity being interpreted as distributive Spec, and its intensitivity as collective H. The question then is why Fill cannot be intensive (intransitive), just like FII. One possibility is to take its distributivity to be encoded "lexically~ much like what happens with quantifiers each and all in English.14 But there is no evidence for such lexical/vocabulary encoding. Another possibility is to explore a difference in marking Pl in Fill (and FIV) on the one hand, and FII, on the other hand. FII geminates the second syllable of the word for encoding Pl; Fill (and its derived "daughter" FIV) encodes Pl gemination on the first syllable. In the nominal system, Pl is only encoded in the second syllable. Suppose then that Pl is normally encoded on the second syllable and, as suggested earlier, verbal plurality in FII is encoded on consonants, whereas nominal plurality is encoded on vowels.IS When plurality is encoded on a single segment

14· See e.g. Beghelli & Stowell (1997) for such a lexical approach. 15. As I explained earlier, the consonant/vowel gemination contrast cannot be taken as a distinctive category feature. For example, finite verbs of FII and their participles have geminated consonants (as in haddada"to threaten': muhaddld, muhaddad), but in action nominalizations the gemination is vocalic (tahdlid "threatening").

40

Key Features and Parameters in Arabic Grammar

H in the second syllable, we get the collective/intensive interpretation. Now the fact that Pl is found in the first syllable in Fill is somehow unexpected. This "displacement" might be in fact an indication that movement has occurred internally, thus creating a two-segment head, which then licenses two Specs (or a distributive Spec), hence the transitivity. The absence of the intensive reading appears then to be correlated with this displacement, much like what happens with transitive FII and FIV. If FII, Fill, and FIV are morpho-syntactically related, then if Fill derives from FII via one step movement inside the consonantal stem, FIV derives from Fill by peripherical movement, outside the consonantal stem. This further step and splitting movement, coupled with presumably more structure (and height), is what is behind double distributivity found in FIV, which excludes both the intensive/collective meaning (by virtue of fissioning like Fill in a first step), and the participation meaning (by virtue of fissioning in a second step to the periphery). 16 2.7

Summary

I have shown that FII gemination expresses multiple or repetitive action, interpreted as ev Pl. Pl of action can also be interpreted as operating on multiple individuals, hence the arg multiplicity. Thus the plurality/multiplicity, which is encoded on the verb, can (ambiguously) apply to evs or to args. Fill gemination expresses plurality of participants. But participants are not treated as "equal': Pl is partitioned, hence the transitivity. In the reciprocal FVI, the participants are assembled, hence the intransitivity, even though the ev is Pl. Glottal prefixation in FIV, analyzed as a fissioned vocalic gemination, also expresses partitioned plurality of participants, hence the transitivity. FII may also express plurality of participants, and it becomes transitive through partitioned plurality (alternating with assembled plurality in the repetitive). FII then contrasts with FIV in this respect, since the latter cannot be an assembled plural. If the inchoative FIV is conceived as a distributive Pl over evs, and the causative as a distributive over args (as well as evs), then the alternating semantics of FIV is limited to distributivityvariation. Complexity of H can be separated from complexity of Spec, and H distributivity separated from Spec distributivityY

It is possible to view the reciprocal participation meaning in Fill coupled with its obligatory transitivity as arising from a sort of (a fissioned) cumulative Pl, which licenses two Specs, but not two events (e.g. in the race case, the race has to be one and the same).

16.

There are no doubt other sources of transitivity (or intransitivity) which I do not address here. For example, apophonic [a] is used to form denominal transitive verbs (compare l;!azlna "to be sad" with l;!azana "to put sadeness into someone"), iambic reduplication to create "attenuation" forms of verbs like 5am5ama "to sniff something': compared with samma "to smell

17.

Chapter 2. Transitivity, causativity, and verbal plurality 3·

Cross-linguistic evidence

Up till now, I have provided reasons to think that the Arabic verb system makes significant uses of plural quantification, which not only parallels that of nominal quantification, but enables us to derive the intensity of verbs, the complexity of ev/arg structure, and to account for alternating behaviors of verbal Num morphology. It would be a welcome step to see how this system carries over to other languages. As a matter of fact, a number of convincing descriptive studies have already established the correspondence between reduplication morphology and (a) causative (or transitive) formation, (b) verbal plurality (or more generally event quantification), and (c) variable behaviours and meanings of the same form. For example, Moravcsik (1978a) has focused typically on the (b) relation, investigating various forms of reduplication and"... the meanings of increased quantity, intensity, diminution, and attenuation which are concepts capable of pulling together many superficially disparate uses of reduplication constructions ...': Similarly, Greenberg (1991) has especially highlighted the relevance of the (b) correlation with respect to Semitic intensive forms. The two studies have in common that they do not have an answer to the question of how properties (a) and (b) are connected. Other compelling work, however, provides rather strong empirical evidence for correlating (a) to (c) properties, typically outlining the role of reduplication (or other morphologies) in causative and transitive formation, and its extensional use for expressing various (b) meanings (cf. in particular Kulikov (1993 & 1999), Dolinina (1999), and the many references cited there)). In 3.1, I first examine some of the relevant data and analyses which provide motivation for this direction of research, and in 3.2. I discuss some of Moravscik's resistant cases to her view of reduplication, which in fact add further support to the correlation of (a) to (c) properties. 3.1

Causatives, transitives, and event quantification

Cross-lingustically, morphological reduplication (in particular) is used to form not only plural verbs (interpreted as repetitive, expressing multiple actions or

something': xanna "to speak through the nose or to nazalize': xan:tana "'to speak through the nose or to nazalize repeatedly': xa5xa5a "'to make rustle or rattle': gargara "'to gargle': etc. 1n these reduplicating forms, each part can be analyzed as a small part of the event, and ev seen as "fissioned" or partitioned over pieces of the root. The fissioning ev can be thought as a sort of paucity plural, in that the whole ev is made of"'small" or"paucal" (parts of) evs (cf. Fassi Fehri (200lb) for detail).

41

42

Key Features and Parameters in Arabic Grammar

participants, distributivity, etc.), but also single causatives or transitives, double causatives, in addition to exhibiting multiple behaviour.

Causativization and transitivization Causative or double causative formation through reduplication is well-documented in the literature. The following constructions are brought up by Kulik.ov (1993) from various sources:

3.1.1

(24)

a.

wanu "to die"

b. wanu-chi "to kill" c.

wanu-chi-chi "to cause to kill"

(25)

a. b. c.

rupit(a) "to work" rupita-pt(u) "to cause to work" rupita-ptu-pt(u) "to ask to work"

(26)

a. b. c.

ut' "to sleep" ut'-k' "to make to sleep" ut'-k' -ek' "to cause to make to sleep"

In Huallaga Quechua (24) (taken from Weber (1989)), the causative morpheme "chi" forms a transitive causative in (24b) and, when reduplicated, a double causative in (24c). In Mansi Vogul (25) (quoted from Robandeeva (1973)), the morpheme "pt" marks the causative in (25b), and is reduplicated to mark double causation in (25c), with a "requestive" meaning. Finally. in Hunsib Daghestan (26) (taken from Isakov (1986)), the causative morpheme "k:" in (26b) reduplicates to form the double causative in (26c). 18 This data provides clear evidence that reduplication is involved in causative formation. When a morpheme is used to causativize a verb, then its reduplication forms a double causative. Reduplication is not of the root. but of the external morpheme. The latter (as described in these studies) does not originate from an intensive or quantificational source. Moravcsik (1978a) also lists (partial) reduplication as a source of transitive formation (a case that she takes as resistant to her view). The following illustrating Sundanese pair is brought up from Robins (1959): (27)

a. b.

narerewas "to frighten" rewas "to be afraid"

Here, the reduplication is root-based, as in Arabic. Root-based and morphemebased reduplications are then used for transitivization or causativization.

t8. See Kulikov (1993) for these references.

Chapter 2. Transitivity, causativity, and verbal plurality

Multiple behaviour

3.1.2

Intimate relationships between causative and "aspectual" meanings of morphemes have been repeatedly noticed in typological studies, and various polysemies observed. 19 In Turkish, either causative or intensive/iterative interpretation of a given form can be acceptable in precisely the same context. Thus double causative morphemes may refer either to double causation, or to intensive/iterative causation:20 (28)

miidiir-e mektu.b-u ac-ti'r-t-ti'-m. director-OAT letter-ACC open-CAUS-CAUS-PAST-lSG a. b.

I had someone make the director open the letter. I made the director open the letter forcefully.

In Yukaghir, the verbal suffix t expresses cumulatively both causative and distributive meanings (Maslova (1993)): (29)

a. b.

sel'ge-t "to break several distinct things" joy@-t "to open (tr.) several times"

Similarly in Aleut, the causative dgu instantiates the cumulation of causative and distributive meanings (Golovko (1993); see (32) below). In Bouma Fijian, the prefix vaa forms causatives with some verbs and intensives with others:21 (30)

a. b.

vuli(-ca) "learn, study" va'a-vuli( -ca) "teach"

(31)

a.

mate"die"

b. va'a-mate-a "kill" Further polysemies are provided by Kulikov (1993), who observes that in e.g. Tuvan, the double causative can function as iterative of the first causative. The author further notes that in some cases "... the boundary between iteratives to causatives and plural causatives is rather vague': In Aleut, there are two causative morphemes: t, which denotes causation, as in (32b ), and dgu, which is used with plural causees, as in (32c): (32)

a.

igluqa-r qaka-ku-r hide-SG dry-NON.FUT-3SG The hide is dry.

19.

See the many references cited in Kulikov (1999), including Nedjalkov (1966).

20.

Cf. Zimmer (197 6), quoted after Kulikov (ibid).

21.

Cf. Dixon (1988), quoted after Kulikov (Ibid)

43

44 Key Features and Parameters in Arabic Grammar

b.

ayagar

igluqa-r qaka-t-i-ku-r

woman-SG hide-SG dry-CAUS-Epenth-NON.FUT-3SG The woman is making/makes the hide dry. c.

ayagar

igluqa-s qaka-dgu-ku-r

woman-SG hide-PL dry-CAUS-NON.FUT-3SG The woman is making/makes the hides dry. But Golovko (1993) argues that dgu "... renders not the plurality of the causee but rather the plurality of the caused events, thus expessing causativity and distributivity at the same time~

Event quantifica.tion

3.1.3

Reduplication is used to form distributive verbs, as documented in various languages. Dolinina (1999) illustrates this by (33) from Classical Nahuatl (quoted after Suarez (1989)), (34) from Cree (after Harden (1934)), and (35) from Moses Columbian (after Kinkade (1977)): (33)

(te t.tla)-maka --+ (te t.tla)-ma?-maka to give sb.sth to give sth to each person

(34)

makiw --+ mu-makiv he gives he gives distributively

(35)

lx lx leq-laqsitsG/RedupVDist-Activity NON-SG/ Agent Each has a position in which to sit.

In Evenki (48), a suffix encodes the multiplicity oflocations:-22 (36)

loko --+ loko-t hang (smth) hang (separately (and in different places))

Similar phenomena occur in Aleut, Eskimo, Turkish, Armenian, Japanese, Indonesian, etc. 23 Likewise, Moltman (1997) observes that the same morphology may be used for iterative (collective) or distributive verbs. Thus in Copanaila Zoque, the idea of repetition is marked by a non-differentiated iterative-distributive prefix. which can have both readings: (37)

min-gezt-u He came again/ he also came.

22.

Cf Nedjalkov (1989), quoted in Dolinina (Ibid).

13.

Dolinina (ibid).

Chapter 2. Transitivity, causativity, and verbal plurality

But in Classical Nahuatl, event plurality, marked by reduplication, expresses (only) iteration:24 (38)

wi.tek wiz-wi.tek to hit to hit repeatedly

In Tepetotula Chinantec, a specialized affix marks repetition in time: 25 (39)

3.2

ki-ka-znauz za haz REP-PAST-seek.TR.ANIM.3 3SG animal She/he repeatedly hunted the animal.

Moravcsik's resistant cases

Moravcsik (1978a) examines what she calls derivational uses of reduplication, and she observes that there is a"... large and indeed disparate set of examples of derivational meanings that cannot be seen 1·elated [my underlining; FF] to any of the non-derivational meaning categories surveyed ...". But the list of resistant cases she establishes is precisely the sort of data which can be used to support our Num theory of adicity (i.e. they come as "good news"). Transitivization and intrantivization uses are typically among the cases which Moravcsik takes as resistant to her more or less quantificational picture (see the Sundanese example in (27) above). But if transitivity is a form of abstract reduplication, as I have shown, these morpho-syntactic alternations represent exactly the kind of verbal behaviour that we expect. As a matter of fact, Arabic counterparts of these pairs are related through glottal prefixation, taken as an abstract form of verbal plural reduplication, as argued earlier: (40) (41)

1-atZa.~a

"to frighten someone" afraid"

fazi~ "to be

In the same direction, and also of great interest, the author mentions, after Harrisson (1973) and Christaller (1875), respectively, that in Mokilese and in Tiwi, intransitives are derived from transitives, as illustrated by the following contrasting pairs: (42)

a. b.

koskos "cut (intr)" kos "cut (tr.)"

(43)

a. b.

didi "eat (intr.)" di "eat (tr.)"

24·

Ct: Suarez (1983), quoted after Dolinina (ibid).

25.

Ct: Westley (1991), quoted after Dolinina (ibid).

45

46

Key Features and Parameters in Arabic Grammar

Such a direction of derivation is also found in Arabic.26 Furthermore, the more general phenomenon of morphologically marking the intransitive with respect to the transitive is not that surprising cross-linguistically. What is important here is that we have substantial evidence from other languages that reduplication is used for marking (in)transitive alternations, whatever the direction of the derivation is. This is exactly the kind of data that we expect, once complex (in)transitives are taken as Pl evs. 27

4·

Conceptual motivations and competing analyses

As observed above, Num is a cross-categorial quantificational semantic category (or feature), which holds for events and objects. Being orthogonal to categories (or types), it enables us to account for event/argument complexity in causatives/transitives, multiplicity of events or objects, well-acknowledged interactions, as well as alternative (or cumulative) uses of argument and event quantificational morphemes (seen traditionally as aspectual). Consequently, Num eliminates potential competitors which cannot account for these properties. Some of these competitors are investigated in this section. 4.1

Little v: Verbalizer or transitivizer?

Consider again little v, taken as a source of transitivity (in e.g. Chomsky (1995 & 1998)). The latter is construed as a transitive head only by stipulation: the stipulation that there is a bidirectional relation between ev and arg complexities. Given this hi-directionality, it is not clear how to capture structural difierences between e.g. unergatives and transitives, or anti-causatives and causatives. Moreover, differences between objective cases (accusative and partitive) are not naturally captured. Finally, if vis a transitive head, it is not clear how it can be equally taken as category forming (or a verbalizer). Nothing in the v theory of transitivity predicts this correlation. In Hale & Keyser (1999), transitivization and verbalization are distinct operations. Transitives arise from two different structural sources: (a) a complement source (e.g. to calve, from "have a calf~ to cake, from "make a cake': to laugh, from "do a laugh"), which allows no intransitive alternant, and (b) a spec!fier source,

This happens. for example, when FIV is used as a resultative intransitive, alternating with a simple transitive Fl. Ct: Fassi Fehri (2001b) for illustratioiL

:16.

27. Other resistant cases to Moravcsick's view are analyzed in Fassi Fehri (2001b) along similar lines.

Chapter 2. Transitivity, causativity, and verbal plurality 47 which is optionally internalized, hence the (in)transitive alternation. According to Hale & Keyser, unaccusatives force the projection of a Spec, and it is this property which permits transitivization (p. 55). Unergatives or transitives, by contrast, do not project (internal) Spec. Differences between the two constructions with respect to transitivization are then related to differences in specifier or complement branching. Transitivization results from embedding a verbal projection as the complement of another verb (p. 53). Unaccusative verbs alternate because both the inner head and the outer head are verbs, and the intransitive is simply the inner projection unmerged with another verb (p. 57). Furthermore, V (or v) is category forming of both unergatives and unaccusatives. In Harley (1995), eventiveness in inchoatives and causatives is introduced by v (Become and Cause, respectively), not the root. The complement of the vP, a rootP, is a predicative structure denoting a state, the end result of the change of state introduced by the v head. But Harley (1999), building on Hale & Keyser's work. adopts the view that roots may also denote things or events, in addition to properties. 28 Likewise, Embick (2000) argues that vis a verbalizer (with e.g. -ize, -en as spell outs), which forces an eventive interpretation. The absence of attachment of v implies the absence of eventive interpretation, hence the structure of pure statives (which are simple states, with no implication of prior event). As for Aspect, it attaches either to Root, or above v. We see then that these authors (among others) posit a verbalizing property of v, but there has been no natural proposal to relate this property to transitivization (via e.g. reduplication or other morphology). Harley's and Embick's "flavoured" views are essentially diacritical, and do not differ radically from Marantz's previously analyzed proposal. In my view, the first vis a verbalizer, and the upper vis a reduplicator, which can then transitivize the structure under appropriate conditions (but not necessarily, as in e.g. iteratives).

42

Aspect

Traditionally, noun quantification and event quantification have been regarded as belonging to different domains of grammar: Num (as an a temporal quantification) is associated with the former, and Aspect (as temporal quantification) is associated with the latter. But as correctly observed by Dolinina (1999), in the case of event

28. Harley (1999) proposes an underspecified view of v, in which vis detennined contextually via a taxonomy of verbal environments: (i)

a. b. c. d.

v + State, no Spec v + State, Spec v + Thing, Spec v + Event, Spec

=Become =Cause =Make =Do.

48

Key Features and Parameters in Arabic Grammar

plurality, quantification is more abstract than Aspect, because event plurality can be triggered by repetition on the axis of time (and thus be related to Aspect), but also by plurality of activities of the participants, and hence lies in other dimensions than Aspect. Distributive constructions, for example, obligatorily contain a component of quantification, but only in a few of them it is due to repetition in time (and hence Aspect). Distributivity then belongs to the (general) cluster of quantifier categories, and not to the temporal cluster. Semantics of distributivity, its typological encoding, and the oppositions it deals with are broader than Aspect. Distributivity is a category with a dual nominaVevent nature, combining the meanings of nominal Pl and event PF9 In observing causative/iterative or intensive polysemies and interdependencies, many authors have identified the latter meaning as "aspectual" (see e.g. Nejdjalkov (1966), Comrie (1985c), Li (1991), Kulikov (1993), among others). The latter author traces the polysemy (or what he calls "causative split") back to intensivity, taking it as the essential component of causation. This is so because, as he puts it "causing someone to do something implies channeling extra force from outside into the situation. The meaning "more forcefullY: "more eftectively" may be thus the common semantic denominator shared by causativity [... ] and intensitivity, iterativity, etc. [... ] It is for that reason that these aspectual meanings can become associated with causativity and, in a sense, appear as its side eftects" (p. 26). But the identification of iterativity or intensivity with Aspect (instead of e.g. Num and Manner), and the mechanical association of causation with Force are questionable. Thus there are Arabic causative doublets, which are precisely distinguished by implying or not implying Force: FII (when causative) cumulates intensity, while FIV does not:

29. Dolinina observes that there have been renewed attempts to unify the area of affiliation of quantification, with some linguists regarding nominal quantification as aspectual (Rijkhoff (1991), Verkuyl (1993)), and others arguing that event quantification belongs to the cluster of quantificational categories (Dressler (1968), Xrakovskij (1989)). But as she puts it, the question is not to choose only Num or only Aspect, but to see which of these features is the "dominant" component. Likewise, Moltmann (1997, pp. 56-57) argues that the ev argument in distributives is a group ev, rather than a single ev, just like what is required in repetitive evs. In the latter case, the ev is a group, but unlike the distributive, it consists of single evs that all have the same participant. If the members of an ev group stand in the relation denoted by the verb to parts of a participant group, the result is not repetitive, but rather distributive. Crosslinguistic evidence supports the assimilation of repetitive and distributive readings: a verb whose ev argument is a group ev is marked with a special distributive morphology, regardless of whether this group event contributes a repetitive or a distributive interpretation.

Chapter 2. Transitivity, causativity, and verbal plurality

(44)

1-axraj-tu l-walad-a CAUS-go.out-PAST.! the-child-ACC I made the child go out

(45)

xannj-tu l-walad-a go.out-gem-PAST.I the-child-Ace I made the child go out (by force; against his will).

Note, however, that the distinction is made only when FII is cumulative. Most instances ofFII are not, and they express either intensive or causative, but not both (see e.g. (5) and (15) above). Note also that in the causative/intensive complementary distribution, the choice between FIV and FII is fairly fixed, FIV being causative, and FII intensive/iterative/distributive. The following contrasts illustrates this fixed meaning opposition: (46)

qattar-a l-walad-u l-lal;Jm-a cutgem-PAST the-child-NOM the-meat-Ace The child cut the meat into pieces.

(47)

1-aqtar-a r-rajul-u l-walad-a l-labm-a CAUS-CUt-PAST the-man-NOM the-child-ACC the-meat-ACC The man made the child cut the meat.

Clearly then, FII is essentially intensive, and FIV causative. FII can also be causative, or cumulative, but FIV is not cumulative. Consequently, no unified account of the two causative forms can be provided through an intensitivity source (seen as Force/ Aspect). Hopper & Thomson (1980) take as an essential evidence for the correlation between transitivity and tense/aspect well-known oppositions found in the Finnish case system, between accusative or partitive objects (the latter contrast correlating with the aspectual perfective/imperfective opposition). Clauses with partitive objects are shown to be less transitive than those with accusative ones. Likewise, in "split ergative" systems (such as Hindi/Urdu, Australian, and Ameridian languages), the ergative construction is limited to perfective and preterite environments, whereas its absolutive counterpart is confined to imperfective or non-preterite (Dixon (1994)). In a similar vein, Kulikov (1999) examines how in Vedic (and Ancient Greek), the transitive-causative constructions are mostly employed with the present, whereas perfect constructions are typically intransitive. But this connection has to do with another ingredient of transitivity, that is Case, and typically accusative or ergative Case.30 While I agree that Tense/Aspect

30. I follow Hopper & Thompson (ibid) in taking transitivity as a complex set of features, rather a simple binary opposition, but these ingredients distribute over heights, and are not "a continuum': as they put it.

49

50

Key Features and Parameters in Arabic Grammar

(typically Perfective) and Accusative Case, for example, are interdependent, I deny a direct connection between Aspect and the argument complexity component involved in transitives or causatives. More crucially, I deny any clear direct connection between intensity and iterativity (seen as aspectual) and transitivity. Summarizing, the fact that verbal Num encodes plurality of actions, or plurality of arguments operating these actions is not accidental. It is rather motivated by semantic considerations. The semantic duality of repetitive/distributive morphemes requires a much broader category than Aspect, and transitivity/causativity requires complex functional ingredients in which Num is playing a major role, yielding common derivational sources and intricate, although principled, polysemies. 31 4·3

Voice

Voice distinguishes (a) active (agentive or unergative) predicates from non-active (ergative or middle) ones, or (b) passive (eventive or stative) verbs from nonpassive ones. Voice can thus be seen as operating hierarchical organization of arguments(= HOA) with respect to predicates, rather than taking care of argument complexity. 32 Other theories of Voice have treated argument complexity, HOA, and even specific properties of external arguments under Voice.33 In my theory, complexity, stated in terms of Num, is the source of (in) transitivity. H and Spec complexities being dissociated, distributive complex Specs turn out to be a typical configuration for transitivity. Cases of HOA, however, are best accounted for via Voice, but not Num. Functional heads expressing Voice (diathesis or HOA) occur at two different heights: (a) a low position, which can be found lower than morphosyntactic categorization (v, n, a, p), and which enables us basically to distinguish arguments of internal ev or external ev (call the internal positioning ergative diathesis, and the external positioning unergative diathesis); (b) a high position, which

The connection between Num and Aspect has been repeatedly noted in the literature. But as far as I can tell, no precise analysis of the connection has been proposed. Holisky (1985), for example, argues that it is the verb Aspect, rather than Num (plurality), which is relevant for determing the (im)perfectivity of the root in the Aorist in Tsova-Thsh, contra Guagua (1962) and Schiefner (1859), who go for the opposite, i.e. Nwn. However, what Holisky shows in fact is that what makes the difference is not the single/plural dichotomy, but rather the distributive/ non-distributive distinction, in my sense. Cf. Fassi Fehri (2001a) for detail.

31.

For various recent views of Voice, and how it connects to transitivity, see, among others, Kratzer (1996),Jelinek (1998), Borer (1998), Embick (1998, 2000), Fassi Fehri (1988b), and also Benveniste (1950).

32.

33· According to Embick (1998) "... voice morphology makes reference to a particular spectfier

of vP (the external argument), as well as the properties of that specifier':

Chapter 2. Transitivity, causativity, and verbal plurality

is projected higher than morpho-syntactic category formation, and which organizes arguments with respect to Case, in particular Accusative Case in the Accusative/Nominative Case systems, and Ergative in the Absolutive/Ergative languages. Low voice applies at L-syntax. while high voice applies at S-syntax. The latter distinguishes e.g. active accusatives from passive non-accusatives. 34 43.1

Anti-transitive rejlexi1'es

FI transitives may have two anti-transitive counterparts (or sorts of reflexives) which correspond to different interpretations, and which should be taken as doublyheaded, although they are monadic with respect to their args/Specs. This is the case of the transitive verb gasal "to wash" with its two detransitivized forms in the following pair: (48)

n-gasala l-makaan-u ref-washed the-place-NOM The place has been washed.

(49)

g-t-asala r-rajul-u ref-washed the-man-NOM The man washed himself.

The two morphologies ([n] and [t]) apply only to transitive roots, and they indicate the "directionality" of derivation: the intransitive is the morphologically marked form. This formal complexity is coupled with complexity of interpretation, yielding a reflexive interpretation in (49), and an "affected" anti-causative in (48). Various tests (control structures, agent-oriented adverbs, passivization) show that (49) is agentive, and (48) is not. Given these properties, both constructions have to be analyzed as having PI distributive Hs (as is the case with normal transitives), but they only have a Sg arg/Spec. The difference between them has then to do with HOA: the subject of (65) is internal (or ergative), that in (49) external (or unergative). 35 Then what about implementation of the difference? I would like

34· The double distinction in Voice systems is not new. Benveniste (1950) proposes two dichotomies: (a) active/middle and (b) active/passive. Burzio's (1981) generalization (associating unergativity and accusativity) reflects the duality of the system. Moreover, Moravcsik (1978b) has established the ntixed nature of languages with respect to ergative/accusative patterns. Various instances of event/argument structure organization can be treated as low voice distinctions, and they have to be kept separate from arg complexity, treated as arising from Num specification. 35· Many other doublets point to the same analysis: n-fatal;a "to open (by itself)" ft-ataha "to open something to his benefit': n-sawaa "to get roasted': S-t-awaa "to roast something to his benefit': n-!agala "to get occupied': s-t-agala "to occupy oneself': etc.

51

52

Key Features and Parameters in Arabic Grammar to introduce a syntactic (and lexeme-based) approach of the latter. Suppose that the detransitivizing morpheme is expressing an (internal) ev, and it can be placed either internal to the stem, or external to it. When internal to the stem, the subject of the entire stem c-commands immediately a segment of the stem which includes (or c-commands) the (internalized) affix [t]. Consequently, the external unergative subject is a subject of the complex ev, dominated by the unergative (or active) ev. The active reflexive interpretation then results from the fact that there is a unique subject of both evs, in which the higher ev and its subject are active. On the other hand, when the (internal) affix is placed externally, the "internal" subject has to be "externalizecF, presumably to "match" the ergative [n], and the unergative subject is not realized (presumably because it can only be external), hence the "passive" anti-causative interpretation. Note that this process of derivation recalls in part the internal/external distinction between FII and FIV. A similar case of distributed ev/arg complexity can be raised with the benefactive FVIII, compared to the transitive FL as in the following pair:

(SO)

qa(.a.ra l-la.f;lm-a cut the-meat-Ace He cut the meat

(51)

q-t-a-tara l-lal;tm-a cutref the-meat-Ace He cut the meat to his benefit

In (51), the number of args is not increased syntactically, although it is understood that the verb has an extra theta slot. I assume that the verbal H here is three-way complex, although the Spec configuration is only double. 4.3.2 Rejlexi11e causati11es FX is usually agreed to be the reflexive of the causative FIV. Reflexivization comes in two forms: (a) as pure reflexives, i.e. in (52), or (b) as benefactives, i.e. in (53):

(52)

(53)

a.

salima "to be sane"-+ 1-aslama-hu "to deliver it (sane)" -+ st-aslama "to deliver himself; to surrender"

b.

1-ayqaqa-hu "to wake him up"

a.

xaraja "to go out"-+ 1-axraja-hu "to make him go out" -+ st-axraja-hu "to make it go out for one's self; to extract"

b.

~aqara "to be

c.

kataba "to write"-+ 1-aktaba-hu "to make him write" -+ st-aktaba-hu "to make him write for his benefit"

-+ st-ayqa(Ja "to wake (one's self) up

present" -+ 1-aJW.ara-hu "to make it present" -+ st-a1}4ara-hu "to make it present for one's self"

Chapter 2. Transitivity, causativity, and verbal plurality

I take the difference between the structure associated with those forms and those in (48)-(51) above to be that FX is not only morphologically more complex. but it is also structurally more complex. Unergatives transitivize e.g. through FIV, but not through FI, and hence have FX as anti-causatives. 43·3 Agentive and "'expositive" causatives Consider the following pair of FIV causatives: (54)

1-abar-tu

zayd-a.n

l-faras-a

CAUS-sold-I Zayd-Acc the-horse-Ace I made Zayd sell the horse. (55)

1-abar-tu

l-faras-a

cAus-sold-I the-horse-Ace I made the horse to sell (to be sold). The two sentences differ in interpretation. In (54), the horse has been sold, but in (55), it may or may not have been. It is only "exposed" for selling, traditional grammarians say. In (54), the agent of selling is realized, and the object is an object of an "unergative/active" internal verb. In (55), the object is an "ergative" subject of a middle/passive verb. This difterence in interpretation cannot follow from complexity captured through Num. First, there is no ergative (active) verb "sell" in Arabic, and the only way to express the middle is via Passive morphology. Second, the agent of"self' can be omitted, but not its theme. If the (internal) verbal root is dyadic in both cases, then the "expositive" (55) might be taken as instantiating a sort oflow passive Voice. 36 43·4 "Requestive" causatives In FX causatives examined above, the reflexive or benefactive is also the causer. But this is not the interpretation of so-called requestive causatives (talabiyyah), instantiated in the following pair: (56)

st-ajham-tu-hu CAUS .REF-understood-I- him I asked him to explain to me (make me understand).

(57)

st-axbar-tu-hu CAUS.REF-informed-I-him I asked him to inform me.

;36.

Cf. Fassi Fehri (2001 b) for illustration.

53

54

Key Features and Parameters in Arabic Grammar In these constructions, the interpretation is (unexpectedly) not: "I made him understand/inform to my benefit~ but rather something like "I made him make me understand/ be informed': What is the source of this interpretation? One possibility is to take these constructions to be instances of double causatives. In this case, the subject (or causer) of the first (higher) Cause is object of the second (embedded) Cause, and the object of the first Cause is subject of the second Cause. By contrast. in (41) above, the subject of the (unique) Cause is the beneficiary; and the object of the Cause is the subject of the event How can we then reverse this ordering? 1his reversal recalls the double use ofFIV: (a) as expressing agentive causativization, and (b) as the result or effect of the caused event on the object FX here appears to be a combination of both structures. But note that the difference between the two structures of FIV is a difference in Voice, rather than a difference in Num of args. 4·3·5 Ergative Num and intensi11e forms Consider again the intensive Num on verbal and non-verbal forms, as well as the restrictions on its distribution and interpretation. The forms to consider here are verbs and participles (the latter being treated normally as adjectival). Pl Num is distributed according to the ergative parameter. It applies to objects of transitive verbs and subjects of intransitives, and to their participles in the same way: (58)

zayd-un

muqaftir-un

l-laf:tm-a

Zayd-NOM cutting.intens.-NOM the-meat-Ace Zayd is cutting (cutter of) the meat into pieces. (59)

l-laf:rm-u

mu.qa(ta.r-un

the-meat-NOM cutting-NOM The meat is cut into pieces. It cannot apply to subjects of transitives: (60) cannot mean that many people or various groups of people cut the meat: (60)

qattara

n-naas-u

l-laf:rm-a

cutintens. the-people-NOM the-meat-Ace The people cut the meat into pieces. Likewise, the intensive is formed from non -stative intransitive s which are ergative, but not from unergatives: (61)

a.

jawwala

r-rajul-u

took.walks.intens the-man-NOM The man took a lot of walks.

r-rajul-u. b. *sakka.ra drank.intens the-man-NOM Intended to mean: The man got drunk.

Chapter 2. Transitivity, causativity, and verbal plurality

c.

*raqqa~a

1'-rajul-u.

danced.intens the-man-NOM The man danced intensively. These observations suggest that verbal Nb (when intensive or collective) is in fact ergative.37 44

Further empirical motivations

44.1 Ergative and unergative Num in event plurality and transitivity In addition to natural ergativity ofNum observed in the case of intensives, transitive/intransitive alternations also exhibit such a natural ergativity (see e.g. English and Arabic "laugh"/4a~aka (in (71) below), which do not transitivize, compared to Engish "open" or Arabic hamam "pour out': which do). If transitives arise from realization of Pl Num, then the efiects of natural ergativity in the two instances (intensives and transitives) are unified. But English does not have a morpheme for intensives (or repetitive evs). The prediction then is that in English, Pl Num can only be distributive (on the assumption that transitives use distributive Num), an option provided by UG syntax in terms of Pl Merge (of Specs). In other words, I assume that syntactic or analytic Pl Num (which is used in English) is universal, whereas synthetic Pl Num (used in the Arabic intensive) is language specific. Furthermore, synthetic Num when used in L-syntax is subject to natural ergativity, but it is free inS-syntax. hence the following contrast:

(62) *I laughed John. (63) I made John laugh. Morphological causatives do not observe the Ergativity condition, as amply documented in the literature (see e.g. Japanese and Arabic), thus paralleling S-syntactic (or analytic) causatives, not L-syntactic ones. This does not necessarily imply that they are hi-clausal. Differences can be captured in terms of structure heights, typically height of Num placement.

44.2 Adicity, (in)transiti11e alternations, and multiple uses If a dyadic predicate is treated as Pl (distributive), then it is expected to have two different realizations, depending on whether Num is analytic, or synthetic. Reflexives

37· Contrary to what happens with participles, exaggeration augmentative forms can be unergative intensive adjectives, as in ?akkaal, raqqaa$, sikkiir. For their derivational source, see Fassi Fehri (200lb).

55

56

Key Features and Parameters in Arabic Grammar and reciprocals instantiate the analytic/synthetic variation cross-linguistically. Reflexives are syntactically dyadic in English, but they are dyadic or monadic in Arabic, depending on roots: (64)

The man washed himself.

(65)

a. g-t-asala r-rajul-u b. *gasala r-rajul-u nafsa-hu

(66)

a.

wabbaxa r-rajul-u nafta-hu The man blamed himself.

b. *ta-wabbaxa r-rajul-u ref-blamed the-man The man blamed himselt~ Reflexives can be seen as arising from a distributive (plural) event, with one (and the same) argument, which can be partitioned in syntax, but unified in morphology. With reciprocals, similar (though not identical) phenomena and contrasts arise: (67)

a.

haddada

n-naas-u

barcj-u-hum barcj-an

threatened the-people-NOM some-NOM The people threatened each other.

some-Ace

b. *ta-haddada n-naas-u (68)

a.

ta-"/:laabba n-naas-u rec-loved the-people-NOM The people loved each other.

barcj-u-hum ba.rt}an b. */:laabba n-naas-u loved the-people-NOM some-NOM some-Ace (69)

a.

~aarar-a

n-naas-u

bart}u-hum bart:}-an

fought the-people-NOM some-NOM The people fought each other. b.

t-~ara.ra

some-Ace

n-naas-u

rec-fought the-people-NOM The people fought each other. Here both the event and the argument are plural (distributive), but the latter can be partitioned in syntax, or assembled through morphology. The morphology often unifies various constructions which may have different structures, but share adicity building properties. For example, Moroccan Arabic has the same morphology for reflexives, reciprocals, and passives. The same is true of Romance. This state-of-affairs is expected if the morphology is associated with an assembling plural. Likewise, causatives alternate with passives in Japanese/Korean, and with inchoatives or (resultative) anti-causatives in Arabic, again a manifestation of partitioning or assembling plurality.

Chapter 2. Transitivity, causativity, and verbal plurality 57

Num theory and Num heights

5·

5.1

Sg and Pl Merge

Suppose that the source of verbal plurality is a functional head Num, which is an extended projection of verbs (or eventive Rs). Then plural Num can be morphologically realized on the head, and it is interpreted as repetitive. It can also be interpreted as the same action operated on many objects (when plurality is distributed over arg/Spec, rather than H/ev). Second, Pl Num can be morphologically realized on H, but Pl interpretation is induced through multiple Specs. Pl is then read off somehow cumulatively, through additioning Sg Specs, as in transitive FII or FIV. Third, there are cases where H is not overtly realized, although transitivity is visible through multiple Specs. This is the case of alternant doublets (cf. e.g. in Fassi Fehri (1996a)), exemplified by the following contrasting pair: (70)

a.

hama.ra

d-damr-u.

poured.out the-tears-NOM Tears poured out b.

hamara

d-damr-a

poured.out the-tears-Ace He poured out tears. Finally, there are cases of (in)transitivization, which appear to be "lexicalized~ in the sense that no alternation appears to be possible. Leaving the morphology aside, I take the four instances of plurality to be subsumable under two subcases of Merge: (a) Sg Merge (with "intensive" Nb ), and (b) Pl Merge (with transitives or causatives). Let us then see how languages make use of the two options. 5.2

Language variation

Arabic, English, and Japanese (and presumably quite many- if not all-languages) make use of the (b) option to form transitives. They differ, however, in that Arabic and Japanese can transitivize unergatives, but English cannot. This difierence can be attributed to the "richness" of morphology. Compare the following Arabic pair with the English (62) and (63) above: 38 (71) *tjal:ra.ka

r-raju.l-u

l-walad-a

laughed the-man-NOM the-child-Ace The man laughed the child.

Richie Kayne (p.c.) informed me that The doctor coughed John is not that bad, in which case there is a more complex story to be said about English. For more on causativity, see Lidz (1999), Pylkldtnen (1999), Ramchand (2003), among others.

j8.

58

Key Features and Parameters in Arabic Grammar (72)

r-raju.l-u

1-a4/:laka

l-walad-a.

CAUS.laughed the-man-NOM the-child-Ace The man made the child laugh. I take the rich morphology (implemented in S-syntax) to allow more structure than does the poor (or lexical) one, hence the difference between English and Arabic, and in fact between Arabic (71) and (72). That means that Num is Ergative (= ENum) in English intransitive "open" and the Arabic (70a). I assume that the latter is projected at the first root level. Complex roots can be formed through Unergative Num (= UNum) in transitives (as in Arabic (70b)), or unergatives (as in English or Arabic "laugh"). UNum is a kind ofPl Nb, which I assume to be typically distributive (applying to H, as in unergatives, or to Spec, as in transitives). UNum is projected higher in the structure of the L-root, as represented (informally) in (73): vP

(73)

~

John

v+UNum

~

the door

v

I

opened The "rich" morphology, on the other hand, which allows the transitive/causative of unergatives, is nothing else but functional Num. The structure of (72) can then be roughly formulated as (74): vP

(74)

~

man

vP+Pl

~

child

vUNum

~

laugh

v

6. Summary and conclusion In this Chapter, I have argued that verbal Num plays a central role in characterizing transitive or causative constructions, seen as instances of (distributive) Pl configurations, generated through Pl Merge. Thus when a Pl morpheme occurs on verbal Hs, it can be interpreted collectively (as an iterative or intensive ev), or distributively (as a distributive action operated on distributive args). This ambiguity of verbal Pl recalls that found in the nominal plural quantification, which

Chapter 2. Transitivity, causativity, and verbal plurality

suggests that Num is orthogonal to verbal and nominal categories. IfNum is generalized to verbs, and extended to transitivization, then there is no need to appeal to a conceptually distinct category (such as Aspect, Voice, or v) to account for verbal complexity and quantification. The choice of competitors turned out to be both empirically and conceptually deficient Arabic is an instance of languages in which Pl morphology in verbs and nouns is essentially uniform: it is based on vocalic or consonantic gemination. Verbal reduplication is used ambiguously for the iterative (or intensive) and the causative meanings. Fissioning or partitioning reduplication is a typical mechanism which derives causatives, or in the case of iambic reduplicating roots, "attenuative" or "paucar partitioned event forms. Reduplication as a mark of verbal plurality is amply documented in various languages. Furthermore, its exclusive or ambiguous use is a matter of cross-linguistic variation. The correlation of uses is expected under my general Num theory, but not under Aspect, v; or Voice. The latter categories play various specific roles in functional structures, but their roles (although interacting with transitivity and Num) have to be kept separate. Typically, Voice as a functional head encoding hierarchical organization of arguments, is different from Num, which accounts for ev/arg complexities. Likewise, Aspect and Tense exhibit interdependencies with Case, but they interact with ev/arg complexities only indirectly.

59

CHAPTER3

Synthetic/analytic asymmetries in voice and temporal patterns*

Among the issues that any linguistic theory has to deal with are the principles and limitations on the asymmetrical mappings between sound and meaning in terms of words and phrases, and typically how grammatical functions (or functional categories; GFs henceforth) arising in lexical or syntactic derivations are mapped into morpho-phonological segments or sequences. Lexeme-based morphology (cf. Beard (1995)) was primarily designed to account for a number of these asymmetrical mappings through the Separation Hypothesis. 1 Distributed Morphology (cf. Halle & Marantz (1993)) has refined the picture to include even more complex relationships between lexica-syntactic structure, morphology, and actual

*

This article is dedicated to the memory of Ken Hale, a great linguist, a great man, and a great friend. For useful discussions, I would like to thank Alec Marantz, Morris Halle, Jim Higginbotham, Anna Maria Di Sciullo, Sabine Iatridou, Sylvain Bromberger, and for help with the Ancient Greek data, Donca Steriade. I also thank the audiences at the Semitic Conference at USC (May 2001) and the Asymmetry Conference at UQAM (May 2001), as well as an anonymous reviewer of the Proceedings for helpful remarks. The usual disclaimers apply. LBM theory is claimed to have arisen in order to solve the problems of a wmber of asymmetric mappings of sound and meaning (at the morphological level), which cannot be solved in morpheme-based grammars (= MBG). The latter take morphemes to be signs. Among those problems are the following: 1.

i.

ii. iii. iv.

Matthews (1972)'s"cuniUlative exponence': i.e. the fact that a single form expresses multiple GFs simultaneously (e.g. Latin verbal-o expressing: l.s.pres.ind.); M's "extended exponence': with multiple forms expressing a single GF (English "be" + -lng making the progressive); Multiple functionality (English -ingwith v, n, a GFs) Multiple expression of a single GF (English -ity, -ness, -ce marlting deadjectival transposition).

To these asymmetries, null and empty morphologies should be added. The Separation Hypothesis amounts to separate the terminal elements in syntax from their phonological realization. For detail, see Beard (ibid), as well as Beard (2001). See also Anderson (1992) for general discussion.

6:1

Key Features and Parameters in Arabic Grammar

vocabulary. 2 But many questions remain to be solved. For example, even though the traditional classification of languages into isolating, agglutinative, or fusional is presumably not directed to focalize on the existence of purely discrete types, the relevance of these classifying properties has not received notable attention. 3 Thus the one-to-one mapping found in isolating (or agglutinative) Ls, as opposed to the many-to-one mapping found in fusional (or cumulative) Ls, is commonly treated as if it were of no empirical (or theoretical) significance. If a (relatively) isolating-agglutinating-fusional continuum is true of most (if not all) languages, then the question that arises is: which natural principles and constraints operate in the organization of the continuum? Best studied traditional mapping relationships have been usually confined to the domain of morphological words (be they roots, stems, simple or complex words), with a one-to-many mappings (when a single function is associated with multiple discrete expressions, or "synonyms"), or with a many-to-one mapping (when multiple discrete functions share only a single expression (which is "polysemous" or "homonymous"). But clearly, the mapping contrasts are not to be found only within word structure, but do distribute simultaneously and naturally on phrase structure, with Ls varying as to whether they use multiple (discrete) words to express GF complexity, or they use only a single word for such expression. I will call the first type of expression analytic, and the second, synthetic. 4 For

2.

DM combines insights of both LBM and MBG. In DM,

syntactic features and phonological features form separate sets, affixes are lexical items, which combine with lexemes or other affixes to form complex words, iii. vocabulary items compete for insertion into syntactic nodes, and iv. morphological structure is a separate component of the grammar.

i. ii.

Later development of DM has modified some of these assumptions, namely (iv). For detail, see in particular recent work by Marantz (2000), among others. 3· Comrie (1985a), for example, is rather pessimistic about the utility of these typologies, typically for the purpose of correlating with other non-morphological properties of L, when he writes: "Although morphological typology does serve the useful purpose of presenting an overview of the morphological structure type of a language, it remains unclear whether it can be considered a significant typological parameter [... ] in the sense of correlating with other parameters outside morphology [my underlining; FF]." (p. 52) 4· These morphological types have been proposed originally by Friedrich von Shlegel (1808), August Wilhem von Shlegel (1818 ), as well as Wilhem von Humboldt. See Bussman (1996) for definitions as well as references.

Chapter 3. Synthetic/analytic asymmetries in voice and temporal patterns

example, the various components of a Mood-Tense-Aspect-Voice-Verb chain can be grouped around a single thematic verb (properly inflected), or they may distribute also on one (or more than one) extra auxiliary verb. It is the characterization of this kind of variation that this Chapter will be devoted to. Most (if not all) languages resort to various mixtures of the synthetic and analytic strategies, but the choices are not free. Evidence will be provided, in fact, for the following claims: a.

options of analysis or synthesis are disjunctive, in the sense that no language makes use of both strategies to express the same GFs (and/or meanings). The following disjunctive "global" asymmetric constraint appears to limit the set of possible morphological or formal objects available in any language; call it

Formal Disjunctivity: 5 (1)

Fonnal Disjunctivity For any formal relation R, established between x andy in a language L, where x andy are morphemes or words expressing GFs, R in Lis either synthetic or analytic.

b.

the choice available in (1) is not L-free, in the sense that it is correlated with other morphologically and/or syntactically detectable properties of L. the level of analysis (or word split) in L complies with the hierarchical organization of GFs (assumed to be aUG property; cf. e.g. Cinque (1999)), and it operates from bottom to top, so that when a split occurs at the level of GF1, all GFs that GF 1 dominates occur in the word where the latter is found.

5· I use "formal" as a cover term for both morphological and syntactic relations, which deal with form, as opposed to mealling or semantics. FD parallels similar asymmetric constraints proposed namely by Di Sciullo (for recent instances, see Di Sciullo (1999) ). Apparent counterexamples to (1) appear to be numerous at first sight, but at a closer examination, the disjunctivit:y will hopefully be confirmed. For example, during discussion at UQAM, Jean-Yves Pollock has raised the case of the French future pair Je vais dormir and]e dormirai as equivalent, and Denis Bouchard has not objected to the judgment. Edwin Williams has put forth the case of periphrastic and synthetic comparative adjectives like much more clever and cleverer, which appear to be problematic at first sight. Presumably, there are subtle differences between these pairs that I will not go into here. Another very discussed case in the literature is the difference between "lexical': "morphological': and "periphrastic" causative doublets or triplets. My guess is that the problem of these apparent counterexamples can be solved in a major part through height of attachment of the GF in the structure, from which differences in meanings result. Furthermore, L may be at some intermediate stage of change where doublets appear to be equivalent, before the change is stabilized.

63

64

Key Features and Parameters in Arabic Grammar

d. Formal Compb:"ity (= FC), based on Temp(oral) and Agr (eement) complexity essentially, plays various roles in creating verbal analysis in the same L, or across Ls. Analysis occurs only in Ls that exhibit an AgrS/ AgrO split. e. Voice belongs to the class of Temp GFs, which can be dominated by a Temp (verbal) projection, or by a (nominal) Agr projection. f. Nominal Agr may be used in Ls to form complex tenses, and the nominal or temporal content of verbal auxiliaries provides variation in the analytic/ synthetic continuum.6 g. Choices made by L become peculiarities of L, in the sense that they create limitations or exclusivities on what can be expressed in L through grammatical combinations. In compliance with (1 ), the article investigates the limitations and constraints on analysis or synthesis through examining the patterns of variation in the expression of Voice and Temporality found in a number of Ls, compared typically with Arabic. It is shown that analysis finds its source in the specific degrees of FC, arising typically in the Temp and Agr systems, and leading to appropriate Splits (in Temp or Agr chains). Voice may also be playing a role in increasing FC (and hence triggering analysis), depending on whether it belongs to finite or non-finite chains. In Section 1, analytic/synthetic patterns of expression of Perfect Tense and Passive voice in a number ofLs are examined and contrasted with their Arabic counterparts. It is then argued that the contrasts in the Perfect expression can be attributed to an Agr Split (into Agrl and Agr2), which is or is not available for L. The latter variation depends on the nomina.l nature of the auxiliary (in the analytic type), or its verba-temporal nature (in the synthetic type). On the other hand, analytic complex tenses are shown to originate from a Temp Split, due to Temp complexity (interacting with Agr complexity). The same is shown to be true of Voice expression, once it is admitted that Pass(ive) belongs to the Temp chain, and it can be either (immediately) dominated by a nominal Agr (i.e. a non-finite participle Agr2), or by a verba-temporal projection (with no availability of Agr2, and no split in the Temp chain). In Section 2, a more precise content is given to the sources of FC, which triggers analysis. Some peculiarities of Ls are then addressed, including those of analytic Pass Perf in Classical Latin, which provide support for the view that Voice increases Temp complexity. Likewise, the analytic character of complex tenses in Arabic (as opposed to synthesis in Passive and Present Perfect) is contrasted to that of the synthetic Ancient Greek, by taking into account the peculiarities of Arabic

6.

On the role of nominality in forming complex tenses, see Fassi Fehri (1996a).

Chapter 3. Synthetic/analytic asymmetries in voice and temporal patterns

verbal inflection. The latter constantly bears Pers finite morphology, and it does not instantiate Agr2. The section also addresses the proposal of deflexion as a direct source of analysis (through reanalysis), and rejects it as a potential explanation. The Splitting analytic theory is defended instead. In Section 3, further peculiarities of Ls are investigated, including e.g. the existence of imperfective Pass in Arabic, and its absence in French. The latter are related to other peculiarities in the Temp systems of the two Ls. The dear-cut division in Arabic between verbal and adjectival voice expressions is also examined and compared to the ambiguous categorization of English Pass participles. It is proposed that various attachment Heights of the same form can be used to express multiple GFs, without resorting to multiple morphologies.

1.

1.1

Analysis, voice, and temporality The problem

Consider the following Voice contrasts between Arabic, English, and French, respectively: (2)

qutila r-rajul-u killed.PASS.PAST the-man-NOM a. The man was killed. b. The man has been killed.

(3)

a. b.

(4)

I.:homme a etetue.

The man was killed. The man has been killed.

In (2), there is only a single verb, which expresses Pass, ambiguously, either in the Past or the Present Perfect tenses (P in the gloss marks Past or Perfect). The internal vowels of the verbal stem u-i mark the passive, and alternate apophonicallywith the active (marked as a-a. in (5) below, for example). The a. vowel suffix on the verb marks Past or Perfect. Thus Voice and Tense (and/or Aspect) are fused in a single synthetic verbal form (although in distinct positions). This form of expression dearly differs from that found in English or French. In English (3), two distinct analytic passive constructions are needed for each Tense. In French (4), the passive expression is also analytic, and it makes use of two auxiliaries, although this form of passe compose translates ambiguously either the English Past (3a) or the Perfect (3b ). There are no other means to express these tense and voice combinations in the three languages, hence providing an instantiation of the disjunction stated in (1).

65

66

Key Features and Parameters in Arabic Grammar

Note that a similar contrast is found with active present Perfect:7 (5)

kataba

r-risaalat-a

l-zaan-a (*gad-an)

(has) written the-letter-Ace now He has written the letter now.

(*tomorrow)

(6)

John has written the letter now ("yesterday).

(7)

Jean a mange la pomme en ce moment (*hier).

Observe again that the active Present Perfect in Arabic is synthetic, while its expression requires an auxiliary in both English and French. Why is then such a difference? Why can't Arabic make use of a verbal auxiliary in the Present Perfect, and why is the Passive expression confined within the finite verb (like active Voice is), in contrast to English and French, which must resort to periphrastic verbality, in which the auxiliary carries Tense and Agr (finite) features, and the thematic verb Voice features?

1.2

Nominal auxiliaries

The absence of the auxiliary with the Arabic present Perfect recalls the absence of the present verbal copula in so-called nominal sentences. Note that in English and French, its presence is obligatory: (8)

r-rajul-u. (""y-akuun-u) fii d-daa.r-i The man (is)

in the house.

7. I leave for the moment the contrasts in the past tenses. which are more complex, and do not distribute in the same way. 1hus French contrasts with both Arabic and English in requiring an auxiliary for the expression of its simple past: (i) (ii)

Jean a mange la pomme hier.

kataba r-risaalat-a

1amsi

wrote the-letter-Ace yesterday He wrote the letter yesterday. (iii)

John wrote the letter yesterday.

French has also a synthetic imperfect past (the "imparfait"), which is comparable to the (analytic) past progressive in English: (iv)

Jean mangeait la pomme hier. John was eating the apple

yesterday.

I return to some of these contrasts in Section 2.

Chapter 3. Synthetic/analytic asymmetries in voice and temporal patterns

(9)

The man *(is) in the house

(10)

rhomme *(est) ala maison.

These sentences have exactly the same (present tense) interpretation, although the Arabic one must be verbless, whereas its English or French counterparts must contain a copular verb. Since it is normally agreed that English and French have no overt temporal morphology in the present, it is reasonable to think that the overt realization of the copula in this case is simply motivated by the necessity to make visible the subject AgrS features. 8 The finite auxiliary here is then basically of nominal (subject agreement) origin.9 In Arabic, nominal (subject agreement) features are not sufficient to make the verbal auxiliary overt, hence the existence of only verbless present finite sentences (the so-called nominal sentences in the grammatical tradition). 10 But the overt/covert realization of the nominal copula in French/English (or more generally Romance/Germanic) does not in itself directly explain why an auxiliary is needed to mark finite phi-features of the subject in English or French Perfects or Voices, if we take copular sentences to be structurally distinct from auxiliary ones. I will show that in fact the two instances are not distinct, in spite of auxiliary selection differences. 1.3

S/0 Agr split and auxiliary selection

Consider again the sentences in (6) and (7). Following early work by Richie Kayne, I assume that the participle there is carrying Part(iciple) (or O(bject)) Agr, and the auxiliary S(ubject) Agr. 11 These two (distinct) Agrs can be made visible when the 0 raises higher than V, as in (11), where the clitic les (for pommes), triggers

8. 0 n the absence of present Tense morphology (in English), see in particular Kayne (2000, Chapter 10, originally published in 1989), and En~ (1991, 1987). For a different view, see Fassi Fehri (1991/1993) and Gueron (1995), among others. 9· I assume that copula and auxiliary"be" do not differ significantly as regards the conditions on their "visibility': The copular sentence presumably differs from the non-copular one in that the latter contains "be" plus a temporal thematic participle. The auxiliary qualifies as nominal because it carries only features of the (finite) SubjectAgr (=Agr1), T1 being null. I assume that participles are dominated by a nominal and/or adjectival (non-finite) Agr (= Agr2), although they may be also headed by a null abstract Aspect or T head (= T2, needed for appropriate interpretation). Participles may then be taken as "weakly" temporal, because they lack overt temporal marks in all contexts, apart from Voice. See Section 3 below.

to. For the conditions on the realization of null, pronominal, or verbal copulas in Arabic, see Fassi Fehri (1993) in particular.

u. Ct: Kayne 2000, Chapter 2 (originally published in 1985).

67

68

Key Features and Parameters in Arabic Grammar

Number and Gender Agr on the participle, and the auxiliary agrees with the high subject: (11)

Je les

ai

mangees.

I them have.lsG eaten-PL.F I have eaten them (f). Note that the S/0 Agr distinction or split correlates with another distinction or split in the auxiliary system, namely, the avoir/have vs ~tre/be split. One feature of this split is that in the be configuration, the subject agrees with both the auxiliary and the participle, but in the have configuration, it agrees only with the auxiliary, as in the following minimal pair: (12)

Les tasses sont casstes. The cups are broken (F.PL)

(13)

Les tasses ont

casse

(*cassees).

The cups have broken (*F.PL). In both cases, the cups are understood as "logical" Os, but what matters is that the agreement chain breaks with a.voir, but not with etreP The S/0 agreement split (coupled with the auxiliary split) is subject to parametric variation. In Ls with a unique copula/auxiliary "be" (the majority, it seems), the agreement chain does not break, even in complex tenses. Consider the following Arabic examples, which illustrate the uniformity of the Agr chain: (14)

kaana-t l-banaat-u

(qad)

katab-na

r-risaalat-a

was-F the-girls-NOM (already) wrote-F.PL the-letter-Ace The girls had (already) written the letter. (15)

kaana-t r-risaalat-u

(qad)

kutiba-t

was-F the-letter-NOM (already) wrote.PASS-F The letter had been (already) written. In this pair, the auxiliary is invariably "be" with the active and passive past perfects alike. At the same time, both the auxiliary and the thematic verb agree with the same structural subject. Fassi Fehri (1993) argues that this configuration obtains through anaphoric agreement (and/or anaphoric T for other purposes). At any rate, in both copular and auxiliary sentences, "be" is used, and there is no S/0 Agr split. There is only an Agr chain (which includes two AgrSs for the thematic

12.

On the analysis of "have" as be+P/D, see Kayne (2000, Chapter 7, originally published

in 1993), relying on previous work by Szabolcsi, Freeze, and Benveniste (see the references

cited there).

Chapter 3. Synthetic/analytic asymmetries in voice and temporal patterns 69 and auxiliary verbs, or for the auxiliary and the participle or adjective). The Agr chain is uniform, and controlled by anaphoric subjects. One can then conclude that there is no AgrO in Arabic, and the absence of this split is correlated with the absence of auxiliary splitP If this reasoning is correct, then we have another source for analysis, in addition to the nominality of the copular/auxiliary. The S/0 Agr split causes a verbal split (in terms of the auxiliary and the thematic verb), in order to support separately the two Agrs (in tact Agr1 and Agr2). Interestingly enough, the Latin data shows that it is the combination of the two properties (nominal Agr and Agr1/ Agr2 complexity) which triggers analysis. Before presenting this case, I will first examine how Temp complexity can yield analytic expressions, through what I call Temp auxiliaries. 1.4

Temp auxiliaries

We have seen that the overt realization of the auxiliary in sentences like (8), denoting a state which holds at the present, is not allowed by the grammar. But a version of (8), with the present form of the auxiliary, is possible, provided a different temporal interpretation from the actual present is induced. For example, (16) can express a future state (the equivalent of"The man will be in the house"), or a habitual state (e.g. in the context "The man is usually in the house after 4 oelock"): 14 (16)

r-rajul-u. y-akuun-u the-man-NOM is a. b.

fii d-daa.r-i

(gadan; bard-a. 1'-ra.birat-i)

in the-house-GEN (tomorrow; after four)

The man will be in the house tomorrow. The man is (usually) in the house after four.

Thus temporal/aspectual specifications like these appear to be marked (compared to the unmarked simple present), and thus force the auxiliary to become overt, to support these temporal features (cf Fassi Fehri (1993)). That specified temporality is behind the use of the verbal auxiliary is further supported by its obligatory

In fact, Participle Agr may coincide with AgrO, as in (11 ), but not necessarily. Kayne was originally dealing with Participle Agr, but Chomsky ( 1995) generalizes AgrO to include it. See Siloni ( 1997) for evidence that AgrO is absent in Hebrew participles. As for the uniformity of the Agr chain with "be~ but not "have': a similar conclusion has been reached by Iatridou, Anagnostopoulou, & Izvorski (200 1) for e.g. Bulgarian, based on a personal observation made to them by Peter Svenonius. For these reasons, preference will be given to the use of (finite) Agr1 vs. non-finite (participle) Agr2, instead of AgrS and AgrO, although they are sometimes used interchangeably.

13.

Interestingly enough, Marantz (p.c.) observes that these are the essential readings that obtain with thematic verbs in the present in English (but not with the copula).

14.

70 Key Features and Parameters in Arabic Grammar

occurrence with specified complex tenses, be they perfects or imperfects, active or passive. Thus (17) and (18) illustrate the case of Past Perfect and Future Perfect, respectively: (17)

ka.ana qad

kataba 1'-risaa.la.t-a

lammaa daxal-tu

l-qaarat-a

was already wrote the-letter-ac when entered-I the room-Ace He had already written the letter when I entered the room. (18)

sa-yakuunu qad

kataba r-risaalat-a

gadan

FUT-is already wrote the-letter-ac tomorrow He will have already written the letter tomorrow. Note that the occurrence of the modal future prefix sa- in (17) is not necessary to force the appearance of the auxiliary; only the future meaning is, as noted earlier. Analytic complex tense constructions are not limited to past or future Perfects. They extend to Imperfects. The construction (19) illustrates the case of Past Imperfect, whereas (20) instantiates either a Habitual Imperfect or a Future Imperfect: 15 (19)

kaana y-aktubu r-risaalat-a was 3-write the-letter-Ace He was writing the letter.

(20)

yakuunu y-aktubu r-rasaaZil-a is a. b.

3-write

fli l-rasiyy-i

(gadan)

the-letters-Ace in the-evening (tomorrow)

He is usually writing the letters in the evening. He will be writing the letters tomorrow.

These contrasts indicate that (isolating) verbal auxiliaries are clearly playing a role in the realization of complex temporality (provided it is marked). Complex temporality then leads to analysis, and Temp GFs are split on two finite verbs, both of which carrying T marks (i.e. typically prefixing or suffixing Person morphology, for Imperfect and Perfect, respectively), in addition to finite Agrl. Temp auxiliaries are widely spread among Ls, to express T splits (caused by Temp complexity). They are not only found in Germanic/ Romance type Ls, which exhibit the Agrl/ Agr2 split, but also in Ls like Arabic (or Semitic), which do not instantiate the latter split. It is worth noting that the T/ Agr splits can also be contrasted in the domain of Voice, to account for the different sources of the synthetic/analytic options. 16

15.

See in particular Fassi Fehri (1993) for detail.

t6. These splits obtain via fissioning (or delinking) mechanisms in syntax, along the lines proposed in Fassi Fehri (1996b, 2000).

Chapter 3. Synthetic/analytic asymmetries in voice and temporal patterns

1.5

Voice

In this section, I will argue that Voice is also sensitive to the occurrence (or nonoccurrence) of Temp (verbal) split or Agr (nominal) split in L. Despite the influential paper of Baker, Johnson and Roberts ( 1989; BJR), where the Pass morpheme was treated as an argument (Case absorber) affix generated under I, early parallel work on the matter (see especially Fassi Fehri (1988b) for Arabic, Tsimpli (1989) for Modern Greek, and Rivero (1990) for Albanian and Modern Greek) has argued that Pass is a head of aspectual nature, which interacts and clusters with other GFs in the verba-temporal chain (see also Ouhalla (1991) for motivation of the headedness and verbality of Pass, as well as penetrating criticisms of BJR's analysis). Later work on the matter has acknowledged the existence of Voice as a verbal head projection (see e.g. Kratzer (1996), to be ordered among the components of the aspectuo-temporal-modal chain (see e.g. Cinque (1999)). More recently, Embick (2000 a & b) also assumes the aspectual character of the Pass morpheme, and he resorts to Height of Attachment to derive the various (aspectual) interpretations of the passive participle (eventive, stative, etc.). I will not repeat here the arguments for the nature of Voice as temporal, but rely on the many references pointing to this direction. If Voice is verba-temporal, then it is natural that it selects (or dominates) only VPs (see e.g. Bresnan (1982), Rivero (1990), Ouhalla (1991), Levin and Rapaport (1986)), be modified by adverbs (of Manner; cf. Cinque (1999)), interacts with other aspects (Beedham (1982), Fassi Fehri (1988b)), and for our purposes, yields synthesis or analysis, depending on other components of the M/T/Asp/Vo/V chain (M for Mood, Vo for Voice). The two triggering factors exploited earlier in active Voice will be equally exploited here, namely Temp and Agr splits. 1.5.1 Arabic and anaphoric Agr Consider first Arabic, a simple case. Here Voice in complex tense constructions does not behave differently from that of simple tense cases. For example, in the Passive Past Perfect and the Passive Future Perfect. illustrated in (20) and (21), respectively, Pass occurs on the embedded thematic verb, which is both temporal and finite (compare with (2) above):

(21)

kaana-t r-risaalat-u (qad) kutiba-t was-F the-letter-NOM (already) wrote.PASS-F The letter had been (already) written.

(22)

t-akuunu r-risaalat-u (qad) kutiba-t F-is the-letter-NOM (already) wrote.PASS-F The letter will be (already) written.

71

72

Key Features and Parameters in Arabic Grammar

Thus Pass is still synthetically temporal, and the complexity of tenses has no impact on its nature, its position, or the rise of analysis. Note also that the two finite Agrs are identical or anaphoric. 1.5.2

Latin and split Agr

Classical Latin shares synthetic properties with Arabic in the active verbal system, but differs from it in that analytic auxiliaries arise in the former from complexity created by Passive voice, interacting with tense. Thus, like Arabic, Latin uses no auxiliary in the Present Perfect. The latter is synthetic, as can be seen in (23): (23)

lau.da.-vi-t praise-PERF- he

He has praised. This synthesis contrasts with the Romance analysis in this case (see e.g. French (7) above, or Italian ha Iodato). Likewise, the past perfect in Latin is also synthetic, although it incorporates a past tense form of the auxiliary esse (contrary to the present): 17 (24)

lau.da.-v-er-am praise-PERF.be-PAST- I

I had praised. Giorgi and Pianesi (1991; = GP), on whose data I am relying, argue that Latin differs from Italian in that it has verba.[ active Perfect (or T2, without Agr2), contrary to the latter in which the Perfect (T2 and Agr2) is adjectival. Consequently, T2 in Latin, which does not need a specific Agr2, can be compatible with Agrl. Furthermore, no auxiliary is needed to realize finite features. In the Past Perfect, the auxiliary must appear for temporal reasons, but T2 and T1 are associated with Agrl (in the absence of Agr2), hence the synthesis. However, the Latin Passive Perfect behaves difterently from the active Perfect in that it is periphrastic: (25)

la.uda-tus

sum

praise-PART.M.SG.NOM be.PRES.lSG

I have been praised. According to GP (ibid), the contrast with the active is in fact expected, given that the needed past or passive participle is adjectival, hence carrying its own Agr (or

Morris Halle (p.c) has informed me that the analysis of-er-as auxiliary "be" is disputable. For discussion, see GP (1997).

17.

Chapter 3. Synthetic/analytic asymmetries in voice and temporal patterns

Agr2). Consequently, the overtness of the auxiliary esse becomes necessary, to support (finite) Agrl and Tl. Our interpretation of the Latin data differs from that ofGP's. The fact that the auxiliary used in Latin past perfect is "be': and that it "shares" with the thematic verb the same Agr (or Agrl) is an indication that Latin uses no Agrl/Agr2 split to construct its active Perfect. The auxiliary ("be,) can be omitted with the present, but not with complex marked perfects. But although the auxiliary is realized in e.g. Past Perfect, Temp complexity is not sufficient to yield analysis. The emergence of an Agr split (into a non-finite participle Agr2 and a finite Agrl) typically in the Pass Perfect case calls then for an explanation. If the Pass Perfect participle contributes Voice and Aspect, and assuming that present Tis empty (as in verbless or present Perfect sentences), then the only information which the auxiliary contributes is AgrS (or Agrl), and typically the Pers feature. But the present auxiliary, as we have seen earlier, does not (usually) occur only to support Agrl, as seen with the active Perfect. It must then be that something special is going on with Pass Perfect. Suppose that the latter form is temporally complex, with a sufficient complexity so as to disallow the occurrence ofT on it. Then the Pass Perfect complex (with no finite T) will surface with non-finite Agr or Agr2 (i.e. with Num and Gr features, and without Pers, which is licensed only with finite T). Since a noncopular sentence requires a T/Pers association to be finite, an independent auxiliary is needed in this case. I think then that the analysis is dictated by the degree of (Temp) complexity allowed on the thematic verb, and that the explanation has nothing to do with the participle being verbal or adjectival, as in GP. The Pass participle is normally treated as verbal, typically when it is eventive and (implicitly) agentive. The synthesis/analysis has to do with whether Agr dominates a finite or non-finite Temp, a choice which is or isn't available, depending on the degree of complexity allowed by L. Thus Latin is forced to use analysis (and Agr split) in the Pass Perfect, but not with Active Perfect, assuming that the former is (temporally) much more complex than the latter. Support for this view comes from the fact that the Pass Imperfect in Latin is also synthetic, presumably because the Imperfect is not marked, and it does not reach the degree of complexity of the Pass Perfect. But unlike what happens with Latin, Arabic Pass Perfect is synthetic. This difference can be attributed to a difference in the nature of the Pass configuration: the latter is (completely) temporal in Arabic (in the sense that it is dominated by other temporal elements, including T) and, as such, allows more complexity than Latin does. Consequently, Latin resorts to a non-finite Agr (Agr2), to dominate the lower temporal complex allowed, and the remaining Temp GFs, associated with finite Agrl, have to find support on the auxiliary. In contrast, the Arabic Agr being the highest in the tree, after all temporal heads are placed, the Arabic synthetic passive (2) ends up having basically the same form (and internal structure) as the

73

74

Key Features and Parameters in Arabic Grammar

synthetic Latin perfect (23). Note that such an analysis corroborates the view that Voice is part of the Temp chain, and contributes to its complexity. It does not rely on a categorization view of Voice, to derive the differences, a view which is clearly unsupported by the distinct behavior of Latin Imperfect and Perfect Pass. The following example illustrates the synthetic nature of the Latin Pass Imperfect: 18• 19 (26)

laud-aa-ba-tur praise-TH.PAST-PASS- 3SG

He/she was being praised.

Modern Greek

1.5.3

Modern Greek(= MG) patterns like Arabic in being synthetic with simple Tense/ Voice combinations. Thus the Aorist Definite Past Passive in (27) is synthetic, with an internal th affix indicating Pass, and an external one indicating T, Asp, and Agr:2° (27)

Ta paidia

plUthikan

ap6 ton Gianni

the children were-washed by

the John

MG differs from Arabic, however, in that its active Present Perfect (in addition to other perfects) is analytic, as in (28):21 (28)

e:w

yrtipsi

I have written Likewise, Passive is analytic with compound tenses: (29)

Ta paidia eXozm pltUhei ap6 ton Gianni the children have been-washed by the John

As proposed by Rivero (ibid), thin (29) occurs on the main V with the same role, and the auxiliary eXoun fails to reflect Voice. Assuming a hierarchical structure

This example is taken from Embick (2000a), but the glosses are mine. As the author explains. Pass arises below T/Agr, but must readjust to be placed in T/Agr. If Imp Asp is realized as zero, Past realized by -baa, and the ending realizes a mixture of Pass and Agr, then the glosses give an approximate segmentation.

18.

Note that the fact that Agr2/T2 is adjectival in GP and that it dominates Voice does not question its internal aspectuo-temporal nature, just as the dominance of finite temporal Voice by Agr in the complex Arabic tenses does not question its temporality. I return below to the role played by Height in determining language variation in Voice.

19.

The same is true of the present passive, which I omit here. The examples are taken from Rivero (1990), except when signaled otherwise.

20.

11.

This example is taken from Joseph & Philippaki-Warburton (1987).

Chapter 3. Synthetic/analytic asymmetries in voice and temporal patterns

T>Asp>Voi>V, she argues that Voice is always an affix head immediately above VP. In simple tenses, Pass attaches to V, and the complex to Asp, then to T. But in perfect tenses, Asp is the stem eX- "have~ rather than an affix. The aspectual Aux raises (separately) toT (and Agr), and the V to Pass (only). However, Rivero does not provide an explanation for the split In my view, these facts can be interpreted as follows. Pass is (totally) verba-temporal in simple tenses because no complexity arises there, in terms ofT or Agr, combined with Voice. The existence of an analytic Present Perfect is a diagnostic indicating that MG exhibits the Agr split (and/or complexity) even with actives. As for complex tenses. MG is not different from Arabic in being analytic. Furthermore, the Pass affix in MG behaves like the Arabic one in that it does not play any role in analysis, unlike what happens e.g. in Latin. 1.5.4

Albanian

As described by Rivero (ibid), Albanian represents an original case of variation in Pass morpho-syntax, and typically how it interacts with structural complexity. The Pass morpheme appears to be mobile, and it occurs as an affix, a clitic, or part of Aux, depending on contexts. The first two cases are synthetic, in that Pass is part of the thematic V compound, while the last case is analytic, appearing on the auxiliary. Consider first the simple past passive (or non-active; Nact) in (30): (30)

La.-he-sh-a wash-Nact- Past-1 SG I was washed.

The internal structure of this construction is (essentially) similar to that of MG (27), except that no Aspect projection can be justified in this case. Pass is adjacent to (or dominates) V, and is dominated by T. This is an instance of affixal Pass. Second, when internal Mood such as the Optative (for wishes) is used, it occupies the position of Voice, and Pass is displaced to a ditic position, as in (31): (31)

U la-fsh-a Nact wash-OptPres-1s. May I be washed.

Third, Perfect tenses indicate Voice in the auxiliary, regardless of Mood. Thus the auxiliary is kam "have" in the active Future Perfect (32), and jam "have been" in the passive (33): (32)

Dote kam

liire

have.Pres.1sG washed I will have washed (something).

FUT

75

76

Key Features and Parameters in Arabic Grammar

(33)

Dote jam liire have+been.PRES.lSG washed I will have been washed.

FUT

In order to account for the mobility of Pass, Rivero proposes that the Albanian thematic verb be restricted to carrying three affixes only. In the affix case, the three are: Vo, T, and Agr, in this order, simple Albanian verbs normally lacking aspectual distinctions (except the Aorist Definite Past, which behaves like Mood, and leads to ditic Pass; see her example (13)). In the ditic case, she assumes that Mood, Aspect, and Voice are generated under the same node dominating VP. When Mood or Aspect are realized, Voice is not part of the three affixes, and it is raised to a ditic position. As for the fusion with Aux. Rivero explains that that operation makes it adjacent to V, the essential requirement. 22 It is clear then that the interpretation of the facts about synthesis and analysis in Albanian can be understood first in terms of complexity. The facts are essentially like those ofMG: synthesis with simple tenses, and analysis with compound perfect tenses. The existence of a Present Perfect points to the existence of an Agr split, which is reinforced by temporal complexity. The mobility of Pass, typically when it fuses with Aux, if correct, opens up an original option for the expression of Voice, also based on complexity. But Voice is not behind analysis. Yet the fusion of Voi, Asp. and T provides further support for the hypothesis that Voice belongs to the aspectuo-temporal chain, rather than the nominal (Agr) chain. 1.5.5

Moroccan Arabic

I will use Moroccan Arabic(= MA) as a prototypical example of spoken Arabic dialects to examine Pass voice and M-T-Asp interactions in terms of microvariation. The MA Temp chain, despite appearances, is very similar to that in SA, except that MA, like other Arabic dialects, has lost its apophonic marking of Pass, due to an extensive loss of stem vowels. Consequently, the morphemes [n-] or [t-] (or their combination [nt-, tn-], depending on regional or individual micro-variation), which were used to express inchoative, reflexive, or reciprocal meanings (through a kind of detransitivization), were extended in MA (and other dialects) to express also Pass (or medio-passive). So the difference in mophological expression between the inchoative (34) and the passive (35) in SA turns out to be expressed by the same [t-] prefix in MA (36):

11. Rivero does not explain, however, why Pass does not stay as an affix on the thematic V, like in MG. It seems to me that the absence of this option is dictated by the necessity for Pass to interact with Asp. Similar interactions are found in Arabic, which led me to label the projection Vasp (for Voice and Asp) in Fassi Fehri (1988b).

Chapter 3. Synthetic/analytic asymmetries in voice and temporal patterns

(34)

ta.-kassar-a

l-kazs-u.

inch-broke.intens-3.s .M. the-glass-NOM The glass broke into pieces (came to break). (35)

kussir-a

l-kazs-u

broke.intens.Pass-3.s.M. the-glass-NOM The glass was broken into pieces (by someone). (36)

t-herres

l-kaas

inch-broke.3.s.M. the-glass-NOM a. b.

The glass was broken. The glass came to break.

0 ne essential difference between the two members of the SA pair is that no implicit agent is inferred in (34), but there must be one in (35). But in (36), the difference is neutralized, and in fact, the (b) interpretation appears to be the most natural one, unless an extra PP-like by-phrase is added. Also, the morphology does not lead to a different conjugation, and appears to behave like the active SA morphology. But before going further with the non-active verb, let us see how its inflectional morphology is built. MA does not differ essentially from SA in this respect. Consider the following two pairs of verbal constructions. the first one being active indicative, and the second one being non-active indicative: (37)

kteb-t

l-ebra

wrote- I the letter a. b. (38)

I wrote the letter. I have written the letter.

ta.-y-ketb-u.

l-ebra.

IND-3-write-PL the letter a. b.

They are writing the letter. They write the letter.

As in SA, the past in (37) is marked by a (continuous) suffix indicating Pers, Num, and Gr ([- t] there). The form expresses also the Present Perfect, without auxiliary use, while Past and Future Perfects resort to auxiliaries. In the present (38), tense is expressed by a discontinuous affix, with a prefixing Pers, and a suffixing Num and Gr, as in SA. The form is also ambiguous between a progressive reading and a habitual reading, among other potential meanings. The first prefix [t-] there expresses indicative (or realis) Mood, as opposed to subjunctive or jussive (or more generally irrealis), marked by its absence. The [t-] position can also be occupied by the prefix [ga.-] (a short form of gaadi), which I take to be a modal for future. This is a further similarity with SA, which puts modal affixes like the future

77

78

Key Features and Parameters in Arabic Grammar

[sa-] in this position. Note that [t-] cannot be analyzed as progressive (or durative) Asp marker, since it can appear with stative verbs:23 (39)

ta-n-rn;( IND-1-know I know.

(40)

ta-y-kuun hna IND-3-is here He is usually here.

Clearly, neither (39) nor (40) have progressive reading. In (41), without [t-], the interpretation is hypothetical (irrealis): (41)

y-kuun hna 3-is here He might be here.

The [t-] prefixed form is then expressing a marked realis Mood, conditioning the actual or habitual present, whereas the bare prefixed form expresses a non-real present.24 Consider now the passive case: (42)

t-ebreq

l-xubz inch-burned the-bread

a. b. c. (43)

The bread was burned. The bread has been burned. The bread has burned.

ta-y-t-ebreq l-xubz IND-3-PASS-burn the bread a. b. c.

The bread is burned. The bread burns. The bread is burning.

Leaving the details of ambiguities aside, the order of prefixing morphemes in (43) is the following: M-T-Voi-V, with a Pass prefix being the closest to V, and lower than T/Asp, an ordering also found in SA. I assume tentatively that in the reflexive/reciprocal/inchoative interpretations, [t-] is prefixed to the root

23. Benmamoun (2000b) mistakenly analyzes this prefix as Asp progressive, despite evidence to the contrary. For an analysis dose to mine, see Brustad (2000).

14. The bare prefixed form can also be used in embedded contexts, with an anaphoric Realis mood.

Chapter 3. Synthetic/analytic asymmetries in voice and temporal patterns

(before category labeling), as in SA, and in the media-passive case prefixed to the stem (after category labeling). This difference in attachment height mirrors the clearly overt morphological distinction found in SA, where the first class of affixes (which are consonantal) attach to the root (to form a complex root), while the vocalic passive infix contributes to category stem formation. At any rate, the passive affix is found over V, and below T and M, in theM-T(Asp)-Voi-V chain. Itis therefore part of a non-split temporal chain, and hence synthetic.

2.

Formal complexity and categorization

In the previous section, I have argued that analysis results from two types of splits: A. A nominal (Agr) split, found typically in Ls that have either (a) an analytic Present Perfect (Romance or Germanic), or (b) an analytic Pass Perfect (Classical Latin) or (c) both (Germanic or Romance). B. A tempora.l (verbal) split, found in Ls like Arabic (SA and Arabic dialects) which (a) do not have an analytic active Present Perfect nor (b) an analytic Pass Perfect. They do exhibit analysis, however, in complex tenses (in e.g. Past or Future Perfect tenses). C. Voice (Pass) has been argued to be part of the Temp chain, and it triggers analysis by interacting with Temp complexity (e.g. in Latin). The source or cause of analysis assumed here can be called Formal Complexity (= FC), that is complexity in morpho-syntactic expression of GFs, and typically formal Temp complexity, conjoined with properties of (formal) Agr. FC makes predictions about the kind of analytic types to be found in Ls, to the exclusion of others, and the nature of limitations to be expected on choices, such as e.g. the Formal Disjunctivity stated in (1) above. Other approaches provided in the literature (including those based on Cognitive complexity or pattern change) do not make the right predictions.25 Before returning to this issue in subsection 2.3, I first address further relevant questions that remain to be answered.

See e.g. Heine (1993). My purpose here is to give appropriate motivation for the type of FC I am dealing with, as well as the structural mechanisms available for analysis. Cognitive factors may be playing a role in changes, but it is not clear how a cognitive approach would predict the right formal available changes in Ls, to the exclusion of others. 15.

79

so Key Features and Parameters in Arabic Grammar 2.1 2.1.1

Further analytic. and synthetic questions

Pass and additional complexity

Why there seem to be no Ls that exhibit only the Aa property, without exhibiting the Ab property (i.e. Ls which would be the reverse of Classical Latin)? The premium answer seems to be that Pass adds an extra level of complexity to the active Present Perfect. So if the active is already analytic, its passive counterpart should also be so (but the reverse is not true). The added complexity of Pass appears to be natural if Pass is marked, compared to the active, and it adds extra temporal specification (two-degree specification here). But if Pass is temporal, as I have argued, why should it trigger analysis in Classical Latin Pass Present Perfect (26), but not in active Past Perfect (25), the two cases appearing to exhibit the same degree of complexity? Various answers are possible here, all of which would lead to add an extra factor that would either make (26) much more complex than (25), i.e. 3-level complex, or make the extra factor contributing to increase general complexity. I will adopt tentatively the former solution.26 Then the T-chain would be associated with one (finite) Agr in (25), but two Agrs in (26). In Ls that exhibit the have/be split in auxiliaries, the Pass non-Pass distinction is usually correlated with auxiliary selection. In actives, the auxiliary (have) carries the Agr of the external subject (or Agrl), whereas the perfect participle is carrying an (abstract) Agr2, which is disjoint from Agrl (and does not agree with it). With the Pass (Perfect), however, Agr2 must be anaphoric with Agrl (and the auxiliary used is be). But given the relevant complexity in Classical Latin, and also the fact that there is only a single auxiliary (be), the only way to deal with Pass!femp complexity is to introduce a distinct Agr (or Agr2) for the passive participle subject (the underlying object of the thematic verb), which cannot be directly associated with Agrl (normally linked to the finite active subject). Moreover, anaphoricity between Agrl and Agr2 is established, given that the auxiliary is be, and the subject of the participle is also the subject of the finite (auxiliary) verb. 2.1.2

Two finite A~

If Arabic present Perfect is synthetic, why can't its past Perfect be, as in Latin? I would claim that Arabic does not behave like Latin in this case presumably because of the conjunction of two factors. First, Arabic has no distinction (in its

16. If the present Petfect is analyzed on a par with other petfects as a temporal complex, as I will assume, then Pass Perfect is of 3-level complexity. I disregard here the possibility of taking the source of analysis in Pass Petfect to be, for example, the avoidance of Agr mismatches, because this does not apply to Pass Imperfect.

Chapter 3. Synthetic/analytic asymmetries in voice and temporal patterns

inflectional inventory) between finite and non-finite verbal morphology (such as participial or infinitive). All temporal verbs are finite (as exemplified above), and deverbal adjectives have a different morphology, as we will see below. In order to form a complex tense, Arabic resorts only to the tensed auxiliary "be" (kaana), and to a finite verb which carries the same Agr form, the latter being anaphoric to the upper Agr. In other words, Agrl and Agr2, which enter in the formation of complex tenses in this language are identical. Second, Agr morphology is hardly dissociable from T because it is essentially the positioning of the Agr morpheme on the verb (as prefix or suffix) which indicates past/non-past (Tl) or perfect/ imperfect (T2); for detail, see Chapter 1. 2.1.3

Ancient Greek as .fu.lly synthetic

Are there Ls which are completely synthetic, in the sense that they exhibit no Agr split, and no Temporal split? Ancient Greek(= AG), as described by Ackema ( 1999), appears to provide an instance of such a language. Consider present active forms, where the Perfect inflection and the T/Agr inflection are all affixed to the thematic verb luo "loosen": (44)

Sg 1. Ieluka 2. Ielukas 3. leluke

Pl lelukamen Ielukate lelilkaasi

In the conjugation of e.g. lelukamen, Perfect morphology consists of reduplicating the stem (of the non-passive voice), joined by a k- suffix; T and Agr then follow. The active Perfect behaves essentially like simple active tenses, except for the (internal) introduction of Perfect morphology in adjacency with the verbal stem. The paradigm for the perfect passive remains synthetic, although three (instead of two positions) are filled, with Pass (or the media-passive) generated in a position higher than Perfect, and adjacent to I (or T/Agr)?-7 (45)

Sg 1. lelumai 2. lelusai

3. lelutai

Pl lelumetha Ielusthe leluntai

17. In fact,Ackema argues that Pass inAG is higher than I; see his structures (135) and (137) for simple and perfect passives respectively (pp. 149-150). I leave the question of ordering Pass and I aside since it does not bear on the question I am addressing here.

81

82

Key Features and Parameters in Arabic Grammar

If this is correct, then the double complexity induced both by Pass and Perfect in AG does not lead to a split (of Agr or Temp).

2.2

Reanalysis as the source of analytic pass or perfect

A number of studies agree that a reanalysis took place in Germanic and Romance to yield periphrastic perfect passives, and then generalized to perfect actives (and non-perfects as wel1)28 • Ackema (ibid, among others) argues that dejlexion (or the loss of inflectional endings), as well as the number of mismatches and ambiguous mappings between the number of affix positions in morpho-syntax and the number of morphonological affixes showing up on the verb make parsing opaque, and hence yields to analysis (through reanalysis). Taking synthesis to be a parameter in terms of the number of positions that a verb takes (with a maximum of three, Spec, Camp, and Adjunct), he then postulates a possible decrease in synthesis, with the Spec being the last position to undergo analysis. Deflexion is then taken to be the cause of a reanalysis of an already existing construction in the language. The target in the periphrastic passive case is a copular structure, consisting of the copula be and the deverbal adjectival passive. The (reanalyzed) adjective as participle is a true adjective by standard tests. Thus in victus est, the original meaning is that the subject is in a conquered state. When the construction was reanalyzed as perfect passive, the new meaning became "it has been conquered by someone': The new passive does not have a distinction perfect inflection (like synthetic perfects have), the new auxiliary be being construed as inherently perfect, with the completion of the action (expressed by the adjective's verbal base) implied. The original structure is not lost, and a structural ambiguity is introduced in the syntax. If deflexion persists, then only one position is left on the verb. Consequently, after the passive becomes analytic, synthetic forms of the non-perfect passive and active perfects start disappearing, and the same deverbal adjective construction is used, with a usual combination with have for the present perfect. Moreover, the reanalysis introducing the periphrastic verbal structures also introduces the present day ambiguity between the latter and their adjective ancestors. One claimed advantage of the analysis/reanalysis proposed is that in Germanic and Romance, Pass and Perfect participles are predicted to be identical in form (as well as their deverbal adjective counterparts), whereas synthetic Pass and Perfect have different forms. The author rightly criticizes alternative approaches like that of Ouhalla (ibid) or GP (ibid), based on categorial parametrization of Perfect or Pass. For example, GP analyze T2 as verbal in the active, and adjectival

18.

See the references cited in Ackem.a (ibid).

Chapter 3. Synthetic/analytic asymmetries in voice and temporal patterns

in the passive, contrary to the evidence that Pass can be verbal as well. The same objection applies to Ouhalla's nominal feature of Pass in analytic cases. Furthermore, these analyses (which Ackema terms "head movement analyses"), do not predict any relation between richnesss of inflection and the synthetic/periphrastic distinction. In sum, the analysis proposed establishes a direct link between deflexion and replacement of synthesis by periphrasis. In the criticized structural analyses, the distinction is characterized via the category of GFs, but it is unclear why there should be a relation between deflexion and the categorial feature change in Pass, Asp. or T. Embick (2000a) also adopts a head movement approach to the synthetic/ analytic distinction in Latin Perfect. Given a unified structure for both cases and a list of GFs, including v; Asp (under which he places Perf or Imperf), Pass (which has no determinate position), and T, he then simply stipulates the condition (46) to prevent synthetic Pass Perf (=his (29)): (46)

[Perf] Asp does not move to Twhen [Pass] is present.

But this stipulated statement can hardly be used to predict the analytic types available across Ls, or account for how the variation arises. For example, (46) does not apply to Classical Greek, and we do not know why. On the other hand, one could change Perf with Imperf and obtain an analytic type which does not exist, or change passive with non-passive, to make an equally wrong prediction. 2.3

A splitting analysis (of Temp and Agr categories)

Although valuable on certain points and criticisms, Ackema's analysis suffers from a number of drawbacks. Assuming that the analysis/reanalysis proposed accounts correctly for the patterns it deals with in essentially Romance, Germanic, Latin, and Greek, (a) it is also based on a ca.tegory change explanation, (b) it is limited to a single type ofLswhere the same path is used to move from synthesis to analysis, without covering a wide range of distinct crosslinguistic paths, (c) it does not take care of peculiarities in Ls that make a simple formulation of the Synthesis Parameter workable in terms of the hierarchy and number of the (argument or adjunct) affixes visible on the verb, and (d) it does not address the question of how predicate affixes, inchoatives, causatives, reflexives, or modal affixes (to cite few caces) fit into the picture. For example, deflexion cannot explain why English developed an analytic present imperfective, whereas French does not, hence the contrast in (47): (47)

a. b.

Jeanmange. John is eating.

83

84 Key Features and Parameters in Arabic Grammar

The same is true of English Past progressive, as opposed to the synthetic French "imparfait': The lack of analytic imperfective in French is also found in the passive patterns, as we will see. Likewise, deflexion can hardly be behind the replacement of punctual simple past in French "passe simple" with an analytic ambiguous "passe compose~ serving also as general past. The reanalysis simply stipulates a category change, independently of the other morpho-syntactic properties of Ls. If deflexion were sufficient for reanalysis, then we expect MA (in which apophonic Pass morphology was lost, due to a general loss of internal stem vowels) to resort to a reanalysis of the adjectival passive construction along similar lines. The reanalysis has simply extended the use of the [t-] inchoative affix to Pass, keeping verbal Pass and adjectival Pass separate. Finally, it is not clear why Ls like Classical Latin do not undergo a reanalysis like Italian in the active perfect. Keeping these objections in mind, we can now go back to the link I have established between morpho-syntactic complexity and morphonological realization. Limitations on the mappings are language specific, just as the resort to analysis is. I do not pretend that FC leads automatically to analysis, but I assume that peculiarities of the language aside, it does. For example, Voice occurs as verbal or adjectival in a number ofLs, but the variation is whether it occurs in finite or non-finite (participial) verbs. The option is then whether Voice is dominated by aT/Asp head associated with a finite Agr (or Agrl), or whether it is dominated by a non-finite T2/Agr2. The former option is found in synthetic passives, and the latter in analytic ones. Assuming then the role of complexity in leading to splitting (or fissioning) in syntax, let us designate the analysis by what we take to be its key feature, and call it the Splitting Analysis, and let us restate its parameters (with respect to Agr-T-Vo complex chains) as follows: ( 48)

Splitting Analysis a. Split Agr into finite Agrl and non-finite Agr2 if Aux is of nominal nature. b. Split the T chain into Tl and T2 if Aux is verbal. c. Split a finite Agr/T chain into a finite Agr/Tl and non-finite Agr/1'2 (and attach Pass to non-finite Agr/1'2).

These split operations take into account the peculiarities of Ls, as we have seen, with the absence of the present copula in Arabic and Latin (the negative instance of (48a)), the non-separability ofT and Agr in SA (a case of (48b)), as well as the high level of analysis in Germanic and Romance (a case of (48c)). In the next section, I address some further interactions between L peculiarities and these splits, as well as the role played by Height of Attachment of the Temp or Agr affix

Chapter 3. Synthetic/analytic asymmetries in voice and temporal patterns in deriving differences inside a single L, or across Ls, instead of attributing them to the nature of categorization.

Peculiarities and structural heights

3· 3.1

Imperfective passive

Arabic imperfective passives exhibit significant properties, and like their active counterparts, undergo no limitations on their formation or meaning. This is not the case for example in French, in which there is no such passives. On the other hand, English does have such imperfectives, although not with exactly the same distribution of meanings. Thus Arabic forms both past and present passive imperfectives, as exemplified by (49) and (50), respectively: (49)

kaa.na l-xi(.aa.b-u y-uktabu was the-speech-NoM 3-write.PASS The speech was being written.

(50)

yakuunu l-xitaab-u y-uktabu is the-speech-NoM 3-write.PASS a. b.

The speech will be being written. The speech is usually written (at four).

Consider now how the past or future progressive passives are expressed in French. French has no grammatical way to express these meanings, and must resort to lexical periphrasis, as illustrated by (51) and (52), respectively: (51)

Le discours etait en train d'~tre ecrit

(52)

Le discours sera en train d'@tre ecrit.

This lack of grammatical parallelism between French and SA can be traced back to the fact that French (a) lacks a progressive participle (which can enter a compound Temp chain, as in English), and (b) lacks the possibility of combining two finite verbs, the lower of which can be interpreted as imperfective/progressive (a possibility available in SA). Let us turn then to English. The English translation of SA (49) is acceptable without using the expression "in the process of': Then there is no significant difference with SA, except that the embedded passive is a synthetic verb in SA, but an analytic participle in English. With regard to (50), the use of the expression "in the process of' appears to make the acceptability easier, but the sentence is

85

86

Key Features and Parameters in Arabic Grammar

nonetheless acceptable. Now, acceptability apart, the two sentences do not share the same meanings.29 SA (SO) can only mean a future passive progressive. But a real present imperfective passive interpretation is available in English: (53)

The speech is being written (now).

Likewise, SA (SO) is ambiguously habitual, but its English present counterpart is not. These peculiarities recall the temporal conditions under which the present of the auxiliary and/or copula may be overtly realized in SA, English, or French. Moreover, non-available interpretations are attributable to the limitations arising from the absence of the imperfective passive (in the French morphological inventory), or the limitations on non-finite (past participle) Passive interpretations (in English and French). The availability of finite verbal Voice and finite imperfective in SA appear to allow more combinations. 3.2

Verbal and adjectival voices

Another striking difference between Arabic, English, and French is how Voice combines with verbs or adjectives to express e.g. Passive, and how the eventive/ stative reading split in passives is captured in their various grammars. Recall that in the early literature on English passives (and Ls of similar type like Romance), the debate was first centered around the catego1y of the passive participle. Thus the eventive agentive interpretation of (54), for example, given in (54a), was attributed to the verbality of the participle (the verb undergoing no category change there), while its stative non-agentive interpretation, given in (54b), was supposed to follow from its adjectival nature, derived through a category-changing operation (cf. in particular Wasow (1977) and Bresnan (1982); cf. also Levin & Rapaport (1986)): 30 (54)

The glass was broken. a. Someone broke the glass. b. The glass was in a state ofhavingbeen broken.

But as argued e.g. by Embick (2000b ), relying on a comparison ofEnglish, Chichewa and Creek passive expression, category is not the deciding factor. For example, there is no direct correlation between the syntax/semantics of stativization

29.

Thanks to Ken Hale and John Lumsden for these judgments.

30. Another issue was whether the derivation is lexically or syntactically driven, a distinc-

tion that Borer (1998), among others, have argued to be dispensible.

Chapter 3. Synthetic/analytic asymmetries in voice and temporal patterns

and the surfacing category31• It must then be that it is the distinct syntacticosemantic properties of the aspectual category heading the participle which make the difference. Thus Embick resorts to Height of Attachment of the aspectual head to derive the difference (in lines with Kratzer (1996) and Marantz (2000), among others). With statives, Asp is directly attached to the Root, and with eventives, it dominates small v. If there is no v, then no event interpretation is available. In sum, Embick appeals to structure heights to derive the ambiguous behaviour of an otherwise identical participial passive affix. 32 But note that Arabic does not exhibit ambiguities in Voice expression, similar to those found in English or Romance. This is so because Arabic resorts to distinct divisions in Voice expressions. The essential dichotomy used here can be identified as finite Voice, found with temporal verbs, and non-finite Voice, found with a temporal adjectives. To get directly to the distinguished flavour of the picture, observe that the two passive interpretations in (54) are expressed by two distinct constructions, (55) and (56), respectively: (55)

kusir-a

l-kazs-u

broke.PASS.-past.3 the-glass-NOM The glass was broken. (56)

l-kazs-u

kaana maksuur-an

the-glass-NOM was broken-Ace The glass was broken. Here, there is no room for ambiguity. In (55), the passive vocalic morphology is internal to the stem of the finite verb, which is marked forT and Agr, and which is both agentive and eventive (with the implicit agent, assessed by standard tests, like agent oriented adverbs and control structures). But in (56), the passive has adjectival pattern morphology, it receives Case, and whatever Agr adjectives receive. Moreover, it is stative, like any other adjectives. Clearly then, the passive in (55) is

31.

The essential facts are the following:

i.

In English, eventive and stative passives both surface as participles (with identical suffixes). In Chichewa, both eventive and stative can be just as verbal as anything else. In Creek, stative is as verbal as anything else.

ii. ill.

See also Embick (2000a) for similar arguments based on Latin. He also argues for the necessity of using Selection for appropriate characterization, a question that I will not address here.

32.

87

88 Key Features and Parameters in Arabic Grammar

dominated byv and T (among other functional projections), but the one in (56) is dominated by A (or a small a), and whatever stative aspectual or functional heads are found in the structure of adjectival phrases (cf. Fassi Fehri (1988b, 1993) for detailed proposals). Then if Pass (or Voice) is a functional head (as proposed e.g. in Fassi Fehri (1988b) and Kratzer (1996)), then Pass dominates v, and is dominated in turn by Asp and T, when it is verbal. A close examination of deverbal SA passive and active adjectives suggests further refinements (cf. Fassi Fehri (1993) for detail). Consider the active and passive adjectives exemplified in (57) and (58), respectively: (57)

r-rajul-u maani"/:J-un l-hadiyyat-a (gad-an, *zamsi) the-man-NOM giving-NoM the-gift-Ace (tomorrow, *yesterday) The man is giving the gift (tomorrow, *yesterday).

(58)

r-rajul-u mamnuub-un 1-hadiyyat-a (gad-an, *Zamsi) the-man-NOM given-NOM the-gift-Ace (tomorrow, *yesterday) The man is given the gift.

In both sentences, the adjective is (internally) eventive, assigns accusative Case to its object, and the event is temporally interpreted as taking place either in the present, or the future, as shown by the adverbial tests. But T is not overtly realized. These properties appear to mirror those of progressive and passive participles in English, auxiliary realization aside: (59)

The man is giving the gift.

(60)

The man is given the gift.

The question that arises then is the following: what is the difference between the translation of (59) or (60) by active or passive (internally eventive) adjectives such as those in (57) or (58), and its translation by finite verbs (in the present form), as illustrated by the following pair: (61)

r-rajul-u

y-amnal:Ju l-hadiyyat-a (gad-an, *zamsi)

the-man-NOM 3-give the-gift-Ace (tomorrow, *yesterday) The man is giving the gift (tomorrow, *yesterday). (62)

r-rajul-u

y-u-mnabu l-hadiyyat-a (gad-an, *zamsi)

the-man-NOM 3-PAss-give the-gift-Ace (tomorrow, *yesterday) The man is given the gift (tomorrow, *yesterday). It appears at first glance that the two constructions are equivalent, but upon closer examination, they are not. One striking difference between the two cases is that only verbs, but not adjectives, take part in complex tense formation, as shown by the following contrasting pair brought up in Fassi Fehri (ibiti):

Chapter 3. Synthetic/analytic asymmetries in voice and temporal patterns 89 (63)

ka.ana 1'-rajul-u.

y-ajlisu. rindamaa. daxal-tu l-qaa.rat-a

was the-man-NOM 3-sit when entered-! the-room-Aee The man was (in the process of) sitting when I entered the room. (64)

ka.ana 1'-rajul-u.

ja.alis-a.n

rindamaa. daxal-tu l-qaa.rat-a

was the-man-NOM sitting-Ace when entered-! the-room-Ace The man was sitting when I entered the room. As observed there, the sentence (63) is ambiguous in a way that (64) is not. In the latter, the adjective is stative, and when I entered, the man was already (in the state ofbeing) sitting. But (63) can have a reading in which he has not reached that state yet, i.e. a progressive reading. Clearly then the adjectives in embedded contexts are necessarily stative, whereas verbs are not. Likewise, the verbal passive (62) is progressive, whereas the adjectival (58) is not. These contrasts indicate that there is a clear separation in Arabic between finite verbs and (deverbal) active or passive adjectives. The former are temporal, whereas the latter are not Furthermore, the fact that those adjectives do not take part in complex tense combinations indicates clearly that they cannot count as verbal participles, i.e. they possess no "mixed" or variable behaviour category (paralleling that of English or Romance), which would enable them to alternate between a category which does take part in the temporal system (i.e. participles dominated by Perfect or T2) and a category which does take part in it (i.e. stative adjectives). 33 3·3

Multiple functions across heights

It is widely spread property of Ls that a morphological form is often ambiguously used to express different GFs, or the same GF at difierent levels of structure. As I have amply argued elsewhere (see Fassi Fehri (1993), and 1996b/2000, in particular), verbal morphology encodes not only Tense, Mood, and Aspect ( = MTA there), but also Voice and Agr. These functional traits (or categories) are organized hierarchically, according to (65), leaving Agr aside: (65)

Mood> Tl > T2 >Aspect> Voice

Tl stands for the Past/non-Past distinction, T2 for the Perfect/Imperfect, Aspect for Perfective/Imperfective. Unlike the Present/Imperfect prefixed form, which

33· This distinction appears to be clear-cut in the case of passive adjectives, but not in the case of active adjectives. For example, (58) embedded under kaana is solely interpreted as a stative complement of the copula, whereas (57) may have a compound tense interpretation, similar to that in (60 ). But in Modem Standard Arabic, the dominant tendency is to use finite verbs for complex Ts.

90

Key Features and Parameters in Arabic Grammar marks Mood, the Past/Perfect suffixed form does not. However, this form does have modal uses, in wishes, or hypothetical situations, as illustrated by the following sentence: (66)

ra1;tima-ka

l-laah-u

blessed-you Allah-NOM May Allah bless you! These ambiguous uses of suffixed temporal morphology can be accounted for via either movement or appropriate placement of the same synthetic word under the relevant category. The following gives a rough sketch of how multi-functionality can be implemented (I use the word kataba "wrote" as an instance for representing this ambiguity): (67)

MoodP/Irrealis kataba

~

Tl/Past kataba

~

T2/Perf kataba

~

Aspect/Pfv kataba

The lower part of the tree, which is irrelevant for the point, is the following: (68)

VoiP

~

Voi

vP

~

v

VP

~

V

NP

As for Voice contrasts, they can be represented in the same way. Three cases are to be represented here: (a) the completely stative adjectival passive, which can be instantiated by (54), the (internally) eventive adjectival passive in (58), although with no implicit agent, and the eventive agentive verbal passive in (53). Following the lines of Marantz (2000) and Embick (2000b) in particular, these voice contrasts can be represented roughly as follows (R for Root): (69)

aP/AspP

~

VoiP

~

RP

~

R

NP

Chapter 3. Synthetic/analytic asymmetries in voice and temporal patterns

(70)

aP/AspP

~

VoiP

~

vP

~

NP

v

(71)

TP

~

VoiP

~

vP

~

NP

vP

~

v

NP

These structures are given for the sake of illustration. In the totally stative (69), Vo is heading RootP (before category formation), and the category (with its aspectual properties) dominates Vo. In the partly eventive (70), Vo dominates an unaccusative vP, with an internal argument only. The small v makes it eventive, but no Agent can be implied. The Asp/ap converts it to an adjective externally. In the verbal eventive and agentive (71), the vP is transitive, dominated by Voi, but the latter is dominated by TP. which makes the verb finite.

4·

Summary and conclusion

In this article, I have shown how temporal and Voice expressions in a number of Ls distribute with respect to the synthetic/analytic continuum, and what kinds of limitations and constraints operate across Ls, and inside a particular L, taking into account its peculiarities. Formal Complexity has been argued to be behind changes from synthetic to analytic expression, once a degree level of complexity is reached in the Temp chain. Appropriate predictions are then made about potential analytic types to be found. The Splitting Analysis operates on Temp chains, as well as on Agrs, to create two separate, though anaphoric chains. Variation splits have been argued to depend on the nominal or temporal nature of the auxiliary features in L, the existence (or non-existence) of non-finite (temporal) Agr in L, and the ability ofVoi or Pass to be (or not be) dominated by a Temp or an Agr projection (see the parameters in (48)). Finally, Formal Disjunctivity has been proposed to limit the choices in Ls to either synthetic or analytic expression of GF in L, but not both, hence making the choices typically asymmetrical.

91

CHAPTER4

Arabic Perfect and temporal adverbs* Languages use grammaticalized temporal inflections (Tp), auxiliaries (Aux), and temporal adverbials (Adv) to express various kinds of temporal reference (TR). The descriptive program of language specific or cross-linguistic temporalities appears then to be to identify which ingredients ofTR grammar and meaning are contributed by the various Tp, Aux, or Adv expressions, or their combinations. Common Tp categories which contribute to (and organize) linguistic TR include Tenses (T), Aspects (Asp), and Aktionsarten (Akt), which project as functional categories in scopal hierarchical syntactic structures. It is largely acknowledged in the literature that Tp categories are ambiguous, or underspecified with respect to TR (see Chapter 1). Moreover, their distinct or identical forms may lead to distinct interpretations within the same language, or across languages. Likewise, temporal Adv can be equally ambiguous in the same language, and they may also produce distinct temporal meanings across languages. The main aim of this chapter is to investigate more closely the salient properties of the Arabic Perfect, and those of temporal Adv that collocate with, or coerce its interpretation. First, various kinds of Perfect (Perf) documented and studied in other languages need to be examined here, to see whether they are instantiated in Arabic, or whether they exhibit common or distinct properties. Second, Perfect interacts significantly with Past and Perfective (Pfv), as already investigated in Chapter 1, exhibiting various TR ambiguities, which need to be properly re-characterized once the disambiguation role of collocation Adv is taken into account. My approach to general semantic and morphological questions is based on the assumption that there are natural (or canonical) mappings between temporal/ aspectual forms and their semantic interpretations. Forms are productively ambiguous, but they are associated with abstract syntactic and selective properties, which then provide room for language specific differentiations. The latter are captured properly only when the semantics involved is made precise, to allow

*

This Chapter has its source in Fassi Fehri (2003c). It elaborates on previous worl<: of mine (cf. Fassi Fehri (2000c), as well as (1993)). It is particularly infiuenced by Arnim von Stechow's (2002; = AvS) work on the semantics of the Perfect and typically the German seit.

94 Key Features and Parameters in Arabic Grammar

accurate comparative work. Elaborating on the results reached in Chapter 1, my grammatical description of TR relations, expressions, and distributions will assume a hierarchical architecture in which Perf is generated higher than Asp (which is basically[± Pfv]), as some 'relative' T, or T2. The latter is lower than the 'deictic' or 'absolute' T1 (denoting Past, Present, or Future), as in the following core temporal architecture: (1)

T1 (±Past)

~

T2 (±Perf)

~

Asp(± Pfv)

~

VP (±Tel)

In Section 1, I recapitulate first some salient properties of Tp inflections, then provide rough semantic characterizations of aspects and tenses. In Section 2, I examine the contribution of deictic, modal, and durative Adv in coercing these interpretations. In Section 3, I turn in particular to investigating the various meanings of mundu, and the kinds of Perf found in Arabic, as well as their interactions. In Section 4, I provide a summary of the results found with respect to interpretations of Tp inflections, Adv, and particles, and conclude with general remarks about the advantages of this dual and interactive description ofTR.

1.

Salient properties of the Arabic TR system

Various salient properties of the Arabic TR system have been addressed in Chapters 1 & 3, and they are worth recalling here. They include the polyfunctionality of the tense/aspect forms, the synthetic/analytic splits, the Past/Perf ambiguity, as well as some properties of durational Adv. 1.1

Polyfunctionality ofT I Asp forms

The T/Asp system is dominated by the polyfunctionality of its temporal forms. First, recall that a temporally inflected verb serves either as a temporally finite verb, or as a 'participle'. In other words, the form of expression of T1 (and Agr1) and T2 (and Agr2) is identical, and the morphology is not discriminatory as far as the Tl/T2 distinction is concerned. This uniformity is exemplified in the two following pairs of constructions. The first pair contrasts the two basic simple tenses available in the system: Past and Present (or non-Past), which express semantic PAST and PRESENT, respectively:

Chapter 4. Arabic Perfect and temporal adverbs

(2)

ka.tab-a r-risaalat-a wrote-3 the-letter-Ace He wrote the letter.

(3)

y-aktu.b-u

r-risaalat-a

3-write-IND the-letter-Ace He is writing the letter. The second pair contains the same forms embedded under auxiliaries, and functioning as Perf and Imperf participles: (4)

ka.an-a katab-a r-risaa.la.t-a was-3 wrote-3 the-letter-Ace He had written the letter.

(5)

kaan-a y-aktub-u

r-risaalat-a

was-3 3-write-IND the-letter-Ace He was writing the letter. Second, the duality between Past/Present and Perfect/Imperfect is further complicated by the fact that these forms can express a Perfective/Imperfective opposition. For example, Past in (2) is interpreted as PFV, and Present in (3) or (5) as IPFV. To be neutral and for ease of comprehension, I will designate the two forms as Past/ Pres and Perf/Imperf interchangeably and depending on contexts. I will also refer to their varied semantics by capital letters, when necessary. Clearly, the terminological dispute about morphological forms is meaningless in the absence of precise and clear-cut semantic associations. 1.2

The PresPerf split: Synthesis and analysis

A second characteristic feature of the Arabic TIAsp system is the synthetic!analytic split of its PresPerf forms. The synthetic form (which is homophonous with that of the Past) is limited to PRES interpretation, which will be detailed in this chapter. It is exemplified in (6), which can also have a Past interpretation: (6)

jaraa ran He has run.

Analytic forms, however, in which the Pres auxiliary kaana 'be' is overtly realized, do not have PRES interpretation. They normally express a futurate Perf ( = FUTPERF), as in (7), and they collocate with future time adverbs: (7)

1-akuunu zanhay-tu r-risaalat-a

gadan

l-am finished-! the-letter-Ace tomorrow-Ace I will have finished the letter tomorrow.

95

96 Key Features and Parameters in Arabic Grammar

In other words, the "tense" forms can also function as "aspect" forms. They can also express "iterative" reading, as in (8): (8)

bi-haadaa y-akuunu l-fariiq-u with-this

xams-a

3-is

qad

sajjala

the-team-NOM already scored

marraat-in

five-ACC times-GEN With this, the team has already scored five times. This situation is partly comparable to that of Portuguese, where the PresPerf is expressed by a simple form (which is also ambiguous with the Past), as in (9), but the analytic form expresses only a habitual Perfect, as in (10): 1 (9)

Comi ate I have eaten.

(10)

Agora ja tem comido now I have eaten Now I have taken the habit to eat.

It is unlike the situation in Germanic or Romance, where the PresPerf, in addition to the Past and further meanings, is expressed by an analytic Perf. In German, for example, the PresPerf cooccur with Past, Pres, and Fut adverbs (cf. Musan (2001, 2003)): (11)

Hans ist

gestern

um zehn weggegangen

Hans is/has yesterday at (12)

Hans ist

ten

left.

jetz weggega.ngen

Hans is/has now left. (13)

Hans ist

morgen

um zehn weggegangen.

Hans is/has tomorrow at

ten

left.

As we will see, this split is not without consequences for dividing lines of interpretations and/or ambiguities, as well as crosslinguistic characterizations. 1.3

The Past split: Simple Past Pfv and complex Past Impfv

A third important feature of the T/Asp system is that it exhibits also an aspectual split with respect to the form of expression of aspectual values for Past. Simple Past is associated with Perfective, whereas Past Imperfective must be expressed by two

See Giorgi & Pianesi (1997) as well as Schmitt (2001) for di1ferent analyses of the Portuguese Perf.

t.

Chapter 4. Arabic Perfect and temporal adverbs

separate finite forms: the auxiliary in the Past, and the thematic verb in the form of the Pres. The two contrasting forms exhibiting the split are given in (14) and (15), respectively: (14)

za.nhaa. r-risaalat-a. finished the-letter-Ace He finished the letter.

(15)

kaana y-unhii r-risaalat-a was 3-finish the-letter-Ace He was finishing the letter.

The letter is finished in (14), but not in (15). This split is partly comparable to that of English simple Past and Past Progressive. It contrasts with the double synthetic nature of this opposition found in e.g. French ("passe simple" vs. "imparfait") or Italian ("passato remoto" vs. "imperfetto"). There are various ways to assess the perfective nature of the simple Past. One argument comes from the behaviour of the Past in SOT contexts. In complement clauses, the simultaneity with the past is solely acceptable if the Present/Imperfect form is used (i.e. no SOT efiect is possible, just like the situation in Japanese or Hebrew):2 (16)

qaal-a l-ii 1inna-hu y-aktub-u r-risaalat-a said-3 to-me that-him 3-write the-letter-Ace He said to me that he was writing the letter.

The use of the Past/Perfect does not yield that interpretation: (17)

qaal-a l-ii 1inna-hu katab-a r-risaalat-a said-3 to-me that-him wrote-3 the-letter-Ace He said to me that he wrote the letter.

In (17), there is no overlapping of the writing and the saying, unlike the situation in (16). The writing is rather anterior to the saying (with a shifted Past reading). This situation is in contrast with SOT behaviour in complement clauses in English, where the embedded verb can be interpreted as anaphoric (or simultaneous, with the meaning "Mary is ill"): 3

2. Cf. Ogihara (1995) and Abush (1997), among others. Cf. also Higginbotham's (2000a) analysis of SOT, based on anaphoridty.

3·

The Arabic stative counterpart to ( 18) read as anaphoric is (i), where no copula surfaces: (i)

qaal-a 1-ii

1inna-hu marii{j-un

said-3 to-me that-him sick-NoM He said to me that he is sick.

97

98

Key Features and Parameters in Arabic Grammar

(18)

John said Mary was ill.

Giorgi & Pianesi (1995) compare languages which possess Imperfect (like Italian) with languages that do not (e.g. German or English). They observe that Italian Imperfect is a dependent anaphoric tense, which denotes Present-under-Past tense, the Past being provided normally by the matrix verb. Moreover, the (simple) Past cannot be used in this language as dependent. In embedded contexts, only Imperfect can be used, with a SOT effect. 4 In contrast, languages like German use Past (ambiguously) as dependent or non-dependent, the latter being neutral with respect to perfectivity. It can be read either way, as simultaneous (Imperfective) or non -simultaneous (Perfective): (19)

Hans sagte, dap Marie einen Apfel ap. a. Hans said that Mary was eating an apple. b. Hans said that Mary ate an apple.

In (a), the simultaneous reading is available, and Past is interpreted as Imperfective. In (b), Past is Perfective, and the simultaneous interpretation is excluded. Thus (simple) Past is necessarily Perfective in Italian, but not so in German. The conclusion then is that if a language has an Imperfect Past, as opposed to a simple Past, the latter cannot be used as dependent in embedded contexts. On the other hand, the Imperfect is dependent and used to denote simultaneity. In languages in which the Past is unspecified, it can be used as dependent/simultaneous. Arabic is close to Italian in this respect. Past is Perfective and non-dependent, contrary to Imperfect, as illustrated above. A second argument comes from the availability of continuous readings. Consider e.g. (6), expressing a Past activity (repeated here as (20) for convenience), and its Past Imperfective counterpart, given in (21): (20)

jaraa ran He ran.

(21)

kaana r-rajul-u

y-aJm

was the-man-NOM 3-run The man was running.

When the past copula is used, no "present -under-past" reading is possible: (ii)

qaal-a 1-11

?inna-hu

kaan-a mari/4-an

said-3 to-me that-him was-3 He said to me that he was sick. 4·

sick-Ace

Compare their examples (17) and (2l),p. 349.

Chapter 4. Arabic Perfect and temporal adverbs

In (20), the event is completed or terminated. It cannot be further extended. This is in contrast with (21), which is non-terminated and can be extended. Hence the sequence in (22) can be used after (21), but it leads to ungrammaticalityafter (20): (22)

[ ... ] wa. maa zaa.la. y-ajrii and still 3-run and he is still running.

In fact, the combination of (20) and (22) can be acceptable, but only if two events are involved, one terminated to make the first clause true, and one non-terminated and ongoing at liT, to make the second sequence possible. 5

2.

The Perfect/Past ambiguity

A significant property of the system consists in a genuine Perfect/Past ambiguity of the temporally suffixed form. As already observed, the temporally suffixing morphology in Arabic can express either PRESPERF or PAST meanings. Thus (2) or (6) above are ambiguous between what I represent in a very sketchy way as (23) and (24), using a simplified Reichenbachian representation of time: (23)

ET, RT < UT

(24)

ET < RT, UT

In the first representation, the Perf form is expressing PAST tense, and in the second PERFECT "aspect': The representation in (23) raises no significant problem. That in (24) is more disputable. I will ignore the potential objections for the moment, and focuse only on the fact that PAST specifies a reference time (RT) which is prior to the utterance time (UT), while PERFECT does not. To the extent that an anteriority component is involved in the PERF interpretation, that anteriority should involve other argument times than UT. This subsection provides grounds for assessing that the Past/PresPerf ambiguity represented here is genuine, rather than thinking of it as an ambiguity arising within the PresPerf semantic configuration itself, the interpretation differences being associated with the various temporal adverb specifications. My task is then basically to provide arguments establishing that, in various constructions that contain the

5· Giorgi & Pianesi (2001), who analyze quite similar contrasts in Italian, reach a similar conclusioiL Italian simple Past and Perfect are Perfective, and they contrast with Past Imperfect, which is Imperfective. The former tenses cannot be extended (unless the interpretation is different, as indicated). See e.g. the contrasts in their examples (22) and (23); see also Chapter 1 for more detail.

99

100

Key Features and Parameters in Arabic Grammar

Perf form, RT cannot be (~) UT, but what is involved is rather RT < UT, which is basically what PAST expresses. But before going on providing such arguments, I will sketch, for concreteness, a reasonable view of what Tenses and Aspects are, although I will not address the details of the new developments of Tense and Aspect theories, the points to be made being neutral with respect to any viable version of them. 6 2.1

Aspects and Tenses

As stated above, Reichenbachian (1947) terminology will be used. The time denoted by a semantic tense is called reference time (= RT). Aktionsarten are located in time by means of a relation that connects RT to an event time or event state (= ET), instantiating the aktionsart This relation can be called an ASPECT relation, after Klein (1994). RT may INCLUDE g) ET in the "Perfective': RT may be INCLUDED(~) in ET in the "Imperfective~ or RT may follow ET, being POST (>)in the "Perfect'? Asp relations, viewed as "temporalisers~ can be defined, after Arnim von Stechow (2002b; = AvS), as follows: 8 (25)

ASPECTS a. b. c.

II PERFECTNE II= A.PA.t3e.t;;;2 't (e) & P (e), P of type (v,t) II IMPERFECTIVE II =A.PA.t3e.t ~ 't (e) & P (e) II PERFECT II =A.PA.t3e.t > 't (e) & P (e)

6. If we follow e.g. Klein (1994, 1999), then Perf is treated as Asp, on a par with Pfv, although Perf is normally higher in structure than Pfv, as in English John has been eating an apple. In other theories, it is treated as "relative tense" in the sense of Comrie (1985), or T2 (ct: e.g. Giorgi & Pianesi (1997) ). In such a case, two Rs can be postulated, R1 and R2, as proposed explicitly in Cinque (1999), following an approach initiated by Sten V!kner (1985). The two Rs approach has various advantages, among which is the fact that it makes room for the representation of the future perfect of the past (as in He would have worked). That approach is implicitly adopted here, although details will not be addressed. I will only place Perf here as higher than Pfv (the genuine Asp). Cf. Cinque (ibid) for cross-linguistic motivation, as well as Chapter 1. 7· Klein's (1994) aspectual relations can be simplified as follows: RT (=his TT, or Topic Time) is a subinterval ofET, or is (properly) induded in it; RT contains ET; RT follows ET; RT precedes ET. 8. e is an event or a state; 't(e) is the time of e, i.e. the interval from the beginning of e till the end of e; v is the type of events, i that of states.

Chapter 4. Arabic Perfect and temporal adverbs

Elaborated definitions for TENSES are also provided by AvS (ibid), in line with Partee's (1973) reference tense theory, and Heim's (1994) proposal that tenses be viewed as restrictors of temporal variables: 9 (26)

TENSES are symbols of type i which bear time variables as indices. Let c be the context of the utterance, with tc the speech time, g is a variable assignment, then a. b. c.

II NOW II g,c is the speech time conceived as a point II PASTj II g,c is defined only if g(j) precedes the speech time tc. If defined, II PASTj II g,• = g(j). II FUTRj II g,c is defined only if g(j) follows the speech time tc. If defined, II FUTRj II g,c = g(j).

These definitions (or simplified versions of them) will be adopted here, except for Perf, which needs further elaboration in order to account for its various meanings, but also its (usually) higher position in the tree architecture, compared to Pfv/Ipfv (which I take to be the core Asp opposition), in conformity with (1). As an illustration of the PAST/PERFECT contrast, consider (27). The latter can be interpreted as PAST, or PRESPERF, hence giving rise to the two LF forms (28) and (29), respectively: (27)

sakan-a barliin-a lived-3 a. b.

Berlin-Ace

He lived in Berlin. He has lived in Berlin.

(28)

1P[PAST AsPP [ PFV yp[He live in Berlin]]]

(29)

1P[NOW PERPP[be AsPP[ VP[He live in Berlin]]]]

The representation (29) reflects the overall architecture of the system, in which PerfP (or T2) is higher than PfvP (or AspP). I have introduced "be" as a head of PerfP, although it is not realized. It could be that this introduction is not necessary, and the "participle" licenses both PERF and PRES. 10 If we follow AvS (ibid,

9· Simple definitions are provided e.g. by Klein (1994): (i) (ii) (iii)

PAST: (some subinterval of) RT is BEFORE UT; FUTURE: RT is AFTER UT; PRESENT: RT is SIMULTANEOUS TO, or CONTAINS UT.

Note that the "participle" of the thematic verb may be taken as the source for licensing PRES, PERF, and even PFV when the PresPerf is interpreted as such. This observation is based on the fact that the sole auxiliary "be" is not selective, and has presumably no "Perf content':

10.

101

102

Key Features and Parameters in Arabic Grammar

Footnote 16, p. 10), then PERF here is an existential quantifier that introduces a time/event prior to RT. Its interpretation is as in (30): 11 (30)

3t t
Note that the construction (27) can also have an extended-now (XN) interpretation, in fact a natural one. Consider the following variants of (27): (27')

a.

sakan-a barliin-a

mundu 1990

lived-3 Berlin-Ace since 1990 He has lived in Berlin since 1990. b.

sakan-a barliin-a

xams-a

marraat-in mundu 1990.

lived-3 Berlin-Ace five-Ace times-GEN since He has lived in Berlin five times since 1990.

1990

Like English since, mundu can introduce a time span through oc, its complement, i.e. a position in time given by a date or some other temporal description. Mundu oc modifies the XN introduced by Perf, and it indicates that oc is the left boundary (LB) ofXN. The construction (27b) is best analyzed as an XN-Perf, in this case an Existential Perf (E-Perf). The quantificational adverb xams-a marraat-in has to be confined within XN. The meaning of the £-Perfect can be represented as follows: (31)

3tXN (t, NOW) & LB (1990, t) & 35t' ~tHe live in Berlin at t'.

NOW is the meaning of the semantic PRES, the speech time conceived as a moment. XN (t',t) means that tis a final subinterval oft'. LB (1990, t') means that 1990 is LB oft'. The construction (27a) can be interpreted as a Universal Perf (U-Perf). The U-reading can be formulated as follows: 12

unlike what has been proposed for "have': See Wunderlich (1997) for a quite illuminating discussion on the contributions of auxiliaries and participles to temporal meanings, depending on cross-linguistic lexical differences.

u. AvS claims that this is the simplest analysis, going back at least to Prior (1967). Cf. the reference there. For further discussion, see Paslawska & von Stechow (2003). For the U/E distinction of Perfects, see McCawley (1971), McCoard (1978), Dowty ( 1979), M ittwoch ( 1988), Vlach ( 1993), among others. Iatridou, Anagnostopoulo, & Izvorski (2001) provide formulations of the two readings and their paraphrases which, if applied to this example, read as follows: 12.

U-reading (i)

a. b.

There is a time span (Perl tsp) whose LB is in 1990 and whose RB is UT, and throughout that tsp He lives in Berlin. 3i (LB = 1990) & RB = Now & \tt E i (Ev (t))).

Chapter 4. Arabic Perfect and temporal adverbs 103

(32) 2.2

3tXN (t, NOW) & LB (1990, t) &"i:lt' ~tHe live in Berlin at t'.

Positional "deictic" adverbs

To disambiguate PRESPERF and PAST meanings, positional deictic adverbs can be used. The contrast is provided in (33) and (34), where past and present adverbials are used. Note that the temporal morphology is not compatible with future adverbs: (33)

katab-a r-risaalat-a

zamsi

(*gad-an)

wrote-3 the-letter-Ace yesterday (*tomorrow) He wrote the letter yesterday. (34)

ka.tab-a r-risaalat-a l-zaan-a (*gad-an) wrote-3 the-letter-ac now (*tomorrow) He has written the letter now (*tomorrow).

The positional adverb diagnostic suggests that the adverb is taken to modify RT. It has been used for English successfully, at least since Jespersen (1924). But this test appears to be disputable, once the behaviour of PresPerf in other languages is taken into account Consider the following pair of German examples (=(13a) and (14) ofLobner's (2002)): (35)

Karla ist gestern

hier eingezogen.

is yesterday here move.in.PART Karla (has) moved in here yesterday. (36)

Jetzt, wo

Ka.rla gestern

hier eingezogen

ist,

now where Karla yesterday here move.in.PART is brauchen wir einen Schliissel furs Klo. need we a key for the toilet Now that Karla has moved in here yesterday, we need a key for the toilet The first construction is ambiguous between a Past reading and a PresPerfect reading. But the Perfekt can have only a PRESPERF reading in (36), although it combines with a past adverbial. This suggests that the adverb is modifying the aspect phrase or some lower phrase, the VP (denoting ET). Thus the combinability

B-reading (ii)

a. b.

There is a tsp (Perftsp) whose LB is in 1990 and whose RB is UT, and in that tsp is an ev of his living in Berlin. :3i (LB = 1990) & RB = Now & ::It E i (Ev (t))).

On further refinements, as well as a characterization of English "since': see Von Fintel & Iatridou (2002).

104

Key Features and Parameters in Arabic Grammar

of Perf with past and non-past time adverbials cannot count as an argument in favour of the ambiguity of the Perf, at least in the German type languages (see Klein (1999) and Musan (2001), among others, for arguments). If a similar reasoning were to extend to Arabic, then we have to look for other diagnostics to establish the existence of a genuine PAST interpretation. 2.3

Perf and modal qad

The particle qad (meaning "already': i.e. precedence, or "just': immediate precedence) is typically acknowledged in traditional literature to be collocating with Perf readings (see Chapter 1, subsection 1.2, on which I elaborate here). For example, qad can occur immediately before the "participle" of a complex Perfect tense, as in (37): (37)

kaan-a qad

katab-a r-risaalat-a

was already wrote-3 the-letter-Ace He had already written the letter.

It can also occur in front of a simple tensed verb, interpreted as PresPerf. Thus in (38), qad can have one of the two precedence readings: (38)

w~ala

qad

already/just arrived a.

b.

He has already arrived. He has just arrived.

But in (38), qad can be further ambiguous in a way that it cannot be in (37). It can mean "indeed': or "in fact~ to stress the factual certainty of the event, and the sentence is translated as "He indeed came" or "He did come': In the latter case, it serves as a modal. Modals are projected higher than Tense (or T1). Unsurprisingly then, when qad serves as a modal it takes T1 projections as its complement, namely Past projections, cooccurring with past time adverbials, as illustrated in (39): (39)

qad

w~al-a

zamsi

indeed arrived-3 yesterday He did arrive yesterday. Likewise, it takes also Present projections, as in (40), but the interpretation is that of "possibility" or "probability': rather than certainty: (40)

qad

y-~il-u

l-zaan-a

may 3-arrive now He may arrive now.

Chapter 4. Arabic Perfect and temporal adverbs The T1 nature of the complement of qad is further corroborated by its occurrence in front of the auxiliary of a complex tense: (41)

qad

kaan-a

y-~aUii

indeed was 3-pray He was indeed praying. The question now is the following: why can't the complement of modal qad in (39) be taken as Perf (or T2), instead of Past (or T1)? In other words, what motivates restricting the selection of modal qad to T1? The answer is straightforward. Consider the following minimal pair of constructions: (42)

ka.an-a qad

wa~a.l-a

was-3 already arrived-3 He had already arrived. (43)

qad

kaan-a

w~al-a

Indeed was- 3 arrived- 3 He had indeed arrived. Both constructions are typically non-ambiguous. In (42), qad is only a Perfect Level specifier, and in (43) a Modal specifier. This absence of ambiguity is due to the fact that qad is unambiguously positioned, with respect to T1 (Past), or T2 (Perf). In (38), the ambiguity arises due to ambiguity placement. Note that no such ambiguity arises in (41), and (39) has at least one reading in which it is a modal structure like (41), which implies that Past, rather than PresPerf is involved there. Note in passing also that if modal (39) were to be interpreted as Pres, then two problems arise. First, we predict wrongly that the interpretation is "possibility~ rather than certainty. Second, qad would be selecting an empty auxiliary (or copula), which it normally does not, as shown by the ungrammaticality of(44): (44)

*qad r-rajul-u musaa.fir-un may the-man travelling-NoM Intended to mean: The man may be traveling.

In this case, the realization of the copula in the Pres form is obligatory ( Cf. Fassi Fehri (1993) for detail). The particle Perf qad has a partly similar German counterpart. Schon, a case of RT or Perf level specification,can be used with Past or with Perfek.t. With the former, the interpretation is normally that the action only started, or it is imperfective, but with the latter it is ambiguous. It has a PresPerf meaning, with the action understood as completed, or it may have an interpretation similar to that

105

106

Key Features and Parameters in Arabic Grammar of Past imperfective. The examples (45) are from Musan (2001. p. 363), and (46) is from Lohner (2002, pp. 386-7):13 (45)

a.

Er hat schon

gegessen.

He has already eaten(= He finished his meal). b.

Er af3 schon He ate already (= He was already eating).

(46)

Sie hat schon

gefruhstuckt.

she has already have breakfast a. b.

She has already had breakfast (Present Pertect) She was already having breakfast. (Past Irnpertective)

Likewise, qad interpretation differs depending on the temporal entity it collocates with. When it means "already': it is a Perf detector. When it is a modal, it collocates with Tl, meaning "indeed" with Past, and "possible" or "probable" with Pres. 14 Thus qad is uniformly a modal when it is higher than T, and a Perf (or Asp) specifier otherwise. Its interpretation is then a true diagnostic for helping us determining whether the Perf form is to be interpreted as PAST or PERF. 2.4

Adverbs and simple vs. complex tenses

Positional adverbs are known to be ambiguous with complex tenses (typically Perf tenses), because their structure allows for two positions that the adverb can specify. 15 For example, this ambiguity is found in (47), where the letter could have been finished at four, or before four: (47)

ka.ana qad

1anhaa r-risaalat-a

fi.i r-raa.bira.t-i

was already finished the-letter-Ace at the-four-GEN He had already finished the letter at four. Surprisingly, however, this ambiguity is not found with PresPerf, be it analytic, as in (48), or synthetic, as in (49):

Musan attributes her examples to Wolfgang Klein (p.c.). Her analysis of the PresPerf differs obviously from LBbner's (2002), who argues that the tense semantics of the latter is not uniformly PRES (Perf), but can be PAST (non-Pert) as well

13.

14.

See Fassi Fehri (1993) for detaiL

The two positions are normally taken to be ET or RT, but they could be two RTs, as pointed out in Footnote 9 above. I will assume the latter approach, which takes into account the "tense" dimension of Perf. 15.

Chapter 4. Arabic Perfect and temporal adverbs (48)

1-akuunu. 1anhay-tu r-risaalat-a

ga.da.n

fi.i r-raa.bira.t-i

I-am finished-! the-letter-Ace tomorrow-Ace at the-four-GEN I will have finished the letter tomorrow at four. (49)

za.nhaa. r-risaalat-a.

(l-zaan-a, 1amsi) fi.i r-raabirat-i

finished the-letter-Ace (now, yesterday) at the-four-GEN He (has) finished the letter (now, yesterday) at four. In (48), the adverb specifies only RT, i.e. the tense carried by the auxiliary. In (49), read as PRESPERF, with a Now adverb, the same is true. The problem arises with (49), when collocating with a past adverb. If(49) is read as PRESPERF in this case, then the past adverbial has to be interpreted as specifying ET, contrary to what happens with other PresPerfs. If (49) is read as PAST, no such a problem arises, because then the temporal Adv would be modifying (uniformly) RT. Since the minimal hypothesis is to assume a uniform RT specification, I will assume that this is the case, unless evidence is provided for the contrary. But note that if the tense were PresPerf, then the (covert) auxiliary/copula denoting PRES would have to be modified by the positional.fii phrase. But such a possibility is independently excluded, as shown by the ungrammaticality of the following PRES sentence: (SO)

r-rajul-u.

marii4-un ("fii r-raabira.t-i)

the-man-NOM sick-NOM (*at four) The man is sick (*at four). Zero-Pres then appears to be not an option with a past adverbial, which suggests in turn that the tense involved can only be PAST. If that is true, then positional temporal adverbs play a role in T/A disambiguization, contrary to the conclusion reached in subsection 2.2, although not directly. 16 2.5

Durational adverbs

Indefinite temporal nominals can function as durational adverbials. They are either complements of mundu "since~ or they take (what looks like) an accusative case. Call them mundu-d and Ace-d, d for durative. The two phrases do not distribute in the same way with tenses and aspects they occur with, nor do they specify them in the same way.

Note that the ambiguity problem appears to be solved differently in languages having a general analytic Perf, such as German or French. Musan (2001), for example, who discusses similar facts in German, acknowledges an asymmetrical behaviour of adverbs with PresPerf: Pres and Fut Advs target RT, whereas Past Advs target only ET. But this limitation of the Past Adv to ET is not motivated in view of the data discussed by L5bner (2002), on the one hand, and the RT level of Past Advs occurring with PAST, on the other hand.

16.

107

108

Key Features and Parameters in Arabic Grammar

Semantic PRES and PRESPERF are compatible with mundu followed by an indefinite complement: (51)

y-aktubu r-risaa.la.t-a

mu.ndu xams-i

saaraa.t-in (l-zaana)

3-write the-letter-Ace since five-GEN hours He is writing the letter since/for five hours (now). (52)

katab-a r-risaa.la.ta

(now)

mu.ndu xams-i saaraa.t-in (l-zaana)

wrote-3 the-letter-Ace since five hours He has written the letter since five hours (now).

(now)

But PAST does not combine with such adverbs: (53)

..kata.b-a. r-risaa.la.t-a l-baa.ri1;tat-a. mundu. xamsi saaraat-in wrote-3 the-letter-Ace yesterday since five hours

The advervial mundu-d introduces an interval that reaches up to RT. But RT has to be UT, basically NOW. If the adverb has RT,UT as its RB, then its meaning is not compatible with that of Past (which is RT
saafara

r-rajul-u

l-baari1;tata xams-a saaraat-in

travelled the-man-NOM yesterday five The man travelled yesterday five hours.

17.

hours

Constructions like (i) are also interpreted as PresPerf: (i)

qul-tu haadaa mundu zaman-in tawiil-in said-1 this since time-GI!N long-GEN I have said this a long time ago.

Note that PAST is compatible with mundu taking definite complements, but these cases are analyzed differently, as positional adverbs, which introduce a time span located somewhere in the past. See subsection 3.1 below.

Chapter 4. Arabic Perfect and temporal adverbs

But PRES cannot combine with Ace-d: (55)

*y-azkulu r-rajul-u 3-eat

(l-zaan-a) xams-a

the-man-NOM (now)

saaraat-in

five-Ace hours-GEN

The latter sentence is possible only with a habitual meaning ("The man has the habit now to eat for five hours"). The same contrast holds with statives: (56)

za.msi

ka.an-a r-rajul-u

marii4-an xams-a saa.raat-in

yesterday was-3 the-man-NOM sick-Ace Yesterday, the man was sick for five hours . (57)

..r-rajul-u

ma.rii4-un xamsat-a. zayyam-in

the-man-NOM sick-NOM five (58)

five-Ace hours-GEN

..r-rajul-u

musa.afir-un

days

xamsat-a zayyam-in

the-man-NOM travelling-NOM five

days

The latter construction is possible as a futurate, not PRES. The contrast between PAST and PRESPERF is easier to detect with statives: (59)

zamsi

kaan-a r-rajul-u

yesterday was-3

jaalis-an,

f:w.ziin-an

the-man-NOM sitting-Ace sad-Ace

xams-a saaraat-in five

hours

Yesterday, the man was sitting, sad for five hours. (60)

""kaa.n-a r-rajul-u was-3

jaalis-a.n,

f:raziin-a.n l-zaa.na xamsa saaraat-in

the-man-NOM sitting-Ace sad-Ace

now

five

hours

These contrasts suggest that there is a genuine TENSE ambiguity of the Past/Perf form. The latter can be PASTor PRES. IfAcc-d is a quantizing adverb, then PAST is compatible with it, since it is also PFV. Simple PRES cannot combine with it because it is IPFV. As for stative PRESPERF, it behaves like PRES in that it is also IPFV (and atelic). 18

Comparable facts in Italian have been reported to me by Fabrizio Arozio. Per adverbials are only compatible with Passato Remoto (PAST PFV), but not with Pres or Imperf Ace-d is then like Per. The incompatibility of PAST and mundu-d also resembles in part the behaviour of Italian da: da can occur with Pres and Imperf, but not with Passato Remoto. Ct: also Giorgi & Pianesi (200 1) for relevant discussion. Note that the PresPerf behaviour with Ace-d depends onAktionsart. See below subsection 3.1.3.

18.

109

no Key Features and Parameters in Arabic Grammar

Temporal adverbs and kinds of Perfect

3· 3.1

Positional mundu

We have seen that mundu which takes an indefinite complement is interpreted as durational in the sense defined. Furthermore, mundu-d is incompatible with Past. On the other hand, mundu can take a definite complement. It then introduces not a length or duration, but rather a date or a time span, which is the left boundary of an extended-now (XN) interval. I will call it positiona.l mundu (following the terminological use of Musan and AvS, among others), and refers to it as mundu-t. Unlike mundu-d, mundu-t appears to be compatible with all tenses. But in analyzing the various interactions of mundu-t with tense/aspect phrases, the PresPerf appears to be specified in typical ways that make this adverbial a diagnostic test for distinguishing it from other tenses, typically PAST. These interactions also show the role played by Aktionsart and Telicity. 3.1.1

Imperfective tenses

Consider first the case of atelic imperfective tenses, in their Pres and Past versions, exemplified in the following pair of constructions: (61)

1-ajrii mundu r-raabirat-i I-run since four I am running since/from four.

(62)

kun-tu "l-ajrii mundu r-raa.bira.t-i was-I 1-run since four I was running since/from four.

In (61), mundu-t provides LB of RT, P is homogeneous, and RB is not expressed, nor presupposed. The same is true of (62), although LB is situated in the Past. As for telic imperfectives, the interpretive situation does not appear to be different. Consider e.g. the Present in ( 63):

(63)

1-aktubu r-risaalat-a

mundu r-raabirat-i

I-write the-letter-Ace since four I am writing the letter since/from four. Like in (61), the LB of RT is specified by mundu-t in (63), P is homogeneous, and RB is not expressed, nor presupposed. In sum, with imperfective temporal predicates, mundu-t specifies uniformly LB (of RT), and RB is not asserted, nor presupposed. Let us then call it Ipfv-mundu-t, and specify its meaning as follows:

Chapter 4. Arabic Perfect and temporal adverbs

(64)

Ipfv-mundu-t

II Ipfv-mundu-t II (t) homogeneous. 19 3.1.2

= A.PA.t'.:Jt"[XN (t",t') & LB (t,t") & P (t'')], Pis

Perfective tenses

Consider now perfectives. Let us start with simple Past telics, the achievement (65), or the accomplishment (66): (65)

waja.d-tu l-f:aall-a.

l-baari"/:la.t-a mundu r-raabirat-i

found-! the-solution-Ace yesterday since I found the solution yesterday since four. (66)

katab-tu r-risaalat-a

four

l-baarif:tat-a mundu r-raabirat-i

wrote-! the-letter-Ace yesterday since I wrote the letter yesterday since four.

four-GEN

In both (65) and (66), mundu-t is interpreted as specifying the RB of the Asp phrase, the point at which the event E culminates (in a telic perfective situation). The solution is found by four, and the letter terminated by four. LB is not expressed, but it is presupposed, and "pushed" backwards into the Past, so that the preparatory phase (including LB) takes place before four, although it has to be located within yesterday. The interpretive situation is not different (in relevant respects) for simple Past atelics, exemplified in (67): (67)

zakal-tu l-baarif:aat-a mundu r-raabira.t-i ate-I yesterday since I ate yesterday since tour.

four

In ( 67), mundu-t modifies also the RB of the Aspect phrase, Pis non -homogeneous, and LB is not expressed, although it is presupposed. It has to be some time in the Past, included in yesterday. Even in (67), the eating has to take place at an interval which ends up at four, and the eating has been taking place before four. Complex perfectives behave in the same way as simplex ones. For example, the complex form (68) behaves in the same way as its simplex counterpart (66) in the relevant respects, namely the fact that the interpretation involves also RB specification: (68)

kun-tu qad

ka.tab-tu r-risaalat-a.

l-baari/:aat-a mzmdu 1'-ra.abirat-i

was-I already wrote-! the-letter-Ace yesterday I had already written the letter yesterday since four.

since

tour

P is homogeneous if it has the subinterval property, i.e. for anyt and t', if t' is a subinterval oft and Pis true oft, then Pis true oft' as well.

19.

111

111

Key Features and Parameters in Arabic Grammar These contrasts indicate that perfective tenses differ in their interaction with mundu-t, compared to imperfective tenses. With the former, mundu-t adverb uniformly specify the RB of the AspP, its terminating point, while with the latter they specify its LB. Call the adverbial Pfv-mundu-t, and let us specify its meaning as follows: (69)

Pfv-mundu-t II Pfv-mundu-t II (t) =A.PA.t'3t"[XN (t",t') &RB (t,t'') &P (t")], Pis non-homogeneous.

PresPerf tense Consider now the PresPerf interpretation of (66) above, and (70):

3.1.3

(70)

j