Perspectives on Arabic Linguistics XX
AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTIC SCIENCE General Editor E.F.K. KOERNER (Zentrum für Allgemeine Sprachwissenschaft, Typologie und Universalienforschung, Berlin) Series IV – CURRENT ISSUES IN LINGUISTIC THEORY Advisory Editorial Board Lyle Campbell (Salt Lake City); Sheila Embleton (Toronto) Brian D. Joseph (Columbus, Ohio); John E. Joseph (Edinburgh) Manfred Krifka (Berlin); E. Wyn Roberts (Vancouver, B.C.) Joseph C. Salmons (Madison, Wis.); Hans-Jürgen Sasse (Köln)
Volume 290
Mustafa A. Mughazy (ed.) Perspectives on Arabic Linguistics XX. Papers from the twentieth annual symposium on Arabic linguistics, Kalamazoo, Michigan, March 2006.
Perspectives on Arabic Linguistics XX Papers from the twentieth annual symposium on Arabic linguistics, Kalamazoo, Michigan, March 2006
Edited by
Mustafa A. Mughazy Western Michigan University
JOHN BENJAMINS PUBLISHING COMPANY AMSTERDAM/PHILADELPHIA
4-
The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences — Permanence of Paper for Printed Library Materials, ANSI Z39.48-1984.
Perspectives on Arabic linguistics XX : Papers from the twentieth annual symposium on Arabic linguistics, Kalamazoo, Michigan, March 2006 / Mustafa A. Mughazy. (Amsterdam studies in the theory and history of linguistic science. Series IV, Current issues in linguistic theory, ISSN 0304-0763 ; v. 290) Includes bibliographical references and index. ISBN 978 90 272 4805 3 (Hb; alk. paper) © 2007 – 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
CONTENTS
Editorial Note
vii
Introduction Mustafa Mughazy
ix
I. Phonology and Morphology Stretching Ablaut: Morphological adaptation of new *CCu and *CCi stems in Moroccan Arabic Jeffrey Heath
3
Hypocoristics Revisited: Challenging the primacy of the consonantal root Samira Farwaneh
25
Morphology and Thematic Arity Operations: Evidence from Standard Arabic Lior Laks
51
II. Syntax On the Individual-Property Contrast in Free State Possessive Nominals in Egyptian Arabic Usama Soltan
71
Reconstruction and Islandhood in Jordanian Arabic Nouman Malkawi and Nicolas Guilliot
87
An Arabic Wackernagel Clitic? The morphosyntax of negation in Palestinian Arabic Frederick Hoyt
105
CONTENTS
III. Psycholinguistics The Split-INFL Hypothesis: Findings from English and Japanese L2 learners of Arabic Mohammad T. Alhawary
135
Lexical Processing in Two Language Varieties: An even-related brain potential study of Arabic native speakers Reem Khamis-Dakwar and Karen Froud
153
IV. Pragmatics A Cognitive Approach to Analyzing Demonstratives in Tunisian Arabic Amel Khalfaoui
169
A Prosodic Feature That Invites Back-Channels in Egyptian Arabic Nigel Ward and Yaffa Al Bayyari
187
V. Computational Linguistics Implementing an Open Source Arabic Resource Grammar in GF Ali Dada and Aarne Ranta
209
A Link Grammar Parser for Arabic Warren Casbeer, Jon Dehdari and Deryle Lonsdale
233
Subject Index
245
EDITORIAL NOTE The papers included in this volume were presented at the Twentieth Annual Symposium on Arabic Linguistics, held on March 3-5 at Western Michigan University in Kalamazoo, Michigan. The symposium was sponsored by the Arabic Linguistics Society, Western Michigan University, and the Center for Language Education and Research (CLEAR) at Michigan State University. Mustafa Mughazy was the local organizer of the symposium. The twenty-eight papers presented at the symposium were selected on the basis of anonymous reviews of the abstracts submitted to the Symposium Committee. The twelve papers in this volume were selected through another process of peer reviewing and editing. The transcription of the non-English data in this volume follows the International Phonetic Alphabet or standard equivalents. Arabic emphatics are transcribed with a dot underneath the symbol, and long vowels are represented as a sequence of two vowels. Some contributors use special symbols that are theoretically or computationally motivated; these are explained and justified in the text. This volume and Twentieth Annual Symposium on Arabic Linguistics would not have been possible without the help of many people. I would like to thank Elabbas Benmamoun, Mushira Eid, and Dilworth Parkinson for their guidance and efforts. Special thanks to Cynthia Running-Johnson for her continuous support and invaluable assistance. Tammy Betz and her assistants, Marlon Boyd and Victor Demetry, coordinated many of the symposium activities and events. Marwa Elzouka provided assistance in formatting the manuscript. I am indebted to the anonymous reviewers for their insightful comments and suggestions. The preparation and printing of the final manuscript was done using the facilities of Western Michigan University.
INTRODUCTION
Mustafa Mughazy
The papers in this volume address a broad range of theoretical issues pertaining to Arabic, particularly in the areas of phonology, morphology, syntax, pragmatics, psycholinguistics, and computational linguistics. These contributions represent the emerging trend of interface analyses, where linguistic phenomena are investigated using the techniques, methodologies, and theoretical frameworks of different linguistic research areas. The paper by Jeffery Heath on ablaut in Moroccan Arabic presents an interesting puzzle for the established views on Arabic morpho-phonology. He identifies six nouns and adjectives of the forms CCi and CCu involving a stem template that does not exist elsewhere in Muslim or Jewish Moroccan dialects. The final high vowel in these forms is quite problematic for the ablaut processes that derive diminutives and plurals. This paper surveys the diminutives and plurals of these forms in a wide variety of Moroccan dialects and proposes theoretical models of the historical pattern of upgrading from triliteral to (pseudo-)quadriliteral stems for ablaut purposes. Samira Farwaneh addresses the long standing issue of the theoretical nature of Semitic roots and templates. She develops a purely word-based surface-to-surface analysis of hypocoristics in Spoken Arabic within the frameworks of Optimality Theory and Correspondence Theory. This analysis successfully accounts for the grammatical forms and rules out ungrammatical ones without making any reference to roots or templates. In fact, templates are treated as well-formedness constraints that unify hypocoristics with intensive nouns and verbs, with the potential of extending to other forms as well. The paper on thematic arity operations by Lior Laks investigates the interaction between morpho-phonolgy and the theta system. Laks examines five Standard Arabic derivational processes that generate verbal predicates, such as passives, causatives, and reciprocals in order to motivate a distinction between syntactic and lexical derivations. He demonstrates that lexical operations, such as causativization, are bi-directional with less than full predictability, as they involve a wide range of complex morpho-phonological processes. On the other hand, syntactic operations, such as passivization, are unidirectional, predictable, and often restricted to simple morpho-phonological processes. This analysis argues against root extraction views by establishing an interface between morphology and the syntax as well as the lexicon.
x
MUSTAFA MUGHAZY
Usama Soltan approaches the diverse syntactic properties of free state possessives in Egyptian Arabic from a semantic perspective. His paper establishes a semantic distinction between individual-denoting and propertydenoting free state possessives. This distinction is coupled with syntactic arguments that individual-denoting possessives occupy a higher position (adjunct to DP) than property-denoting possessives (adjunct to NP). The result is a straightforward analysis of the distribution patterns of the two types of possessives. Soltan’s analysis has significant theoretical implications, as it suggests that cross-linguistic variations in possessive constructions can be reduced parametric settings. Moreover, it supports a theory of grammar with transparent mapping between syntax and semantics. Nouman Malkawi and Nicolas Guilliot present new types of data from Jordanian Arabic that motivate reconsidering the standard assumptions regarding reconstruction and islandhood. They propose an alternative finegrained account based on two main claims: (a) the reconstruction effects of an XP indicate the existence of a copy of this XP, and (b) reconstruction with weak resumption (clitics and double clitics) follows from the NP deletion of pronouns. Reconstruction with strong resumption (strong pronouns and epithets), on the other hand, is triggered by A’ movement with two possible outcomes: apparent resumption, if movement is available, and true resumption, if movement is disallowed. This analysis has far reaching theoretical implications, as it involves redefining the basic concepts of reconstruction and islandhood. The paper on negation in Palestinian Arabic by Fredrick Hoyt introduces new directions for research on linguistic phenomena traditionally assumed to be purely syntactic. He argues that the distribution patterns of the negation morphemes maa- and -R are mainly determined by prosodic rather than syntactic constraints. This analysis is based on the claim that these morphemes are special clitics, as they are unselective affixes that attach to words already hosting other clitics and whose distribution is influenced by non-syntactic factors. Hoyt proposes a view where the domain of negation is defined in terms of phonological or prosodic phrases rather than syntactic constituents. Mohammad Alhawary uses L2 Arabic production data from L1 English and Japanese speakers to test the psycholinguistic reality of the Split-INFL Hypothesis. Two sets of elicitation tasks are used to collect tense and subjectverb agreement data from both groups. The data is statistically analyzed to investigate whether the agreement features associated with these forms develop at the same stage, i.e., as a single projection, and whether L1 transfer affects interlanguage systems. The results indicate that the two groups follow different developments paths, as the L1 Japanese group acquires tense and verbal agreement at the same stage suggesting a single maximal projection, while the L1 English group follows a different pattern. The fact that the Japanese
INTRODUCTION
xi
participants acquire tense and subject-verb agreement at the same stage suggests that there is no L1 transfer effect, since their L1 exhibits tense, but not subject-verb agreement. The paper by Reem Khamis-Dakwar and Karen Froud is a breakthrough in the debate over the relation between Modern Standard Arabic and colloquial varieties, an issue that has dominated Arabic Linguistics for over half a century. They use the event-related potential method to measure neural responses to controlled stimuli involving code-switching between MSA and Palestinian Arabic. The results indicate that educated Palestinians process switching between MSA and Palestinian Arabic in a fashion similar to that of switching between different language pairs such as English and Spanish. These results confirm the hypothesis that Palestinian Arabic is processed as a first language, whereas MSA is processed as a second language. Amel Khalfaoui challenges the traditional assumption that demonstratives are pointing words that are interpreted according to the spatial or temporal distance of their referents from the speech context. She uses data from Tunisian Arabic to argue that the felicity of demonstratives depends on the cognitive statuses conventionally associated with these forms. The analysis, which is based on native speakers’ responses to a questionnaire, leads to a pairing of demonstratives and the cognitive statuses that are necessary and sufficient for their felicitous use. She extends this analysis to phrases with “double demonstratives” such as ha-l-ktab haCa (literally: this-the-book this) to show that there are no differences between their felicity conditions and those of their single demonstrative counterparts. Nigel Ward and Yaffa Al Bayyari examine the prosodic contours that prompt back-channeling, a discourse pragmatic phenomenon that has been largely understudied in Arabic. Their analysis, which is based on an integrated qualitative and quantitative approach, yields two main acoustic cues for backchanneling in Egyptian Arabic: a phrase-final pitch upturn and a sharp pitch down-slope. Ward and Al Bayyari use their findings to develop a set of conditions that reliably predict the appearance of back-channeling behavior. The open source resource grammar described by Ali Dada and Aarne Ranta involves the development of a library of constructs and rules for Arabic using Grammatical Framework. The system implements two types of rules: abstract syntax, which is a formal language-independent set of rules or algorithms, and concrete syntax including the grammar of Arabic. An application programming interface can use this resource grammar to generate grammatical Arabic sentences as well as sample lexical and morphological paradigms. Warren Casbeer, Jon Dehdari, and Deryle Lonsdale tackle the challenging task of developing a wide-coverage cost-effective Link Gramar parser for Arabic. One of the main advantages of this system is that it can handle lexical
xii
MUSTAFA MUGHAZY
and syntactic ambiguities, as it provides multiple parses that are ranked according to a cost vector. It can even guess the category of words that are not included in the lexicon based on their syntactic environments. This system can be used as a grammar checker or as a tool for corpus linguistics.
I
PHONOLOGY AND MORPHOLOGY
STRETCHING ABLAUT MORPHOLOGICAL ADAPTATION OF NEW *CCu AND *CCi STEMS IN MOROCCAN ARABIC
JEFFREY HEATH University of Michigan Introduction1 As the result of regular sound changes, six noun and adjective stems of the shapes CCu and CCi made their appearance at some point in Moroccan Arabic. These shapes were quite unlike any existing stem shapes. Their final high vowels posed a problem for the morpho-phonology, specifically for the ablaut processes that produce plurals and diminutives from noun and adjective stems. However, plurals had to be formed, and with senses like ‘pup’, ‘goat kid’, and ‘sweet’ it was also necessary to provide for diminutives. This paper is about how plurals and diminutives were constituted in a wide range of Moroccan dialects, both Jewish and Muslim. The data are chiefly from fieldwork carried out between 1980 and 1986, in Morocco and in communities of ex-Moroccan Jews in Israel. The larger project culminated in a comprehensive dialectology (Heath 2002), but that volume omitted coverage of CCu/CCi stems and a handful of other topics that require separate, articlelength treatment. The six key CCu/CCi stems were inherited from Classical Arabic (CA). It would be technically more correct to speak of proto-dialectal Arabic, but CA is a suitable proxy for this. The stems, in Sg form, are given in (1). 1.
(1) a. nouns
b. adjectives
1
Gloss ‘pup’ ‘bucket’ ‘goat kid’
CA (Sg) *jarw-, *jirw*dalw*jady-
MA (Sg) žru, žru dlu ždi
‘sweet’ ‘clean’ ‘fresh’
*ħilw*naqiyy*tari¿-
ħlu nqi tri
Fieldwork (around 1982) on Jewish dialects was supported by the National Science Foundation (BNS 82-19685, “Judeo-Arabic dialects of Morocco”). The primary dialectological survey of Muslim dialects was supported by a grant from the Fulbright Foundation in 1986.
4
JEFFREY HEATH
2.
The historical sound shifts Two of the CA stems, *naqiyy- and *tari¿-, already had second syllables with *i as nucleus. The CA glottal stop *¿ was lost across the boards in MA, so the syllabic structure of *tari¿- may have fallen together at an early stage with that of *naqiyy-. In the four remaining cases, the CA stem was of the shape *Cv·Cw- or *Cv·Cy-. The notation *v· denotes any short vowel from the set *{u, a, i}. In CA, such stems were routinely followed by vocalic case suffixes (e.g., nominative). These vocalic suffixes disappeared without trace in MA. As a result, the final *w and *y were in a position favoring syllabification, regardless of whether the immediately preceding C was a sonorant or obstruent. As a result of these processes, each of the six stems in question took the bi-syllabic shape *
with an initial short open syllable. In MA, a short vowel in this position regularly syncopated without a trace, resulting in *C1C2u and *C1C2i.
3.
Non-syncopating North African dialects The scenario described above is valid for most but not all dialects spoken within the traditional boundaries of Morocco. It is possible to distinguish three primary dialectal strands, reflecting the complex settlement history of the country. (2) a. “Northern” type: northern Muslim dialects--Tangiers, Tetuan, and Chaouen; archaic Muslim dialects of the southern fringe of the Rif mountains; archaic Muslim urban dialects (Rabat, Fes, Sefrou, Taza); all Jewish dialects. b. “Saharan” type: bedouin tribes of the far southern oases (Tata, Guelmine, M’hamid) and of the hinterlands of Rabat; very close to Hassaniya Arabic of Mauritania, Mali, and the Western Sahara. c. “Eastern-central” type: dialects of the east (Oujda area), rural dialects of the central plains, and the urban dialects of Meknes and Marrakesh. The northern type represents the oldest stratum, having taken shape in the Roman garrison towns (Volubilis, Tangiers, and perhaps Salé) after the Arab conquest. The Saharan type represents the massive influx of Arabian bedouin into the Maghreb beginning in the 11th Century. The eastern-central type is something of a hybrid of the two, with both bedouin and urban features but also a heavy Berber substratum, probably having taken shape initially in western Algeria.
5
STRETCHING ABLAUT
Saharan dialects have fewer cases of CCu/CCi than do the others, due to lexical erosion (adjectives ‘clean’ and ‘fresh’) and also because syncope is less thoroughgoing (da·lw ‘bucket’, and žərw or ža·rw ‘pup’). In these dialects, the only case of CCu is the adjective ħlu ‘sweet’, and the only CCi stem is the noun ždi ‘goat kid’. In North Africa as a whole, syncope in the relevant set of stems becomes more systematic going roughly from east to west. The examples in (3) are based on occasional data I have collected from informants, plus (for Algeria) some colonial literature2. (3)
‘sweet’
‘bucket’
‘pup’ ‘goat kid’
ħĭlw
da·lw
žərw
—
b. optional syncope for ‘sweet’ Libya ħəlw, ħluuw Tunisia (Gabès) ħəlw, ħluuw
da·lw da·lw
žərw žərw
ža·dy ža·dy
c. systematic syncope for ‘sweet’ Tunisia (Tunis) ħluuw Algeria (Ulad Brahim) ħluuw
da·lw da·lw
žərw žərw
ža·dy ža·dy
d. syncope for ‘goat kid’ Algeria (Oran) — Algeria (Algiers-J) ħluuw
— —
žərw —
ždi g;diiy
e. syncope in all forms Algeria (Tlemcen)
“dlû”
jrû
jdî
a. no syncope Sudan (Khartoum)
—
Reading (3) top down, i.e., as the dialects go westward from Sudan and Libya to Algeria, we see that ‘sweet’ is the first to syncopate, then ‘goat kid’, though there are gaps in the dialectology (due in part to lexical erosion). We must be cautious with respect to the ‘goat kid’ data, since some cases of ždi may really involve /ždiyy/ or /ždəyy/. This is suggested by Algiers-J (Jewish Algiers) g;diiy, with possessed forms like g;diiy-ək ‘your …’ rather than #g;dii-k. Here we are probably dealing with an original Diminutive of *ja·dy, rather than with a syncopated reflex of the non-diminutive simplex. At any rate, by the time we get to coastal western Algeria (i.e., Tlemcen), syncope is systematic in all stems.
2
Ulad Brahim: Marçais (1908). Algiers-J[ewish]: M. Cohen (1912). Tlemcen: Marçais (1902).
6
JEFFREY HEATH
4.
Adaptation strategies The singular stems themselves posed no great phonological problem. The shape CCi, though new for nouns and adjectives, was familiar from imperfective verbs, directly inherited from the CA simple “weak” imperfective *-(a)CCii or measure IV causative imperfective *-u-CCii, as in MA bki ‘weep’ and ški ‘complain’, respectively. The shape CCu was not as well-established. Many MA dialects have a single imperfective verb stem of this shape, namely ħbu ‘crawl, walk on all fours’. However, Saharan dialects (and some others influenced by them) have imperfective ħba or ħba, and even the ħbu dialects generally have suffixally inflected perfective forms like ħbi-t ‘I crawled’ with i instead of u. For a full MA dialectology of this stem, see Heath (2000). The problem was, therefore, not with the singular shapes as such (dlu, žru, ždi, ħlu, nqi, tri). Rather, the difficulty was how to produce plurals and diminutives from them, given that there was no pre-existing system for feeding such inputs into ablaut derivation. In any such situation, three basic strategies are available, a priori: (4) a. preserve the inherited derived forms (subject only to regular sound shifts), even though these forms may be archaic-looking; b. replace inherited forms with entirely new derivatives, based on synchronic cutting-edge ablaut models; c. compromise between (a) and (b) by applying “updates” to make inherited derivatives look reasonably up-to-date. 5.
Plurals of CCu and CCi adjectives We can dispense with the plurals of the adjectives ħlu ‘sweet’, nqi ‘clean’, and tri ‘fresh’ quickly. These stems take suffixal rather than ablaut plurals in MA: ħluw-in, nqiy-in, triy-in. The productive adjectival plurals are CCaC (for singulars of the CCiC adjective class, plus one instance of CCuC, namely sxun ‘hot’, Pl. sxan or swxan), and CuCC (for “color-defect” adjectives, Sg. shape CCəC, plus µma ‘blind’). None of the CCu and CCi adjectives is of the “color-defect” type semantically or morphologically. They are arguably special cases of the regular CCiC (or CCuC) type; if so, we might expect a plural #CCa, i.e., CCaC minus the final C. However, the shape CCa is still in use as an elative (ħla ‘sweeter’) in some dialects, and since this directly reflects a CA weak elative type it is reasonable to think that CCa was not available for adjectival plurals.
7
STRETCHING ABLAUT
6.
Inherited plurals of CCu and CCi nouns Things are quite different for the nouns. The original CA plurals are shown in the second column in (5). The third column shows the expected MA reflexes, some of which are unattested (#). (5) CA Sg. *jady*dalw-
*jarw-/*jirw-
CA Pl. *jidaa¿*jidy-aan*¿adlii*dilaa¿*¿adlaa¿(*¿adliy-at-) *¿ajrii- #(ə)žRi *jiraa¿*¿ajraa¿*¿ajriy-at-
Expected MA Pl. #žda ždy-an (ə)dli dla dla or (ə)dla #(ə)dlya
Gloss ‘goat kid’ ‘bucket’
‘pup’ žRa (“R” = r or r) žRa or (ə)žRa #(ə)žRya
For ‘goat kid’, the only MA plural in common use is ždy-an. This type was not originally in use for ‘bucket’ or ‘pup’, for which we find a plural type CCa, in one case (‘bucket’) competing with CCi. The type CCa is actually composite etymologically, and in Saharan-influenced dialects it may still be possible to distinguish two forms (one with initial short vowel) on the basis of the phonology of the definite prefix, e.g., d-dla versus l-ədla. The original stem-initial vowel is lost in most dialects, but where still alive it prevents definite l- from assimilating to a stem-initial coronal consonant, and sometimes a belated syncope still leaves an unassimilated lateral (l-dla). The plural variant *¿ajriy-at- has left behind no direct reflex (early MA *žrya would likely have been reinterpreted as a feminine singular). There may have been a parallel variant plural *¿adliy-at- for ‘bucket’. 7.
Productive MA nominal plural ablaut Such plurals as CCa and CCi are isolated archaisms. Current productive MA nominal ablaut patterns for short (“triliteral”) singular stems are in (6), though there are numerous lexicalized plurals that do not fit any such pattern. (6) Plural pattern(s) a. CCuCa, CCaC b. CCaC c. CiC-an d. CCaCi e. CCaCa
Corresponding singular pattern CəCC, CCəC CuC, CiC CaC CCC-a (feminine) CCCi, CVCi
8
JEFFREY HEATH
f. CCaC-atCəCC-t- etc. (female kin terms) Examples: kəlb ‘dog’, Pl. kluba or klab; bir ‘well’, Pl. byar; far ‘mouse’, Pl. fir-an; klw-a ‘kidney’, Pl. klawi; fas-i ‘Fessi (resident of Fes)’, Pl. fwasa, µmm-t-i ‘my paternal aunt’, Pl. µmam-at-i. There are some CCa nouns in MA that potentially could have provided models for the ablaut pluralization of the new CCu and CCi stems. However, in practice they are not helpful. Many CCa nouns have a suffixal plural: mraw-at ‘women’ (alongside suppletive plurals), dwa-y-at ‘medications’, sla-w-at ‘prayers’, qfa-w-at or qfa-y-at ‘napes’, bra-w-at ‘letters’. CCa masculine verbal nouns like kra ‘rental’ are only awkwardly pluralizable. Most MA dialects preserve inherited plurals of the shape CCi only for µsa ‘stick, club’ and rħa (or rħa) ‘grinding mill’, hence µsi and rħi (rħi). This CCi plural is isolated in the overall context of MA ablaut plurals, and even for these stems it has been supplanted in several other dialects, especially by the plural type rħawi (rħawi), µsawi. 8.
MA ablaut plurals of nouns (ždi, dlu, žru) For ždi ‘goat kid’, in nearly all dialects the plural is the inherited ždy-an or a predictable variant (žədy-an, Jewish zdy-an, Twn-M jəðy-an). An alternative ždawi, cf. (6d), was recorded only once each for M and J dialects. The real problem was therefore the two new CCu nouns. They fit awkwardly into the patterns in (6), though each of the latter was utilized in one dialect or another. For ‘bucket’ and ‘pup’ the MA plurals that I recorded are organized into the sets in (7), excluding purely suffixal plurals dlu-y-at/ dlu-wat and žru-w-at. The forms within a set like (7a) have a family resemblance to each other, but the forms on different rows within a set have differentiating features. (7) ‘bucket’ ‘pup’ (a) a·dla dla žRa, jra dli (inherited, now more common for ‘pup’ than for ‘bucket’; dla was recorded once each in El Jadida-M and the oasis town Tata-M, and a·dla with initial short vowel in Guelmine-M; dli was recorded twice in the oases (Guelmine-M, Tata-M) and once for Fes-M; žRa is dominant in the northern M dialects and along the Atlantic coast from Casablanca-M to Safi-M, less often in the oases; one questionable attestation of dli for Sefrou-J may be a Hebraism) b) (ə)dlaw dlaw
(a·)žRaw žRaw
STRETCHING ABLAUT
(mutates CCa of (7a) into the MA CCaC plural type, which conflates Classical *CiCaaC- and *¿aCCaaC-, see (6a-b); this set is common in the southeastern (Tafilalt) M dialects, and attested in parts of the M urban belt; a·žraw and ədlaw with initial short vowels were recorded in the oasis town Tata-M; no J-dialect attestations of this set) c) dlawa žRawa dlawi žRawi (further development of (7b) adding a final vowel, modeled on plurals of CCC-a and CCCi stems; see (6d-e); M distribution overlaps with that of set (7b) but attestations are less dense, with dlawa in the oasis town Tata-M, dlawi attested but sparse in the north (Tetuan-M, Chaouen-M), in coastal Azemmour-M, and in the oasis town M’hamid-M, and both dlawa and dlawi in Oujda-M; dlawi is common and widespread in J dialects, most strongly in the south, giving way to dlawa in the north- and southeast (Oujda-J to Tafilalt); žRawa is attested in northeastern J dialects and in Beni Mellal-J and Tiznit-J, while žRawi was recorded in southwestern Taroudant-J and El Kelâa-J) d) dlw-an dly-an
žRw-an žRy-an i-žrw-an (analogy from ždy-an to CCu stems, as in some other dialects; in the case of ‘bucket’ perhaps also fed by an inherited dialectal dəlya not in my data but reported by Premare (1993); the shape CCw/y-an also occurs elsewhere in the morphology in the form of CCw/y-an verbal nouns; the Pl. suffix -an is also supported by Berber nominal plurals in -an or -ən ; for ‘pup’ a similar plural is reported for YemenDathina: jiryaan/jirwaan (Landberg 1905-13:1706, 1920-42:280); attestations: dlw-an is regular in Marrakesh-M and fairly common in the oases with a few attestations farther north; dlw-an also occurs in some eastern J dialects and is recorded for Fes-J and Essaouira-J; dlyan is uncommon but attested here and there in the same general area; žRw-an has the same basic distribution in M dialects as dlw-an but is additionally regular from Fes-M and Sefrou-M to Taza-M and in the Rifi villages; žrw-an also in Debdou-J (east); Berberized i-zrw-an for Toulal-J in the southeast; žry-an occurs sporadically in the M oasis dialects and was attested once in Souk Larba-M)
e) dlula dluwa
žruwa
9
10
JEFFREY HEATH
(sporadic recourse to CCuCa, the productive MA plural for strong triliterals, but rare and awkward when C3 = w, note the doubling of input C2 in the variant dlula; distribution: dlula has three scattered M attestations; dluwa once each in M’hamid-M and Azemmour-M; žruwa once each in Azemmour-M and Oujda-M) f) dlaw-at (-aS) žraw-in žriw-an (has medial a due to ablaut, as in (7b-c), but adds a Plutal suffix; CCaw-at may have been a mutation from CCawa (7c), perhaps influenced by CCaC-at feminine kin-term plurals such as µmam-at‘paternal aunts’ from Sg. µmm-t- ; dlaw-at is moderately common in the M north and in Meknes-M and is attested in Oujda-M and Safi-M; it is usual in Rabat-J and attested in Meknes-J, Casablanca-J, Beni Sbih-J, and Debdou-J; žraw-in, recorded once in Oujda-M, reflects a minor ablaut-suffixal pattern CCaC-in seen also in e.g., Pl. dyaf-in ‘guests’ from dif, while Berber-influenced žriw-an was recorded once for Sefrou-M) g) dlw-at — (rare pseudo-F.Pl., attested Aoulouz-J, probable mutation from *dlwa or *dlw-an) h) dlaym — (rare, Hebraized, attested once Oujda-J) One can imagine stepwise historical sequences like those in (8), each mutation involving a slight phonological increment or substitution. However, the constant potential for dialect mixing and the ready availability of analogical models for the fuller plurals would have permitted “jumping” over intermediate stages. (8) a. CCa > (7a) b. žra > (7a) c. dlw-an > (7d) d. žrw-an > (7d)
CCaw > (7b) žraw > (7b) dlw-at (7g) i-žrw-an (7d)
CCawa > (7c) žrawi > (7c)
CCaw-at (7f) žraw-in (7f)
STRETCHING ABLAUT
9.
11
Productive MA nominal and adjectival diminutive ablaut The diminutives of nouns and adjectives are based on slightly distinct ablaut templates that can easily be confused. The basic nominal diminutive pattern is CCiCX*, with a rigid CCiC onset (dialectally CwCiC) followed by a tail X* of variable shape, but obligatorily non-null and extendible to more than one segment. For long input noun stems, the first few segments of the input fill the consonantal positions of CCiC. Then what is left of the input stem (i.e., its rightmost segment or segments) is transferred onto X*, with some modifications, notably to an input non-final full (“long”) vowel that immediately follows the cut-off point. In northern-type dialects including the J dialects (2a), this vowel is reduced to schwa in X*, and in some dialects can be syncopated. In Saharan dialects (2b), and (probably under their influence) most eastern-central type dialects (2c), the input full vowel remains full but shifts to i. For example, brrad ‘tea kettle’ has these diminutive variants: “northern” brir(ə)d, elsewhere bririd. Except for the first vowel, these diminutives resemble non-diminutive quadriliteral Pl. brar(ə)d and brarid, respectively, but the dialectal distribution of Pl. brarid is much narrower than that of bririd, undoubtedly because of the sound-symbolic value of repeating the i-vowel in the diminutive (cf. teensy-weensy, etc.). When the input to nominal diminutive CCiCX* is too short to provide segments to fill both output C3 and X*, the final input segment normally transfers to X*, leaving output C3 to be filled by a non-lexical filler. For noun stems, this is w or y, sometimes geminated to ww or yy. For example, kəlb ‘dog’ has diminutives like kwliyyəb. In early MA, the diminutive of adjectives may well have been very similar to that of nouns, but with a smaller range of input shapes. The basic adjectival shapes are CCiC for adjectives of quality and state (we may add sxun ‘hot’ with u), and CCəC (Saharan a·CCa·C) for adjectives of color and defect. In both cases, the dominant MA diminutive is CCxiCx(ə)C, which could be regarded as a special case of the nominal diminutive CCiCX* (the next section will demonstrate that this is no longer synchronically correct). However, adjectival diminutives are often of the output shape CCxiCx(ə)C with input C2 doubled (appearing as output C2 and C3). Specifically, C2-doubling is standard in colordefect adjectives and occurs at least dialectally with many quality-state adjectives: kħəl ‘black’, dimin. kwħiħ(ə)l, and kbir ‘big’, dimin. kwbib(ə)r. We will see just below that C2-doubling is a feature of diminutives of CCu and CCi stems, not only for adjectives (where we expect it) but also for nouns. This most likely reflects interaction between CCu/CCi nouns and CCu/CCi adjectives, since there are almost no cases of C2-doubling in other nouns (the only clear case is northern dialectal mwim-a ‘a little bit of water’ from ma ‘water’, where the input has only one consonant; in some dialects, mmwim-t‘mother-dimin.’ from mm- may be best analyzed as another instance).
12
JEFFREY HEATH
The dimin. C3 is sometimes secondarily geminated (=lengthened). This jumps out in dialectal forms like kwbibb(ə)r ‘big-dimin.’. It is probably much more widespread, though harder to hear, when dimin. C3 is y, as in sFiy(y)(ə)r ‘small-dimin.’ from sFir (cf. the more easily audible gemination in Hassaniya sFa·yya·r). In the purer Saharan dialects, an entirely different diminutive (m.sg. CayCəC or variant) is used with color-defect adjectives. Diminutivization ranges from extremely productive (Saharan) to quite productive (central-eastern) to fairly productive (northern dialects). However, the inherited system gave few clues as to how to diminutivize the new CCu and CCi stems. The major morpho-phonological issues were: (a) whether the final input high vowel was mapped onto output C3, output X*, or not at all; and (b), since C3 and X* must be non-null, how they are filled. Speakers of early MA would have sought guidance for how to diminutivize CCu and CCi by looking at what happened with CCa stems. The only CCV nouns subject to diminutivization, prior to the emergence of CCu and CCi stems, were of the shape CCa. However, the fact that some of them were always feminine, others variably feminine or masculine (depending on dialect), and still others always masculine, made the data difficult to interpret. Gender is not an issue in ablaut plurals, but it is very relevant to diminutives, which often overtly express f.sg. suffix -a even when the input noun is “covertly” feminine (revealed only by agreement and concord): ħanut ‘shop’ (covertly feminine: ħanut kbir-a ‘a big shop’), dimin. ħwinit-a (with overt f.sg. suffix). For mra ‘woman’ we get diminutives like mriw-a and mri(y)y-a (sometimes differentiated semantically, with one being used as an insult directed at men). For µsa ‘stick, club’, covertly feminine in most dialects, the usual diminutive is µsiw-a or µsi(y)y-a. None of the consistently masculine CCa nouns (such as verbal nouns) seems to have a high-frequency diminutive. The new CCi and CCu nouns were therefore on their own. Likewise, the only CCa adjective in MA is µma ‘blind’. I found it difficult to elicit a diminutive for this stem, partly because of the respect accorded to the blind in Moroccan culture. µma is often supplanted entirely by µwər (original sense ‘one-eyed’), and in cultivated speech by bsir (an antonymic taboo replacement originally meaning ‘clear-sighted’). So the new CCu and CCi adjectives would have gotten sparse help from µma in creating their diminutives. 10. MA diminutives of nouns (dlu, žru, ždi) The many variant diminutives that I recorded for the CCu and CCi nouns are presented in (9). For most Jewish dialects except those in the east (e.g., Oujda), “ž” is really z due to merger of palatoalveolar and alveolar sibilants.
13
STRETCHING ABLAUT
In certain rows, two sets of diminutive forms for ždi (one with w, the other with y) are displayed side by side. In such cases the w variant (e.g., ždiww) has exactly the same form as a diminutive of the CCu stems except of course for the initial cluster (ždiww, dliww, žriww). In such a dialect, the output w is not dependent on an input u, and is either a non-lexical filler or the result of phonological conflation of the two semivowels. By contrast, when y occurs (e.g., in ždiyy) in a pattern where the CCu stems have w (e.g., dliww, žriww), we have a dialect that clearly maps the lexical high vowel u or i onto the output consonant (here C3) position. In (9a-c), we observe C2-doubling, the input C2 appearing both as dimin. C2 and either dimin. C3 (9a-b) or C4 (9-c). These C2-doubling patterns are almost certainly an extension to nouns of the regular adjectival pattern (see the following section). (9)
‘bucket’ ‘puppy’ ‘goat kid 1’ ‘goat kid 2’ a) dluw(w)l dliwəl ždiwəd dliwwəl dliy(y)əl žra·yya·r ždiyəd (input C2 spreads to dimin. C2 and C4; dimin. C3 is either mapped from input V, or default; distribution: dliwl is the only widespread form, occurring in and near the main urban belt: Chaouen-M, Meknes-M, Fes-J, Beni Mellal-J, Souk Larba-M, Taounate-M, Oujda-M; for ‘puppy’, žra·yya·r once Casablanca-M; for ‘goat kid’, eastern ždiyəd attested once in Oujda-M and northern ždiwəd attested once each in Chaouen-M and Tangiers-M, but none of the speakers in question had diminutives from this set for ‘bucket’ or ‘puppy’) b) dlil — — (infrequent reduction of (9a), attested once Marrakesh-M) c) dlili ždidi dlilu žRiRu ždidu (input C2 spreads to dimin. C2 and C3; dimin. V2 is variably copied from the input V, or templatic; this set is concentrated in and near the middle of the urban belt: Fes-M, Sefrou-M, and Taza-M; dlili once Casablanca-M, versus much more common dlilu; ždidi likewise once each Fes-M and Sidi Kasem-M, versus more common ždidu) d) dliww dlew dliw
žriww žRew(w) žRiw
ždiww ždey(y) ždiw
14
JEFFREY HEATH
dliwu dliy(y)u dliyya·w
žriwu žriy(y)u
ždiyyu
ždayu (no input-C2 spreading; various treatments of ending, including nonlexical semivowels and mappings from the input; the forms ending in w(w) and y(y) are common in the southern oases, along the Atlantic coastal towns south of Casablanca, and around Rabat-M; dliwu, dliyu, etc. are phonetically slight adaptations; Saharan-looking dliyya;w once Zagora-M; for ‘goat kid’ ždiw(w) was recorded in El Jadida/ Azemmour-M, Souk Larba-M, and Tafilalt-M, while the other oasis dialects-M have ždew and ždey(y); ždayu once Safi-M) e) dliwi žRiwi ždiwi ždiyi (related to (9d), vocalism templatic, dimin. C3 either a non-lexical semivowel, or mapped; this set is densely attested in northern dialects (Tangiers-M, Tetuan-M, Chaouen-M, Ouazzane-J) and for Oujda-J, dliwi also Meknes-J, žRiwi and ždiwi also sporadically elsewhere; ždiyi once each Chaouen-M and Oujda-M; an ablaut diminutive plural ždawa was recorded for Oujda-J) f) dliw-a žRiw-a ždiw-a dluy-a (feminine in form; CCiw-a is the regular dimin. of f.sg. input CCw-a or CCy-a ; dliw-a is very common in J dialects of the Atlantic coast (Casablanca south) and the entire south, and is also attested Tangiers-M, Ouarzazate-M, Zagora-M, Rissani-M; žRiw-a has a fair number of attestations, but it is primarily the dimin. of žRw-a ‘bitch’; ždiw-a (Meknes-M, Azemmour-M), variant jdiw-a (Tangiers-M), and scattered J dialects with zdiw-a ; some speakers use such forms only in the f.pl. with -at. g) dliwn — — (incorporates n from Pl dlw-an, perhaps as a mutation from dliwl; attested once Marrakesh-M) h) — žwiyr — (looks like (9a) but with input C2 linked only to dimin. C4; attested once, Marrakesh-M)
15
STRETCHING ABLAUT
11. MA diminutives of adjectives (µma, ħlu, nqi, tri) Adjectival diminutives from my data, including very limited data from µma ‘blind’, are given in (10). I did not usually elicit these forms from Jewish speakers so my information is largely limited to Muslim dialects. Of the CCV adjectives, ħlu ‘sweet’ clearly plays the central role here. Its sense lends itself to (hypocoristic) diminutivization with human reference, just as in English (sweetie, cf. sweatheart, my sweets, etc.). For El Jadida-M I recorded a nominal use ħlillu ‘water drunk on the second day of the Feast of the Ram’. In some rows, I have again placed w and y variants for the diminutive of nqi side by side, where the w variant suggests a non-lexical (or phonologically merged) semivowel and the y variant points to mapping of the lexical high vowel onto an output C position. Since the limited data for tri ‘fresh’ generally track those of nqi, the former are omitted except where they flesh out a set. (10)
µma ‘blind’ ħlu ‘sweet’ nqi ‘clean 1’ nqi ‘clean 2’ (dimin. rare) (dimin. common) a) ħlilu nqiqu ħlillu ħlullu µmimi ħlili nqiqi ħlilwi nqiqqiw (input C2 spreads to dimin. C2 and C3; vocalism is partly templatic; regular suffixal forms of type f.sg. ħlilw-a, Pl. ħlilw-in; this set is generally dominant for M dialects of the north and the main urban belt, and attested in Marrakesh-M; the usual forms are ħlilu and nqiqi, with rarer attestations of ħlillu ∼ ħlullu once Marrakesh-M, ħlili once Safi-M, ħlilwi once Casablanca-M, nqiqu once in Taza-M and once in a nearby Rifi village Taounate-M (triru ‘fresh-dimin.’ has a similar distribution), nqiqqiw once Erfoud-M; µmimi was recorded once in Oujda-M) ħliw
b) µmiww
nqiw nqew
nqiy nqey
ħliww
nqeww ħliy(y)u ħliyya·w (no spreading of input C2; dimin. C3 is either mapped from the input vowel, or is a default w; this set is common in Atlantic coastal towns from El Jadida/Azemmour to Safi, and in the southern oases)
16
JEFFREY HEATH
c) µmiwi — nqiwi (no spreading of input C2; Dimin C3 default w; µmiwi attested once Rissani-M, nqiwi attested several times in the north: Tangiers-M, Tetuan-M, Chaouen-M) d) — ħlayu nqayu (occasional development of (b) with medial V dissimilating to a; both forms recorded for one Safi-M speaker) e) µmiym
ħliwl ħliwwəl
triwr
nqiyəq triyyər
(input C2 spreads to dimin. C2 and C4; dimin. C3 usually patterns as a default w or y; distribution: ħliwl is regular in Oujda-M (east) and recorded around Marrakesh-M and Ouarzazate-M, and for Fes-J and Sefrou-J, while geminated ħliwwəl was recorded once each for Meknes-M and Rabat-M; nqiyəq once Ouarzazate-M, likewise triyyər ‘fresh-dimin.’ for one Marrakesh-M speaker; triwr for one Oujda-M speaker who also gave nqiqi, ħliwl, and µmiym ∼ µmimi) f) µa·yma ħa·ylu — (Hassaniya-type “color-defect” m.sg. dimin. pattern Ca·yCə/a·C, recorded once in the oasis town Tata-M) 12. Modeling issues I: Creeping quadriliteralization A formal model of MA plural and diminutive ablaut was presented, for a mainstream koine-ized Muslim dialect (Fes-Meknes area), in (Heath 1987). Outputs are constituted by selectively mapping input stem segments onto templates that generally have vacant C positions and pre-specified V positions, and in some cases end in a variable tail. The nominal plural data given above bring out a historical trend to upgrade triliteral to (pseudo-)quadriliteral stems for ablaut purposes. Elsewhere in MA, this is most clearly seen in the treatment of (feminine) CCCa and CVC-a nouns, and of (masculine) CCCi and CVCi nouns (including -i nisba adjectives used as nouns). For such stems, the final stem vowel is now counted as a full segment, so instead of typical strong triliteral plurals we normally get a quadriliteral plural based on CCaCX*, as in qrµ-a ‘bottle’, Pl. qraµ-i and sbsi ‘(smoking) pipe’, Pl. sbasa (within X*, input i and a are normally switched). This upgrading has spread (dialectally) to CCV nouns (CCa, plus the new CCu and CCi stems), resulting in plurals of the shape CCawi (or CCawa), see (7c) and the end of Section 7 (rħawi, µsawi). (7c) gives both CCawi and
STRETCHING ABLAUT
17
CCawa plurals for CCu noun stems. The type dlawi, žrawi matches the most common quadriliteral-type plural for CCa stems (rħawi, µsflawi), and suggests that these, along with the feminine type CCC-a, were the primary analogical models. On the other hand, the type dlawa, žrawa with final a is what we would expect (given the a/i vowel-switching rule within X*) if input u were treated like its fellow high vowel i, as seen in sbsi ‘pipe’, Pl. sbasa. (11) a. d l u (≈ a) (a/i switching) CCaC X* w output: dlawi
b.
dl u (≈ i) (a/i switching) CCaC X* w output: dlawa
13. Modeling issues II: C2-doubling The diminutive data for CCu and CCi adjectives (10a, f) show extensive C2-doubling (i.e., mapping onto two output C positions, or C and X*, with at least one intervening segment). This is expected, since C2-doubling is productive in other adjectival diminutives, especially in the color-defect class. More surprising, we also observe considerable C2-doubling in diminutives of CCu and CCi nouns (9a-c). C2-doubling is substantially absent in ablaut plurals of these same nouns; Pl. dlula (7e) is a rare and isolated example. The two basic mapping strategies involving C2-doubling can be illustrated with the diminutives in (12). (12)
a.
d l u CC i C X*
output: dlilu
b.
d l u CC i C X* w output: dliwl
(12a-b) show how the same input can combine with the same template in different ways depending on how the mappings are constrained. In both cases, the requirement that X* be non-null is satisfied. In the more straightforward (12a), this involves transferring the stem-final u to X*. After dimin. C1 and C2 are filled (dl...), dimin. C3 remains vacant. It could be filled by a non-lexical semivowel y or w, and indeed dliyu and dliwu (among other variants) are attested dialectally (9d). However, in other dialects with dlilu and similar forms, input C2 is doubled on dimin. C3. In (12b), the stem-final u is not transferred to X*. Instead, the double of input C2 is transferred into X*, leaving dimin. C3 to be filled by a non-lexical semivowel.
18
JEFFREY HEATH
One can easily imagine a variation on (12b) such that the dimin. C3 w is in fact lexical, being mapped from the stem-final u. This would be the correct analysis in any dialects where we got w from stem-final u but y from stem-final i (dliwl, ždiyd). None of my informants clearly displayed such a dialect, but it would be worth looking farther. Of course such a dialect would involve “crossing” of C-to-C and V-to-C association lines, a phenomenon observable in Sierra Miwok (Smith 1985). A more extreme mapping is the rare variant žwiyr (9h) from žru. Here input C2 is transferred to X* but is not mapped onto dimin. C2 or C3, leaving both of these positions blank until non-lexical semivowels are inserted. 14. Maps The schematic “maps” in the appendix show the distribution of the key variant types for Muslim dialects (the data are too sparse for Jewish dialects to make mapping useful). The maps are of the same type used in Heath (2002) and are mainly for Moroccan specialists. The large squares are, from left to right: (a) top row: Tangiers, Tetuan; (b) middle row: Rabat, Meknes, Fes, Taza, Oujda (I frequently refer to the Rabat-Taza sequence as the “urban belt”); (c) lowest row: Marrakesh. The circle and the three triangles roughly above Taza are the Arabic dialects of the western flank and southern fringe of the Rif. The three triangles above Rabat-Fes are rural-type dialects. The three circles and one triangle in the left center are the Atlantic coast communities from Casablanca to Safi. The triangles in two rows on the bottom are the oasis dialects, including Tafilalt (Erfoud, Rissani) at the far right. Shading indicates the density of the variant in question, e.g., 2 out of 5 informants =40% is shaded less densely than a higher percentage.
STRETCHING ABLAUT
19
REFERENCES Cohen, Marcel. 1912. Le parler arabe des Juifs d’Alger. Paris: Société de Linguistique de Paris. Heath, Jeffrey. 1987. Ablaut and ambiguity: Phonology of a Moroccan Arabic dialect. Albany: SUNY Press. ----------. 2000. “Crawling toward enlightenment: The verb HBU in Moroccan Arabic”. In Naturally! Linguistic studies in honour of Wolfgang Ulrich Dressler, ed. Christiane Schaner-Wolles, John Rennison and Friedrich Neubarth, 183-193. Torino: Rosenberg & Sellier. ----------. 2002. Jewish and Muslim dialects of Moroccan Arabic. London: Curzon Press. Landberg, Carlo. 1905-13. Datina. Leiden: Brill. ----------. 1920-42. Glossaire datinois. Leiden: Brill. Marçais, William. 1902. Le dialecte arabe parlé à Tlemcen: Grammaire, textes et glossaire. Paris: Leroux. ----------. 1908. Le dialecte arabe des Ulad Brahim de Saïda. Paris: H. Champion. Premare, A. L. de. 1993. Dictionnaire arabe-français. Paris: l’Harmattan. Smith, Norval. 1985. “Spreading, reduplication and the default option in Miwok nonconcatenative morphology”. In Advances in nonlinear phonology, ed. Harry van der Hulst and Norval Smith, 363-396. Dordrecht: Foris.
20
JEFFREY HEATH
APPENDIX Maps (Muslim dialects)
Maps 1-6: Adjectival diminutives
1. ħliw(w), ħlew(w)
2. nqiw(w), nqew(w), triw(w), trew(w)
3. C3=C2 ħlilu (ħlillu, ħlili)
4. C3=C2 nqiqi, triri, etc.
5. C4=C2 ħliw(w)l, ħliw(w)əl
6. C4=C2 nqiwq, nqiyq, triwr, triyr
STRETCHING ABLAUT
Maps 7-18: Nominal diminutives 7. dliwi < dlu (excluding da·lw)
8. žRiwi < žRu (excluding ža·rw)
9. ždiwi < ždi
10. dliw, dliww¸ dliwu
11. žriw(w), žriwu < žRu (excluding ža·rw)
12. ždiw(w)
21
22
JEFFREY HEATH
(Nominal diminutives, cont.) 13. ždey, ždiyi
14. C3=C2 dlilu < dlu (excluding da·lw)
15. C3=C2 žRiRu
16. C3=C2 ždidu¸ ždidi
17. C4=C2 dliwl¸ dliyl< dlu (excluding da·lw)
18. C4=C2 ždiwd, ždiyd
STRETCHING ABLAUT
Maps 19-28: Nominal Plurals 19. dli, dla
20. žRa
21. dlaw
22. žRaw
23. dlawi, dlawa
24. žRawi¸ žRawa
23
24
JEFFREY HEATH
(Nominal Plurals, cont.) 25. dlaw-at
26. dlw-an
27. žRw-an, žRy-an
28. dlula, dluwa
HYPOCORISTICS REVISITED CHALLENGING THE PRIMACY OF THE CONSONANTAL ROOT*
SAMIRA FARWANEH The University of Arizona
1.
Introduction Research on Semitic morphology within current theoretical frameworks focuses on the lexical status of the consonantal root and the template, according Semitic languages an “exotic” status. The root/pattern dichotomy was motivated by the observed common semantic features among words or outputs sharing identical consonants, which in turn led to the hypothesis that the root consonants constitute a meaningful unit, hence, a morpheme, independent of the canonical shape of the form or its template, itself an independent morpheme. Capitalizing on the non-linear architecture of Autosegmental Phonology, McCarthy (1981) proposes a representation of the morphemic components of stems in Arabic, as in (1).
(1) Morphemic composition of the word [katab] k t b Root | | | CVCVC Template | | a a Vowel melody Proponents of a root-primary analysis derive their arguments for the primacy of the root in word formation from external data such as language games (McCarthy 1981, 1982), speakers with aphasia (Prunet et al. 2000), speech errors (Abd-El-Jawad & Abu-Salim 1987), and hypocoristics (Davis & Zawaydeh 1999a, 2001), all of which seem to indicate that different morphological processes are capable of extracting and operating on root consonants alone while bypassing vocalic and affixal segments (see examples in (Davis 2001)).
*
This work has benefited greatly from discussions with and input from the following to whom thanks are due: Stuart Davis, Bert Vaux, Adam Ussishkin, and the audiences at the University of Wisconsin, Milwaukee and the 20th Symposium on Arabic Linguistics held in Kalamazoo Michigan, March 2006. Errors are the responsibility of the author.
26
SAMIRA FARWANEH
An alternative view emerged in recent years challenging the root-oriented model (with or without templates) of Semitic morphology. On this view, the input to morphological processes is the word, itself an output, or a base (a closed syllable or a disyllabic foot), which may not have autonomous existence, with the root as an autonomous morpheme stripped of any active role in morphological derivations, thereby bringing Semitic languages more in line with concatenative languages (Bat-El 1994; Ratcliffe 1997, 1998; Benmamoun 1999; Ussishkin 1999, 2005). The impetus behind the root-to-word shift is the belief that a root-oriented model is not empirically adequate, as it fails to account for certain facts and correlations in the language. Ratcliffe (1997), for example, observes a prosodic congruence between the verbal and nominal systems, where Measure II and III imperfective verbs are derived from Measure I verbs by mapping a prosodically-circumscribed base, a moraic trochee (CVC), onto an iamb (CVCVV or CVCVC), in a manner parallel to the derivation of broken plurals and diminutives in the nominal system; thus, [yukattib] ‘he makes someone write’ from [yaktub] ‘he writes’ is parallel to [makaatib] ‘offices/desks’ from [maktab] ‘office/desk’. Such parallelism cannot be expressed in a root-totemplate mapping approach. Benmamoun (2003) observes a correlation between the quality of vocalic active and passive markers. The vowel in the passive bears the opposite height value of its active counterpart: [katab] ‘he wrote’ vs. [kutib] ‘it was written’. Herzallah (1990) makes a similar observation in the perfective and imperfective paradigms in Palestinian Arabic, where the vowel quality of the imperfective bears the opposite value of that of the perfective, as in [nidim] ‘he regretted’ > [yindam] ‘he regrets’ but [katab] ‘he wrote’ > [yiktib] ‘he writes’. Modern Hebrew presents ample empirical evidence for surface derivation. Bat-El (1994) shows that when denominal verbs are derived from loan words with onset or coda clusters, the cluster position is preserved in the output, as in [flirtet] from ‘flirt’ and [trinsfer] from ‘transfer’. A root-extraction approach would incorrectly over-generate CVCCVC forms, e.g., *[filret] instead of the attested [flirtet] with final spreading. Bat-El (1994) also observes that denominal verbs derived from complex nouns preserve the affixal elements of the base noun: The verb [hixzik] ‘to hold’ yields the deverbal noun [ta-xzuka] ‘a maintenance’ which in turn yields the denominal verb [tixzek] ‘to maintain’ and not *[xizzek]. Similarly, the denominal verb [misper] ‘to enumerate’ is derived from the deverbal noun [mispar] ‘number’ derived from the stem [sapar] ‘to count’. As arguments and counter-arguments over the autonomy of the root and/or template continue to flow, four analytical approaches covering all possible value combinations emerged:
HYPOCORISTICS REVISITED
27
(a) +Root, +Template (McCarthy 1981) (b) +Root, -Template (Davis and Zawaydeh 1999a, 2001) (c) -Root, +Template (Ratcliffe 1997) and (Heath 2003) (d) -Root, -Template (Benmamoun 1999, 2003) and Ussishkin (1999, 2005) As a contribution to this debate, this paper focuses on the most recent chapter in this series, namely, Davis and Zawaydeh’s (D&Z) seminal work on Arabic hypocoristics. In their contributions (Davis & Zawaydeh 1999a, 1999b, 2001) D&Z spotlighted the study of Arabic hypocoristics within Optimality Theory (OT), in which they created a corpus of hypocoristics from first names in Jordanian Arabic. Although they provide a description of three types of nicknames, they focus their analysis primarily on the most salient type, namely, of the shape faµµuul (C1aC2C2uuC3), as the nickname walluud from the name waliid. Their data seems to demonstrate the centrality of the consonantal root with the distinction between input (underlying) and output (surface) root. While they advocate for the root as a morpheme, they dispense with the notion of templates, replacing it with a specific hypocoristic input morpheme on par with the truncation morpheme postulated for truncated hypocoristics in German (Ito & Mester 1997). Accordingly, they derive the desired templatic effects of the faµµuul hypocoristic form through the interaction of markedness and faithfulness constraints of OT. In this paper, I present new hypocoristic examples that challenge both tenets of D&Z’s analysis: the postulation of a hypocoristic morpheme and the morphemic status of the root. The data forming the empirical basis for the research is drawn primarily from hypocoristics derived from weak (glide-final) and hollow (glide-medial) roots, as well as reduplicated hypocoristics from sound roots. I argue that all iambic (stress-final) hypocoristic patterns are one and the same, and can be accorded a unified account under output-output (relation between surface forms) correspondence (McCarthy & Prince 1995; Benua 1997). The paper is organized as follows: Section 2 provides an illustration of hypocoristic formation (HF), focusing in particular on hypocoristics from sound roots, the main target of study in D&Z’s work, with a summary of their observations and analysis. The root analysis is then tested against a variety of hypocoristic forms from non-sound roots. Section 3 presents an alternative word-based approach to HF, followed by a description of the set of wellformedness conditions underlying the hypocoristic template faµµuul. In Section 4 the constraint hierarchy developed in Section 3 is tested with all types of hypocoristics cited in this paper, which concludes with Section 5.
28
SAMIRA FARWANEH
2. Hypocoristic formation in Arabic 2.1 Hypocoristic forms Arabic hypocoristics (or nicknames) come in various shapes and forms. Some maintain the Classical Arabic fuµayl forms, albeit with dialectal phonological alternations, e.g., the name saaliħ gives rise to the hypocoristic form sleeħ or sweeleħ. Another type involves truncation of a name into a long monosyllable augmented to disyllabicity by adding the feminine marker /-a/, as in tuuħ-a or suum-a, which became the popular nicknames for the renowned dancer Tahiyya Karyoka and the singer Um Kulthoum, respectively. The most common type of nicknaming found in many Arabic dialects is the stress-final disyllabic pattern of the shape CaCCuuC, where the C slots are filled by three or four root consonants extracted from the name. This pattern acquired high productivity in the dialects to the extent of replacing the earlier primary pattern fuµayl1. The following examples serve to illustrate the pattern. (2) Stress-final disyllabic hypocoristics (D&Z 1999a:117) Root Name Nickname Gloss /xld/ xaalid xalluud “eternal” /wrd/ warda warruud “flower” /ħsn/ ħasan ħassuun “good” /smr/ samiira sammuur “nightly conversational partner” HF is such a highly productive morphological process in Arabic dialects that its target is no longer limited to proper nouns. Any common noun becomes a viable target for nicknaming. Cosmic and fauna terms figure prominently in the list of common nouns susceptible to nicknaming, and are used metaphorically to refer to humans with affectionate or pejorative connotations2. (3) Hypocoristics from common nouns Root Noun Nickname />mr/ >amar >ammuur /šms/ šams šammuus-a /klb/ kalb kalluub
1
Gloss “moon” “sun” “dog”
The faµµuul form is particularly productive in Egypt and Eastern Mediterranean countries. The fµeel form is more productive in the Gulf and North Africa. 2 An anonymous reviewer stated that common noun diminutives should be distinguished from proper noun hypocoristics, as they involve an entirely different process. It is noteworthy to mention here that diminution in my Palestinian dialect, and perhaps in others, is expressed syntactically and not morphologically; e.g., bissa zgiira for ‘kitten’; the faµµuul form is used exclusively as a nickname connoting endearment or derogation. It would be interesting to research the current status of true morphologically-derived diminutives across the dialects.
HYPOCORISTICS REVISITED
29
Hypocoristic forms derived from common nouns can denote positive or negative emotional relationships to the addressee. Viewed positively, nicknames serve to denote affection and endearment, and they are most often used by children addressing their pets or animal toys, or by caregivers addressing their children. They may be used metaphorically to indicate positive physical traits, e.g., referring to a woman as >ammuur-a, which indicates “cuteness”. Viewed negatively, nicknames are used pejoratively to denote negative or socially unacceptable physical traits, as in referring to an obese person as dabduub, which capitalizes, through diminution, on a negative attribute of the animal, in this case the size of the bear. Hypocoristics as address or reference terms can also be formed from adjectives, in which case the hypocoristic serves an intensive function. (4) Hypocoristics from adjectives Root Adjective Hypocoristic /ħb/ ħabiib ħabbuub /fhm/ fahiim fahhuum /štr/ šaatir šattuur /nµm/ naaµim naµµuum
Gloss “beloved” “smart” “diligent” “soft”
Interestingly, all hypocoristic forms, whether derived from proper or common nouns or adjectives, display common structural characteristics to be detailed in the following section. 2.2 The structural features of hypocoristics I begin this section with an overview of D&Z’s observations based on their data drawn exclusively from proper name hypocoristics. Additional examples from my own collection of nicknames are then introduced in Section 2.3 to serve as empirical evidence for the non-root correspondence-based analysis detailed in Section 4. Among D&Z’s observations is that despite the variable size of the nouns and adjectives serving as targets for hypocoristic formation, the hypocoristic output or nickname is invariable in structure. Nicknames are maximally disyllabic, consisting of two closed syllables varying in duration: CVCCVVC. While the vowel quality in a nickname remains constant, viz. /a/ in the first syllable and /uu/ in the second, the consonants of the hypocoristic are transferred from the noun or adjective which serves as its base, with the medial consonant geminated to fill the two medial slots. In the case of quadri-literal roots, all four root consonants emerge in the output, as in the nickname maryuum and barhuum from maryam and brahiim respectively, thus obviating the need for medial consonant gemination.
30
SAMIRA FARWANEH
D&Z further observe that not all consonants in the base are eligible for transfer to the nickname; only the root consonants are. Affix exclusion is exemplified by the nickname bassuum from [ib-t-saam] and ħammuud from [mu-ħammad], where the infix /-t-/ and the prefix /mu-/ of the name do not participate in HF. Another interesting observation credited to D&Z is that hypocoristic outputs are based on the root consonants as they appear in the actual name, and not the underlying lexical root. Their observation is based on hypocoristic outputs such as the forms in (5), which are more faithful to the phonetic realization of the root than to its underlying segmental features. (5) Hypocoristics from output roots Root Name Hypocoristic /µwd/ µayda µayyuud /rwd/ raa>id ra>>uud
Gloss “return” “pioneer”
Thus a name like µayda from the root /µwd/ yields the output µayyuud not *µawwuud. Examples such as these recognize a distinction between an input (underlying) root and an output (surface) root. This distinction plays an important role in D&Z’s root analysis. However, in the account proposed here they are taken as an indication of the important role of the base or surface form in HF. I shall return to the role of the root consonants in Sections 2.4 and 4. Thus far, D&Z’s observations suggest that a descriptively adequate analysis must take into consideration the following empirical facts about HF: (a) the disyllabicity of the hypocoristic, (b) the exclusion of affixal consonants from the hypocoristic output, (c) the distinction between input and output roots, and (d) the invariability of the vocalic melody. D&Z’s root-based analysis sketched in the following section provides a descriptively adequate account for hypocoristics derived from sound roots. 2.3 Davis and Zawaydeh’s analysis of hypocoristics Davis and Zawaydeh’s formal analysis of Arabic hypocoristics adopts Ito and Mester’s (1997) OT account of truncated names in German, which is premised on two notions of correspondence relation: input-output and outputoutput, especially base-output correspondence. They posit a hypocoristic input morpheme, which they label H, analogous to Ito and Mester’s T (truncation) morpheme, whose phonetic realization is determined by the base-output constraints. The vocalic melody is qualitatively and quantitatively specified in the input as two long (bi-moraic) vowels phonetically specified as /a/ and /u/. The prosodified name serves as the base for the hypocoristic. Thus, the actual shape of the hypocoristic is determined by input-output constraints which regulate the size and vowel quality of the hypocoristic and base-output constraints which specify the consonantal quality.
HYPOCORISTICS REVISITED
31
The organization of the consonants in hypocoristic forms is regulated via a few undominated base-output alignment and faithfulness constraints stipulating the alignment of the first and last consonant of the output root with the left and right edge of the form respectively (ALIGN L (OR, Hy) and ALIGN R (OR, Hy)). Moreover, these constraints preserve the identity of the output root consonants by stipulating that every output root consonant must have a correspondent in the hypocoristic (MAX (OR-Hy)), and vice versa (DEP-C (OR-Hy)). The main markedness constraint that plays a crucial role in their analysis is *V:V: which bans long vowels from occurring in adjacent syllables. The interaction of the faithfulness and markedness constraints ensures that the emerging output is the one that displays all the characteristics of the faµµuul hypocoristics pattern without reference to specific templates. As we explore more hypocoristics data beyond the regular tri- and quadri-consonantal strong forms, new observations come to light, challenging the root analysis and pointing instead toward a surface derivation, as will be addressed below. 2.4 Hypocoristics from non-sound roots The examples to be examined in this section are drawn primarily from hypocoristics of names, common nouns, or adjectives derived from triconsonantal hollow (glide-medial), weak (glide-final) or bi-consonantal roots. Nouns derived from hollow roots and their corresponding hypocoristics present an interesting array of facts that may pose a challenge in formulating an adequate analysis. An analysis of these forms must be flexible enough to accommodate all the phonological alternations surveyed by these forms, and parsimonious enough to preempt over-generation. The root glide assumes a number of phonological identities in the hypocoristic; it may appear with all its phonological features intact, as in (6a), it may assume a different phonological shape, as in (6b), or it may disappear altogether, as in (6c) and (6d). (6) Hypocoristics from hollow roots /CGC/ Root Name Hypocoristic (6a) /nwr/ nuur nawwuur /µyd/ µiid µayyuud (6b) /µwd/ µayda µayyuud /rwd/ raa>id ra>>uud (6c) /nwl/ manaal mannuul /nwr/ muniira mannuur /rwm/ maraam marruum /nwr/ manaar mannuur (6d) /zyn/ zeena zanzuun /rym/ riim ramruum
Gloss “light” “feast” “return” “pioneer” “aim” “bright” “goal” “lighthouse” “beautiful” “gazelle”
32
SAMIRA FARWANEH
In (6a), the hypocoristic glide is identical to that of the name and the lexical root. This is not the case in (6b), where the lexical labiovelar glide appears as a palatal glide in both the surface name and the hypocoristic. The glide in the nickname is identical to that of the name but not the underlying root. In (6c) and (6d), the glide, which is vocalized in the name, has no manifestation in the hypocoristic. Two strategies are used to compensate for the loss, or vocalization, of the input glide. In (6c), the prefixal consonant /m/ participates in the HF process, thereby challenging D&Z’s assumption that affixal consonants are categorically banned from occupying a consonantal slot in hypocoristics. The second strategy, exemplified in (6d), shows reduplication of the root consonants that appear in the name; thus in the name zeena from the root /zyn/, only the consonants /z/ and /n/ are reduplicated, since the medial glide is vocalized into the mid long vowel [ee], yielding the nickname zanzuun. Hypocoristics derived from weak glide-final roots show the same pattern of reduplication observed in (6d). Alternatively, the final output root consonant, the medial consonant in the input root, spreads to occupy all the consonantal positions in the hypocoristic template. Thus, the hypocoristic patterns from weak root derived nouns are the following: C1aC2C1uuC2, as in (7a), or C1aC2C2uuC2, as in (7b). (7) Hypocoristics from weak roots /CCG/ Root Name Hypocoristic (7a) /hdy/ huda hadhuud /smw/ saamya samsuum (7b) /hdy/ huda hadduud /ndy/ naadya nadduud
Gloss “enlightenment” “eminent” “enlightenment” “morning dew”
Both hollow and weak root-derived nouns converge on the fact that glide vocalization in the base output gives rise to reduplication. They differ in that the realization of the medial glide of the root in the base yields a correspondent in the hypocoristic form; the realized glide of a weak base has no correspondent in the output form, which triggers the spreading of the previous consonant; thus, the name µayda produces the hypocoristic µayyuud while nadya does not yield *nadduuy but nadduud. The ill-formedness of *nadduuy, as D&Z have observed, is attributable to a prohibition on syllable-final long vowel plus glide combinations, for which they propose the markedness constraint *VVG. This constraint is operative stem-internally and not in stem+affix combinations, where a high glide may follow a long vowel, as in >abuu-y ‘my father’ and >axuu-y ‘my brother’. The constraint must also be restricted to long high vowels: *uuy, *iiy, *uuw, and *iiw. Long low vowels may be followed by a glide syllable internally within stems, as in the intensive nouns šakkaay ‘complainer’ from /šky/, and bakkaay ‘crier’ from /bky/.
HYPOCORISTICS REVISITED
33
A similar pattern of consonant spreading is observed in the hypocoristics of nouns derived from bi-literal roots, especially if the surface noun contains two instantiations of the root-final consonant, as in (8). (8) Hypocoristics from bi-literal roots /CC/ Root Name Hypocoristic /dl/ dalaal dalluul /ħb/ ħabiib ħabbuub /ħn/ ħanaan ħannuun
Gloss “coquettishness” “beloved” “tenderness”
The examples presented in this section pose a challenge to the root-primary analysis of D&Z, particularly the alignment constraint that monitors the alignment of root consonants in the output. The affixed hypocoristic mannuul places the consonantal prefix [m-] in an output-initial position, thereby blocking the root initial consonant /n/ from occupying this position, as stipulated by ALIGN-L(OR, Hy). Outputs displaying final consonant spreading or consonant reduplication may constitute violations to ALIGN-R(OR, Hy), as they place a non-root consonant, a reduplicant, in the output-final position. Finally, it is not clear how a root analysis would handle the array of output possibilities that non-sound roots exhibit, the hollow root preference for reduplication, as in zanzuun, the bi-literal root preference to spreading, as in dalluul, and the weak root tolerance of both, as in hadhuud or hadduud. The word-based account detailed in the following sections overcomes these challenges with no reference to template, hypocoristic input, or root. 3.
A word-based analysis of hypocoristic formation In this section, I propose an alternative OT analysis to HF that takes as its input a prosodically circumscribed base, which may be a word (an output) or a part thereof. The analysis makes no reference to roots, underlying or surface, nor does it refer to specific templates. Templatic effects are derived via the interaction of the general faithfulness and markedness constraints necessary for deriving other lexical categories. D&Z (1999a) show the expressive power of OT that through a number of markedness and faithfulness constraints is capable of deriving the templatic shape of the hypocoristic without recourse to an ad hoc template. To ensure descriptive adequacy, D&Z are compelled to postulate a unique hypocoristic morpheme. I will show below that this postulation is an unnecessary complication of Arabic grammar, as well as an inaccurate description of Arabic hypocoristics. Arabic hypocoristics are augmentative rather than truncatory; thus imposing a reductionist analysis proven successful for German, French, or English hypocoristics to an augmentative process common in Semitic languages would miss the target.
34
SAMIRA FARWANEH
3.1 Hypocoristic formation within OT HF is analyzed in this paper as a purely output-to-output process which takes as its input a prosodically circumscribed base, a full word, or a part thereof, and not a consonantal root, underlying or surface. The base to HF is specified as a minimal word, a disyllabic or bimoraic foot (an iamb, a syllabic or moraic trochee). Most nouns and adjectives that lend themselves to HF conform to a member of this foot inventory, as the list in (9) illustrates. (9) Nouns and their foot types Iamb Syllabic trochee CVCVV(C) CVCV(C) waliid samar ħabiib µumar rafiif ħasan
Moraic trochee CVC(C)3 ward-a “flower” kalb “dog” šams “sun”
If the surface structure of a noun or adjective exceeds the structure of the primary foot types, the output is reduced to fit the maximum foot type possible. The reduction is governed by paradigm recognition, where the speaker reduces the oversized name to a form that matches an already recognized paradigm of related names, as in (10). (10) Name muħsen btisaam nuµmaan µiħsaan
Base ħsen, *muħs bisaam, *btis nuµm, *µmaan ħsaan, *µiħs
Related names ħseen, ħusni, ħasan basiima, baasem, basma naµiim, niµma, minµim ħseen, ħasan
Hypocoristics ħassuun bassum naµµuum ħassuun
If the name is ‘too big’ to form an optimal base for HF, it is reduced to one of the three foot types: CVCVC (or CCVC), CVCVVC, or CVCC, given that the reduced base fits some recognizable name paradigm the speaker has internalized. This does not imply root extraction or recognition, since affixed names that fit the optimal foot structure are not subject to reduction. If the oversized name does not yield a recognizable reduced base, either the whole name is taken as a base, despite its size, or HF is abandoned altogether. This is usually the case with non-Arabic names, or archaic Arabic names that have no related derivatives. For example, the Arabized Hebraic names brahiim and maryam do not yield acceptable bases, hence the entire name, analyzed as a quadri-literal stem, undergoes HF yielding [barhuum] and [maryuum]. The old Arabic name µuθmaan, from the now obsolete tri-literal root /µθm/, did not 3
The final consonant in parenthesis indicates the extrasyllabicity of final consonants, following the classic literature on syllable theory.
HYPOCORISTICS REVISITED
35
produce a hypocoristic; neither *µaθθuum nor *µaθmuun were acceptable to the speakers I consulted with. Having thus far defined the base of HF as a trochaic (syllabic or moraic) or iambic foot, the discussion now turns to the faithfulness and markedness constraints that evaluate possible output candidates in search for the optimal output. The relevant faithfulness constraints which regulate identity or correspondence relations between the base of HF and its output are given in (11) along with their definitions. (11) Faithfulness constraints MAX-BO: Every segment in the base must have a correspondent in the output; no deletion INTEGRITY: Every element in the base has one correspondent in the output; no copying ANCHOR: Left and right edges of base and output must match; no misalignment The MAX constraint rejects any output that does not contain all the segments of the base. Taking the name saamer (D&Z’s example), outputs missing a base segment like *sammuu would be blocked by MAX (realization of vocalic segments will be discussed in a later section). INTEGRITY blocks outputs containing more than one instantiation of base segments, e.g., *samruur or *samsuur. ANCHOR ensures that segmental alignments in the base are preserved in the output, thus ruling out possible but misaligned outputs such as *samruu where the base-final [r] is not output-final. The standard analysis of the Arabic nominal and verbal systems entails assigning a prescribed wazn or template for each lexical category to which a triconsonantal or quadri-consonantal root is mapped. Thus, the hypocoristic wazn under investigation is faµµuul or in CV-phonological terms CaCCuuC. We concur with Bat-El (2005) that such templates are sets of well-formedness conditions imposed on lexical and morphological categories and operate on fully-specified forms, i.e., outputs. Thus, a template such as the second wazn faµµal is an amalgam of well-formedness constraints, and not a discrete unit arbitrarily specified in the lexicon and memorized by speakers. The relevant conditions on HF, then, are those that make up the template faµµuul. As these conditions are brought to light, we will see that they are not unique to HF, but operate elsewhere in the language. The first condition extractable from this template concerns the size of stems and syllables. The template faµµuul sets an upper and lower bound of two syllables per stem and two moras per syllable. The literature on Arabic morphology and phonology has already shown that this binarity is observable in Arabic stems and syllables in general, and not limited to hypocoristics (for
36
SAMIRA FARWANEH
stem disyllabicity see McCarthy 1981; McCarthy & Prince 1986, 1990, and for syllable bimoraicity see Ito 1989; Broselow 1992). The two markedness constraints regulating stem and syllable size are stated in (12). (12) Size-regulating constraints SYLL-BIN: Syllables are maximally bimoraic; trimoraic syllables of the shape CVVC or CVCC are disallowed word-internally. DISYLL: Stems are optimally disyllabic; stems of less or more than two syllables are disfavored. The second well-formedness condition accounts for the long vowel that constantly appears in the second syllable of hypocoristic forms. D&Z (1999a) attribute this property to a prosodically specified hypocoristic input morpheme, whose structure is to be preserved by high-ranked input-output correspondence. A survey of verbs and nouns, however, reveals a tendency toward second syllable length, or iambicity, in derived nouns, and syllables of equal length, or trochaicity, in verbs (derived or non-derived). McCarthy and Prince (1986) and McCarthy (1993) have introduced and thoroughly analyzed the observation that foot structure in the verbal and nominal systems is incongruent: verbal stems are always trochaic (stress-initial consisting of equally-long syllables) while nouns show tendency toward iambicity (stress-final with a longer second syllable). Consider the verbal and nominal categories in (13). Structural incongruence between nouns and verbs (13a) Derived nouns and adjectives [CV(C)CVVC] Broken plurals Verbal nouns Derived nouns Form Gloss Form Gloss Form Gloss žibáal “mountains” duxúul “entrance” kitáab “book” wiláad “children” taqríir “report” šaráab “drink” Adjectives Form Gloss kibíir “big” tawíil “long” (13b)
Hypocoristics Form Name sammúur samar hadhúud huda
Derived and non-derived verbs [CV(C)CVC] Measure I Measure II Measure V Form Gloss Form Gloss Form Gloss kátab “write” qárrar “decide” tkállam “talk” dáxal “enter” dárras “teach” tzáwwaž “marry”
37
HYPOCORISTICS REVISITED
Measure VIII Form Gloss stálam “receive” žtámaµ “meet”
Measure X Form Gloss stáslam “surrender” stámtaµ “enjoy”
The absence of iambic verbs and the optimality of iambic nouns can be expressed in terms of an IAMB constraint, which comprises two morphologically sensitive markedness sub-constraints stated in (14). (14a) *IAMB]V: Verbs may not be iambic (14b) IAMB]-V: Non-verbal categories are optimally iambic (right-headed) The first instantiation of the IAMB constraint in (14a) prohibits iambic verbs of the shape [CVCVVC] from ever emerging4. The second constraint in (14b) favors nouns and adjectives of the CVCVVC shape but may allow, under pressure from other constraints, trochaic nouns, such as walad ‘boy’. In fact, most trochaic nouns are lexical or what Holes (2004) terms “primitive”, i.e., non-derived nouns. Most derived nominals including the ones exemplified in (13), e.g., deverbal nouns, intensive nouns, broken plurals, and diminutives are iambic with final stress. The third feature characterizing the prosodic structure of hypocoristics is the gemination of the medial base consonant, unless the base has four consonants. Again, this feature derives in D&Z’s analysis from the prosodically specified and obligatorily preserved input. However, this structural property is not unique to hypocoristics. The hypocoristic pattern faµµuul resembles that of faµµiil of intensive adjectives and faµµaal of nouns of profession, modulo vowel quality. In fact, these three forms are grouped together in the Arabic linguistic tradition under the banner of Siighat mubaalagha (Omar 1974). Below are some examples. (15) Intensive forms faµµiil Form Gloss šarriib “alcoholic” kassiib “wealthy” laµµiib “player”
4
faµµaal Form ħaddaad nažžaar sarraa>
Gloss “blacksmith” “carpenter” “thief”
f aµµuul Form (names) ħassuun ħammuud sammuur
It is possible to rewrite the *IAMB constraint as a positive constraint TROCHEE, as in (BatEl 2005). But the categorical ban on iambicity in verbs indicates that the trochaic foot structure in the verbal system is the consequence of a prohibition on iambicity rather than an enforcement of trochaicity.
38
SAMIRA FARWANEH
Many Arabic dialects have developed a new verbal category of intensive forms, featuring medial gemination. These intensive verbs are morphologically parallel to, but syntactically different from, causative verbs. Verbs whose semantic features involve intensity or frequency assume the shape of the second measure, or wazn faµµal, similar to causative verbs. Examples of intensive verbs in dialectal Arabic are in (16). (16) Intensive verbs Basic Intensive kanas kannas xalat xallat fataħ fattaħ
Gloss “sweep” “mix” “open”
Morphologically, causatives and intensives are identical; syntactically, they are distinct. Forming causatives from basic verbs involves augmenting the complement by one argument, thereby converting Measure I intransitive to Measure II transitive, as in sakat ‘become silent’ to sakkat ‘silence someone’, and Measure I transitive to Measure II di-transitive, as in libis ‘put on’ to labbas ‘dress someone’. Intensive verbs bear the same complements as their basic counterparts, as in the transitive verb xalat or xallat ‘mix’. As we survey the array of intensive verbs, nouns, and adjectives, one common feature emerges as an indicator of intensive formation. All intensive forms involve augmentation of the base by one prosodic unit, a mora. Thus, the intensive of the verb kasar ‘break’ is kassar ‘smash’, and the intensive correspondence of ħabiib ‘beloved’ is ħabbaab or ħabbuub ‘lovable’, with mora augmentation in both cases. Capitalizing on the realizational morphology approach, I propose a constraint REALIZE-MORPHEME, originally proposed by Samek-Lodovici (1993) and expanded by Kurisu (2001), whose function is to ensure that morphological features of the input have certain phonological realization in the output. Below is the formulation of this constraint from Walker (2000:244): (17) REALIZE-MORPHEME (RM) A morpheme must have some phonological exponent in the output. This is a family of constraints regulating correspondence relations between morphological input features and phonological output features. Each morphological feature may assume a unique phonological realization. In this case, the intensive morpheme is realized as an extra mora added to the base, and the constraint is formulated, as in (18). (18) REALIZE-MORPHEME-Intensive (RM-Int) [+intensive] = [mu]
HYPOCORISTICS REVISITED
39
This constraint stipulates that the intensive morpheme is indexed with an additional prosodic unit in the output. In the case of disyllabic stems, be they verbs, nouns, or adjectives, the corresponding intensive forms are realized with an additional mora triggering consonant gemination. The RM constraint does not stipulate the position of the intensive mora, but other considerations determine that a post-vocalic position (a coda) in the first syllable produces the most harmonic output. Inserting the phonetically empty intensive mora initially triggers geminating of the stem-initial consonant, e.g., [katab] > *[µ[k]katab], thereby violating the prohibition on initial geminates. Having discussed the prosodic properties of hypocoristics summarized in bimoraicity, disyllabicity, iambicity, and medial gemination, and their corresponding OT constraints, SYLL-BIN, DISYLL, IAMB, and RM, we now turn to the final property of hypocoristics, namely, the invariable vowel melody of short [a] and long [uu]. Vocalic patterns serve as functional morphemes specifying grammatical features (number, gender, or case) or lexical subclasses (kitaab ‘book’ vs. sariir ‘bed’). We have identified three classes of non-verbal intensive categories: adjectives of the template faµµiil, nouns of professions of the template faµµaal, and hypocoristics of the form faµµuul. If the only distinguishing characteristic among related words is the vocalic affix, affix faithfulness preserves affixal vowels to ensure that the distinction is maintained in surface forms. The effect of affix faith is melodic overwriting, whereby affixal elements overwrite base elements (Steriade 1988; McCarthy & Prince 1990; Ussishkin 2005). The constraint responsible for melodic overwriting is given in (19). This completes the survey of the constraints governing HF; (20) gives a summary of the structural properties of hypocoristics and their corresponding constraints. (19) MAX-AFFIX: Affix elements must be realized in the output (20) Structural features of hypocoristics Feature Constraint a. Hypocoristic stems are disyllabic, like all other stems. DISYLL b. Hypocoristics prefer iambic foot structure, like most [IAMB]-V derived nouns. c. Hypocoristics contain an additional mora, like other RM-INTENSIVE intensives. d. The vocalic marker of hypocoristics is [uu]. MAX-AFFIX e. Hypocoristics preserve the properties of their base. MAX-BO f. Hypocoristics begin and end with a consonant. ANCHOR 3.2 Ordering the constraints The previous subsection shows how the structural properties of the faµµuul template can be derived from constraints necessary for OT and applicable in
40
SAMIRA FARWANEH
Arabic morphology in general. This subsection addresses the order of these constraints. In general, input-output (IO) faithfulness outranks markedness constraints, which in turn outrank output-output (OO) faithfulness, as in (21). (21) IO-Faith >> Markedness >> Affix-Faith >> OO-Faith This ranking allows a distinction between non-derived forms, subject to IO-faithfulness, and derived forms governed by OO-faithfulness. Thus, the IAMB constraint affects only derived nouns, such as deverbal nouns and hypocoristics, but not non-derived nouns, e.g., wálad ‘boy’, which appears faithful to its input without vowel lengthening. MAX-AFFIX outranks MAX-BO to allow vocalic morphemes to appear in the output. Expanding each family of constraints into its members gives the partial ranking in (22). With this partial ranking in mind, I will demonstrate in the following section how the interaction of markedness and output-output faithfulness succeeds in selecting the correct candidate from a number of possible outputs. (22) RM-INT, IAMB]-V >> MAX-AFX >> MAX-BO, ANCHOR, INTEGRITY 4. Testing the constraint hierarchy 4.1 Basic hypocoristics We begin testing the proposed constraints and their hierarchy by considering a common hypocoristic such as sammuur, analyzed in D&Z as a nickname for saamer. Since the name exhibits an anti-iambic structure, i.e., a reverse iamb, it is reduced to the closest optimal base, a disyllabic trochee of the shape sVmVr. Outputs generated from the base are subjected to evaluation by the constraint hierarchy, yielding the evaluation tableau in (23). (23) Name: saamer Base: samar (CVCVC) Possible outputs: *samuur; *saamuur; sammuur; *samruur; *samsuur; *sammur; *samur; ... samer]B+INT+a,u]AFX samuur saamuur sammuur samruur samsuur sammur samur
RM-INT
IAMB]-V
INTEGRITY
*!
* * *!
(B = Base; INT = Intensive; a,uu = vocalic affix)
*! *!
HYPOCORISTICS REVISITED
41
The candidate with short vowels, namely *sammur, is ruled out by the IAMB constraint, which requires a long vowel in the second syllable of derived nominals. The candidate *samuur, which lacks medial gemination, is eliminated by RM, since the intensive feature of the hypocoristic is not realized phonologically. The candidate *samur fares worse than the other two with violations of both constraints. The two candidates with reduplication, namely *samruur and *samsuur, satisfy both the high-ranking constraints RM and IAMB, but they violate the INTEGRITY constraint, due to the one-to-many correspondence between base and output segments. Although violation of the INTEGRITY constraint is not fatal, the two forms in question do not emerge as winners. That is because they are superceded by the more harmonic candidates sammuur and *saamuur which appear to perform favorably with respect to the three constraints. However, the candidate *saamuur is not an attested hypocoristic form. To ensure that the desired output sammuur is obtained, D&Z propose the constraint reproduced in (24) below, eschewing outputs with adjacent long vowels. The constraint expresses a well-known generalization in Arabic dialects (but not in Modern Standard or Classical Arabic) that forms with long vowels in adjacent syllables are subject to a persistent rule of vowel shortening, e.g., /žaamuus/ > [žamuus] ‘buffalo’, or consonant gemination, /šaakuuš/ > [šakkuuš] ‘hammer’. Incorporating the constraint on vowel length in (24) in the constraint hierarchy renders [sammuur] as the most harmonic output among the seven candidates. (24) *VV.VV: Long vowels are prohibited in adjacent syllables. An affixed name like manaal from the root /nwl/ receives the same treatment under the word-based analysis. Satisfying the iambic foot pattern (CVCVVC), the name manaal is eligible to serve as the input to HF without further modification. The tableau in (25) shows similar effects as the previous one: RM rules out candidates lacking an augmented mora, (*manuul), *VV.VV eliminates candidates with adjacent long vowels, (*maanuul), INTEGRITY rules out the candidate with reduplicated segments (*manluul). The tableau in (25) shows further the effect of the MAX-AFFIX constraint, not shown in the previous tableau, which eliminates any candidate where the affixal vowels are not realized, such as *mannaal. The output mannuul emerges as the winning candidate, as it satisfies all constraints, although satisfaction of all constraints is not a requisite to achieve most harmonic status. By allowing the stem, rather than the root, to serve as input to hypocristic formation, the analysis has the advantage of providing a unified account of hypocristic formation from names with or without affixes.
42
SAMIRA FARWANEH
(25) Name: manaal Base: manaal (CVCVVC) Possible outputs: *manuul; *maanuul; mannuul; *manluul; *mannaal; ... manaal]B+INT+a,u]
*VV.VV
RM-INT
IAMB]-V
MAX-AFX
INTEGRITY
AFX
mannuul maanuul manluul
*!
mannaal manuul
* *! *!
The root analysis faces a number of challenges when accounting for affixed hypocoristics: taking the underlying root /nwl/ as input to HF, the evaluation process will incorrectly favor *nawwuul. If, following D&Z’s proposal, an output root [mnl] is extracted from the name, one is likely to arrive at the desired output mannuul. However, the root [mnl] runs afoul of the OCP as it contains two adjacent coronal sonorants. 4.2 Reduplicated hypocoristics If hypocoristics are built directly off the root consonants extracted from names, then the cases of partial and total reduplication discussed below raise another challenging piece of evidence against the root-primary analysis. Consonant-initial reduplication, usually associated with intensive and pejorative meaning, occurs uniquely in Levantine Arabic. McCarthy (1982) brings to light examples, credited to Munther Younes, involving the copying of the initial consonant into the onset of the second syllable in a disyllabic base; the rule may be formulated transformationally as follows: C1C2C3 → C1VC2C1VC3, exemplified by the forms in (26). (26) Partial reduplication Verb Gloss firiħ “become happy” daħal “roll” šaraħ “explain”
Output farfaħ daħdal šaršaħ
Gloss “rejoice” “roll gradually” “criticize”
Such forms lend themselves readily to hypocoristic formation, yielding hypocoristics that preserve the identity of the reduplicant affix, as in (27). (27) Hypocoristics from partial reduplication Form Gloss Output Gloss Hypocoristics lahab “flame” lahlab “burn” lahluub šaraħ “explain” šaršaħ “criticize” šaršuuħ firiħ “become happy” farfaħ “rejoice” farfuuħ
Gloss “active” “lowlife” “merry”
43
HYPOCORISTICS REVISITED
It is clear that the hypocoristic form farfuuħ ‘merry’ cannot be derived from the consonantal root /frħ/ or the Measure I base verb faraħ ‘become happy’; otherwise, an output with medial gemination *farruuħ would ensue. The form farfuuħ ‘merry’ is more likely to be derived from the intensive verb farfaħ ‘rejoice’, which itself is an output generated from the base faraħ ‘become happy’ whose output form is identical to its input /faraħ/. Thus, we assume here, following Ratcliffe (1998) and Ussishkin (1999, 2005), that words rather than roots are primitive lexical entries from which other words are derived (through evaluation and selection among numerous possible candidates). The diagram in (28) shows the correspondence relationship between the three derivational chains. (28) Intensive verbs and their corresponding hypocoristics Input /faraħ/ IO-Faith----> | | Base [faraħ] -----> Output [farfaħ] Base [farfaħ] -----> Output [farfuuħ] | | OO-Faith Examining the second link in the chain (the first being the correspondence of input and output faraħ ‘rejoice’) we generate first the intensive verb farfaħ ‘rejoice’ from the basic form faraħ ‘become happy’ plus a phonetically-empty intensive morpheme (abbreviated in the tableaus as INT). The tableau in (29) shows the evaluation of possible candidates derived from faraħ ‘become happy’ relying on the same constraint hierarchy utilized in selecting hypocoristic forms in earlier tableaus. The two constraints IAMB]-V and MAXAFFIX, featured in previous tableaus, play no relevant role here; instead, we witness the effect of ANCHOR, which monitors the alignment of base and output segments. (29) Reduplicated intensive verbs Base: faraħ+Int Possible outputs: *faraħ; *fafraħ; farraħ; farfaħ; ... faraH]B+INT]AFX faraħ farraħ farfaħ fafraħ
RM-INT ANCHOR *!
INTEGRITY
* *
44
SAMIRA FARWANEH
According to the evaluation of three relevant constraints, the geminated form farraħ ‘make happy’ incorrectly emerges as the winning candidates, while the other candidates, including the correct farfaħ ‘rejoice’, incur some constraint violation. The faithful candidate *faraħ is ruled out for failing to exhibit a phonological sign of intensity, as required by RM; both reduplicated forms *fafraħ and farfaħ ‘rejoice’ incur one violation each of INTEGRITY, which disfavors one-to-many segmental correspondences. Yet farfaħ ‘rejoice’ is more optimal than its apparent equal *fafraħ. Within a root-based generative framework, I accounted for this pattern of internal reduplication via an initial association rule that precedes the regular left-to-right association rules mapping root elements to corresponding skeletal positions (Farwaneh 1990). This rule anchors the reduplicant preserving the syllabic position of its base. Thus, onset segments are anchored in onset positions and vice versa. The effect of this rule can be derived within Correspondence Theory by further specifying the ANCHOR constraint to refer to alignment within syllables, as in (30). (30) ANCHOR ANCHOR-ONSET (BO): All output correspondents of the left edge of the base must occupy onset position in the output. ANCHOR-CODA (BO): All output correspondents of the right edge of the base must occupy coda position in the output. The ANCHOR constraint was proposed in the literature to ensure positional identity of edge segments, requiring an overlap between the left and right edges of the base and the output. Replacing the general ANCHOR with the specific ANCHOR-ONS and ANCHOR-COD, we gain a further advantage by ensuring that in the cases of one-to-many correspondence, the ‘many’ correspondences occupy the same syllabic position of the source; that is, the left-most base segment can only be reduplicated in the onset position, and copies of the right-most base segment are always in the coda position. We can now account for the optimality of farfaħ ‘rejoice’ over *fafraħ, using the expansions of the ANCHOR constraints in (30). (31)
ONS COD | | Base: faraħ
faraH]B+INT]AFX faraħ farraħ farfaħ fafraħ
RM-INT *!
ANCHOR-ONSET
*!
ANCHOR-CODA INTEGRITY
* *
45
HYPOCORISTICS REVISITED
The expanded ANCHOR constraint can yield the desired result of optimizing farfaħ ‘rejoice’ over *fafraħ. However, the geminated farraħ ‘make happy’ still surfaces as the winner. As mentioned above, both intensive and causative verbs are characterized phonologically by medial gemination, but distinguished syntactically by the number of arguments they subcategorize for. However, a geminated form of the stem may not serve both causative and intensive functions. If a geminated causative verb form exists, the intensive form satisfies the RM constraint via internal reduplication, thus producing causative/intensive pairs such as farraħ ‘make happy’ and farfaħ ‘rejoice’, and barrad ‘make cold’ and barbad ‘be cold’. Whenever the geminated form serves the intensive function, then a corresponding reduplicated form is non-existent, e.g., kassar ‘smash-Int’ but not *kaskar. Thus, the phonological manifestation exponent of a morpheme must be unique. Since the optimal farraħ ‘make happy’ is marked as causative, the intensive form takes on the next best alternative, namely, the partially-reduplicated output. Once the intensive verb is generated, it can serve as the base for the corresponding hypocoristic, as in the tableau in (32). Notice that while the verbal output farfaħ ‘rejoice’ from the base faraħ ‘become happy’ incurs a violation of INTEGRITY, the nominal form farfuuħ ‘merry’ from the base farfaħ ‘rejoice’ does not, since each segment in the base has only one correspondence in the output, except the vocalic segments overwritten by the affix morpheme /uu/. Other outputs fail for different reasons, the worst being *farħuuħ which violates MAX-BO for failing to realize the second [f] in the base, ANCHOR-CODA for positioning base-final [ħ] as onset, and INTEGRITY for having two instantiations of [ħ] corresponding to one in the base. The form farfuħ ‘merry’, being a derived nominal, violates IAMB. Finally, the first candidate *farruuħ also fails to realize the second base [f], a consequence not obtainable if hypocoristics are bound to the consonantal root or the Measure I base faraħ ‘become happy’. (32) Reduplicated hypocoristics ONS COD | | Base: farfaħ Possible Outputs: *farruuħ; farfuuħ; *farħuuħ; ... farfaħ]B+a,u]AFX IAMB]-V farruuħ farfuuħ farħuuħ farfuħ
MAX-BO
ANCHOR- ANCHORONS CODA
INTEGRITY
*! *! *!
*
*
46
SAMIRA FARWANEH
Finally, we test the validity of the constraint hierarchy by evaluating hypocoristic outputs from hollow and bi-literal stems. Recall that hollow stems like the names riim and zeena give rise to reduplicated hypocoristics, which are preferable to forms with final consonant spreading. Thus, ramruum and zanzuun are preferred over *rammuum and *zannuun. Conversely, stems from bi-literal roots like the names dalaal and ħanaan yield hypocoristic outputs with final spreading which are more harmonic than reduplicated outputs. Thus ħannuun and dalluul are preferred over *ħanħuun and *dalduul. Considering first hollow-base hypocoristics with syllable reduplication, we examine the name zeen-a whose base is /zeen/, a bimoraic foot CVVC. The tableau in (33) shows some possible outputs and how they fare against the constraint hierarchy. (33) Name: zeena ONS COD | | Base: zeen Possible outputs: *zuun; *zaanuun; *zanuun; *zannuun; zanzuun zeen]B+INT+a,uu]AFX *VV.VV RM- DISYLL ANCHORINT ONS zuun *! *! zaanuun *! zanuun *! zannuun zanzuun
ANCHORCODA
INTGR
* * *
* * * *
Among the five candidates, zanzuun is the most harmonic, incurring only one violation of INTEGRITY. The reverse effect is observed with names with final consonant spreading like dalaal whose outputs are evaluated in (34) below. (34) Name: dalaal ONS COD | | Base: dalaal Possible outputs: dalluul; *dalduul; *dadduul dalaal]B+INT+a,u]AFX RMINT dalluul dalduul dadduul
MAXBO
*!
ANCHORONS
*!
ANCHORCODA
INTEGRITY
* *
HYPOCORISTICS REVISITED
47
The winner is dalluul which preserves all base consonants, one [d] and two [l]s; since the medial [l] is geminated, i.e., its root node is doubly-linked to two adjacent syllabic positions, it does not constitute violation of INTEGRITY. The next best candidate is the one with syllable reduplication dalduul with one nonfatal violation of INTEGRITY; hence, we would predict that this output may be acceptable to some speakers, but its acceptability is lower than the first candidate. The worst is *dadduul, violating MAX-BO for failing to realize the medial [l] of the base; notice that this result is unattainable, were HF to operate on an underlying bi-literal root /dl/, unless we resort to a different interpretation of ANCHOR, namely, STRONG ANCHOR proposed by Ussishkin (2005). 5.
Conclusion In this paper, I present a purely word-based analysis of hypocoristics within Optimality Theory and Correspondence Theory that does not require reference to a root or template, but relies exclusively on the architecture of the model to derive all attested outputs of various types of hypocoristics. With a surface form as a base for HF, this approach provides a unified account for all types of hypocoristics considered: geminated walluud, affixed mannuul partially reduplicated farfuuħ, totally reduplicated zanzuun or marked with final spreading dalluul. Reference to the root fails to capture all these types within one framework, while reference to the template as a morphological unit misses the generalization observed by Ratcliffe (1998) that derived forms of the same category have a consistent phonological shape despite their variable inputs. Another far reaching consequence of viewing the template as a set of wellformedness conditions is that it has the advantage of killing two birds with one stone. First, it unifies hypocoristics and other intensive nouns and verbs under markedness constraints, particularly the IAMB constraint. Second, it takes a step forward toward unification, albeit yet incomplete, of the nominal and verbal system of Arabic under the REALIZE-MORPHEME family of constraints. The examination of different types of hypocoristics in Spoken Arabic undertaken in this article shows that the root as a morphological entity is neither sufficient nor necessary. However, to reach a decisive answer regarding this issue, numerous sets of data from different Arabic languages and dialects need to be explored and analyzed. A wealth of linguistic material still lies hidden within the pages of dictionaries and the repertoires of native speakers, waiting to be ‘mined’ and scrutinized.
48
SAMIRA FARWANEH
REFERENCES Abd-El-Jawad, Hassan & Issam Abu-Salim. 1987. “Slips of the Tongue in Arabic and their Theoretical Implications”. Language Sciences 9.145-171. Bat-El, Outi. 1994. “Stem Modification and Cluster Transfer in Modern Hebrew”. Natural Language and Linguistic Theory 12.571–596. ----------. 2005. “The Emergence of the Trochaic Foot in Hebrew Hypocoristics”. Phonology 22.115–143. Benmamoun, Elabbas. 1999. “Arabic Morphology: The central role of the imperfective”. Lingua 108.175–201. ----------. 2003. “The role of the imperfective template in Arabic morphology”. In Language processing and acquisition in languages of Semitic rootbased morphology, ed. Joseph Shimron, 99-114. Amsterdam & Philadelphia: John Benjamins. Benua, Laura. 1997. Transderivational identity: Phonological relations between words. PhD diss., University of Massachusetts. Broselow, Ellen. 1992. “Parametric variation in Arabic dialect phonology”. In Perspectives on Arabic linguistics IV, ed. Ellen Broselow, Mushira Eid and John McCarthy, 7-46. Amsterdam & Philadelphia: John Benjamins. Davis, Stuart. 2001. The controversy over the grammatical status of root consonants in Arabic. Paper presented at the 15th Annual Symposium on Arabic Linguistics, March, in Salt Lake City, Utah. ---------- & Bushra Zawaydeh. 1999a. “Hypocoristic formation in AmmaniJordanian Arabic”. In Perspectives on Arabic linguistics XII, ed. Elabbas Benmamoun, 113-139. Amsterdam & Philadelphia: John Benjamins. ---------- & Bushra Zawaydeh. 1999b. “A Descriptive Analysis of Hypocoristics in Colloquial Arabic”. Languages and Linguistics 3.83–98. ---------- & Bushra Zawaydeh. 2001. “Arabic Hypocoristics and the Status of the Consonantal Root”. Linguistic Inquiry 32.512-520. Farwaneh, Samira. 1990. “Well-formed association in Arabic: Rule or condition?”. In Perspectives on Arabic linguistics II, ed. Mushira Eid and John McCarthy, 120-142. Amsterdam & Philadelphia: John Benjamins. Herzallah, Ruqayyah. 1990. Aspects of Palestinian Arabic phonology: A nonlinear approach. PhD diss., Cornell University. Holes, Clive. 2004. Modern Arabic: Structures, functions, and varieties. Washington, D.C.: Georgetown University Press. Ito, Junko. 1989. “A Prosodic Theory of Epenthesis”. Natural Language and Linguistic Theory 7.217-259. ---------- & Armin Mester. 1997. “Sympathy Theory and German Truncations”. University of Maryland Working Papers in Linguistics 5.117-138. Kurisu, Kazutaka. 2001. The phonology of morpheme realization. PhD diss., University of California, Santa Cruz.
HYPOCORISTICS REVISITED
49
McCarthy, John. 1981. “A Prosodic Theory of Nonconcatenative Morphology”. Linguistic Inquiry 12.373–418. ----------. 1982. “Prosodic templates, morphemic templates and morphemic tiers”. In The structure of phonological representations, ed. Harry van der Hulst and Norval Smith, 1:191-223. Dordrecht: Foris. ----------. 1993. “Template form in prosodic morphology”. In Papers from the Third Annual Formal Linguistics Society of Midamerica Conference, ed. Laurel Smith Stvan, 187-218. Bloomington: Indiana University Linguistics Club. ---------- & Alan Prince. 1986. Prosodic morphology. Ms., University of Massachusetts & Brandeis University. ---------- & Alan Prince. 1990. “Foot and Word in Prosodic Morphology: The Arabic broken plural”. Language and Linguistic Theory 8.209–283. ---------- & Alan Prince. 1995. “Faithfulness and reduplicative identity”. In University of Massachusetts occasional papers vol. 18: Papers in Optimality Theory, ed. J. N. Beckman, L. Walsh Dickey and S. Urbanczyk, 249–384. University of Massachusetts. Omar, Ahmad Mukhtar. 1974. Min qadflāya al-luāa wannaħw [Questions in language and grammar]. Cairo: Matabi` Subul Al-‘Arab. Prunet, Jean-Francois, Renee Beland & Ali Idrissi. 2000. “The Mental Representation of Semitic Words”. Linguistic Inquiry 31.609–648. Ratcliffe, Robert. 1997. “Prosodic templates in a word-based morphological analysis of Arabic”. In Perspectives on Arabic linguistics X, ed. Mushira Eid and Robert Ratcliffe, 147–171. Amsterdam & Philadelphia: John Benjamins. ----------. 1998. The ‘broken’ plural problem in Arabic and comparative Semitic. Amsterdam & Philadelphia: John Benjamins. Samek-Lodovici, Vieri. 1993. Morphological gemination. Paper presented at Rutgers Optimality Workshop 1, October, in New Jersey. Shimron, Joseph, ed. 2003. Language processing and acquisition in languages of Semitic root-based morphology. Amsterdam & Philadelphia: John Benjamins. Steriade, Donca. 1988. “Reduplication and Syllable Transfer in Sanskrit and Elsewhere”. Phonology 5.73–155. Ussishkin, Adam. 1999. “The Inadequacy of the Consonantal Root: Modern Hebrew denominal verbs and output-output correspondence”. Phonology 16.401-442. ----------. 2005. “A Fixed Prosodic Theory of Nonconcatenative Templatic Morphology”. Natural Language and Linguistic Theory 23.169-218. Walker, Rachel. 2000. Nasalization, neutral segments, and opacity effects. Garland: New York.
MORPHOLOGY AND THEMATIC ARITY OPERATIONS EVIDENCE FROM STANDARD ARABIC∗
LIOR LAKS Tel-Aviv University
1.
Introduction It is commonly assumed that different thematic realizations of the same concept are derived from the same basic entry via thematic arity (valence changing) operations. In this paper I examine five such operations in the verbal system of Modern Standard Arabic (MSA) based on the judgments of native speakers of Arabic as well as data drawn from the works of Wehr (1961) and Wright (1896/1974). These operations are listed and exemplified in Table 1. Operation Passivization Decausativization Causativization Reflexivization Reciprocalization
Examples kataba “wrote” → kutiba “was written” >awqaµa “caused X to fall” → waqaµa “fell” šariba “drank” → šarraba “made X drink” maššata “combed” → tamaššata “combed oneself” µaanaqa “hugged” → taµaanaqa “hugged each other” Table 1: Thematic arity operations
I will shed light on three intriguing generalizations observed in the verbal system of MSA involving the morpho-phonological differences between passivization and the other four arity operations. These generalizations are: (a) Passivization shows unidirectional relations between input and output forms, while the other operations demonstrate bidirectionality, with some forms serving both as a base and as a derived form; (b) Passivization is derived by melodic overwriting, unlike the other operations, which involve different morpho-phonological processes, such as affixation and gemination; (c) The morphological output of passivization can be easily predicted in contrast with the other operations that can have more than one possible output form.
∗
I would like to thank Outi Bat-El and Tal Siloni for their constructive suggestions and comments.
52
LIOR LAKS
I argue for a correlation between arity operations and their morphophonological manifestations. Specifically, I will show that the difference in the component of the grammar where operations take place, lexicon vs. syntax, is what underlies the observed generalizations. This analysis is developed within the framework of the Word-Based Theory developed by Aronoff (1976) and the Lexicon-Syntax Parameter (Lex-Syn Parameter) proposed by Reinhart and Siloni (2005). I will argue that the setting of this parameter dictates certain morpho-phonological properties with regard to derivation as well as the predictability and directionality of such operations. This paper is organized as follows: Section 2 presents the notion of thematic relations and arity operations. Section 3 deals with the verbal system of MSA and discusses the morpho-phonological processes responsible for deriving one verbal form from another. In Section 4, I analyze the morphophonological differences between syntactic and lexical operations based on the processes that manifest the derivation of new predicates. Finally, Section 5 presents the conclusions of my analysis and its implications. 2. Thematic arity operations 2.1 The active lexicon The theta system is the system enabling the interface between the conceptual and computational systems, the syntax, and, indirectly, the semantic interface systems (Reinhart 2000). In this framework, the grammar includes an active lexicon which is more than a mere list of items, and allows the application of derivational operations (Siloni 2002). The theta system consists of lexical entries, which are concepts that define the theta roles of verb entries, and a set of arity operations on lexical entries, which may generate new entries. Arity operations derive different instantiations of the same concept by changing the syntactic valence of a verb, and consequently forming predicates such as passives and reflexives. Reflexive and reciprocal verbs exhibit considerable cross-linguistic variation with regard to nominalizations, idioms, semantic drift, and productivity. Reinhart and Siloni (2005) argue that this variation depends on the level where these operations apply according to a parametric choice. Accordingly, arity operations can apply in the lexicon or the syntax. The syntactic component of the grammar is the engine that builds phrases from elements selected from the lexicon. The question arises as to whether the syntactic component can manipulate the thematic information of these elements. The syntactic machine operates with the selected elements and the lexical-semantic information they bear, but it cannot change their basic properties (Siloni 2002). Once a theta role is part of the theta grid of a predicate, it must either be merged as an argument or have a residue in the syntax or at the level of interpretation. This is formulated in the Lexicon
MORPHOLOGY AND THEMATIC ARITY OPERATIONS
53
Interface Guideline: The syntactic component cannot manipulate theta grids; therefore, it is illicit to eliminate, modify, or add a theta role in syntax (Reinhart & Siloni 2005). While lexical operations apply to theta grids, operations in syntax apply to syntactic structures already associated with semantic representations of events. The Lex-Syn Parameter is applicable only if the grammar includes an active lexicon, i.e., an inventory of concepts, which takes participants (theta roles) and can undergo arity operations. 2.2 Passivization Passivization involves an operation labeled “saturation” which saturates the external theta role by existential closure (Chierchia 2004; Reinhart & Siloni 2005). The external argument is no longer syntactically accessible, but it is still accessible at the level of interpretation. Horvath and Siloni (2005) provide evidence that verbal passivization is syntactic cross-linguistically, based on semantic properties such as drifts, nominalizations, and idioms. 2.3 Decausativization and causativization Decausativization derives decausative verbs by fully eliminating an external CAUSE theta role. This operation is restricted to predicates whose external argument is a CAUSE and their internal argument is a THEME or an 1 EXPERIENCER . Similarly to passivization, the predicate’s valence is reduced, and the verb loses its accusative case. However, unlike passivization, the reduced argument is no longer accessible at the level of interpretation. It is possible to add a by-phrase or an instrument phrase in the case of passivization, as in (1a) and (1b), but not in the case of decausativization, as in (2a) and (2b). (1a) kusira al-šubbaak-u µala yad-i zayd-in “The window was broken by Zayd.” (1b) kusira al-šubbaak-u bi-waasitat-i šaakuuš-in “The window was broken with a hammer.” (2a) *>inkasara al-šubbaak-u µala yad-i zayd-in “*The window broke by Zayd.” (2b) *>inkasara al-šubbaak-u bi- waasitat-i šaakuuš-in “*The window broke with a hammer.”
1
When the internal theta role is a THEME, this operation derives unaccusative verbs, but it derives subject experiencer verbs when the internal theta role is an EXPERIENCER. I term them both decausatives for the purpose of this paper.
54
LIOR LAKS
Causativization adds a theta role to the theta grid of the predicate. This operation applies to verbs whose external theta role is an AGENT. According to the Lexicon Interface Guideline, causativization and decausativization are lexical operations, since the theta grid is manipulated in both cases. Pesetsky (1995) views causativization as having a much broader range. He derives the transitive alternates of decausative, transitive and unergative verbs from one-place entries by causativization. This analysis raises some problems (Reinhart & Siloni 2005). First, this operation adds a new role to a basic entry. This role alternates between an AGENT for transitive and unergative verbs and a CAUSE for decausative verbs. The assumption that the same operation is involved in both cases makes it impossible to predict which new theta role is added. Moreover, the transitive-intransitive alternation is morphologically unsystematic, while the alternation of causativization is. This suggests that the two alternations are derived by two distinct operations. Following Pesetsky’s analysis, we would lose the definition of the set of decausative and unergative predicates, as they all undergo the same operation. Besides, languages such as French do not have lexical causativization (Friedman 2000). There is a lexical alternation between decausative and transitive verbs, but there is no such alternation between unergative and causative verbs. Again, if the two alternations were the result of the same operation, languages such as French would exhibit only part of this predication for no obvious reason. These arguments support the view that different thematic operations are involved in the derivation of causative and decausative verbs. 2.4 Reflexivization and reciprocalization Reflexivization and reciprocalization do not eliminate theta roles. Rather, a theta role is not mapped onto a syntactic argument position in the semantics of such predicates. When these operations apply in the lexicon, they take two theta roles and form one complex theta role (Reinhart & Siloni 2005). This operation, known as “bundling”, associates two theta roles with the external argument. Reciprocals are similar to reflexives, but their semantics are different. In the case of reciprocalization, the argument must consist of at least two participants, while in reflexivization the AGENT and the THEME are the same argument. When reflexivization and reciprocalization apply in the syntax, the operation is different. Following the Lexicon Interface Guideline, manipulating the theta grid is possible only in the lexicon. Thus, bundling in syntax does not apply to the theta grid of the verb, but to unassigned theta roles. An internal theta role is not mapped onto its canonical position due to the lack of case. Upon the merging of the external theta role, the unassigned role is bundled with the external role, resulting in the assignment of two roles to the same syntactic argument.
MORPHOLOGY AND THEMATIC ARITY OPERATIONS
55
3. The verb morphology of Modern Standard Arabic 3.1 MSA verbal system The MSA verbal system consists of nine prosodic templates, as illustrated in Table 2. This does not include inflectional pronoun suffixes, which are concatenated to the stem for agreement purposes. For the purpose of this paper, I do not include the >ifµal template that is restricted to verbs of colors and physical defects. These templates differ from each other with regard to their prefixes (if any) and the length of vowels or consonants. The phonological shape of a verb is essential in determining the shape of the other forms in the inflectional paradigm (Bat-El 1989; Aronoff 1994). A verb which does not conform to one of the existing templates cannot enter the verbal system. Perfect Imperfect faµal yafµa/i/ul faµµal yufaµµil faaµal yufaaµil yufµil >afµal tafaµµal yatafaµµal tafaaµal yatafaaµal >infaµal yanfaµil >iftaµal yaftaµil >istafµal yastafµil Table 2: MSA prosodic templates
I assume that passivization is a syntactic process following Horvath and Siloni (2005), while all other operations are lexical in languages such as Modern Hebrew (Reinhart & Siloni 2005) and MSA (Laks 2004). Table 3 demonstrates the possible lexical operations in MSA with their morphological manifestations.
a. Causativization b. Decausativization c. Reflexivization d. Reciprocalization
Base Derived form raqas “danced” >arqas/raqqas “made X dance” labis “dressed” labbas “made X dress” kasar “cause X to break” >inkasar “broke” >awqaµ “cause X to fall” waqaµ “fell” maššata “combed” tamaššata “combed oneself” >aslam “gave in” >istaslam “gave oneself in” katab “wrote” kaatab “corresponded” qaatal “fought” taqaatal “fought against each other” Table 3: Lexical operations in MSA
Passivization applies to every transitive verb by changing its vocalic pattern, regardless of its prosodic structure, as in Table (4).
56
Base Perfective kasar “broke” qaddam “handed” saaµad “helped” >arsal “sent” tawaqqaµ “expected” tanaawal “handled” >infataћ “opened” >intaxab “elected” >istaqbal “met”
LIOR LAKS
Derived form Base Imperfective kusir “was broken” yaksir “break” quddim “was handed” yuqaddim “hand” suuµid “was helped” yusaaµid “help” >ursil “was sent” yursil “send” tuwuqqiµ “was yatawaqqaµ expected” “expect” tunuuwil “was yatanaawal handled” “handle” -yanfatiћ “open” >untuxib “was elected” yantaxib “elect” >ustuqbil “was met” yastaqbil “meet” Table 4: MSA passivization
Derived form yuksar “is broken” yuqaddam “is handed” yusaaµad “is helped” yursal “is sent” yutawaqqaµ “is expected” yutanaawal “is handled” ---yuntaxab “elected” yustaqbal “is met”
3.2 Deriving the verbal templates The analysis proposed here is based on the Word-Based approach, originally proposed in Aronoff (1976). This approach is based on the notion that the lexicon consists of words rather than morphemes or roots. Aronoff’s main thesis states that a new word is formed by applying word formation rules to already existing words. There are two main approaches to the relation between a consonantal root and a vocalic template in Semitic languages. The traditional approach (McCarthy 1981) views the consonantal root, which consists of 2-4 consonants in a specific order as carrying the core meaning of the word, and thus expresses the semantic relations between words. This view is structurally expressed by the multi-tiered representation, where vocalic patterns are represented independently. Deriving new forms involves the extraction of a consonantal root from a root, a vocalic pattern, and possibly suffixes, then associating it with a template. However, this approach invokes both a theoretical and empirical problem, known as “the problem of transfer” (Bat-El 1994). The information transferred from the base to the derived form not only consists of the order of the consonants, but also the consonants that occupy adjacent positions in the base, i.e., consonant clusters. In addition, properties such as the quality of the base vowels and affixes are also transferred from the base. Such properties cannot be attributed to the consonantal root. Stem Modification is a model that can better account for generalizations regarding morpho-phonological alternations, as it allows for internal stem adjustments. It was first introduced by Steriade (2000) in an analysis of reduplication and by McCarthy and Prince (1990) in an analysis of the Arabic broken plurals and diminutives. Broken plurals cannot be derived by root-totemplate morphology, as there are structural properties drawn from the singular base that cannot be attributed to either the root or the template. This is attested when vowel length is transferred from the singular stem to the plural and when derivational morphemes survive derivational processes, as in (3a)-(3c).
MORPHOLOGY AND THEMATIC ARITY OPERATIONS
57
(3a) qindiil “a lamp” qanaadiil “lamps” (3b) miftaaћ “a key” mafaatiiћ “keys” (3c) µunsur “an element” µanaasir “elements” These examples show that the derivation of broken plurals involves more than just the roots and that broken plurals are established over words. The plural templates in (3) consist of a short vowel in the first syllable, but the second vowel is long only when the second vowel of the singular form is. In (3b), the consonant /m/, which is not a part of a consonantal root, is transferred to the plural form. Bat-El (1994) provides further support for this model within an analysis of denominative verbs in Modern Hebrew; bases which contain consonant clusters yield verbs containing the same clusters, as in (4a) and (4b). (4a) praklit “lawyer” priklet “to practice law” (4b) sandlar “shoemaker” sindler “to make shoes” These patterns provide further support for stem modification, which is motivated by the need to keep derived verbs as faithful as possible to their bases when suitable vocalic patterns exist in a language. The root extraction approach fails to explain why the CV structure varies for different verbs of the same template. Moreover, it does not explain why a multi-consonantal root should be arranged as its base. In stem modification, however, the relevant changes are made to the base itself; thus, it is not surprising that some base properties survive in the derived form. Such an approach shows that the root does not exist as a morphological unit. The analysis proposed in this paper provides further support for the stem modification analysis, as it predicts that the alternation between Arabic verbal templates is manifested via several processes. 3.3 Melodic overwriting The vocalic pattern of every transitive verb can be overwritten by u-i in perfective forms and by u-a in imperfective forms. When a verb exceeds the minimal word size (two syllables), one of the vowels of the passive pattern spreads to the rest of the syllables. In the perfective form >ustuqbil ‘was met’ in (5a), the last stem vowel changes to /i/ and the preceding one to /u/, which spreads to the preceding syllable. In the imperfective form the first vowel turns into /u/ and the second into /a/, which spreads to the rest of the word, as in (5b).
58
(5a)
LIOR LAKS
u
a
>is taq bal → >ustuqbil “meet” “was met”
(5b) u
a
yas taq “meet”
bil → yustaqbal “is met”
The data in (5a) and (5b) raise a question with regard to the different direction of spreading in the perfective and imperfective forms. I assume it stems from the difference between the imperfective suffixes (e.g. /ya-/ in yastaqbil ‘meet’) and the syllable added to some of the perfective forms (e.g. />i-/ in >istaqbal ‘met’). The occurrence of the former is not phonologically conditioned while the occurrence of the latter is; therefore, they are not considered prefixes. The vocalic pattern of the passive voice associates with the first two inherent syllables of the form, and the rightmost vowel in the pattern spreads to the right when there are more syllables and also to the left. The epenthetic syllable is not inherent, and thus the vocalic pattern skips it in association, but then the rightmost vowel of the pattern spreads to the right. The relation between active and passive predicates exhibits only melodic overwriting; the prosodic structure in both forms is identical and thus vacuously assigned2. Melodic overwriting involves no reference to the consonantal root (Bat-El 2003), as it operates directly on the base, which consists of consonants and the vocalic pattern, and it may include affixes as well. Consonants and affixes remain intact in this process, but vowels do not. 3.4 Prosodic circumscription McCarthy and Prince (1990) suggest a process of prosodic circumscription in order to account for the formation of Arabic broken plurals. To derive the plural from the singular, they posit a rule of positive prosodic circumscription that isolates the leftmost moraic foot of the singular base and maps the circumscribed material onto an iambic foot template. The residue is added to the iambic foot, and melodic overwriting follows as well, as in (6). (6) Singular Form: maktuub “a letter” Prosodic Circumscription: Base-mak (µµ) Residue-tuub Mapping: [makµµ]tuub Melodic Overwriting: {aii makaatiib “letters”
2
My analysis consists of mainly sound verbs. I assume that other verbs share the same vocalic pattern in their passive form. However, they are subject to phonological alternations that result in a different form in their phonetic representation, e.g. baaµa ‘sold’, deriving biiµa ‘was sold’.
MORPHOLOGY AND THEMATIC ARITY OPERATIONS
59
McCarthy (1993) extends the circumscription analysis to the verbal system by suggesting a rule of negative circumscription. This rule extracts a prosodic unit consisting of a moraic syllable and adds a mora prefix to the residue, as in (7). (7) Base Negative Circumscription Prefix µ Spread L Output
ћamal “carried” <ћa>mal <ћa>µmal <ћa>mmal ћammal “made X carry”
Reciprocals are derived in a similar way. Instead of gemination, the first vowel is lengthened, and it occupies the position of the new mora. The distinction between the derived forms ћammal ‘made X carry’, in (7), and qaatal ‘fought against each other’, in (8), is attributed to the direction of spreading by which the inserted mora is filled. (8) Base Negative Circumscription Prefix µ Spread R Output
qatal “fought” tal µtal atal qaatal “fought against each other”
The circumscription analysis relies directly on the notion of the word as the base form (Aronoff 1976). As opposed to the root-and-template analysis, one can identify morpho-phonological elements which express the derivation, e.g., a long vowel for reciprocals and a consonant for causatives. 3.5 Affixation The MSA >afµal template is derived by adding the prefix />a-/ to the faµal form. However, the first vowel of the stem is deleted to preserve the prosodic shape of a binary foot, resulting in the >afµal form. The tafaµµal and tafaaµal templates can be derived from faµµal and faaµal, respectively, by adding the prefix /ta-/. In this case, a syllable is added to the stem, but its internal prosodic structure does not change. As opposed to the formation of >afµal, there is no vowel deletion, as that would result in an ungrammatical tri-consonantal cluster (*tafµµal). The >infaµal template is derived from faµal by the affixation of /n/. An epenthetic vowel is then inserted in order to avoid a consonant cluster in the word initial position, and a glottal stop is inserted to avoid having a vowel initial syllable. The >istafµal template is derived by adding the prefix /sta-/. An epenthetic vowel and a glottal stop are inserted for the same reasons explained with regard to >infaµal. The first vowel of the stem is deleted, as in >afµal.
60
LIOR LAKS
The derivation of one verbal form from another can sometimes involve more than one morpho-phonological process. The reciprocal verb tanaataћ ‘thrust each other’, for example, is derived from the verb nataћ ‘thrust’. In this case, the faµal template, which lacks prefixes and long vowels or consonants, serves as the base for the derivation of the tafaaµal template. This derivation is performed both by the affixation of /ta-/ and by prosodic circumscription, resulting in lengthening the first vowel of the base form. 4.
The morpho-phonological manifestation of the Lex-Syn Parameter While some operations are universally lexical (e.g., decausativization) or syntactic (e.g., passivization), some operations demonstrate cross-linguistic variation, such as reflexivization and reciprocalization. This variation can be explained on the basis of the component of the grammar where a given operation takes place. There are languages, such as Modern Hebrew, MSA and Hungarian, whose parameter is set to ‘lexicon’, while other languages, such as French and Romanian, have their parameter set to ‘syntax’. There is a cluster of syntactic-semantic features, which is determined by the value of the LexSyn Parameter (Reinhart & Siloni 2005). I argue that the Lex-Syn Parameter has morpho-phonological consequences as well; once the parameter value is set, a cluster of three morpho-phonological properties follows: complexity of operation, directionality, and predictability. 4.1 Complexity of operation In morpho-phonological terms, lexical and syntactic operations differ in the degree of complexity of their morpho-phonological manifestation. Based on the observed morpho-phonological alternations discussed in Section 3.2, I define a hierarchy of complexity for the observed processes. (9) a. Hierarchy of Complexity: (i) is more complex than (ii) i. Prosodic modification: addition or deletion of syllables or morascircumscription, and affixation ii. Segmental modification: melodic overwriting b. Complexity Evaluation: i. The higher the level of word structure manipulated, the greater the degree of complexity ii. The more levels manipulated in one operation, the greater the degree of complexity The hierarchy of complexity correlates with the structure of phonological words. The modification of the prosodic structure, including syllables and moras, is more intrusive. Greater intrusiveness correlates with greater
MORPHOLOGY AND THEMATIC ARITY OPERATIONS
61
complexity. I assume the levels in (9a), as the higher the level is manipulated, the more complex the operation is (9b-i). In other words, I consider processes, which add or delete moras or syllables, to be complex processes. The more intrusive the operation, i.e., the more it changes the original base form, the more complex it is. This hierarchy is also perceptually grounded, since the prosodic structure is more perceptually accessible than the segmental structure, and thus an alternation in the prosodic structure is more intrusive. Language acquisition studies suggest that children can perceive prominence contrast among syllables within words and among words within a phonological phrase prior to speech production (Sansavini, Bertoncini & Giovanelli 1997). Perception studies on infants provide information about the psychological reality of prosodic information for infants by demonstrating their sensitivity and attentiveness to the prosodic organization of speech input. Furthermore, prosodic packaging may provide the type of pre-categorization that allows infants to segment the input in a way that makes the discovery of syntactic constituents more likely (Jusczyk & Nelson 1996). In addition to the segmental-prosodic distinction, the number of changes that occur within the base plays a role as well. Derivations that involve more than one process, e.g., affixation and circumscription, are also intrusive, as the more processes apply, the more dissimilar the input and the output are. Melodic overwriting applies to the segmental level only. It involves changing the quality of the stem vowels without changing its prosodic shape and is therefore considered less complex. The most prominent morpho-phonological processes that characterize thematic operations in MSA are shown in Table 5. Operation Base Derived form a. Prosodic circumscription Causativization ћamal “carried” ћammal “made X carry” Reciprocalization katab “wrote” kaatab “corresponded” b. Affixation Decausativization kasar “broke” >inkasar “broke” Reciprocalization faawad “negotiated” tafaawad “negotiated with one another” c. Affixation and prosodic circumscription Reflexivization žahil “was ignorant” tažaahal “pretended to be ignorant” tamaadaћ “praised each other” madaћ “praised” Reciprocalization talaatam “slapped each other” latam “slapped” d. Melodic overwriting mawwal “financed” muwwil “was financed” Passivization tanaawal “handled” tunuuwil “was handled” >untuxib “was elected” >intaxab “elected” Table 5: Types of morpho-phonological processes in MSA
62
LIOR LAKS
The pattern that emerges is that syntactic operations are morphophonologically less complex than lexical operations. Passivization, which is syntactic, involves only melodic overwriting. Note that the same vocalic pattern applies in all passive forms regardless of the prosodic structure of the active base form, i.e., the number of syllables and their weight. When the base consists of a long vowel, e.g., µaalaž ‘treated’, it remains long in the passive form and only changes to /u/ in µuuliž ‘was treated’. When the base contains more than two syllables, e.g., >iqtaraћ ‘suggested’, one of the vowels of the passive melodic pattern spreads to the remaining syllable, forming >uqturiћ ‘was suggested’. The other arity operations, which are lexical, involve the addition of moras or syllables via prosodic circumscription or affixation or both. Moreover, passivization involves only one morpho-phonological process, while lexical operations can involve more than one process. 4.2 Directionality Passivization is manifested by changing the vocalic patterns of active verbs, which demonstrate uniformity with regard to the quality of vowels, as they all share the same vocalic pattern. Assuming that passivization is syntactic, the formation of passive verbs is post-lexical. The output of syntactic operations is not listed in the lexicon; therefore, it is not available as a basic entry. The relation between the active and passive forms is unidirectional. The active form is the base and the passive one is derived, followed by a regular change in the vocalic pattern. The picture is different with regard to some lexical operations, such as causativization and decausativization, represented in (10a) and (10b), respectively. (10a) Base: µalim “knew”
Derived form: >aµlam “let X know”
(10b) Base: >a-wqaµ “caused X to fall” Derived form: waqaµ “fell” Following Reinhart and Siloni (2005), I assume that the unergativetransitive alternation (10a) and the transitive-decausative one (10b) are derived by two distinct lexical operations, as each is limited in a particular way. In (10a), the causative form is derived from faµal, resulting in >afµal, while in (10b) the output is faµal and the input is >afµal. Both prosodic templates serve as a base form and as a derived form. This alternation can be explained in terms of paradigm accessibility. It has been long recognized that many exceptions to the regular phonological processes can be explained by proposing that paradigms of morphologically related words influence each other’s pronunciation (Dowing et al. 2005). A paradigm expresses the ways in which linguistic entities may be mutually connected. Languages demonstrate various cases, where a phonologically motivated alternation does not apply in order to achieve paradigm uniformity. For instance, many adjectives in English
MORPHOLOGY AND THEMATIC ARITY OPERATIONS
63
are formed by adding the suffix /-able/ to a verb. In some cases, stress shift is motivated in order to avoid a string of more than two unstressed syllables. For example, the affixation of /-able/ to díscipline should yield disciplínable, but the prevalent form in actual use is disciplinable, as it achieves paradigm uniformity with regard to stress. In the paradigm {díscipline, dísciplinable} stress is uniform, as it is on the same syllable in both forms (Steriade 2000). Accessibility to other forms plays a role in the derivations in (10). As long as an operation takes place in the lexicon, the morphological system has access to all lexical forms. Consequently, it can derive one form from the other, applying to the basic entry listed in the lexicon, in accordance with the relevant thematic operation. When acquiring a language, speakers are exposed to the derivation of such paradigms, i.e., simple-to-complex derivations and vice versa, which they can apply to new predicates. This approach intertwines with Aronoff’s (1976) view of the lexicon as a system of relations that can be active in generating new words. Although bidirectionality does not occur in all operations, when it does occur, it is restricted to the lexical ones. Passivization, in contrast, demonstrates unidirectional relations. 4.3 Predictability The derived output of syntactic operations can be easily predicted, as the only change that occurs is the vocalic pattern. Each passive template in MSA is restricted to a single corresponding template where their transitive counterparts are formed. This is not true for the templates that feed lexical operations, as there is no strict one-to-one relation between pairs of templates. Templates such as fuµil and >ustufµil, for example, do not have an independent existence, as they serve only as the passive forms of faµal and >istafµal, respectively. In contrast, the morphological output of lexical operations is unpredictable, as most operations have more than one possible input and output templates. As demonstrated in Table 6, there are several combinations of input and output forms for the same lexical operation. As stated earlier, the morphological system has access to paradigms of lexical operations. Once a speaker is exposed to a sufficient number of such paradigms, s/he can derive different input forms from different output forms. Interestingly, some verbs in MSA have more than one morphological shape for their derived counterparts. The causative counterpart of xabar ‘knew’ can be found in two templates: faµµal (xabbar) and >afµal (>axbar). Both causative verbs share the meaning of ‘informed’ although there may be a difference in their usage and productivity. I regard the possibility of variation in the morphological form of a predicate as a unique feature of predicates derived in lexical operations. There is no variation in the morphological shape of passive forms, which are derived in syntax. This also correlates with the notion that as long as a predicate is in the lexicon, it is exposed to different changes. I regard
64
LIOR LAKS
this alternation as one of them, in addition to nominalization, semantic drift and idiom formation (Horvath & Siloni 2005). Lexical Operation Decausativization
Causativization Reflexivization
Reciprocalization
Input
Output
Input
Output
faµal faµµal >afµal
>infaµal tafaµµal faµal
>inkasar “broke” tafarraq “separated” saqat “fell”
faµal faµal faµal faµµal
faµµal >afµal >iftaµal tafaµµal
>afµal faaµal
>istafµal tafaaµal
kasar “broke” farraq “separated” >asqat “caused X to fall” šarab “drank” >akal “ate” rafaµ “lifted” žammal “beautified” >aµadda “prepared” saalaћ “made peace” katab “wrote” madaћ “praised”
šarrab “made X drink” >aakal “fed” >irtafaµ “lifted oneself” tažammal “beautify oneself
>istaµadda “prepared oneself” tasaalaћ “made peace with one another” faaµal faµal kaatab “correpsond” faµal tafaaµal tamaadaћ “praised each other” Table 6: MSA possible input/output templates
It is important to point out that I do not claim that the possible input-output forms of lexical operations are totally free. There is a limited set of forms for each operation, e.g., there would be no causative or reciprocal predicates in the >infaµal template. These options vary more than the set of syntactic operations. 4.4 Summary Table 7 below summarizes the differences between syntactic and lexical operations. The analysis provides support for stem modification over root extraction. If we assumed root extraction, there would be no way to explain the morpho-phonological differences between lexical and syntactic operations. Root extraction would apply in all operations, mapping the consonantal root to different vocalic templates, which may consist of affixes. Root extraction could not explain why in many syntactic operations, the root of the base is mapped to a template that differs from the base only in vowel quality, while in the case of most lexical operations, a root is mapped to different templates that can also differ from the base in the weight and number of syllables, in addition to the quality of some of the vowels. Such an analysis gives further rise to a surface-based account, in which forms are derived from actual words, rather than a system in which forms are derived by relating to an entity that never occurs in isolation on the surface (Ussishkin 2000).
MORPHOLOGY AND THEMATIC ARITY OPERATIONS
Property Complexity
65
Syntactic operations limited to simpler processes: melodic overwriting unidirectionality
Lexical Operations complex processes: prosodic circumscription, affixation, a combination of two processes Directionality bidirectionality, no regular template for a specific operation Predictability predictable unpredictable, variation of forms Table 7: Comparison of features between lexical and syntactic operations
These differences are not manifested in every language to the same extent. The two kinds of thematic arity operations are expected to demonstrate differences regarding the features listed in Table 7. The extent of these differences may vary due to various phonological, semantic, syntactic and pragmatic properties. 5.
Conclusion In this paper, I show the correlation between morpho-phonology and the theta system. I describe the morpho-phonological processes, which MSA verbs undergo when deriving new predicates from existing entries in the lexicon. After dividing the thematic operations into lexical and syntactic ones, I demonstrate that these two kinds of operations consistently exhibit different patterns of morpho-phonological processes. The Lex-Syn Parameter is responsible for the cross-linguistic variation exhibited by reflexives (Reinhart & Siloni 2005) and reciprocals (Siloni 2002). It seems that languages are consistent regarding the setting of parameters. This facilitates parameter setting during acquisition, since evidence from various sources converges to set the choice. The morpho-phonological features observed in this paper reveal another aspect of this parameter and strengthen this claim. Lexical operations are characterized by bidirectionality, low predictability and a wide range of complex morpho-phonological processes. Syntactic operations are predictable, mono-directional and limited to relatively simpler morphological processes. The analysis presented within this framework reveals the interaction between arity operations and morpho-phonological processes, supporting the existence of an interface between morphology and both the lexicon and the syntax (Borer 1991). The analysis lends support for the Lex-Syn Parameter and for the Stem Modification approach. The former enables us to make the fundamental distinctions between lexical and syntactic operations, the latter paves the way to implementing this distinction in the domain of morphophonology. In this respect, the root extraction approach is clearly inadequate, as it does not discriminate between lexical and syntactic operations, making it virtually impossible to account for the observed generalizations. The analysis also supports the view of an active lexicon, which seems to be properly construed as a component of linguistic knowledge (parallel in this respect to syntax and phonology) rather than a mere list of arbitrary items.
66
LIOR LAKS
REFERENCES Aronoff, Mark. 1976. Word formation in Generative Grammar. Cambridge, MA: MIT Press. ----------. 1994. Morphology by itself: Stems and inflectional classes. Cambridge, MA: MIT Press. Bat-El, Outi. 1989. Phonology and word structure in Modern Hebrew. PhD diss., University of California, Los Angeles. ----------. 1994. “Stem Modification and Cluster Transfer in Modern Hebrew”. Natural Language and Linguistic Theory 12.572-596. ----------. 2003. “Semitic verb structure within a universal perspective”. In Language processing and acquisition in languages of Semitic root-based morphology, ed. Joseph Shimron, 29-59. Amsterdam & Philadelphia: John Benjamins. Borer, Hagit. 1991. “The Causative-Inchoative Alternation: A case study in Parallel Morphology”. The Linguistic Review 8.119–158. Chierchia, Genaro. 2004. “A semantics for decausatives and its syntactic consequences”. In The unaccusativity puzzle, ed. Artemis Alexiadou, Elena Anagnostopoulou and Martin Everaert, 288-331. Oxford: Oxford University Press. Dowing, Laura, Alan Hall & Renate Raffelsiefen, eds. 2005. Paradigms in phonological theory. Oxford: Oxford University Press. Friedman, Marc. 2000. Experiencing verbs in French. Lecture given at the Lexicon Workshop, October, in Utrecht, the Netherlands. Horvath, Julia & Tal Siloni. 2005. Adjectival passives: Active lexicon. Ms., Tel-Aviv University. Jusczyk, P. & Deborah Nelson. 1996. “Syntactic units, prosody, and psychological reality during infancy”. In Signal to syntax: Bootstrapping from speech to grammar in early acquisition, ed. James Morgan and Katherine Demuth, 389-408. New Jersey: Erlbaum. Laks, Lior. 2004. Lexical causativization: A case study of Arabic and Hebrew. Ms., Tel-Aviv University. McCarthy, John. 1981. “A Prosodic Theory of Nonconcatenative Morphology”. Linguistic Inquiry 12.373-418. ----------. 1993. “Template form in prosodic morphology”. In Papers from the Third Annual Formal Linguistics Society of Mid-America Conference, ed. Laurel Smith Stvan, 187–218. Bloomington: Indiana University Linguistics Club. ---------- & Alan Prince. 1990. “Foot and Word in Prosodic Morphology: The Arabic broken plural”. Natural Language and Linguistic Theory 8.209283. Pesetsky, David. 1995. Zero syntax: Experiencer and cascades. Cambridge, MA: MIT Press.
MORPHOLOGY AND THEMATIC ARITY OPERATIONS
67
Reinhart, Tanya. 2000. “The Theta System: Syntactic realization of verbal concepts”. OTS Working Papers in Linguistics, University of Utrecht. ---------- & Tal Siloni. 2005. “The Lexicon-Syntax Parameter: Reflexivization and other arity operations”. Linguistic Inquiry 36.389-436. Sansavini, Alessandra, Josiane Bertoncini & Giuliana Giovanelli. 1997. “Newborns Discriminate the Rhythm of Multisyllabic Stressed Words”. Developmental Psychology 33.3-11. Siloni, Tal. 2002. “Active Lexicon”. Theoretical Linguistics 28.383-400. Steriade, Donca. 2000. “Paradigm uniformity and the phonetics-phonology boundary”. In Papers in laboratory phonology V: Acquisition and the lexicon, ed. Michael Broe and Janet Pierrehumbrett, 313-334. Cambridge: Cambridge University Press. Ussishkin, Adam. 2000. The emergence of fixed prosody: The verbal system in Semitic. PhD diss., University of California, Santa Cruz. Wehr, Hans. 1961. A dictionary of Modern Written Arabic. Wiesbaden: Otto Harrassowitz. Wright, William. 1896/1974. A grammar of the Arabic language, 3rd ed. Cambridge: Cambridge University Press.
II
SYNTAX
ON THE INDIVIDUAL-PROPERTY CONTRAST IN FREE STATE POSSESSIVE NOMINALS IN EGYPTIAN ARABIC∗
USAMA SOLTAN Middlebury College
1.
Two types of possessives: Introducing the individual-property contrast It has often been noted that nominal expressions can be either individualdenoting, in which case they refer to individuals in the discourse, or propertydenoting, in which case they refer to properties or kinds, rather than to individuals (Chierchia 1982, 1985). To illustrate this distinction, consider, for example, the semantic denotations of the nominal expressions Bonds and a controversial player in (1a) and (1b). (1a) Bonds/a controversial player hit a homerun. (1b) Bill is not Bonds/a controversial player. In (1a) the proper noun Bonds and the NP a controversial player are both individual-denoting, since each picks out an individual in the discourse. By contrast, these same two expressions, as used in (1b), do not refer to a particular individual, but rather to a type or a property, i.e., the property of being Bonds-like (e.g., for having an exceptional homerun-hitting ability) or of being a controversial player. This individual-property distinction shows up in possessive nominals as well, where the possessor could be either individualdenoting or property-denoting, as shown by the English data in (2a)-(2c). (2a) John’s book (2b) A children’s book (2c) The children’s book
∗
(individual-denoting only) (property-denoting only) (individual-denoting or property-denoting)
For their valuable questions, comments, and suggestions, I would like to thank Elabbas Benmamoun, Mushira Eid, Alaa Elgibali, Tomohiro Fujii, Scott Fults, Norbert Hornstein, Howard Lasnik, Mustafa Mughazy, Doris Payne, Juan Uriagereka, the audience at the 20th Arabic Linguistics Symposium, held at the Western Michigan University in March 2006, and the audience at the 37th Annual Conference on African Linguistics, held at University of Oregon, April 2006. Special thanks to Emad Mekki and Sameh Yosef for their valuable help (and patience) with Egyptian Arabic grammaticality judgments. Any errors or shortcomings in this paper remain entirely my responsibility.
72
USAMA SOLTAN
I will call possessives such as that in (2a) “individual-denoting possessives” (IDPs), and those like (2b) “property-denoting possessives” (PDPs), following Kolliakou (1999) and Strauss (2003). The distinction between IDPs and PDPs typically gives rise to a set of interesting asymmetries regarding several syntactic phenomena, as demonstrated by Kolliakou (1999) in a discussion of de-phrases in French and genitive constructions in Modern Greek. In this paper, I show that these asymmetries do obtain in free state possessives in Egyptian Arabic (EA), arguing that an account of the distinct syntactic behavior of IDPs, as opposed to PDPs, is readily available given the general assumptions of the Principles and Parameters framework, coupled with the assumption that the two types of possessive phrases are hierarchically merged in different positions within free state nominals. The paper is organized as follows: Section 2 introduces the EA syntactic strategies used to express adnominal possession, i.e., the construct state and free state constructions. Section 3 illustrates the asymmetries in the syntactic behavior of IDPs and PDPs with regard to linear order, definiteness/specificity restrictions, distribution, relativization, possessivization, anaphora, and scope ambiguity. Section 4 provides a syntactic analysis of these asymmetries that follows from general principles of grammar (such as locality, conditions on anaphora and quantifier scope) and a structure of free state nominals, in which semantically distinct possessors occupy structurally distinct positions. Section 5 summarizes the conclusions of the paper. 2.
Adnominal possession strategies in EA Like several other Arabic dialects, as well as Modern Hebrew, adnominal possession in EA is expressed by two main syntactic strategies: the construct state (CS), as in (3a), and the free state (FS), by means of the morpheme bitaaµ (glossed “Poss” throughout), as in (3b). While the two strategies are interchangeable with regard to alienable possession, only the CS can be used for inalienable possession (e.g., body parts and family members), as in (4)-(5). (3a) madras-it >aħmad school-f. Ahmad “Ahmad’s school”
(3b) il-madrasa bitaaµ-it >aħmad the-school Poss.-f. Ahmad “Ahmad’s school”
(4a) regl >aħmad leg Ahmad “Ahmad’s leg”
(4b) *il-regl bitaaµ-it >aħmad the-leg Poss.-f. Ahmad “Ahmad’s leg”
(5a) >axuu >aħmad brother Ahmad “Ahmad’s brother”
(5b) *il->ax bitaaµ the-brother Poss. “Ahmad’s brother”
>aħmad Ahmad
ON FREE STATE POSSESSIVE NOMINALS IN EGYPTIAN ARABIC
73
IDPs and PDPs can be expressed by either the CS, as in (6a)-(6c), or the FS, as in (7a)-(7c). (6a) madras-it >aħmad school-f. Ahmad “Ahmad’s school”
(IDP reading only)
(6b) madras-it banaat school-f. girls “a girls’ school/a school for girls”
(PDP reading only)
(6c) madras-it il-banaat school-f. the-girls “the girls’ school/the school for girls”
(both IDP and PDP readings)
(7a) (il-)kitaab bitaaµ >aħmad (the-)book Poss. Ahmad “a/the book belonging to Ahmad”
(IDP reading only)
(7b) kitaab bitaaµ >atfaal book Poss. children “a children’s book/a book for children”
(PDP reading only)
(7c) (il-)kitaab bitaaµ il->atfaal (the-)book Poss. the-children “a/the children’s book” “a/the book for the children”
(both IDP and PDP readings)
While CS possessive nominals might raise interesting questions regarding the issue at hand, for the purposes of this paper I focus only on the asymmetry in syntactic behavior between IDPs and PDPs in EA FS possessive nominals, hoping to extend the analysis presented here to CS nominals in future research. 3.
Syntactic asymmetries between IDPs and PDPs in FS possessives in EA As Kolliakou (1999) points out with regard to de-phrases in French and genitive phrases in Modern Greek, several syntactic asymmetries arise between IDPs and PDPs. I illustrate these below with data from FS possessives in EA. 3.1 Linear order Within multiple possessor nominals, the PDP typically precedes the IDP in linear order, as in (8a) and (8b); the reverse order degrades the grammaticality of the sentence, as in (9a) and (9b).
74
USAMA SOLTAN
(8a) il-madrasa bitaaµ-it il-banaat bitaaµ-it >uxt-ak the-school-f. Poss.-f. the-girls Poss.-f. sister-your “your sister’s girls’ school” (8b) il-liµba bitaaµ-it il->atfaal bitaaµ-it the-toy-f. Poss.-f. the-children Poss.-f. “your brother’s children’s toy”
>axuu-k brother-your
(9a) ?*il-madrasa bitaaµ-it >uxt-ak bitaaµ-it il-banaat the-school-f. Poss-f. sister-your Poss.-f. the-girls “your sister’s girls’ school” (9b) *il-liµba bitaaµ-it >axuu-k bitaaµ-it il->atfaal the-toy-f. Poss-f. brother-your Poss.-f. the-children “your brother’s children’s toy” 3.2 Definiteness and specificity A second asymmetry between IDPs and PDPs has to do with their ability (or lack thereof) to occur in FS structures headed by definite head nouns. While IDPs are typically associated with a definiteness requirement, as indicated by the grammaticality contrast between (11a) and (11b), PDPs are compatible with both definite and indefinite contexts, as the data in (10) shows. (10a)
madrasa bitaaµ-it >atfaal school-f. Poss.-f. children “a children’s school”
(10b)
il-madrasa bitaaµ-it il->atfaal the-school-f. Poss.-f. the-children “the children’s school”
(11a) *madrasa bitaaµ-it >aħmad school-f. Poss.-f. Ahmad “Ahmad’s school” (11b)
il-madrasa bitaaµ-it >aħmad the-school-f. Poss-f. Ahmad “Ahmad’s school”
Notice, that, as Kolliakou (1999) argues for comparable examples in French, (11a) is grammatical on a specific reading, where specificity involves
ON FREE STATE POSSESSIVE NOMINALS IN EGYPTIAN ARABIC
75
membership in a presupposed set (e.g., the set of schools owned by Ahmad, built by Ahmad, etc.). It seems, then, that the generalization is that IDPs are possible with possessive nominals whose head noun is either definite or specific1. 3.3 Distribution According to Kolliakou (1999), the distribution of possessive nominals is governed by the Nominal Denotation Hypothesis (NDH), as stated in (12). (12) At most one IDP phrase and at most one PDP phrase is allowed per a possessive nominal. The NDH seems to be supported by the fact that the occurrence of multiple IDPs within the same possessive construction is (presumably universally) prohibited, as shown by the ungrammaticality of (13b). Multiplicity of PDPs, however, seems to be allowed, at least in EA. I come back to discuss this issue with relevant examples in Section 4.3.3. (13a)
il-kitaab bitaaµ il->atfaal bitaaµ >aħmad the-book Poss. the-children Poss. Ahmad “Ahmad’s children’s book”
(13b) * il-kitaab bitaaµ >aħmad bitaaµ Zeinab the-book Poss. Ahmad Poss. Zeinab “*Ahmad’s Zeinab’s book” 3.4 Relativization A further asymmetry between the two types of possessives has to do with the relativization possibilities allowed for each when they co-occur within the same nominal. While the relativization of an IDP possessor in the presence of a PDP possessor is possible, as in (14b), the relativization of a PDP possessor in the presence of an IDP possessor leads to ungrammaticality, as in (14c). Notice that the grammaticality of (14d) shows that there is nothing wrong with the relativization of PDP possessors per se. (14a)
1
il-beet bitaaµ il-µizba bitaaµ the-house Poss. the-farm Poss. “Ahmad’s farm house”
>aħmad Ahmad
Thanks to Mushira Eid for drawing my attention to the grammaticality of (11a) on the specific reading of the head noun.
76
(14b)
USAMA SOLTAN
>aħmad >illi il-beet bitaaµ il-µizba bitaaµ-oh kibiir Ahmad that the-house Poss. the-farm Poss.-him big Literally: “Ahmad, that the house of the farm of him is big” “Ahmad, whose farm house is big”
(14c) * il-µizba >illi il-beet bitaaµ-ha bitaaµ >aħmad kibiir the-farm that the-house Poss.-its Poss. Ahmad big Literally: “the farm, that the house of it of Ahmad is big” “the farm, whose house of Ahmad, is big” (14d)
il-µizba >illi il-beet bitaaµ-ha kibiir the-farm that the-house Poss-its big Literally: “the farm, that the house of it is big” “the farm, whose house is big”
3.5 Possessivization A fifth asymmetry between IDPs and PDPs in syntactic behavior arises with possessivization, i.e., pronominalization of the possessor. As indicated by the grammaticality contrast between (15b) and (15c), a property-denoting THEME can be possessivized only in the absence of the individual-denoting AGENT. By contrast, the possessivization of an individual-denoting AGENT in the presence of a property-denoting THEME is possible, as in (15d). (15a)
il-diraasa bitaaµ-it masraħiyyit haamlit bitaaµ-it Karma the-study Poss-f. play Hamlet Poss-f. Karma “Karma’s study of Hamlet”
(15b)
il-diraasa bitaaµ-it-ha the-study Poss-f.-its/her “its/her study”
Agent/Theme possessivization
(15c) * il-diraasa bitaaµ-it-ha bitaaµ-it Karma the-study Poss-f.-its Poss-f. Karma Intended reading: “its study of Karma”
*Theme possessivization
(15d)
Agent possessivization
il-diraasa bitaaµ-it-ha bitaaµ-it masraħiyyit haamlit the-study Poss-f.-her Poss-f. play Hamlet “her study of Hamlet”
ON FREE STATE POSSESSIVE NOMINALS IN EGYPTIAN ARABIC
77
3.6 Anaphora As noted by Carlson (1977), a nominal expression that has different denotations is typically associated with different anaphoric elements, as in (16a-16c) from Kolliakou (1999). (16a) (16b) (16c)
Kelly is seeking a unicorn. (two readings: individual and property) Kelly is seeking a unicorn, and Millie is seeking it too. (individual reading only) Kelly is seeking a unicorn, and Millie is seeking one too. (property reading only)
Expressions such as another/some more/some others are compatible with an individual interpretation; hence they can occur freely with indefinite singular antecedents. By contrast, the grammaticality of such sentences degrades when their antecedents are bare plurals, as the contrast between (17) and (18) shows. (17) Jack is hunting for a unicorn, and Frank is looking for another/some more/some others. (18) Jack is hunting for unicorns, and Frank is looking for ??another/??others/??some more/??some others. Not surprisingly, IDPs in EA can serve as antecedents for anaphoric elements such as il-tanyiin ‘the others’, but PDPs cannot, as in (19a) and (19b) 2. (19a) ? il-hagamaat bitaaµ-it il-suwwaar dool kaan-it mutawaqqaµ-a the-attacks Poss.-f. the-insurgents those was-f. expected-f. bass il-hagamaat bitaaµ-it il-tanyiin kaan-it mufag>a but the-attacks Poss.-f. the-others was-f. surprise “The attacks of those insurgents were expected, but the attacks of the others were a surprise.” (19b) * il-hagamaat bitaaµ-it il->intiħariyyiin bi-t-dorr il-salaam the-attacks Poss.-f. the-suicide-bombers hurt the-peace bass il-hagamaat bitaaµ-it il-tanyiin bi-tšaggaµ il-µunf but the-attacks Poss.-f. the-others encourage the-violence “The attacks of the suicide-bombers hurt peace, but the attacks of the others lead to violence.” 2 The judgments in (19) are quite subtle, though, probably due to the relative markedness of the expression il-tanyiin ‘the others’ in general. That said, (19b) sounds much worse than (19a), hence the indicated “?-*” grammaticality contrast.
78
USAMA SOLTAN
3.7 Scope ambiguity Negation and quantifiers may take scope over one another, as in (20a), which is ambiguous between the two interpretations in (20b) and (20c), depending on whether all takes scope over negation, or negation taking scope over all. (20a) (20b) (20c)
I didn’t read all the books. “For all the books, it is not the case that I have read them.” “It is not the case that, for all the books, I have read them.”
With regard to EA possessive nominals, IDPs can take scope over negation, hence the ambiguity of (21a), whereas PDPs cannot, as shown in (21b), where the occurrence of the cliticized individual-denoting possessive pronominal forces the first bitaaµ-phrase to have a property denotation, given the NDH constraint in (12). (21a)
>ana ma-šuft-iš il-lawħaat bitaaµ-it kull il-takµiibiyy-iin I not-saw-not the-paintings Poss.-f. all the-cubists Reading 1: “I have only seen the paintings of some of the cubists.” Reading 2: “I have not seen the paintings of any cubist.”
(21b)
>ana ma-šuft-iš il-lawħaat bitaaµ-it kull il-takµiibiyy-iin I not-saw-not the-paintings Poss.-f. all the-cubists bitaaµ-t-ak Poss.-f.-your One reading: “I have not seen the paintings of any cubist.”
3.8 Summary The individual-property distinction gives rise to a number of syntactic asymmetries in free state possessive nominals in EA, in a way similar to what has been observed for comparable structures in other languages. The list of these asymmetries is given in Table 1. Syntactic property PDPs IDPs Precedence in linear order Yes No Occurrence in definite/specific nominals only No Yes Multiple occurrence Yes No Relativization in the presence of the other type No Yes Possessivization in the presence of the other type No Yes Variable anaphora No Yes Scope over negation No Yes Table 1: The syntactic asymmetries between IDPs and PDPs
ON FREE STATE POSSESSIVE NOMINALS IN EGYPTIAN ARABIC
79
The challenging question that arises now is whether the current theory of grammar can account for the asymmetrical behavior of IDPs and PDPs regarding the syntactic properties in Table 1. In the following section, I present an analysis of these facts within the Principles and Parameters framework. 4. The IDP-PDP distinction at the syntax-semantics interface 4.1 Semantically distinct possessors occupy syntactically distinct positions Munn (1995) notes that the IDP-PDP distinction correlates with syntactic differences in English in a fashion similar to that noted in EA in the previous section. For one thing, adjectives precede the possessor in PDPs, but follow it in IDPs, as indicated by the grammaticality contrasts in (22)-(23). (22a) Mary’s new school (22b) * new Mary’s school (23a) * men’s many shoes (23b) many men’s shoes Also, in multiple possessor constructions, the IDP possessor has to precede the PDP possessor, but not vice versa, as shown in (24a) and (24b). (24a) John’s men’s shoes (24b) * men’s John’s shoes To account for these two facts, Munn (1995) proposes that the possessor occupies a different hierarchical position within the nominal structure in each case. Specifically, PDP possessors are in Spec NP (i.e., in the lexical domain), whereas IDP possessors are in Spec DP (i.e., in the functional domain), as shown in (25a) and (25b), respectively. (25a)
DP ru D NP | ru a possessor N man’s | shoe
(25b)
DP ru D' possessor John ru D N | | ’s shoe
Given the structural distinction in (25a) and (25b), the differences between IDPs and PDPs in English regarding the position of the adjective and the order of the possessors follow. Specifically, adjectives, being left-adjoined to the NP projection, will always precede a PDP possessor, and always follow an IDP possessor, as shown by the structures in (26a) and (26b), respectively.
80 (26a)
USAMA SOLTAN
DP ty D NP | ty a Adj NP | ty new NP N | | man’s shoe
(26b)
DP ru DP D' John ty D NP | ty ’s Adj NP | | new shoe
Similarly, in case there is more than one possessor within the possessive nominal, the IDP possessor will always precede the PDP possessor, since SpecDP is higher than SpecNP, as the tree structure in (27) illustrates. (27)
DP ru DP D' John ty D NP | ty ’s NP N | | men's shoes
In addition to accounting for the empirical differences between these two types of possessives in English, Munn’s analysis implies that interpretational differences are actually associated with (or the result of) structural differences. This is predicted by some version of the Syntax-Semantics Transparency Thesis (see Uriagereka (forthcoming) for a discussion of the mapping between the syntax and semantics). The question now, however, is whether it is possible to extend Munn’s analysis to account for the EA facts presented in Section 3. However, we have to first discuss the structure of FS nominals in general. 4.2 The structure of FS nominals in EA Notice that one of the main differences between English nominals and EA FS structures has to do with the position of the possessive phrase (PossP) in each language. In English, possessors typically precede the head noun. In EA FS nominals, on the other hand, possessors follow the head noun. To account for this linear order fact, I will assume here, following Siloni’s (1997) analysis of FS structures in Hebrew, that bitaaµ-phrases (like Hebrew šel-phrases) are right adjuncts.
ON FREE STATE POSSESSIVE NOMINALS IN EGYPTIAN ARABIC
81
Furthermore, in order to account for the IDP-PDP distinction, I will assume, in the spirit of Munn’s (1995) analysis for English and following Strauss’s (2003, 2005) analysis of comparable Hebrew data, that the two types of possessive phrases occupy two different positions within the syntactic structure of nominals. Given these assumptions, the structure of FS nominals in EA is as in (28). Given the structure in (28), the question is whether it is possible to account for the set of syntactic asymmetries between IDPs and PDPs listed in Table 1. I provide an answer to this question in the next section. (28)
DP ty DP PossPIDP ty Spec D' ty D NP ty NP PossPPDP | N
4.3 Deriving the IDP-PDP syntactic asymmetries 4.3.1 Linear order asymmetry revisited. Recall from the data in (8)-(9), repeated below as (29)-(30), that PDPs typically precede IDPs within multiple possessor constructions. (29a)
il-madrasa bitaaµ-it il-banaat bitaaµ-it >uxt-ak the-school-f. Poss.-f. the-girls Poss.-f. sister-your “your sister’s girls’ school”
(29b)
il-liµba bitaaµ-it il->atfaal bitaaµ-it >axuu-k the-toy-f. Poss.-f. the-children Poss.-f. brother-your “your brother’s children’s toy”
(30a) ?* il-madrasa bitaaµ-it >uxt-ak bitaaµ-it il-banaat the-school-f. Poss-f. sister-your Poss.-f. the-girls “your sister’s girls’ school” (30b) ?* il-liµba bitaaµ-it >axuu-k bitaaµ-it il->atfaal the-toy-f. Poss-f. brother-your Poss.-f. the-children “your brother’s children’s toy”
82
USAMA SOLTAN
Given the structure in (28), since the IDP phrase is right-adjoined to DP, whereas the PDP phrase is right-adjoined to NP, it follows that the IDP will always be higher than any PDP within the nominal; hence, the former will always follow the latter. 4.3.2 Definiteness/specificity asymmetry revisited. The second syntactic asymmetry between IDPs and PDPs in EA is that only the former requires the head of the nominal to be definite/specific, whereas the latter is compatible with both definite and indefinite contexts. The data are repeated below as (31)(32). (31a)
madrasa bitaaµ-it >atfaal school-f. Poss.-f. children “a children’s school”
(31b)
il-madrasa bitaaµ-it il->atfaal the-school-f. Poss.-f. the-children “the children’s school”
(32a) * madrasa bitaaµ-it >aħmad school-f. Poss.-f. Ahmad “Ahmad’s school” (32b)
il-madrasa bitaaµ-it >aħmad the-school-f. Poss-f. Ahmad “Ahmad’s school”
To account for this fact, suppose that the difference between definite and indefinite nominals is that only the former contain a DP projection. If this is the case, then the exclusive occurrence of IDPs with definite head nouns follows from the structure in (28). If there is no DP projection, then there is no place for the IDP adjunct in the structure, hence the definiteness restriction. PDPs, by contrast, do not require a DP projection since they are adjuncts to NP, hence their ability to occur within either definite or indefinite nominals. As it turns out, there are two pieces of supporting empirical evidence for this analysis. First, if specificity, like definiteness, is encoded in the D head, then we have an explanation why such sentences are grammatical on a specific reading, as noted earlier (cf. Section 3.2) with regard to (32a), which is perfect under an interpretation in which the head noun denotes a member of a set. Second, EA has a third strategy of expressing possession by means of the preposition li-, which has the lexical requirement that its selecting head noun be “bare”, i.e., occur without a definite determiner. As predicted by the present
ON FREE STATE POSSESSIVE NOMINALS IN EGYPTIAN ARABIC
83
analysis, li-possessives should occur only in PDP contexts, but never in IDP contexts, a prediction that is borne out by the examples in (33a) and (33b)3. (33a)
madrasa li-l->atfaal school-f. for-the-children “a school for children”
(33b) * madrasa li->aħmad school-f. for-Ahmad “a school for Ahmad”
In sum, the definiteness/specificity restriction on the occurrence of IDP possessive phrases follows from the structure in (28), where they are treated as right adjuncts to DP, hence their exclusive dependence on the presence of a D head in the structure of the nominal, which is, by assumption, the locus of definiteness/specificity. 4.3.3 Distribution asymmetry revisited. Can we account for the distribution facts, expressed in Kolliakou’s Nominal Denotation Hypothesis (NDH) that at most one of each type of possessive may occur inside the nominal? There is sufficient evidence from EA that, at least with regard to the number of PDPs inside FS nominals, the NDH cannot be maintained, as shown by the grammaticality of (34). (34) il-kitaab bitaaµ il-fizya bitaaµ talta >iµdaadi bitaaµ >aħmad the-book Poss. the-physics Poss. third preparatory Poss. Ahmad “Ahmad’s ninth-grade physics book” By contrast, as noted in Section 3.3, the part of the NDH concerning IDPs seems empirically correct: No more than one IDP is allowed per a possessive nominal, as in (35). (35) *il-kitaab bitaaµ >aħmad bitaaµ Zeinab the-book Poss. Ahmad Poss. Zeinab “*Ahmad’s Zeinab’s book” An explanation for this fact is possible given general assumptions of feature checking in minimalist syntax (Chomsky 1993, 1995). In particular, following Strauss (2003, 2005), let’s assume that a D head is required to enter into a checking relation with an XP carrying a referential feature in its domain, where domain is any specifier or adjunct within the DP projection. Given this assumption, we can now account for the contrast between (34) and (35) with regard to the NDH. Specifically, if there is at most one referential value per D 3
The sentence in (33b) is grammatical only on the benefactive reading, which is irrelevant here. (33a) also has an IDP reading on the benefactive interpretation.
84
USAMA SOLTAN
head, it follows that D can only support at most one IDP phrase. In other words, the presence of more than one IDP in the domain of D is not allowed at the semantic interface. If D has only one uninterpretable feature to check against the right-adjoined DP, then multiple checking of that feature by more than one such DP is presumably ruled out by considerations of economy. This proposal accounts at the same time for why multiplicity of PDPs is allowed. Since PDPs do not take part in feature checking (due to their non-referential nature), they are allowed to stack, in the same way all adjuncts do. 4.3.4 Relativization asymmetry revisited. Consider the asymmetry between IDPs and PDPs with regard to relativization, as illustrated earlier in (14). An explanation for such an asymmetry follows from the structure in (28), coupled with standard assumptions of locality in syntax, expressed informally in (36). (36) Grammatical operations are strictly local, where local is understood as “applying to the closest target for the grammatical operation in the structure,” where closest is defined as “hierarchically highest”. Consider, for example, the so-called “superiority effects” in multiple whquestions in English, as in Who did what? as opposed to *What did who do? Under locality, the second sentence is ruled out since the grammatical operation fronting wh-questions is not applied locally, targeting the structurally lower object wh-phase what, rather than the structurally higher subject whphrase who. When locality is observed, however, the resulting sentence is grammatical. Since IDPs are structurally higher than PDPs, as assumed in (28), it follows that under the economy condition in (36), the grammatical operation responsible for relativization can only apply to the former, but not to the latter, exactly as borne out by the facts in (14). 4.3.5 Possessivization asymmetry revisited. The same locality-based analysis of the relativization asymmetry can be extended to the possessivization asymmetry between IDPs and PDPs. The relevant data are those in (15) in Section 3.5. If possessivization, like any other grammatical operation, is subject to the locality condition in (36), and given the structure in (28) for FS nominals in EA, it follows that an IDP AGENT will always be structurally higher than a PDP THEME, hence always the closest target for possessivization, thereby accounting for the grammaticality contrast between (15c) and (15d). THEME-possessivization, however, is rendered possible in case there is no AGENT in the structure, as the ambiguous interpretation of (15b) shows, since the locality condition in (36) is satisfied vacuously in this case, again exactly as desired.
ON FREE STATE POSSESSIVE NOMINALS IN EGYPTIAN ARABIC
85
4.3.6 Anaphora asymmetry revisited. One way to account for the inability of PDPs to function as antecedents is to assume, along the lines suggested by Kolliakou (1999), that this follows from their “modificational” nature, coupled with the independently motivated principles governing anaphora in natural language in general, e.g., Chierchia’s (1985) Functor Anaphora Constraint, which states that modifiers cannot engage in pronominal anaphora, as shown by the ungrammaticality of *The Shakespeariani tragedy is at its best in hisi Hamlet on the intended reading of his as co-referential with the modifier Shakespearian. 4.3.7 Scope ambiguity asymmetry revisited. As with anaphora, the difference in the behavior of IDPs and PDPs with regard to scope ambiguity may be argued to follow from the assumption that IDPs are “quantificational” while PDPs are not, given their “modificational” nature. Therefore, only the former can interact with negation, but not the latter. Evidence that this is the case comes again from Carlson’s (1977) contrasts between indefinite singulars and bare plurals, as in (37a) and (37b). (37a) (37b)
A cat is in this room and a cat is not in this room. *Cats are in this room and cats are not in this room.
As the contrast in (37) indicates, while individual-denoting singular NPs can take scope over negation, as illustrated by the possibility of a noncontradictory reading in (37a), generic bare plurals cannot, as shown by the impossibility of any other reading than the contradictory reading in (37b). If this is the case, then the asymmetry between IDPs and PDPs in this regard follows from their inherent properties. In syntactic terms, given the structure in (28), one can assume that IDPs, being quantificational, can move via quantifier raising to a position higher in the structure than negation, giving rise to the lower scope reading of negation. PDPs, by contrast, are modificational, and therefore are not a potential target for quantifier raising, thereby their exclusive occurrence under the scope of negation. 4.3.8 Summary. The syntactic asymmetries between IDPs and PDPs in EA free state nominals receive a natural explanation given (a) the structure in (28), whereby the IDP phrase is merged structurally higher than the PDP phrase, and (b) a set of independently motivated principles of grammar (e.g., the locality condition in (36), Chierchia’s (1985) Functor Anaphora Constraint, conditions on anaphora and quantifier scope, and the syntax-semantics transparency thesis).
86
USAMA SOLTAN
5.
Conclusions I have shown in this paper that the syntactic consequences of the individual-property distinction in FS possessive nominals in EA regarding linear order, definiteness/specificity effects, distribution, relativization, scope ambiguity, anaphora, and possessivization follow from an analysis in which the IDP phrase occupies a higher position (adjunct to DP) than the PDP position (adjunct to NP) within nominal structure, coupled with independently motivated principles of grammar such as locality, feature checking, and constraints on anaphora and quantifier scope. Cross-linguistic variation between English and EA is minimal and reduces to one parametric difference: The adjunct-specifier option for possessive phrases. Finally, if correct, the analysis presented here provides further support for a theory of grammar that assumes transparency of mapping between syntax and semantics.
REFERENCES Carlson, Greg. 1977. Reference to kinds in English. PhD diss., University of Massachusetts. Chierchia, Gennaro. 1982. “Nominalization and Montague Grammar”. Linguistics and Philosophy 5.303-354. ----------. 1985. “Formal Semantics and the Grammar of Predication”. Linguistic Inquiry 16.417-433. Chomsky, Noam. 1993. “A minimalist program for linguistic theory”. In The view from Building 20: Essays in honor of Sylvain Bromberger, ed. Kenneth Hale and Samuel Jay Keyser, 1-52. Cambridge, MA: MIT Press. ----------. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Kolliakou, Dimitra. 1999. “De-Phrase Extractability and Individual/Property Denotation”. Natural Language and Linguistic Theory 17.713-781. Munn, Alan. 1995. “The possessor that stayed close to home”. In Proceedings of WECOL 24, ed. Vida Samiian and J. Schaeffer, 181-195. Siloni, Tal. 1997. Noun phrases and nominalizations. Dordrecht: Kluwer. Strauss, Uri. 2003. “Individual-denoting and property-denoting possessives”. In Possessives and beyond: Semantics and syntax, ed. Ji-Yung Kim, Yury A. Lander and Barbara Partee, 183-199. Amherst. MA: GSLA. ---------- 2005. Possessive position and interpretation in Modern Hebrew. Ms., University of Massachusetts. Uriagereka, Juan. Forthcoming. Syntactic anchors. Cambridge: Cambridge University Press.
RECONSTRUCTION AND ISLANDHOOD IN JORDANIAN ARABIC∗
NOUMAN MALKAWI & NICOLAS GUILLIOT University of Nantes
1.
Introduction The aim of this paper is to show that current analyses of reconstruction and islandhood are inadequate. In particular, the distribution of resumptive elements (REs) in Jordanian Arabic does not follow the pattern suggested by Aoun, Choueiri and Hornstein (2001), henceforth ACH. This pattern is based on the distinction between apparent resumption, when no island intervenes, and the relation between antecedents and REs is established via movement, as in (1), and true resumption (within islands), where no movement is involved, as in (2). (1) Apparent resumption [DP…pronouni…]j…[IP…QPi [CP... [DP…pronouni..]j-resumptive elementj]] (2) True resumption *[Antecedent…pronouni…]j … QPi … [island …. [resumptive element]j …] As the schemas in (1) and (2) show, ACH’s analysis makes two predictions: (a) reconstruction effects should always appear with apparent resumption, as in (1), and (b) reconstruction effects should never occur with true resumption, as in (2). We will use data from Jordanian Arabic to show that ACH’s predictions are not borne out, as we will provide cases where there is no reconstruction with apparent resumption in contradiction with (1) and cases of reconstruction with true resumption in contradiction with (2). In order to account for the reconstruction data in Jordanian Arabic, two hypotheses are adopted. First, we argue that reconstruction with weak resumptives (clitics and doubled clitics) follows from the NP-deletion analysis of pronouns proposed by Elbourne (2001). Second, only reconstruction with strong resumptives (strong pronouns and epithets) follows from the copy analysis of movement, hence corresponding to cases of apparent resumption in ACH (2001). ∗
We would like to thank Hamida Demirdache, Orin Percus, all the other members of LLING, and the audience of ALS 20 for their comments and suggestions. We also thank our Arabic speaking informants for their judgments and intuitions.
88
NOUMAN MALKAWI & NICOLAS GUILLIOT
This paper is organized as follows: Section 2 introduces the standard assumptions regarding resumption and reconstruction; Section 3 provides critical data showing their limitations, while Section 4 investigates the crucial parameters that play a significant role in banning or allowing reconstruction. Section 5 presents our proposal, followed by the conclusions in Section 6. 2.
Resumption A resumptive element is any anaphoric form (e.g., pronouns and epithets) that appears in a position where, under other circumstances, a gap would be expected (Sells 1984; Demirdache 1991; ACH 2001). In other words, an RE is related to a constituent that appears in an A-bar position, as shown in (3). The clitic -uh in (3) is related to the dislocated constituent (hence in A-bar position) kariim, and it appears in a position where a gap could be expected, as in (4). mbariħ] (3) kariimi [šuft-uhi Karim saw.1s.-Cl yesterday “Karim, I saw (him) yesterday.” ti mbariħ]] (4) [kariimi [šuft Karim saw.1s. yesterday “Karim, I saw yesterday.” 2.1 Resumption properties Resumption displays two well-known properties: (a) it can cross islands, and (b) it allows reconstruction. Some complex structures, known as “islands”, ban the movement of their constituents. However, there is still a way to create dependencies that cross islands, as resumption is generally insensitive to islandhood. The Lebanese Arabic data in (5) and (6) show the dependency between a dislocated element and a resumptive pronoun, which can cross islands. The ungrammatical sentence in (5) contains a strong adjunct island banning movement, but the sentence in (6), where the resumptive clitic -a ‘her’ resumes the antecedent ha-l-muttahamme ‘this defendant’, is grammatical. lamma µrifto >enno ħabasu (5) *ha-l-muttahammei tfeeža>to this-the-defendant surprised.2p. when learnt.2p. that imprisoned “This defendant, you were surprised when you learned that they sent to jail.” (6) ha-l-muttahammei tfeeža>to lamma µrifto >enno ħabasuu-ai this-the-defendant surprised.2p. when learnt.2p.that imprisoned-her “This defendant, you were surprised when you learned that they sent her to jail.”
RECONSTRUCTION AND ISLANDHOOD IN JORDANIAN ARABIC
89
2.2 Reconstruction effects under resumption Reconstruction corresponds to the interaction between movement (dislocation, topicalization, interrogation, and relativization) and interpretation procedures, such as binding conditions, Conditions A, B, and C, Bound Variable Anaphora (BVA), and scope interaction. The sentence in (7a) exemplifies binding reconstruction, as the co-reference between John and he is unavailable, suggesting a kind of Condition C violation. Notice, however, that Condition C is not violated on the surface. The sentence in (7b) concerns scope reconstruction, as the moved constituent can be interpreted in the scope of the quantified expression every doctor, giving rise to a distributive reading, where a different patient is mapped to every doctor. That surface order only provides the wide scope or individual reading of patient, where every doctor examined the same patient. These examples demonstrate that displaced constituents are not always interpreted in their surface positions. (7a) *[Which photographi of Johnj]i did hej give ti to Mary? (7b) [Which patient] did every doctor examine t? Reconstruction effects follow from the copy analysis of movement, which posits that movement leaves full copies of the displaced material instead of a trace (Chomsky 1995; Fox 2000). The presence of a copy in (8a) triggers a Condition C violation, but it would allow Condition A or BVA to be satisfied in similar environments. The existence of the distributive reading in (8b) is also predicted, as patient can be interpreted in its thematic position, i.e., within the scope of the universal quantifier. (8a) *[Which photographi of Johnj] did hej give [photographi of Johnj] to Mary? (8b) [Which patient] did every doctor examine [patient]? Notice that the traditional account of reconstruction is crucially based on the existence of movement in the structure. In other words, the assumption in (9) comes as a direct consequence of the analysis. (9) Whenever an XP allows reconstruction, movement of that XP has occurred. Reconstruction is often assumed to be prohibited when a resumptive element resumes a displaced constituent. However, this prediction does not hold, as reconstruction effects can appear with resumption. Consider the following Lebanese Arabic examples.
90
NOUMAN MALKAWI & NICOLAS GUILLIOT
(10) [telmiiz-[a]i l-kesleen]j ma baddna nxabbir [wala mµallme]i student-her the bad Neg. want-1p. tell-1p. no teacher b-l-faħis >inno huwwej/ha-l-mažduubj zaµbar that he /this-the-idiot cheated.3sm in-the-exam “Her bad student, we don’t want to tell any teacher that he/this idiot cheated in the exam.” taµrif [wala mµallme]i (11) *[telmiiz-[a]i l-kesleen]j ma badda student-her the-bad Neg. want.3sf know.3sf no teacher lee l-mudiira šaħatit-oj mn l-madrase why the-principal expelled-him from the school “Her bad student, no teacher wants to know why the principal expelled him from school.” maµ [wala mµallme]i (12) *[telmiiz-[a]i l-kesleen]j ma ħkiina student-her the-bad Neg. talked.1p. with no teacher >able ma huwwej/ha-l-mažduubj yuusal before he /this-the-idiot arrive.3sm “Her bad student, we didn’t talk to any teacher before he/this idiot arrived.” On the basis of these examples, ACH (2001) argue that REs which appear inside islands (e.g., the weak island in (11), and the strong island in (12)) behave differently from REs which are not inside islands, as in (10). They suggest the terms “true resumption” and “apparent resumption” for these two cases, respectively. The possessive pronoun in the left-dislocated DP in (10) can be bound by the negative QP wala mµallme ‘no teacher’. This can be obtained if we suppose that the left-dislocated DP is reconstructed in the scope of the negative QP. Thus, resumption is compatible with reconstruction effects. ACH conclude from (10) that resumption can be derived with movement when no island appears (apparent resumption). The LF representation of (10) is given in (13). Notice that there is a copy of the antecedent (the dislocated DP) adjoined to the resumptive element. (13) Apparent resumption [DP telmiiz-[a]i l-kesleen]j…[IP…QPi [CP…[DP telmiiz-[a]i l-kesleen]j REj]] [DP student-[her]i the-bad]j…[IP…QPi [CP... [DP student-[her]i the-bad]j REj]] REs inside islands seem not to show reconstruction effects, as (11) and (12) show. The wh- and adjunct clauses in (11) and (12) constitute islands for movement. Resumptive pronouns and epithets within islands are true last resort resumptives. This strategy implies the absence of movement, hence the
RECONSTRUCTION AND ISLANDHOOD IN JORDANIAN ARABIC
91
absence of a copy of the antecedent telmiiz-a l-kesleen ‘her bad student’, as shown in (14). Thus, we obtain a violation of BVA: the anaphoric element -a ‘her’ is not in the scope of the negative QP wala mµallme ‘no teacher’ because there is no copy of the antecedent in the embedded clause. (14) True resumption *[DP telmiiz-[a]i l-kesleen] j… … QPi … [island … [RE]j … ] *[DP student-[her]i the-bad]j… … QPi … [island … [RE]j … ] Reconstruction facts appear with the apparent strategy and disappear with the true one. For ACH, if there is reconstruction, even with an RE, there is a copy of the antecedent in the position occupied by the RE; therefore, movement should be involved even with the resumptive strategy. The case of the apparent strategy generalizes the assumption in (9) to resumption, as in (15). (15) Be it via resumption or not, whenever an XP allows for reconstruction, movement of that XP has occurred. 3.
Paradoxes Notice that the generalization in (15) leads to the following predictions:
(16) Reconstruction should never occur within islands; movement is banned in this context. (17) Reconstruction should always occur in the absence of islands; movement is allowed in this context. 3.1 Reconstruction inside islands We argue that ACH’s predictions in (16) and (17) are too strong for Jordanian Arabic, as many cases of reconstruction can occur even when an island intervenes between the antecedent and the RE. Consider the following data with clitics or doubled clitics as REs 1. (18) Clitic/doubled clitic inside strong (adjunct) island ziµlat la>annuh [talib-[ha]i l-kassul]j l-mudiira student-her the-bad the-principal upset.3sf because [kul mµalmih]i šafat-uhj /-uh huj γaš bi-li- mtiħan every teacher saw.3sf-CL/-CL he cheated.3sm in-the-exam “Her bad student, the principal got upset because every teacher saw him cheating in the exam.” 1
A left-dislocated DP in JA can be resumed by four types of resumptive elements: epithets, strong pronouns, weak pronouns (clitics) and doubled clitics, i.e., clitics doubled by strong pronouns.
92
NOUMAN MALKAWI & NICOLAS GUILLIOT
(19) Clitic/doubled clitic inside weak (wh-) island [talib-[ha]i l-kassul]j ma bad-ku tis>alu [ wala mµalmih]i student-her the-bad Neg. want.2p. ask no teacher layš l-mudiira kaħšat-uhj /- uh huj mn l-madrase why the-principal expelled.3sf-CL/-CL he from the-school “Her bad student, you don’t want to ask any teacher why the principal expelled him from school.” The examples in (18) and (19) can both have functional readings. In both examples, the possessive pronoun -ha ‘her’ in the left dislocated DP can be interpreted under the scope of the universal quantifier (QP), leading to the functional reading mapping each teacher to a different student. This can be obtained if we suppose that the left-dislocated DP is reconstructed within the scope of QP. Thus, BVA would be satisfied, but this means that reconstruction effects can appear when resumption is embedded in an island. These examples clearly contradict ACH’s analysis of true resumption, and more generally, any account of reconstruction as an exclusive consequence of movement. If an XP allows reconstruction within an island, as in (18) and (19), movement cannot be the trigger for it. 3.2 No reconstruction without islands ACH’s analysis makes another prediction a priori, viz. that reconstruction effects should always appear with apparent resumption. Thus, their analysis predicts that, whenever movement is available, reconstruction effects should appear equally with BVA and Condition C. The configuration in (20) is predicted to be illicit, as it would be filtered out by Condition C. (20) Apparent resumption: [Dislocated DP…REj]i [IP…pronounj…[[DP…REj]i RE]i] Co-reference between the RE and the embedded pronoun should be banned as a Condition C violation. The RE within the copy is c-commanded by the pronoun, but this prediction is not valid, as indicated by the grammaticality of the Jordanian Arabic example in (21). (21) [µalamitj kariimi], lazim proi/k >iγayyar-haj Note Karim must he change-it “The note of Karim, he must change (it).” Since there is no island in (21), we conclude that it is a configuration of apparent resumption. In other words, the analysis supposes a copy of the antecedent in the c-command domain of the subject, as shown in (22). ACH
RECONSTRUCTION AND ISLANDHOOD IN JORDANIAN ARABIC
93
predict that reconstruction effects with Condition C should appear, but coreference between the RE kariim and pro is licit in (21) (see Guilliot (2006) for similar data from Breton). The absence of reconstruction with Condition C, when no island intervenes, contradicts ACH’s analysis of apparent resumption2.
(22) [µalamitj kariimi] lazim proi/k >iγayyar - [DP[µalamitj kariimi]haj]] Note Karim must hei change- [DP[Note Karimi] itj] 4.
What really matters for reconstruction In this section, we highlight crucial parameters that play a role with respect to reconstruction data. There is strong evidence that the presence or absence of reconstruction effects in Jordanian Arabic depends on three parameters: the type of resumption, the type of island, and the type of binding condition. 4.1 The type of resumption Jordanian Arabic distinguishes between weak and strong resumptives. Clitics and doubled clitics, which are formed of a clitic doubled by a strong pronoun, are weak resumptives, while strong pronouns and epithets are strong resumptives, as in Table 1.
Weak resumption
Strong Resumption
Clitic
Singular masculine feminine - uh “him” -ha “her”
Plural masculine feminine - hum “them” - hin “them”
Doubled clitic
- uh hu “him he”
-ha hi “her she”
- hum humuh “them they”
- hin hinnih “them they”
Strong pronoun
hu “he”
hi “she”
humuh “they”
hinnih “they”
Epithet (ha)-l -NP (demonstrative)- definite article-NP Table 1: Classification of resumptive elements in Jordanian Arabic
The type of resumptives (weak or strong) plays an important role in allowing or banning reconstruction within islands, as in (23) and (24).
2
This contradiction in ACH (2001) can be overcome if we follow Aoun and Li (2003) who argue for two mechanisms: a head-raising analysis and a matching analysis. The former gives rise to reconstruction with BVA, whereas the latter would account for the absence of reconstruction with Condition C. However, we argue that it cannot account for the reconstruction data with Condition C when strong resumption is at stake (See Section 4.3.2).
94
NOUMAN MALKAWI & NICOLAS GUILLIOT
(23) Clitic/doubled clitic inside an adjunct island maµ [wala mµalmih]i [talib-[ha]i l-kasul]j ma ħakyan student-her the-bad Neg. talked.1p. with no teacher gabl ma tšuf-uhj/-uh huj l-mudirah before saw.3sf-Cl /-Cl he the-director “Her bad student, we didn’t talk to any teacher before the director saw him.” (24) Strong pronoun/epithet inside an adjunct island *[talib-[ha]i l-kasul]j ma ħakyan maµ [wala mµalmih]i student-her the-bad Neg. talked.1p. with no teacher gabl ma huj/ha-l-γabij yesal before he/the-idiot arrive.3sm “Her bad student, we didn’t talk to any teacher before he/this idiot arrived.” Although weak resumption (clitics and doubled clitics) allows reconstruction within strong islands, as in (23), strong resumption (strong pronouns and epithets) bans reconstruction in the same environment, as in (24). The functional reading is not available when resumption is embodied by an epithet or a strong pronoun. 4.2 The type of island: Weak vs. strong The nature of island (weak vs. strong) plays a role in allowing or banning reconstruction, but only with strong resumption. Although reconstruction is banned with strong resumption when embedded within a strong island, as in (24), it surprisingly reappears when strong resumption is embedded in a weak island, as in (25). The reconstructed functional reading is available when strong resumption is embedded in a wh- island. (25) Strong pronoun/epithet inside weak (wh-) island [talib-[ha]i l-kasul]j ma badku tis>alu [wala mµalmih]i student-her the-bad Neg. want.2p. ask.2p. no teacher layš huj/ha-l-γabij γaš b-l-mtiħan why he/the-idiot cheated.3sm in-the-exam “Her bad student, you don’t want to ask any teacher why he/this idiot cheated in the exam.” 4.3 The types of binding condition: Positive vs. negative The distinction between positive binding conditions: (BVA and Condition A, which allow some interpretations) and negative ones (such as Condition C,
RECONSTRUCTION AND ISLANDHOOD IN JORDANIAN ARABIC
95
which excludes some interpretations) plays a crucial role in the availability of reconstruction, at least with weak resumption. 4.3.1 Weak resumption and binding conditions. Whether an island appears in the structure or not, weak resumption always gives rise to reconstruction effects with positive binding conditions (PBC), but never with negative ones (NBC), as in (26). The availability of reconstruction with PBC is shown in the (a) and (b) examples for BVA and Condition A, whereas the absence of reconstruction with Condition C appears in the (c) examples. (26) BVA, Condition A and Condition C with no island (a) [talib-[ha]i l-kasul]j ma biddna nxabbir [wala mµalmih]i student-her the bad Neg. want-1p. tell-1p. no teacher /uh huj mn l-madrase >inno l-mudiirah kaħšat-uhj that the-principal expelled.3sf-Cl /-Cl he from the-school “Her bad student, we don’t want to tell any teacher that the principal expelled him from school.” (b) [surit ħaluhi]j kariimi mazaµa-haj/-ha hij picture himself Karim tore-CL /-Cl she “Picture of himself, Karim tore it.” >innu proi lazim >iγayyar-haj (c) [µalamit kariimi]j bitfakir Note Karim think.2sm that-he must change- it “The note of Karim, you think that he must change (it).” (27) BVA, Condition A and Condition C in weak islands (a) [talib-[ha]i l-kasul]j ma bidd-ku tis>al-u [wala mµalmih]i student-her the-bad Neg. want-2p. ask-2p. no teacher layš l-mudiirah kaħšat-uhj /uh huj mn l-madrase why the-principal expelled.3sf-Cl /-Cl he from the-school “Her bad student, you don’t want to ask any teacher why the principal expelled him from school.” kariimi mazaµa-haj/-ha hij (b) [surit ħaluhi]j sa>alt-u layš picture himself asked-2p. why Karim tore-CL /-Cl she “Picture of himself, you asked why Karim tore it.” (c) [>axu laylai]j proi sa>alat layš l-mudirrah tardat-uhj brother Layla asked.3sf why the-principal expelled-Cl “The brother of Layla, she asked why the director expelled him.”
96
NOUMAN MALKAWI & NICOLAS GUILLIOT
(28) BVA, Condition A and Condition C in strong islands la>annuh [kul mµalmih]i (a) [talib-[ha]i l-kasul]j l-mudiirah ziµlat student-her the-bad the-principal upset.3sf because every teacher šafat-uhj /-uh huj γaš bi-li-mtiħan saw.3sf -CL/-CL he cheated in-the-exam “Her bad student, the principal got upset because every teacher saw him cheating in the exam.” la>annuh kariimi mazaµa-haj/-ha hij (b) [surit ħaluhi]j zµilt-u picture himself upset.2p because Karim tore-CL /-Cl she “Picture of himself, you got upset because Karim tore it.” (c) [>axu laylai]j proi ziµlat la>annuh l-mudirah tardat-uhj/-uh huj brother Layla upset.3sf because the-principal expelled-Cl/-CL he “The brother of Layla, she got upset because the director expelled him.” Whether the structure has no island, as in (26), a weak island, as in (27), or a strong island, as in (28), PBC allows reconstruction with weak resumption, as BVA in the (a) examples and Condition A in the (b) examples seem to be satisfied. However, reconstruction is absent in the same environment when Condition C is at stake, as co-reference in the (c) examples is available. The reconstruction of a displaced constituent with an embedded RE would give rise to a Condition C violation. Reconstruction data with weak resumption are summarized as follows: PBC allows reconstruction with or without island, while NBC does not allow reconstruction. 4.3.2 Strong resumption and binding conditions. Surprisingly, the contrast noticed with weak resumption regarding binding conditions is not present with strong resumption, as PBC and NBC follow exactly the same pattern. In other words, reconstruction effects appear when there is no island, as in (29), or when a weak island intervenes, as in (30), but they disappear when a strong island intervenes, as in (31). (29) BVA and Condition C with no island (a) [talib-[ha]i l-kasul]j ma biddna nxabbir [wala mµalmih]i student-her the bad Neg. want-1p tell-1p. no teacher >inno huj/ha-l-habilihj γaš bi-li- mtiħan that he/this-the-idiot cheated.3sm in-the-exam “Her bad student, we don’t want to tell any teacher that he/the idiot cheated in the exam.”
RECONSTRUCTION AND ISLANDHOOD IN JORDANIAN ARABIC
97
(b) *[>axu laylai]j proi galat >innu huj/ha-l-habilihj safar brother Layla said.3sf that he/this-the-idiot left.3sm “The brother of Layla, she said that he/the idiot left.” (30) BVA and Condition C in weak islands tis>al-u [kul mµalmih]i (a) [talib-[ha]i l-kasul]j ma biddk-u student-her the bad Neg. want-2p. ask-2p. every teacher layš huj/ha-l-habilihj γaš bi-li-mtiħan why he/this-the-idiot cheated.3sm in-the-exam “Her bad student, you don’t want to ask every teacher why he/the idiot cheated in the exam.” l-mµalmih layš huj/ha-l-habilihj (b) *[>axu laylai]j proi sa>al-at brother Layla asked.3sf the-teacher why he/this-the-idiot γaš bi-li-mtiħan cheated.3sm in-the-exam “The brother of Layla, she asked the teacher why he/the idiot cheated in the exam.” The BVA constraint is satisfied in (29a) and (30a). The clitic -ha ‘her’ within the left-dislocated DP has a functional reading, as it can be reconstructed under the scope of the QP wala/kul mµalmih ‘no/every teacher’. Moreover, the same reconstruction effects appear with Condition C, as (29b) and (30b) are ungrammatical with co-reference between the RE Layla and the embedded subject pro, suggesting a Condition C violation. In contrast with contexts with weak or no islands, reconstruction effects do not appear with strong resumption when the latter is embedded in a strong island, as in (31). Notice again that BVA and Condition C follow the same pattern. In (31a), BVA is not available indicating the absence of reconstruction. In (31b), reconstruction does not hold either, as Condition C is not violated. (31) BVA and Condition C in strong islands (a) *[talib-[ha]i l-kasul]j ma ħakyan maµ [wala mµalmih]i student-her the-bad Neg. talked with no teacher yesal gabl ma hui/ha-l-γabij before he/this-the-idiot arrive.3sm “Her bad student, we didn’t talk to any teacher before he/this idiot arrived.” (b) [>axu laylai]j proi ziµlat li>annuh huj/ha-l-habilihj safar brother Layla upset.3sf because he/this-the-idiot left.3sm “The brother of Layla, she got upset because he/the idiot left.”
98
NOUMAN MALKAWI & NICOLAS GUILLIOT
4.4 Summary In Section 4, we discussed the following crucial parameters in relation to reconstruction in Jordanian Arabic: (a) reconstruction with weak resumption is sensitive to the type of binding condition (present with PBC, but absent with NBC), and it is insensitive to islandhood (available with PBC even in strong islands), and (b) reconstruction with strong resumption is sensitive to islandhood (present in weak island contexts and when there are no islands, but absent in strong island contexts), and it is insensitive to the type of binding condition. The results are given in Table 2. Weak Resumption Strong Resumption (Clitic / Doubled Clitic) (Strong Pronoun/Epithet) PBC NBC PBC NBC No island Reconstruction No reconstruction Reconstruction Reconstruction Weak island Reconstruction No reconstruction Reconstruction Reconstruction Strong island Reconstruction No reconstruction No reconstruction No reconstruction Table 2: Reconstruction, resumption, islandhood and binding conditions
5.
Analysis To account for the complex array of reconstruction data in Jordanian Arabic, we argue for an analysis based on the central claim in (32). This claim provides a direct way to distinguish reconstruction with weak resumption from reconstruction with strong resumption. Copies may be provided by two distinct mechanisms: movement and ellipsis. We argue for two claims. First, reconstruction with weak resumption follows from ellipsis, and more precisely from the NP-deletion analysis of pronouns proposed by Elbourne (2001). Second, reconstruction with strong resumption follows from the copy analysis of movement. (32) Whenever an XP allows for reconstruction, a copy (rather than movement) of that XP is present. 5.1 Weak resumptives as definite determiners Elbourne (2001) analyzes third person pronouns as definite determiners, assuming that the structures in (33a) and (33b) correspond to two different interpretations. In (33a), the pronoun takes an NP-complement as an argument, which undergoes NP deletion under identity with a preceding antecedent. In (33b), the pronoun takes only an index variable as an argument. (33a) [[the/it] NP] (33b) [the/iti]
RECONSTRUCTION AND ISLANDHOOD IN JORDANIAN ARABIC
99
We propose to extend this analysis of pronouns to weak resumptives (Clitic and Doubled Clitic). Weak resumptives are thus definite determiners in the sense of Elbourne (2001), taking either the NP-complement argument, as in (33a), or an index, as in (33b), leading to the structures in (34). (34) Weak resumption Clitic DP g D' 3 D° NP/ i -uh
Doubled Clitic DP 3 DP DP g 4 D' hu 3 D° NP/ i -uh
5.1.1 Insensitivity to islands. Since we are dealing with ellipsis, not movement, the insensitivity to islands is now predicted. The fact that reconstruction with weak resumption is allowed within strong islands, as in (28a) and (28b), repeated as (35a) and (35b), now follows, as weak resumptives can be analysed with the structure proposed in (33a). The pronoun takes an NP-complement as an argument, which is deleted under identity with the antecedent NP in an Abar position, hence giving rise to the schemas in (36). (35) BVA and Condition A in strong islands la>annuh [kul mµalmih]i (a) [talib-[ha]i l-kasul]j l-mudiirah ziµlat student-her the-bad the-principal upset.3sf because every teacher /-uh huj γaš bi-li-mtiħan šafat-uhj saw.3sf -CL/-CL he cheated in-the-exam “Her bad student, the principal got upset because every teacher saw him cheating in the exam.” (b) [surit ħaluhi]j zµilt-u la>annuh kariimi mazaµa-haj/-ha hij picture himself upset.2p because Karim tore-CL /-Cl she “Picture of himself, you got upset because Karim tore it.” (36a) [talib-[ha]i l-kasul]j...[kul mµalmih]i...[DP-uh/-uh hu [NP talib[ha]i l kasul]] the bad student of heri…every teacheri….[DP him [bad student of heri] (36b) [surit ħaluhi]j … [kariim]i … [DP-ha/-ha hi [NP surit ħaluhi]] picture of himselfi… Karimi…[DP it [picture of himselfi]
100
NOUMAN MALKAWI & NICOLAS GUILLIOT
In (36a), having an elided copy as an argument of the weak RE accounts for the availability of the functional reading in (36a), i.e., there is a different student for each teacher. The pronoun ha ‘her’ can be interpreted as a variable bound by the universal quantifier, since a copy of it is provided under the scope of that quantifier. The same argument goes for Condition A, as (36b) shows. The anaphor ħaluh ‘himself’ can be interpreted where the elided copy is created, i.e., within the scope of its antecedent Karim. 5.1.2 Sensitivity to the type of binding condition. Our analysis, which is based on the two structures in (33), also accounts for the sensitivity of weak resumption to binding conditions, i.e., the fact that reconstruction is available with PBC, but not with NBC. In contrast with PBC, Condition C does not trigger any reconstruction effects with weak resumption; otherwise, coreference between Layla and the embedded subject in (37) would be banned. la>annuh l-mudirrah tardat-uhj/-uh huj (37) [>axu laylai]j proi ziµlat brother Layla upset.3sf because the-principal expelled-Cl/-CL he “The brother of Layla, she got upset because the principal expelled him.” Weak resumption does not force reconstruction with Condition C, as it can be analysed with the other structure proposed by Elbourne (2001). This structure, applied to (37), gives rise to the schema in (38), where the clitic pronoun -uh ‘him’ does not take an NP-complement as an argument, but an index referring to the displaced constituent >axu layla ‘the brother of Layla’ instead. (38) [>axu laylai]j … proi … [DP -uhj] brother Layla himj Our account of unexpected reconstruction within islands (with PBC) is on a par with Elbourne’s analysis of “paycheck sentences”, such that in (39). The pronoun it in (39) has often been considered problematic for current analyses of pronouns, as it can have a distributive reading (a different paycheck for each person), a reading that is often attributed to bound pronouns. Technically, there is no binding relation in (39) between the universal quantifier and the pronoun; therefore, it is not clear where the distributive reading of the pronoun comes from. This same analysis is based on the NP deletion account of pronouns, as in (40), where the pronoun takes the NP paycheck of him as an argument, which undergoes NP deletion under identity with its antecedent. We get a covariant reading for the pronoun it because the elided copy provides the pronoun him, which can be bound by the universal quantifier.
RECONSTRUCTION AND ISLANDHOOD IN JORDANIAN ARABIC
101
(39) John gave his paycheck to his mistress. Everybody else put it in the bank. (40) John gave [DP the [NP paycheck of him]] to his mistress. Everybody else put [DP it [NP paycheck of him]] in the bank. To summarize, weak resumptives can be interpreted in the same way as co-variant pronouns in “paycheck sentences”. The presence of a bound pronoun within the elided copy gives rise to the co-variant reading of the resumptive pronoun. 5.2 Strong resumption: Reconstruction through movement Reconstruction with strong resumption, in contrast to reconstruction with weak resumption, is sensitive to islandhood, but insensitive to the type of binding condition. To account for these properties, we argue that strong resumption should be derived via movement when available. In other words, we propose that ACH’s distinction between apparent and strong resumption should be preserved, but only for strong resumption. ACH’s analysis of apparent resumption is repeated in (41). (41) [DP …pronouni…]j…[IP… QPi ..[CP..[DP….pronouni..]j -REj]] In (41), the dislocated NP is first merged with the RE in an adjunction structure, and then fronted to an A-bar position. This structure is proposed by ACH for all kinds of resumption when no island intervenes. However, Elbourne (2001) points out that this proposal runs into difficulty, since weak pronouns can never be cliticized onto DPs on the surface3. The sentence in (42) is ungrammatical with the RE adjoined to a clitic pronoun, but both strong pronouns and epithets can appear overtly adjoined to a DP, as in (43) and (44). (42) *kariim-uh illi šuft-uh Karim-Cl that saw.1s.-Cl “Karim that I saw is dead.”
mat dead
(43) hu kariim illi šuft-uh he Karim that saw.1s.-Cl “Karim that I saw is dead.”
mat dead
(44) šuft kariim ha-l-habilih saw.1s. Karim the-idiot “I saw Karim, the idiot.” 3
This observation is credited to Benmamoun and Choueiri (personal communication).
102
NOUMAN MALKAWI & NICOLAS GUILLIOT
The contrast between (42) and both (43) and (44) gives strong support to our claim that ACH’s notion of apparent resumption should be restricted to strong resumption. We, therefore, adopt ACH’s distinction between apparent and true resumption, but only for strong resumptives. This proposal accounts for the behaviour of strong resumption with respect to reconstruction, i.e., the fact that reconstruction with strong resumption is sensitive to islandhood (blocked by strong islands), but insensitive to the type of binding conditions, as PBC and NBC follow the same pattern. 5.2.1 Sensitivity to islands. Sensitivity to islands is predicted, as reconstruction is available when no or a weak island intervenes but not when a strong island intervenes in the derivation. This contrast is predicted as reconstruction follows from movement, and more precisely adjunction of a copy of the displaced material to the strong resumptive. Reconstruction with strong resumption is now expected whenever movement is available, as it gives rise to the schema in (45), based on the notion of apparent resumption. A-bar Movement
(45) [talib-[ha]i l-kasul]j…QPi ..[no/weak island…[[ DP talib[ha]i l-kasulj] strong RPj]] With strong resumption, BVA is satisfied by the presence of a copy created by movement and adjoined to the resumptive element, as in (46). The interpretation of that copy gives rise to the reconstructed reading. As predicted, introducing a strong island in the derivation, blocks this mechanism of reconstruction, leading to true resumption, as in (47). BVA cannot be satisfied and the reconstructed reading vanishes. Antecedent
Epithet
(46) *[talib-[ha]i l-kasul]j … [strong island…QPi [DP ha-[NP l-γabij]] Antecedent
Strong pronoun
(47) *[talib-[ha]i l-kasul]j …[strong island…QPi [DP h-D° [NP i/u] j] 5.2.2 Insensitivity to the type of binding condition. As reconstruction with strong resumption follows from movement, we also account for the fact that Condition C follows exactly the same pattern as PBC. Reconstruction applies when movement is available, but not with strong islands. Consider the data in (48a)-(48c). >innu huj/ha-l-habilihj safar (48a) * [>axu laylai]j proi galat brother Layla said.3sf that he/this-the-idiot left.3sm “The brother of Layla, she said that he/the idiot left.”
RECONSTRUCTION AND ISLANDHOOD IN JORDANIAN ARABIC
103
(48b) * [>axu laylai]j proi saµal-at l-mµalmih layš huj/ha-l-habilihj brother Layla asked.3sf the-teacher why he/this-the-idiot γaš bi-li-mtiħan cheated.3sm in-the-exam “The brother of Layla, she asked the teacher why he/the idiot cheated in the exam.” (48c)
[>axu laylai]j proi ziµlat li>annuh huj/ha-l-habilihj safar brother Layla upset.3sf because he/this-the-idiot left.3sm “The brother of Layla, she got upset because he/the idiot left.”
Both (48a) and (48b) ban co-reference between the RE and the embedded subject, suggesting a Condition C violation. Our account of reconstruction with strong resumption straightforwardly accounts for this Condition C effect, as both cases involve apparent resumption, leading to the schema in (49). Movement => Condition C violation
(49)
[>axu laylai]j….proi… [no/weak island [>axu Laylai] Strong RPj]] brother of Layla brother of Layla
The movement of the left-dislocated DP leaves a copy adjoined to the strong resumptive, leading to a Condition C violation. Again, inserting a strong island reverses the judgement, as co-reference is now available in (48c), suggesting the absence of reconstruction. The absence of Condition C violation follows from the fact that true resumption is at stake: movement is blocked by strong islands, resulting in the absence of any copy to trigger a Condition C violation. Notice that we assume that strong resumptives cannot be treated like weak resumptives, i.e., as definite determiners taking either an index or an elided NP as an argument. The reason for this distinction might be found in the fact that strong resumption involves more complex internal structure, as independently suggested by Benmamoun (2000) and ACH (2001). They indeed propose the structures in (50) for strong resumption. (50) Strong resumption Strong pronoun DP 3 D' h3 D° NP 4 [Ф-morpheme]
Epithet DP 3 D' ha3 D° NP 4 lγabi
104
NOUMAN MALKAWI & NICOLAS GUILLIOT
Both cases of strong resumption are full DPs, banning any elided NPs as complements. It is obvious for epithets, as the NP-complement is given, but it is assumed for strong pronouns, where the pronominal -h, which bears Фfeatures, occupies the specifier position of that DP, number and gender features being provided by the Ф-morpheme, which is generated as an NP complement. 6.
Conclusion This paper shows that current analyses of reconstruction with resumption cannot account for data in Jordanian Arabic, particularly the fact that reconstruction can occur within strong islands when weak resumption is used. We argue for a fine-grained analysis of reconstruction with resumption, based on two claims: (a) the reconstruction effect of an XP signals the presence of a copy of that XP, and (b) reconstruction with weak resumption follows from ellipsis; more precisely from NP deletion of pronouns following Elbourne (2001). Reconstruction with strong resumption follows from A’-movement, leading to apparent resumption if movement is available, and true resumption if it is blocked.
REFERENCES Aoun, Joseph, Lina Choueiri & Norbert Hornstein. 2001. “Resumption, Movement, and Derivational Economy”. Linguistic Inquiry 32.371-403. ---------- & A. Li. 2003. Essays on the representational and derivational nature of grammar: The diversity of wh-constructions. Cambridge: MIT Press. Benmamoun, Elabbas. 2000. The feature structure of functional categories: A comparative study of Arabic dialects. Oxford: Oxford University Press. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Demirdache, Hamida. 1991. Resumptive chains in restrictive relatives, appositives and dislocation structures. PhD diss., MIT. Elbourne, Paul. 2001. Situations and individuals. Cambridge, MA: MIT Press. Fox, Danny. 2000. Economy and the semantic interpretation. Cambridge, MA: MIT Press. Guilliot, Nicolas. 2006. “A Top-Down Analysis of Reconstruction”. Lingua 116.1888-1914. Malkawi, Nouman. 2004. Sur la syntaxe de quelques expressions anaphoriques: Epithètes et pronoms résomptifs. Mémoire de DEA, Université de Nantes. Sells, Peter. 1984. Syntax and semantics of resumptive pronouns. PhD diss., University of Amherst.
AN ARABIC WACKERNAGEL CLITIC?
THE MORPHOSYNTAX OF NEGATION IN PALESTINIAN ARABIC∗
Frederick M. Hoyt University of Texas at Austin
1.
Introduction Many dialects of Arabic express negation with a combination of the morphemes maa- and -R. This paper studies how cognates of these morphemes are used in Palestinian Arabic (PA). I argue that in PA, maa- and -R are nd SPECIAL CLITICS (Zwicky & Pullum 1983) and that -R is a 2 -position clitic (Wackernagel 1893). 1.1 Data sources The data used in this study are from the following sources: (1) a. a two-volume collection of folktales collected in 1910 (Schmidt & Kahle 1918, 1930: hereafter SK18 and SK30, respectively); b. theoretical work (Awwad 1987; Mohammad 1998, 2000); c. internet data containing Palestinian-specific isoglosses such as iRi ‘anything’ (identified with ‘WWW’); d. the Levantine Arabic QT Training Data Set 4 from the Linguistic Data Consortium (LDC2005S14); e. examples elicited from native speakers (identified as ‘Elicited’) The data from SK18 and SK30 were collected in 1910 and so are nearly 100 years old1. For this reason, data from both the older and contemporary sources have been verified with native speakers. As such, any data included from the 1910 sources are in accord with contemporary intuitions and usage. A note on transcription is in order: the conventions used here are based on source texts and on impressionistic transcription of elicited data. Transcription ∗
Thanks to Ghassan Hussein-Ali for his help with Palestinian data, and to Peter Abboud, Abbas Benmamoun, Kristen Brustad, Mona Diab, Mushira Eid, Nizar Habash, Ernest McCarus, Mustafa Mughazy, Jerry Sadock, Usama Sultan, and other participants of ALS 20 for their comments.
1
Data from SK18 and SK30 are cited according to text and paragraph. For example, SK18:§1.1 is the first paragraph of the first text in (Schdmidt & Kahle 1918).
106
FREDERICK HOYT
of internet data approximates the orthography used in the source document. However, PA is a network of speech varieties differing between regions and socio-economic strata. For this reason, the transcriptions given vary in terms of certain phonemes. In particular the phoneme /q/ is given as [kfl] for the SK data to reflect the pronunciation in this dialect, in which /q/ is pronounced as [k]. Likewise, elicited data from speakers of urban dialects have [>] for /q/. Data from electronic sources are shown with the standard [q]. Likewise, the dialect depicted in the SK data substitutes the voiceless palatal affricate [tR] for the phoneme /k/. This is shown in the transcriptions. The paper is organized as follows: Section 2 describes how maa- and -R are used to form negative sentences in PA; in Section 3 the implications of the comparison for theoretical approaches to Arabic negative sentences are discussed. Section 4 concludes. 2.
Negation in Palestinian Arabic Like many dialects of Arabic, PA uses the morphemes maa- and -R in various permutations to express sentential negation, as in (2a)-(2c). (2a) miR raaħ aktib kull laħðfla not fut. write every moment “I’m not going to write every moment.” (WWW) (2b) wallaahi ma-nii-R µaarif innu heyk b-ysfliir by-God not-1s-neg know.actpart.sm that that happen.3sm “By God, I didn’t know that that happens.” (WWW) (2c) ma-ħabbeyt-iR azµaǰ-ak not-liked.1s-neg annoy.1s-you “I didn’t like to annoy you.” (WWW) However, in PA, either maa- and -R can be omitted in certain contexts (SK18; Blau 1960; Awwad 1987): (3a) bass al-Ruγla zeiy heyk, maa-bidd-i >iyyaa-ha but the-work like this not-want.1s obj-it “...but work like this, I don’t want it.’’ (WWW)
NEGATION IN PALESTINIAN ARABIC
107
(3b) >aa, bidd-haa-R tiħki maµ-ak yeah, want.3sf-neg speak.3sf with-you “Yes, she doesn’t want to speak with you.” (LDC2005S14: fsa_25620:246.88) Which is used seems to have to do with prosody and speaker choice (SK18: 93; Blau 1960:193). In morphological terms, maa- and -R are special clitics (Zwicky & Pullum 1983) because they have the following properties: (4) a. b. c. d.
they are affixes; they unselective: they attach to words of different lexical classes; they attach to words already hosting other clitics; their distribution is influenced by idiosyncratic non-syntactic factors.
These properties are examined in the remainder of Section 2. 2.1 Affixal properties Both maa- and -R are affixes because they trigger word-internal phonological interactions between stress placement and vowel length (Brame 1971; Kenstowizc & Adbul-Karim 1980; Younes 1995). First, maa- is pronounced with a long vowel when stress falls on it, as the case when it is preceded by one of the adverbial expressions wallaahi ‘by God!’ or µumr- ‘ever, never’ (Blau 1960), as in (5a) and (5b). (5a) [wal.la9.hi !ma9.RTf.tu] by-God not-see.perf.1s-him “By God I didn’t see him!”
(5b)
[µTm.ri !ma9.RTf.tu] ever-me not-saw.1s-him “I didn’t ever see him.”
The use of these expressions coincides with focus intonation on the negation particle. The use of -R is rare or unacceptable in such cases. If stress falls later in the word, the [a] in maa- is pronounced short: (6a) [ma.RTf.!tu9.R] not-see.perf.1s-him-neg “I didn’t see him.”
(6b)
[ma.ħa.!ke9.tǺl.hTm] not-tell.perf.1s-to-them “I didn’t tell them.”
Similarly, -R closes word-final syllables, blocking a constraint in the Levantine dialects that shortens long vowels in word-final open syllables (Younes 1995). For example, the object clitic -ni ‘me’ has an underlyingly long vowel /-ni:/ that is pronounced as short [-ni] word finally, as in (7a).
FREDERICK HOYT
108
Closure of the syllable with -R and the emergent length of the vowel create a super-heavy syllable that attracts stress, as in (7b) (Brame 1971; Kenstowizc & Abdul-Karem 1980). This shows that -R is like object clitics in closing wordfinal syllables. (7a) [bǺt.!ħǺbb.ni] love.3sf-me “She loves me.”
(7b)
[bǺt.ħǺbb.!ni:-R] love.3sf-me-neg “She doesn’t love me.”
In contrast, stem-final long vowels are pronounced as short vowels in word-final open syllables, even in close phrase groups such as the construct state possessive construction. For example, >abu ‘father’ has an underlying long final vowel /abu:/. In (8a) and (8b) it occurs in construct with l-banaat ‘the girls’. Since the two words are in a close phrasal group, resyllabification applies across word boundary, causing the article on l-banaat to close the final syllable of >abu. Nonetheless, the /u:/ is pronounced short and stress remains on the initial syllable: This is because syllabification is a phrasal phenomenon while stress placement is purely word-internal. (8a) [>a.bul.bæ.!næ9t]
(8b)
*[>a.!bu9l.bæ.!næ9t]
In contrast, addition of a possessive clitic to >abu either closes the final syllable or adds an additional syllable to the word. In either case, the stem final [u:] is pronounced long and attracts stress: (9a)
*[!>a.bu.k ]
(10a) *[!>a.bu.hTn]
(9b)
[>a.!bu:k]
(10b) [>a.!bu:.hTn ]
Because stress placement is a word-internal process, these data show that clitics form part of the word that they are attached to. The fact that -R causes final vowel lengthening and stress shift indicates that it is also a clitic and therefore is part of the word to which it attaches. In sum, both maa- and -R are affixes. Assuming that the PROSODIC WORD (Selkirk 1980) is the domain to which vowel-shortening and stress placement apply, then maa- and -R form prosodic words with their host. However, despite being affixes their distribution within a clause is largely determined in terms of syntactic position.
NEGATION IN PALESTINIAN ARABIC
109
2.2 Distribution of maaThis section begins with a look at the position of maa- relative to the LEFTPERIPHERY of the clause (Rizzi 1997). The left-periphery is a set of positions occupied by clitic-left-dislocated NPs, fronted constituents, and question words. For descriptions of the form and function of the left periphery in PA and other dialects, see (Blau 1960:204-206), (Cowell 1964:429-435), (Brustad 2000:Ch.10), and (Holes 2004:257-264). For theoretical approaches see (Demirdache 1991; 1997), (Lalami 1996), (Aoun & Benmamoun 1998), (Doron & Heycock 1999), (Aoun, et al. 2001), and (Alexopoulou, et al. 2004). For expository convenience, I assume that a clause containing leftperipheral elements is labeled CP, and that the left-periphery is outside an IP constituent containing the clausal predicate and tense-aspect-mood marking (Mohammad 2000). Clitic-left-dislocation involves NPs appearing in a position outside of the IP-constituent from where they bind a resumptive pronoun inside the IP (indicated in the schemata with subscripts). Fronting involves a constituent of any category being moved to a position immediately to the left of the IP and leaving a trace or gap in the position in which it is interpreted: (11)
[CP (NPi)* [C’ XP/Q-word [IP …proi…tj… ] ] ]
A sentence can contain multiple clitic-left-dislocated NPs (indicated by the Kleene-star on NP in (11)) and a single fronted constituent (including question words). A clause that lacks either clitic-left-dislocated or fronted elements is assumed to project just an IP node (Aoun & Benmamoun 1998; Aoun et al. 2001; Alexopoulou et al. 2004). In general, maa- appears to the right of left-peripheral elements. For example, in (12), maa- follows the clitic-left-dislocated NPs >ana ‘I’ and haldiin il-µwaaǰ ‘this crooked religion’: (12a) >ana hal-diin l-iµwaaǰ ma-bidd-i yyaa I this-religion the-crooked not-want-me obj-it “[As for] me, this crooked religion, I don’t want it.” (SK30:§) (12b) [CP [NPi >ana] [. [NPj hal-diin l-µwaaǰ ] [IP ma-bidd-ii iyyaaj ] ] ] In (13), the fronted question word lēR ‘why’ precedes maa-: (13a) lēR ma-ǰaawabt µala l->as>ila why not-answered.3sm upon the-questions “Why didn’t you answer the questions?” (WWW)
110
FREDERICK HOYT
(13b) [CP lēRi [IP ma-ǐaawabt µala l->as>ila ti ] ] In (14) maa- follows clitic-left-dislocated NP l-muγaariba wa-l-tuwaanisa ‘the Moroccans and the Tunisians’ and a fronted adjective phrase >aħsan minhum ‘better than them’: (14a) wa-l-muγaariba wa-l-tuwaanisa >aħsan min-hum maa-fii and-the-Moroccans and-the-Tunisians better from-them not-exist “and the Moroccans and the Tunisians, there’s none better than them!” (WWW) (14b) [CP[NPil-muγaariba wa-l-tuwaanisa][[APj >aħsan min-hum] [IP ma-fii tj]]] Native speakers reject examples in which maa- precedes left-peripheral elements: (15a) lēR ma-µaad ħada radd µalai-y why not-return.3sm one.sm answered.3sm upon-me “Why didn’t anyone answer me anymore?” (WWW) (15b) * ma-µaad ħada radd µalai-y? not-return.3sm one.sm answered.3sm upon-me “Did anyone answer me anymore?” (Elicited) These data suggest that maa- cannot attach to a word which is any further to the left of the clause than the left-edge of the IP-string: (16)
Generalization 1: maa- must appear no further left than the left edge of the IP-string.
Generalization 1 suggests that maa- is attached to a sub-constituent of IP. The question then becomes what position maa- takes relative to IP-internal elements. This is considered in 2.3. 2.3 maa- attaching to verbal elements With respect to IP-internal elements, maa- often attaches to the main verb in clauses with simplex tense-aspect structure: (17a) lamma faaq ma-ħakaa-l-ii-R subaaħ il-xeer when awoke not-said.3sm-to-me-neg morning the-good “When he woke up he didn’t tell me ‘Good Morning.’” (WWW)
NEGATION IN PALESTINIAN ARABIC
111
(17b) ǰuµaana, maa-kalt-iR iRi l-yoom hungry.sf not-ate.1s-neg thing the-day “[I’m] hungry! I haven’t eaten anything today.” (WWW) With compound tense-aspect, maa- attaches to the left-most auxiliary: (18a) abū-y u-µamm-i [IP ma-bakaa-R yiǰii-him ulaad] father-me and-uncle-me not-was.3sm-neg come.3sm-them children “My father and my uncle, they hadn’t had any children.” (SK18:§51.9) (18b) wallaahi haaði l-luγa l-gadiida ma-kunt-iR aµrif-ha by-God this the-language the-new not-was-neg know.1s-it “By God, this new language, I didn’t know it.” (WWW) (18c) [IP maa-kaam-iR yaµtii min ǐraab-e abadan] not-stood.3sm-neg give.3sm from pocket-his ever “He didn’t ever give him [anything] from his pocket.” (SK18:§85.3) (18d) [IP ma-raaħ yiðall wala filistiini fi-l-balad] not-fut remain.3sm even.one Palestinian in-the-country “There won’t be a single Palestinian left in the country.” (WWW) However, some auxiliaries, including kaan-yikūn ‘be’ and the SERIAL µad ‘again’ and kflaam ‘so, thereupon’ sometimes precede negation2 (Blau 1960; Husseini 1990; Mitchell & Al-Hassan 1994):
AUXILIARIES
(19a) law maa-fii Rabaab aw maa-fii banaat kaan maa fii ħayya if not-exist boys or not-exist girls was.3sm not exist life “If there were no boys or no girls there wouldn’t be life.” (WWW) (19b) sūri µad ma->akdar aradd µalē-kum bi-surµa sorry anymore not-be-able.1s answer.1s upon-you with-speed “Sorry, I can no longer answer you quickly.” (WWW) If these auxiliaries form part of the IP-constituent, then the examples in (19) indicate a class of exceptions to Generalization 1 in which maa- appears
2
Mitchell and Al-Hassan (1994:77) claim that, in both Egypt and the Levant, serial auxiliaries are not negated: >aam raaħ ma-kal-R ‘suddenly he refused to eat’.
112
FREDERICK HOYT
after the first word in the IP-string rather than at its left edge. This suggests the following modification of Generalization 1: (20)
Generalization 1': maa- must appear no further left than the left edge of the IP-string, except when preceded by an auxiliary verb;
2.4 maa- attaching to non-verbal elements In addition to verbs, maa- also attaches to certain kinds of non-verbal expressions. These include inflected prepositions, the existential particle fii (itself derived from an inflected preposition), indefinite pronouns, indefinite noun phrases, and the adverb µumr ‘ever, never’: (21a)
haaða bakii-l-e faras ma-l-haa-R uxt this.sm be.actpart.sm-to-him mare.sf not-to-her-neg sister “He had a mare that was without compare.” (SK18:§39.6)
(21b)
ma-fii-R samak fii l-baħr wa->ana sflayyaad not-exist-neg fish in the-sea and-I fisherman “There aren’t [any] fish in the sea and I am a fisherman.” (WWW)
(21c)
lammin istawat atlaµ il-zalame >arbµiin ǰaddaad when ripened.3sf had-climb.3sm the-fellow forty picker µa-ðahir-ha u-ma-ǰaddaad yismaµ la-ǰaddaad takk on-back-it and-not-picker heard.3sm to-picker sound “When it ripened, the fellow had forty pickers climb it, and no picker heard the sound of another.” (SK18§33.9)
(21d)
ma-µumr-ii-R Ruft-u not-ever-me-neg saw.1s-him “I never saw him.” (Elicited)
These are generally single words, meaning that they have atomic (nonbranching) syntactic objects. However, in some cases maa- attaches to some constituents which appear to have branching structure: (22a) kaamat haaði takkat ma-[PP fi-µēn-ha ] balle u-maatat stood.3sf this fell.3sf not in-eye-her drop and-died.3sf “Then she fell without a drop in her eye and died.” (SK18:§45.10)
NEGATION IN PALESTINIAN ARABIC
(22b)
wallaah ma-[PP fi-hal-lēle ] b-anaam µind-ak by-God not in-this-night sleep.1s at-you “I won’t sleep with you this night.” (SK30:§90.6)
(22)
ma-[PP fi-l-yadd ] ħiile not in-the-hand trick “Have no trick in the hand [idiom].” (WWW)
113
The preposition fii- is frequently pronounced as a prefix on the following word, so in these examples the expressions hosting maa- may not be branching at all. Mohammad (1998) reports that prefixing maa- to other branching prepositional phrases is unacceptable: (23a)
mona, ma-µand-ha ktaab Mona not-at-her book “Mona doesn’t have a book.”
(23b) *ma-µand mōna ktaab not-at Mona book “Mona doesn’t have a book.” In other cases, maa- prefixes to an expression preceding the initial verb in the clause (although there is some variation among native speakers as to the acceptability of such examples): (24a) ma-fiR-R kaan µind-na >aiy maqamaat not-exist was.3sm at-cl1P any possessions “We didn’t have any possessions.” (LDC2005S14: fsa18404: 554.27-558.66) (24b) ma-l-iR-R baaki walad not-to-him-neg was.sm son “He didn’t have a son.” Mohammad (1998) presents examples like these as being unacceptable. Mohammad’s examples are from a variety of PA spoken in rural areas of the Galilee region (Mohammad Mohammad, p.c.). It may be that there is variation within regions or varieties of Palestinian Arabic regarding the position of negation relative to auxiliary verbs. As such, the generalizations concerning the position of maa- should be taken as describing the varieties in which examples like (24a-b) are acceptable.
114
FREDERICK HOYT
Generalization 1 is further complicated by certain expressions that can appear on either side of maa-, raising the question of whether they are varying position or whether maa- is. One such expression is the dative clitic preposition l- ‘to’ when it hosts clitic pronouns: (25a) >il-i maa-kaan maqbūl >inn-hum tahaǰǰamu µala >amani to-me not-was.3sm agreement that-they attacked.3mp upon Amani “I had no acceptance for them attacking Amani.” (WWW) (25b) >il-u ma-kaan ulaad to-him not-were.3sm children “He didn’t have children.” Other inflected prepositions cannot precede maa-: (26a) mōna ma-fiR-R µind-ha ktaab Mona not-exist-neg at-her book “Mona doesn’t have a book.” (26b) * mōna, µind-ha ma-fiR-R ktaab Mona at-her not-exist-neg book “Mona doesn’t have a book.” Therefore, l- presents another exception to Generalization 1, suggesting the following refinement. (27)
Generalization 1'': maa- must appear no further left than the left edge of the IP-string, except when preceded by an auxiliary verb or an inflected dative clitic.
The word that seems to precede maa- most frequently is the adverb µumr ‘ever, never’: (28a) ma-µumr-ii-R Ruft-u not-ever-me-neg saw.1s-him “I never saw him.” (Elicited) (28b) µumr-i ma-Ruft-u ever-me not-saw.1s-him “I never saw him.” (Elicited)
NEGATION IN PALESTINIAN ARABIC
115
It frequently appears in what looks like a construct-state possessive with a following nominal that corresponds to the subject of the clause: (29a) [µumr il-masaayib ] ma-saabat-ni ever the-catastrophes not-struck.3sf-me “… never have catastrophes struck me.” (SK18:§62.11) (29b) ana ħaqqaqt-l-ik illi [µumur ħada ] maa-ħilim fii I realized.1s-to-you rel. ever one not-dreamed.3sm in-it “I have made real for you what no one has ever dreamed of.” (WWW) Sometimes µumr hosts a clitic pronoun coreferential with a subject NP: (30a) haaða µumr-u ma-naam bala sirka this.sm ever-him not-slept.3sm without theft “He never went to sleep without stealing [something].” (SK18:§22.2) (30b) µumr-u ma-ħada simiµ µan-hum γēr kull xēr ever-him not-one.sm heard.3sm on-them other-than every good “No one has ever heard about them other than all the best.” (WWW) Other times the pronoun and the subject are not co-referential: (31a) hal-kuliyya µumr-ha ma-kaan fii-ha the-college ever-her not-was.3sm in-her >ansaaf wala µadl la-l-taalib-a impartiality or fairness to-the-student-sf “In this college, there was never justice or fairness for the female student.” (WWW) (31b) µumr-i maa kaan µand-i muRkiila bi-kawn-i filastiiniyya ever-me not was.3sm at-me problem.sf with-being-me Palestinian.sf “I have never had a problem with my being Palestinian.” (WWW) Additionally, µumr can appear without a clitic or possesor NP: (32a) µumr ma-ħad Raaf waǰh-i wa-µaql-u ðall maµ-u ever not-one.sm saw.3sm face-my and-mind-his stayed.3sm with-him “No one has ever seen my face and kept his wits about him.” (WWW)
116
FREDERICK HOYT
(32b) ma-µumr Raddat-ni qussa miθil il-qussa haaði not-ever affected.3sf-me story.sf like the-story this.sf “Never has a story affected me like this story.” (WWW) Adverbial µumr is derived from the noun µumr ‘age’, as in (33a). In its ‘age’ meaning, µumr appears very frequently in construct with a following noun, with the whole expression meaning ‘(in) X’s life’, as in (33b). (33a) baakfli µumr-e yimtRin µiRriin sane be.part.sm age-his perhaps twenty years “It was maybe twenty years old.” (SK18§31.5) (33b) bidd-i atǰawwaz-ha law >axir yōm µumr-i want.1s marry.1s-her if final day life-my “I want to marry her even if it’s the last day of my life.” (WWW) The adverbial use probably developed with ‘X’s life’ in negative sentences where it implies the meaning of ‘ever’: µumr-i ma-kalt-ū-R ‘in my life I have not eaten it’ → ‘I have never eaten it.’ Nominal µumr can precede negation in a left-peripheral position or follow it in an IP-internal position, explaining how it can appear on either side of negation. The “bare” use of adverbial µumr is likely to be a morphological reduction of adverbial µumr in construct that retains the same syntactic distribution as its etymological source. (34)
[CP [NP (µumr-NP)] [S maa-(µumr-NP)...] ] → [CP [NP (µumr)] [S maa-(µumr)…] ]
This suggests that µumr preceding negation is in a left-peripheral position, while µumr following negation is in an IP-internal position. The distribution of µumr is therefore not an exception to Generalization 1. Another complication for Generalization 1 is the position of subjects in the SV word order. There are two ordering possibilities S-Neg-V and Neg-S-V. Which is used depends on several morphological, prosodic, semantic and pragmatic factors. When the subject NP follows negation, it is generally an indefinite noun or a pronoun (Mohammad 1998, 2000). This can be seen in (21c) and (30a) above. Subject NPs in SV order are either definite NPs, or indefinite NPs that are interpreted as “specific” in a widely noted if poorly understood sense (Khan 1988; Mohammad 1998, 2000):
NEGATION IN PALESTINIAN ARABIC
117
(35a) >ana ma-bidd-ii-R aquul-ak Ru >axtaa>-ik fi-l-tasmiim I not-want-me-neg say.1s-to-you what errors-you in-the-design “I don’t want to tell you what your errors [are] in the design.” (WWW) (35b) bass >umm-i ma-bi-taµmil miθil hay il->aRyaa> but mother-me not-make.3sf like these things “But my mother doesn’t make things like these.” (WWW) The two sentences mean different things although they contain the same words, as in (36a) and (36b). (36a) describes a situation in which no one came, whereas (36b) describes a situation in which a particular individual did not come while still allowing that other people might have done so. (36a) ma-waaħad aǰa not-one.sm came.3sm “No one came.” (36b) waaħad maa->aǰa one.sm not-came.3sm “One [person] didn’t come.” As noted above, subjects that precede negation are subject to the same specificity condition that applies to clitic-left-dislocated NPs. Accordingly, Generalization 1 might be taken to imply that the S in a negative sentence with SV word order is not a subject at all (in the sense of occupying a dedicated IPinternal subject position), but rather a left-peripheral element. This is in keeping with a traditional analysis that treats pre-verbal subjects as clitic-leftdislocated NPs that are resumed by the agreement marking on the verb. However, Mohammad (2000) argues in detail that preverbal subjects really are grammatically subjects, meaning that they show the grammatical characteristics of occupying an IP-internal position. According to Mohammad, the subject NPs in (35a) and (35b) would all be in the IP-internal subject position, and therefore the negation marker is not marking the left edge of the IP, contrary to Generalization 1, but rather the left edge of the what one might call the “I'-string”. This would imply yet another refinement of Generalization 1: (37)
Generalization 1''': maa- must appear no further left than the left edge of the IP-string, except when preceded by an auxiliary verb, an inflected dative clitic, or a subject NP.
118
FREDERICK HOYT
Mohammad’s argument raises questions about the positions of other expressions that precede maa-: if maa- can vary its position relative to subject NPs, then it can also vary its position relative to the dative clitic and to µumr. This suggests that Generalization 1 is not correct, as the exceptions to it are systematic and therefore indicative of some other missing generalization. In Section 3, I suggest that a version of Generalization 1 might be correct if the domain in which maa- is located is defined in purely prosodic terms, rather than as a word-string which is isomorphic with the IP-constituent. It was noted above that maa- can be omitted in certain contexts. This is only possible with stems beginning with labial obstruents [b] or [f], and only in the presence of -R. Early 20th-century grammars of Lebanese (Feghali 1928) and PA (SK18; Blau 1960) note reduction of ma- to a- before the b-imperfect: (38a) kflaal a-b-ixuss-nii-R said.3sm not-concerns.3sm-me-neg “He said ‘It doesn’t concern me’.” (SK18§25.8) (38b) kflaalat a-bidd-ii-R axassr-ak said.3sf not-want-me-neg harm.1s-you(sm) “She said ‘I don’t want to harm you’.” (SK30§129.4) Total reduction of maa- is rare in the 1910 data in (SK18) and (SK30), but is more pervasive in contemporary PA. It also occurs with existential fii. This may be the result of analogical extension from verb stems with indicative prefix bi- to [b]-initial stems (such as bidd- ‘want’) more generally and then to stems beginning with labial obstruents, of which PA has only two. When maa- is omitted, -R is still constrained to attach to the word to which maa- would attach if it were present. In other words, -R is constrained by the distribution of maa- even if maa- is not pronounced: (39a) b-ikūn-R fii Ritaa miθl il-iyaam illi raaħ tiǰi be.3sm-neg exist rain like the-days rel. fut come.3sf “There won’t be any rain like the days that are coming.” (WWW) (39b) * b-ikūn fiR-R Ritaa miθl il-iyaam illi raaħ tiiǰi be.3sm exist-neg rain like the-days rel. fut come.3sf “There won’t be any rain like the days that are coming.” (Elicited) (40a) ma-b-ikūn-R fii makaan not-be.3sm-neg exist space “There won’t be any space.” (WWW)
NEGATION IN PALESTINIAN ARABIC
119
(40b) * ma-b-ikūn fiR-R makaan not-be.3sm exist-neg space “There won’t be any space.” (Elicited) This suggests that when maa- is not pronounced, a word-initial labial obstruent can stand proxy for it. Accordingly, clauses in which maa- is omitted are still in keeping with Generalization 1. 2.5 Distribution of -R The -R morpheme is subject to a well-known constraint that requires it to attach to one of a very restricted set of stem types: (41)
i. Verbs ii. Inflected prepositions iii. Existential fii iv. µumr ‘ever’ v. ħada ‘one’
Each of these must already be hosting maa- or begin with a labial obstruent in the left most position in the IP-string (modulo the exceptions noted above). Therefore, -R inherits the positional distribution of maa- and applies only to a subset of it. Except for ħada, each of these kinds of expressions contains a morpheme which expresses person features or which has an etymological source which expressed person features (c.f. Eid 1993; Jelinek 2002): (42)
Generalization 2: -R attaches to a word which is marked with a negation morpheme and which is inflected for person features.
While ħada is not inflected for person features, it is idiosyncratic in being able to host negation. The synonymous waaħad ‘one’ cannot, although it has an otherwise identical distribution: (43a) ma-ħadaa-R ħaka iRi not-one.sm-neg said.3sm thing “No one said anything.” (WWW) (43b) * ma-waaħad-iR ħaka iRi not-one.sm-neg said.3sm thing “No one said anything.” (Elicited)
120
FREDERICK HOYT
(44a) ma-ħada raaħ yanǰaħ not-one.sm fut succeed.3sm “No one is going to succeed.” (WWW) (44b) ma-waaħad raaħ yinǰaħ not-one.sm fut succeed.3sm “No one came.” (WWW) Mohammad (1998) suggests that ħada is a negative polarity item and that it has an “intrinsic” association with negation that lets it host -R. However, while ħada has a negative polarity use, waħad does as well: (45a) ma waaħad b-ifakkir ixatib >aw yitǰawwaz not one think.3sm engage.3sm or marry.3sm >aw Reiy min hal-nuwµ or thing from this-kind “No one thinks [about] getting engaged or getting married or anything of that kind.” (LDC2005S14: fsa25780: 576.11) (45b) >ana µumr-i ma-Ruft waaħad miθlu I ever-my not-saw.1s one like-him “I have never seen anyone like him.” (WWW) ħada can be used as a positive polarity item or as a referential pronoun: (46a) bidd-i ħada aħki maµ-u µaRaan want.1s one speak.1s with-him because ma-fii ħada b-iħki maµ-i not-exist one speak.3sm with-me “I want someone to talk to because there isn’t anyone who talks to me.” (WWW) (46b) il-ħamdu li-llaah saar maµ-i ħada yiRidd maµ-i the-praise to-God began.3sm with-me one.sm stand-firm.3sm with-me “Thanks to God I have someone with me to stand firm with me.” (WWW) This indicates that although ħada is usually used as an NPI and waaħad as a PPI or a referential pronoun, these are tendencies rather than rules. Similarly, if ħada has an association with negation, then µumr should as well, given that the kinds of sentences in which they occur overlap almost
NEGATION IN PALESTINIAN ARABIC
121
completely. However, µumr cannot host -R while ħada can, except in those cases in which µumr hosts a clitic pronoun and is therefore marked with person features. A possible explanation for the exceptional ability of ħada to host -R is that it is a pronoun and belongs to the determiner (D) category, while waaħad is a noun stem. This difference would be supported by the fact that waaħad can host the definite article while ħada cannot: (47a) il-waaħad the-one “the one”
(47b) * il-ħada the-one “the one”
This follows if ħada and the definite article are both members of category D and therefore in complementary distribution. Another possibility is that ma-ħada is actually a compound comparable to English no-one or nobody. Arabic has a number of negative compounds, including the so-called pronouns of negation (Awwad 1987; Mohammad 1998) found in most dialects of Arabic (Eid 1993; Brustad 2000; Jelinek 2002) and ma-µad ‘no longer’. The question of which of these possibilities is more correct is beyond the scope of this paper. However, either would imply that the ability of ħada to host -R is not an exception to Generalization 2. If ħada is treated as a pronoun, then Generalization 2 can be refined to say that -R must be right-adjacent to a pronoun or to a morpheme marked with person features (Eid 1993; Jelinek 2002). Pronouns are necessarily marked for person, so the second possibility implies the first and is therefore more general. (48)
Generalization 2': -R must attach to the right edge of a word which is marked with a negation morpheme as well as a morpheme expressing person features.
On the other hand, if ma-ħada is treated as a compound, then Generalization 2 can be retained in its original form. Generalization 1 as given does not exclude -R attaching to nouns hosting possessive clitics, since these are word-sized constituents and the possessive clitics express person features: (49a) >ibn-u miR mniiħ son-his not good “His son isn’t good.” (WWW)
122
FREDERICK HOYT
(49b) * ma-bn-ū-R imniiħ not-son-his-neg good.ms “His son isn’t good.” (Elicited) However, the unacceptability of examples like (49b) is not an exception to Generalization 2 if we follow Benmamoun (2000) in assuming that pronouns belong to a +D category and by treating agreement morphology as expressing a +D categorial feature. Generalization 2 can then be further refined as follows3: (50)
Generalization 2''': -R must attach to the right edge of a +D word that is marked with a negation morpheme and that expresses person features.
Because the distribution of -R is a subset of the distribution of maa-, Generalization 2''' inherits the various exceptions to Generalization 1. 2.6 Summary The distribution of maa- and -R in PA is as follows: (51)
maa- and -R are special clitics (Zwicky 1977; Zwicky & Pullum 1983): a. They are affixes, forming prosodic words with their hosts; b. They unselectively attach to words from several different classes; c. They attach to words already hosting other clitics; d. Their distribution is influenced by non-syntactic factors.
(52)
maa- attaches to the left-most word in the IP-string except when preceded by: a. a subject NP; b. kaan-yikuun ‘be’, µaad-yµuud ‘again’, qaam-yiquum ‘so then’; c. the adverb µumr ‘ever’; d. the dative preposition l- hosting a clitic pronoun.
(53)
-R is a phrasal enclitic which attaches to the following provided that they are hosting maa- or begin with a labial obstruent: a. ħada ‘(any)one’; b. stems marked with person agreement features.
3
This solution would entail treating construct-state noun phrases as being of category -D. This would be a theoretically controversial assumption to make.
NEGATION IN PALESTINIAN ARABIC
123
The distribution of -R is therefore conditioned by the distribution of maa-. Given that maa- generally attaches to the left-most word-sized constituent in the IP-string, it follows that -R attaches to the end of the left-most word-sized constituent in the IP-string. Therefore -R has a tendency to appear as a 2ndposition clitic in the IP-string, where positions are understood in terms of prosodic words. This tendency is obviated in sentences in which the word hosting -R is not the first word in the IP-string, but rather the 2nd. This raises the question of whether the IP-string is the correct characterization of the phrasal domain to which maa- and -R attach, or whether the phrasal domain should be characterized in prosodic terms without reference to syntactic categories such as IP. This is discussed briefly in Section 34. 3. Analytical approaches 3.1 Previous approaches Perhaps the most widely adopted strategy for analyzing negation in Arabic clauses follows Pollocks’s (1989) analysis of French negation (Benmamoun 1992, 1997, 2000; Ouhalla 1993, 2002). According to this approach, maaheads a functional projection NegP that immediately dominates the verbal complex, with -R filling the specifier of NegP. The main verb raises to adjoin to maa-, and then further to I0, “stranding” -R in the specifier of NegP, deriving the desired word order: (54)
[IP NP [I' [I ma- verb ] [NegP -R [Neg' tverb [VP tNP tverb ] ] ] ] ]
In a clause with a compound tense-aspect structure, the auxiliary verb originates in a functional projection below NegP and then raises to Neg0 and on to I0, once again deriving the desired morpheme ordering: (55)
[IP NP [I' [I ma- AUX ] [NegP -R [Neg' tma-aux [AuxP taux [VP tnp tverb ] ] ] ] ]
The Pollock-type approach successfully models examples in which maaand -R attach to the tensed verb (see 17, 18, and 19 above) given the assumption that tense-aspect-mood marking occurs on I0. However, this fails to predict the positions of the negation morphemes when they attach to a pre-verbal word such as ħada, inflected prepositions, or µumr. A similar problem arises with the “serial auxiliaries” noted above. These are a class of auxiliated verb stems used in PA and other Levantine dialects
4
For reasons of space, the negative auxiliary miR ‘not’ and the negative pronouns ma-nii-R ‘I’m not’, ma-huu-R ‘he’s not’ are not discussed here.
FREDERICK HOYT
124
essentially as aspectual adverbs (Blau 1960; Hussein 1990; Mitchell & alHassan 1994; Khalaily 1997). Stem
Lexical meaning
Auxiliary meaning
Negated meaning
kaam-ykuum µaad-yµuud, µaawad-yµaawid raaħ-yruuħ >aǰa-yiiǰi
‘rise, stand’ ‘return’
‘so then, and then, so’ ‘again’
‘ever, at all’ ‘anymore’
‘go’ ‘go to do X’ ‘come’ ‘come to do X’ Table 1: Serial auxiliaries in Palestinian Arabic
-
In non-negative sentences, serial auxiliaries precede the tensed verb and agree with it in tense-aspect-mood form as well as in person, number, and gender. This gives them the appearance of being tensed verbs: (56a) kaamat θaani ǰimµa µaawadat ðabħat-l-e wazze stood.3sf second Friday returned.3sf slaughter.prf.3sf-to-him goose “Then the second Friday she slew a goose for him again.” (SK§60.4) (56b) qaam raǰaµ naam stood.3sm returned.3sm slept.3sm “Then he went to sleep again.” (WWW) Serial auxiliaries are marked as expressing tense or aspect, but are interpreted as adverbial modifiers or as conjunctions. Because they neither contribute tense information nor have the distribution of a tense head, I treat them as adjuncts which adjoin to the projection of I0 and which agree with I0 in terms of its inflectional features. For example, the derivation of (56b) would have a structure like the following (ignoring the time adverbial θaani ǰimµa ‘the second Friday’). (57)
[IP qaam [IP raǰaµ [IP naam ] ] ]
Additional grammatical mechanisms would have to be invoked to ensure that the serial auxiliaries concord with the main verb in tense-aspect form and in subject agreement marking. In negative sentences with serial auxiliaries, the main verb is more frequently in the imperfect: (58a) ma-kaam-iR yixllii-hin yitlaµin not-stood.3sm-neg allowed.3sm-them go-out.3fp “He never let them venture out.” (SK§46.1)
NEGATION IN PALESTINIAN ARABIC
125
(58b) ma-µadt-R tiµrif iRi not-returned.2sm know.2sm thing “You don’t know anything anymore.” (WWW) However, there are rare instances in which the main verb is in the same tense-aspect form as the serial auxiliary: (59a) u-ma-µaawadat-iR baiyanat and-not-returned.3sf-neg was-clear.3sf “...and it was no more to be seen.” (SK§64.3) (59b) ma-µad-R >al-l-i >innu Rtara sayyara not-returned.3sm-neg said.3sm-to-me that-he bought.3sm car “He no longer told me that he bought a car.” (Husseini 1990:344) Given that the main verb expresses the tense-aspect information for the clause, I assume it to be in the I0 position. This entails that the serial auxiliary is attached above it, and hence that the negation marker is as well. If serial auxiliaries are adjuncts, then a Pollock-style analysis would incorrectly predict that these examples would be unacceptable because the main verb would be predicted to host negation by virtue of raising through the Neg projection. In sum, an approach to modeling PA negative sentences that follows Pollock (1989) incorrectly predicts that maa- and -R can only attach to the verb stem occupying the I0 position in the clause. 3.2 Strategy two Another analysis proposed for negative sentences in dialectal Arabic places the negation marker in a functional projection which dominates the IP constituent in the clause (Diesing & Jelinek 1995; Shlonsky 1997; Jelinek 2002): (60a) [FP ma- [IP [I VERB I ] [VP pro tverb (OBJ) ] ] ] (60b) [FP ma- [IP [I AUX I ] [AuxP taux [VP pro tverb (OBJ) ] ] ] ] This analysis correctly predicts a wider range of facts than does the Pollock-style analysis, in particular predicting Generalization 1, but makes no predictions about the distribution of the -R morpheme. The distributions of maa- and -R can be schematized in Table 2.
126
FREDERICK HOYT
a. maaverb -R b. maaaux -R verb/aux c. maaP-cl -R verb/aux d. maafii -R verb/aux e. ma-ħada -R verb/aux f. ma-µumr -R verb/aux g. ma-µad -R verb/aux Table 2: Distributions of maa- and -R
Items (a) and (b) in Table 2, in which maa- and -R attach to a verb or auxiliary, could be captured in several ways, depending on one’s assumptions about the position of the verb itself. If one assumes that the verb raises to I0, then one could stipulate that -R is the head of I0, as in (61a). Alternately, one could claim that the verb raises to F0 (Diesing & Jelinek 1995), and therefore that maa- and -R are both in F0, as in (61b): (61a) [FP ma- [IP [I0 VERB -R ] [VP pro (OBJ) ] ] ] (61b) [FP [F' ma-_-R [I0 VERB I0 ] ] [IP [VP pro verb> (OBJ) ] ] ] For (c)-(f) in Table 2, in which the negation morphemes are hosted by an expression to the left of the tensed verb, it will not do to place -R in either I0 or F0. This is because the word in these cases are not verbal heads but rather phrasal categories such as PPs or NPs that do not adjoin to F0. To capture (c)-(f) in Table 2, one might claim that the linear order of -R and the verbal head is left unspecified in the syntax, so that the morphophonological grammar will make -R branch to the left when attached to preverbal elements, as in (62a), and to the right when attached to verbal elements, as in (62b). (62a) [FP ma- [IP ħada [I' [I0 -R VERB ] [VP pro (OBJ) ] ] ] ] (62b) [FP ma- [I' [I0 VERB -R ] [VP pro (OBJ) ] ] ] However, -R can attach to an expression that is separated from I0 by an intervening XP. In (63), -R is attached to ma-ħada which is then followed by a prepositional phrase. The PP has the semantics and distribution of an NPinternal modifier and separates -R from I0: (63a) haði l-as>ila ma-ħadaa-R min µumr-i these the-questions not-one-neg from age-my yiqdar yiħill-l-i yyaa-ha can.3sm solve.3sm-to-me obj- them “These questions, no one of my age can answer them for me.” (WWW)
NEGATION IN PALESTINIAN ARABIC
127
(63b) ma-ħadaa-R min il-lusūs illi ħakamū-na not-one-neg from the-thieves rel. ruled.3mp-us tiliµ µala l-maµaaR went-out.3mp upon the-pension “Not one of the thieves who ruled us went into retirement.” (WWW) Assuming that the PP is internal to the NP headed by ħada, then -R must also be internal to the NP: (64)
[FP ma- [IP [NP [N ħadaa -R ] [PP min-hum ] ] [I' kaan [VP µind-u flūs ] ] ] ]
If this is the correct structure for examples like (63a), then a constraint on the distribution of -R cannot refer to the spine of the clause. 3.3 A prosodic analysis? In (63a) and (36b), -R is attached to the first word-sized constituent within the IP-string. This shows that a generalization which captures the distribution of -R in terms of linear order in the word string is more robust than one which states its distribution in phrase-structural terms. Instead, a grammar which relies on phrase-structural constraints would have to rely on a filtering mechanism based on prosodic constraints. For example, the -R morpheme could be treated simply as the “spell-out” of a negation or polarity feature which is specified on I0. Constraints or operations on the phonological form of the sentence would then be used to derive the correct position of -R within the string. However, as was discussed in detail above, there are a number of systematic exceptions to Generalization 1 which need to be accounted for. The problem for an analysis like (61a) is that the phrase structure anchors maa- at the left edge of the IP-string. A promising approach to resolving the exceptions might be to argue that the domain in which the distribution of maa- is defined in purely prosodic terms, rather than making reference to the IP. For example, assume the prosodic hierarchy of Selkirk (1980) in which syllables are grouped together in feet, feet are grouped as prosodic words (“p-words”), prosodic words as phonological phrases (“p-phrases”), and phonological phrases as intonation phrases (“i-phrases”). Generalizations 1 and 2 might then be revised a last time as follows: (65)
Generalization 1 (final): maa- appears at the left edge of a phonological phrase.
128 (66)
FREDERICK HOYT
Generalization 2 (final): -R appears at the right edge of a prosodic word that is: (i) aligned with the left edge of a phonological phrase; (ii) marked for negation; (iii) marked a morpheme expressing person features.
This is the kind of approach advocated by Truckenbrodt (1999) and Chung (2003), according to whom principles of prosodic construction (whether rules or constraints) make no direct reference to syntactic structure. Formulating an analysis along these lines would be a non-trivial undertaking and will have to be left to further research. 3.4 Summary of theoretical implications The distribution of maa- and -R in PA is not easily characterized in phrase-structural terms. In particular, -R gravitates toward the second position in the clause, in some cases intruding into another constituent in order to do so. Therefore, the distribution of -R is more accurately described in terms of prosodic constituents rather than syntactic constituents. 4.
Conclusion This paper has been a detailed examination of negation morphology in Palestinian Arabic. This examination shows that the negation morphemes maaand -R behave as special clitics in Zwicky and Pullum’s (1983) sense, and in particular that their distribution is conditioned largely by prosodic factors. There is a strong tendency for them to be hosted by the left-most word in the IP-string in a phrase-structural representation of a clause. This suggests that -R is a second-position clitic. However, exceptions to this generalization call into question whether the IP-string is the correct characterization of the domain according to which they are positioned. It is suggested that the domain would be more accurately characterized in prosodic terms, for example as a “phonological phrase”. This needs to be the basis of further research, but should it turn out to be an accurate characterization, the distribution of maa- and -R could be characterized robustly. This raises interesting questions about how negation morphology is represented in other Arabic dialects. Studies of negation in Egyptian Arabic by Woidich (1968), Eid (1991, 1993), and Jelinek (2002) suggest that Egyptian and Palestinian are very similar in terms of how negation is realized, although a conclusion to that effect awaits a detailed comparison. In contrast, detailed descriptions of negation in Moroccan Arabic (Harrel 1962, 1965, 1966; Marçais 1977; Benmamoun 1992, 1997, 2000; Ouhalla 2002) suggest that
NEGATION IN PALESTINIAN ARABIC
129
Moroccan maa- and -R are affixes rather than clitics in Zwicky and Pullum’s (1983) sense, because they selects verbal stems as their hosts (Benmamoun 2000) and because the distribution of -R is affected by the syntactic grammar. It seems likely that there is significantly more variation between the dialects than has been previously acknowledged in terms of how negation is expressed.
REFERENCES Alexopoulou, Theodora, Edit Doron & Caroline Heycock. 2004. “Broad subjects and clitic left dislocation.” In Peripheries: Syntactic edges and their effects, ed. David Adger, 329-358. Oxford: Oxford University Press. Aoun, Joseph & Elabbas Benmamoun. 1998. “Minimality, Reconstruction, and PF Movement”. Linguistic Inquiry 29:4.59-597. ----------, Lena Choueiri & Norbert Hornstein. 2001. “Resumption, Movement, and Derivational economy”. Linguistic Inquiry 32:3.371-403. Awwad, M. A. 1987. “Free and Bound Pronouns as Verbs in Rural Palestinian Colloquial Arabic”. Journal of Arabic Linguistics 16.108-118. Benmamoun, Elabbas. 1992. Inflectional and functional morphology: Problems in projection, representation, and derivation. PhD diss., University of Southern California. ----------. 1997. “Licensing of Negative Polarity Items in Moroccan Arabic”. Natural Language and Linguistic Theory 15.263-287. ----------. 2000. The feature structure of functional categories: A comparative study of Arabic dialects. Oxford: Oxford University Press. Brame, Michael. 1971. Stress in Arabic and Generative Phonology”. Foundations of Language 7.556-591. Brustad, Kristen. 2000. The syntax of Spoken Arabic. Washington: Georgetown University Press. Chung, Sandra. 2003. “The Syntax and Prosody of Weak Pronouns in Chomorro”. Linguistic Inquiry 34:4.547-599. Cowell, Mark. 1964. A reference grammar of Syrian Arabic. Washington: Georgetown University Press. Demirdache, Hamida. 1991. Resumptive chains in restrictive relatives, appositives and dislocation structures. PhD diss., MIT. ----------. 1997. “Dislocation, resumption, and weakest crossover”. In Materials on left-dislocation. ed. Elena Anagnostopoulou, Henki von Riemsdijk & Frans Zwarts, 193-231. Amsterdam & Philadelphia: John Benjamins.
130
FREDERICK HOYT
Diesing, Molly & Eloise Jelinek. 1995. “Distributing Arguments”. Natural Language and Linguistic Theory 3.123-176. Doron, Edit & Caroline Heycock. 1999. “Filling and licensing multiple specifiers”. In Specifiers: Minimalist approaches. ed. David Adger, Susan Pintzuk, Bernadettet Plunkett & George Tsoulas, 69-89. Oxford: Oxford University Press. Eid, Mushira. 1991. “Verbless sentences in Arabic and Hebrew”. In Perspectives on Arabic linguistics III. ed. Bernard Comrie & Mushira Eid, 31-61. Amsterdam & Philadelphia: John Benjamins. ----------. 1993. “Negation and predicate heads in Arabic”. In Principles and predication: The analysis of natural language. ed. Mushira Eid & Gregory Iverson, 135-152. Amsterdam & Philadelphia: John Benjamins. Feghali, Michel. 1928. Syntaxe des parles arabes actuels du Liban. Paris: Édouard Champion. Harrel, Richard. 1962. A short reference grammar of Moroccan Arabic. Washington: Georgetown University Press. ----------. 1965. A basic course in Moroccan Arabic. Washington: Georgetown University Press. ----------. 1966. A dictionary of Moroccan Arabic. Washington: Georgetown University Press. ----------. 2004. Modern Arabic: Structures, functions, and varieties (Revised Edition). Washington: Georgetown University Press. Husseini, Lutfi. 1990. “Serial verbs in Colloquial Arabic”. In When verbs collide: Papers from the 1990 Ohio State Mini-Conference on Serial Verbs. ed. Brian Joseph & Arnold Zwicky, 340-354. Columbus: The Ohio State University Press. Jelinek, Eloise. 2002. “Agreement, clitics and focus in Egyptian Arabic”. In Themes in Arabic and Hebrew syntax. ed. Jamal Ouhalla & Ur Shlonsky, 71-105. Dordrecht: Kluwer Academic Publishers. Kenstowizc, Michael & Abdul-Karimm, K. 1980. “Cyclic Stress in Levantine Arabic”. Studies in the Linguistic Sciences 10.55-76. Khalaily, Samer. 1997. One syntax for all categories: Merging nominal atoms in multiple adjunction structures. The Hague: Holland Institute of Generative Linguistics. Khan, Geoffrey. 1988. Studies in Semitic syntax. Oxford: Oxford University Press. Lalami, Laila. 1996. “Clitic left dislocation in Moroccan Arabic”. In Perspectives on Arabic linguistics IX. ed. Mushira Eid & Dilworth Parkinson, 115-129. Amsterdam & Philadelphia: John Benjamins. Marçais, Phillip. 1977. Esquisse grammatical de l’arabe maghrébin. Paris: Librarie d’Amérique et d’Orient.
NEGATION IN PALESTINIAN ARABIC
131
Mitchell, Terrence & Shahir al-Hassan. 1994. Modality, mood, and aspect in Spoken Arabic. London: Keegan Paul International. Mohammad, Mohammad. 1998. The syntax of indefinite subjects in equative sentences in palestinian Arabic. Ms., University of Florida. ----------. 2000. Word order, agreement, and pronominalization in Standard and Palestinian Arabic. Amsterdam & Philadelphia: John Benjamins. Ouhalla, Jamal. 1993. Negation, Focus, and Tense: The Arabic maa and laa. Rivista di Linguistica 5.275-300. ----------. 2002. “The structure and logical form of negative sentences in Arabic”. In Themes in Arabic and Hebrew syntax. ed. Jamal Ouhalla & Ur Shlonsky, 299-320. Dordrecht: Kluwer Academic Publishers. Pollock, Jean-Yves. 1989. “Verb Movement, Universal Grammar, and the Structure of IP”. Linguistic Inquiry 21:4.365-424. Rizzi, Luigi. 1997. “The fine structure of the left periphery”. In Elements of grammar: A handbook of Generative Syntax. by Lilliana Haegeman, 281337. Dordrecht: Kluwer Academic Publishers. Schmidt, Hans & Paul Kahle. 1918. Volkserzälungen aus Palästina, v.1. Göttingen: Vandenhoek und Ruprecht. ----------. 1930. Volkserzälungen aus Palästina, v.2. Göttingen: Vandenhoek und Ruprecht. Selkirk, Elisabeth 1980. “The Role of Prosodic Categories in English Wordstress”. Linguistic Inquiry 11.563-605. Shlonsky, Ur. 1997. Clause structure and word order in Hebrew and Arabic. Oxford: Oxford University Press. Truckenbrodt, Hubert. 1999. “On the Relation between Syntactic Phrases and Phonological Phrases”. Linguistic Inquiry 30.219-255. Wackernagel, Jacob. 1892. “Über ein Gesetz der indogermanischen Wortstellung”. Indogermanische Forschungen 1.333-436. Woidich, Manfred. 1968. Negation und Negative Sätze im ÄgyptischArabischen, PhD diss., München: Ludwig-Maximilians-Universität. Younes, Munther. 1995. “On vowel shortening in Palestinian Arabic”. In Perspectives on Arabic linguistics VII. ed. Mushira Eid, 157-171. Amsterdam & Philadelphia: John Benjamins. Zwicky, Arnold. 1977. On clitics. Bloomington: Indiana University Linguistics Club. ----------. & Geoffrey Pullum. 1983. “Cliticization vs. Inflection: English n’t”. Language 59:3.502-513.
III
PSYCHOLINGUISTICS
THE SPLIT-INFL HYPOTHESIS
FINDINGS FROM ENGLISH AND JAPANESE L2 LEARNERS OF ARABIC∗
Mohammad T. Alhawary The University of Oklahoma
1.
Introduction Following recent formulations of the Minimalist Program, Chomsky (1995, 1998, 1999, 2001) abandoned earlier proposals of the Split-INFL hypothesis (Pollock 1989) and the Agr-based theory of clause structure (Chomsky 1991, 1993) and proposed instead a single INFL projection (TP) within which Agreement, Case, and the EPP feature of T are checked1. This is illustrated in structures (1) and (2), where (2) now replaces (1), with respect to subject-verb agreement. (1) [AgrsP NPi Agrs [TP [VP ti V …]]] (2) [TP NPi T [vP ti v [VP …]]] However, findings from the L1 acquisition literature, including German, French and English suggest that tense and the grammatical properties of subject-verb agreement (as well as the conditions licensing null subjects) develop at different times, i.e., as two independent projections. With respect to tense, early L1 child German data (before the age of 2) exhibit knowledge of the finite-nonfinite (or tensed vs. untensed) distinction through the systematic placement of finite verbs in first and second position and nonfinite verbs in final position (Meisel & Müller 1992; Verrips & Weissenborn 1992; Poeppel & Wexler 1993). Early L1 child German data also show that the production of post-verbal negation occurs consistently with finite verbs and preverbal negation with nonfinite verbs (Clahsen et al. 1993/1994). Similarly, early L1 child French data show that preverbal negation (i.e., with the negation particle pas before the verb and the subject) occurs consistently with untensed verbs
∗
I would like to thank the students who participated in this study. My sincere thanks also go to Robert Ratcliffe and Tominaga Masato for their help in recruiting the Japanese participants. The study was supported by funding from the College of Arts and Sciences, the School of International and Area Studies, and the Research Council at the University of Oklahoma. 1 Chomsky suggests that his account should be restated as: “Agr and not T/v that is the locus of f-features, Case and EPP” (1999: footnote 14).
136
MOHAMMAD T. ALHAWARY
and post-verbal negation occurs consistently with tensed verbs (Meisel & Müller 1992; Déprez & Pierce 1993). As for subject-verb agreement, there seems to be a disagreement among researchers as to when subject-verb agreement emerges in L1. For example, Clahsen (1990) and Clahsen and Penke (1992) claim that, unlike tense properties, subject-verb agreement is not present at the earliest stages of syntactic development. Others claim that subject-verb agreement is initially present (in early German and French L1 development) although children have not yet learned the complete subject-verb agreement paradigm (Verrips & Weissenborn 1992; Poeppel & Wexler 1993; Clahsen et al. 1993/1994). Others claim that subject-verb agreement emerges first and tense develops significantly later (Meisel 1994; Wexler 1994). Whatever the situation may be, the findings seem to converge on one conclusion: tense and agreement properties do not seem to develop at the same time2. TP and AGRP seem to be independent projections. Chomsky’s current reformulations would; therefore, inaccurately predict that tense and agreement properties are developed simultaneously (see Griffin 2003). Most recently, this issue has received renewed interest on similar grounds. Based on L1 English data, Guasti and Rizzi (2002) show that agreement and tense in negative do constructions develop clearly earlier than agreement in the interrogative use of the auxiliary do. Accordingly, they claim that Agreement (AGRP) and Tense (TP) are independent maximal projections with agreement higher than tense (Guasti & Rizzi 2002; cf. Roeper 2003). The situation is less obvious in L2 acquisition. Most findings on tense and agreement in L2 acquisition were presented in the morpheme order studies of the 1970’s. Although such studies suffer from a number of methodological limitations, tense and agreement are claimed to emerge around the same stage, where emergence of tense (regular and irregular past) and the 3rd singular {-s} subject-verb agreement suffix seem to cluster together at a later stage, with respect to other morphemes. However, the exact ordering of emergence of tense and agreement in those studies is mixed. Findings reported include: (1) tense and agreement emerge at the same time (e.g., Krashen 1977), (2) tense emerges before agreement (e.g., Bailey et al. 1974; Anderson 1978), and (3) either regular past tense emerges before agreement, followed by irregular past tense, or irregular past tense then agreement followed by regular past tense (e.g., Larsen-Freeman 1975; Zobl & Liceras 1994). A more recent study on L2 German, based on one participant (José SL) from the ZISA project (Meisel et al. 1981), show that subject-verb agreement and tense emerge at the same stage (Eubank 1992). Most other L2 acquisition studies have focused on the 2
The findings also indicate that there is a correlation between the emergence of subject-verb agreement and the decrease of empty/null subjects (e.g., Clahsen & Penke 1992; Pierce 1992; Meisel 1994).
THE SPLIT-INFL HYPOTHESIS
137
emergence of tense and aspect and do not include any comparative findings with respect to the emergence of subject-verb agreement. The purpose of the present study is to investigate whether collapsing AGR and Tense under one maximal projection (TP) is motivated based on Arabic L2 acquisition data from L1 speakers of English and Japanese. The significance of the present study is twofold. First, Arabic L2 data from such a typological constellation has not been reported on before. Second, the study is an attempt to contribute to the current renewed interest in the Split-INFL debate. 2. Methodology 2.1 Participants Fifty-six Arabic L2 learners (L2ers), from two different native language backgrounds, American English and Japanese, were invited to participate in the study in their home institutions in the US and Japan, respectively. The participants were grouped according to their placement by their home institutions and the length of exposure to Arabic as part of their academic programs. Table 1 summarizes the details of the participants. Length of Exposure English L1 Group1 (n=9) Year 1 Group2 (n=9) Year 2 Group3 (n=9) Year 3
Credit Hours Weekly 6 5 4
M/F
Ages Range
Ages Means
4/5 5/4 6/3
18-21 20-29 22-34
19.22 22.22 29.11
18-20 19-21 20-23
19 20 21.11
Japanese L1 Group1 (n=10) Year 1 12 8/2 Group2 (n=10) Year 2 12 5/5 Group3 (n=9) Year 3 4 1/8 M/F= Total Males/Total Females Table 1: Participants
The participants had little or no exposure to Arabic prior to joining their academic institutions and were not heritage speakers. Beginning students of both groups had zero exposure and had made no trips to Arabic-speaking countries. A few of the participants in the other groups made brief trips to Arabic speaking countries but did not stay for a significant period of time. The English L1 group received formal instruction in Arabic with focus on grammatical forms, using (Abboud et al. 1983, 1997). As for the Japanese L1 group, some belonged to classes that used (Badawi & Yunis 1983; Badawi et al. 1992), which focuses on grammatical forms from early on, and others belonged to classes that used (Brustad et al. 1995a, 1995b).
138
MOHAMMAD T. ALHAWARY
2.2 Target forms The present study investigates the emergence of subject-verb agreement and tense (present and past). The investigation of subject-verb agreement is restricted to 3rd person singular masculine/feminine to make the comparison across all three proficiency levels possible and the forms accessible to all participants. Sentences (3)-(5) below are examples of subject-verb agreement and tense contexts in the target L2 language (Arabic) and in the L1 backgrounds (English and Japanese) of the participants. (3a) ya-šrab(u) al-qahwa(ta) fii al-sabaaħ 3sm-drink the-coffee in the-morning “He drinks coffee in the morning.”
(Arabic)
(3b) ta-šrab(u) al-qahwa(ta) fii al-sabaaħ the-coffee in the-morning 3sf-drink “She drinks coffee in the morning.” (3c) šariba al-qahwa(ta)3 >ams drank.3sm the-coffee yesterday “He drank coffee yesterday.” (3d) šarib-at al-qahwa(ta) >ams drank-3sf the-coffee yesterday “She drank coffee yesterday.” (4a) He drinks coffee in the morning. (4b) She drinks coffee in the morning. (4c) He drank coffee yesterday. (4d) She drank coffee yesterday.
(English)
(5a) asa kare-wa kōhī-o nomimasu in the morning he-TOP coffee-ACC drink “He drinks coffee in the morning.”
(Japanese)
(5b) asa kanojo-wa kōhī-o nomimasu coffee-ACC drink in the morning she-TOP “She drinks coffee in the morning.”
3
The parentheses indicate that mood markings are not focused on in the present study.
THE SPLIT-INFL HYPOTHESIS
139
(5c) kinō kare-wa kōhī-o nomimašita yesterday he-TOP coffee-ACC drank “He drank coffee yesterday.” (5d) kinō kanojo-wa kōhī-o nomimašita she-TOP coffee-ACC drank yesterday “She drank coffee yesterday.” The examples above illustrate that while Arabic and English exhibit subject-verb agreement, Japanese does not. Additionally, Arabic exhibits rich subject-verb agreement morphology: the verb shares with the subject the feature agreement of person, number and gender by means of a prefix in the present tense and a suffix in the past tense. English, on the other hand, exhibits an impoverished morphological system of subject-verb agreement where the verb agrees with the subject with the 3rd person singular and only in the present tense. However, all three languages exhibit tense morphology, as the examples show. I adopt tentatively here the standard analysis that tense is expressed in Arabic by means of the stem, rather than the root-vocalic melody analysis, since Arabic L2ers are not usually introduced to the notion of root and pattern at the early stages of L2 acquisition4. I follow the standard Minimalist assumption of attributing parametric variation to the strength of functional features. Due to its rich verbal agreement features, Arabic is analyzed with the functional feature strength set to [+strong], while the functional feature strength in English and Japanese is set to [-strong]. Accordingly, the typological constellation of the target and source languages of the participants generates the following (6) and (7) pairings: (6) English participants who are speakers of a [+Tense], [+AGR] and [-strong] L1, learning a [+Tense], [+AGR] and [+strong] L2 (7) Japanese participants who are speakers of a [+Tense], [-AGR] and [-strong] L1, learning a [+Tense], [+AGR] and [+ strong] L2 2.3 Research questions The present study attempts to address the following questions: (a) Do Arabic L2ers, who are speakers of a language (such as Japanese) that exhibits tense but does not exhibit verbal agreement, acquire tense and verbal agreement on par with speakers of a language (such as English) that exhibits both tense and verbal agreement markings? 4
It can also be argued that Arabic verbs are inflected for the perfective and imperfective aspect stems, roughly corresponding to past and present tenses. For ease of reference, the term “tense” is used here rather than aspect. Arabic does not exhibit an infinitival form of the verb.
140
MOHAMMAD T. ALHAWARY
(b) Do such L2ers exhibit L1 transfer in their interlanguage systems? (c) Do Arabic L2ers acquire tense and verbal agreement simultaneously (i.e., at the same stage) that makes collapsing Tense and Agr into one single maximal projection (TP) tenable in L2 acquisition? 2.4 Data collection and coding Data collection aimed at eliciting spontaneous/semi-spontaneous production data of the target forms from the L1 English and L1 Japanese participants. Elicitation took place in one-on-one interview sessions; one interview per participant (30-45 minutes). Elicitation consisted of four narrative tasks as well as a random task. The production data were transcribed and coded. Certain items were not coded. These included hesitations, repetitions, and self-corrections except the last attempt. In coding subject-verb agreement tokens, agreement was determined by considering the verbal form and whether it was inflected properly, not by first identifying the subject then the verb it agrees with. This is significant because the verb may agree with a discourse referent subject and the L2ers may be producing the wrong subject, especially when the subjects are the pronouns hiya ‘she’ and huwa ‘he’ which are close in pronunciation (Meisel 1991; Poeppel & Wexler 1993; Prévost & White 2000). 2.4.1 Narrative and random tasks. Four narrative tasks were used for the purpose of elicitation: two in the past tense and two in the present tense divided equally between a female and a male character. The two narratives in the past tense were each about a female and a male character. The participants were requested to describe the planned vacation activities (on a calendar) carried out by each of the characters during their vacation (which each took the previous month for a period of 10 days) day by day. As a distracter, the participants were asked to figure out and to comment on whether the characters in the stimuli made a compatible couple based on what they did on their vacation. The two narratives in the present tense were also about a female and a male character. The participants were requested to describe the daily routines and activities of each character at different times of the day. As a distracter, the participants were asked to figure out where the characters are from based on his/her activities. The two sets of narratives (past vs. present) were not presented sequentially. Rather, the past set was presented towards the middle of the interview and the present towards the end, with tasks of other unrelated structures used in the beginning of the first and second half of the interview, serving as distracters (see Appendix A for a narrative sample in the past tense of a male character).
THE SPLIT-INFL HYPOTHESIS
141
Additionally, a random task was included to control for the present tense narratives due to responses on pilot elicitations. An artifact of the task was found when administering the present narrative tasks. Some participants misunderstood the task as a narrative in the past (i.e., as a story) and accordingly produced past tense verbs. Therefore, such a task was thought necessary to control for the unintended effect of the present narratives. The random task elicited production data of present tense as well as verbal agreement through discrete picture description items, each of which shows an individual engaged in a common everyday activity. Hence, the random task was included to allow for more immediate and straightforward contexts than the present tense narrative tasks did. 2.5 Results 2.5.1 Tense (narrative tasks). The data reveal neither between nor within group differences with respect to the production of tense, past and present. In other words, no difference was found between the performance of the L1 English participants and their L1 Japanese counterparts, or within the groups with respect to tense generally, or the specific production of past tense vs. present tense. Overall, although both the English and Japanese groups seem to have higher ratios of correct rule application on the past tense than the present tense, no significant statistical difference was found (see Tables 2 and 3 below, Section 2.5.3). Individual participants of both L1 backgrounds exhibited variable performance patterns ranging from correct production of tense forms in the past and present tense contexts, consistent correct use of tense with occasional wrong use of the past tense forms in present contexts and conversely, consistent incorrect use of tense with and without occasional correct use of past tense forms in present contexts and conversely, and random or mixed use of correct and incorrect tense. Incorrect production of tense is illustrated in examples (8)-(10) below: (8) IL: *fii al- sabaaħ hiya šarib-at5 in the-morning she drank-3sf TL: fii al-sabaaħ hiya ta-šrab in the-morning she 3sf-drink “In the morning, she drinks…”
(L1 Japanese: Group 16)
5 For the present tense narrative tasks, recall that the participants were requested to describe the daily activities of each character at different times of the day from the morning to the evening of every day. Thus, the starter fii al-sabaaħ ‘In the morning …’ occurs clearly in the habitual, present tense (i.e., in the morning, of everyday…) context. 6 IL = Interlanguage use; TL = target language use.
142
MOHAMMAD T. ALHAWARY
(9) IL: hiya ta-drusu >ilaa µašara wa *faµal-at tennis she 3sf-study till 10 o’clock and did-3sf tennis TL: hiya ta-drusu >ilaa al-µaaširah wa ta-fµalu tennis she 3sf-study till 10 o’clock and 3sf-do tennis “She studies until 10 o’clock and does [plays] tennis.” (L1 Japanese: Group 1) (10) IL: *baµda raħlata-hu ya-šrab šaahii7 trip-his 3sm-drink tea after TL: baµda riħlati-hi šariba al-šaay trip-his drank.3sm the-tea after “After his trip/travel, he drank tea.”
(L1 English: Group 2)
In addition, a few tokens of past tense overproduction were found in the entire data sets. These were produced in embedded clauses, as in (11) and (12). (11) IL: džalas-at *li-šaahada al-tilivizyuun sat-3sf to-watched-3sm the-T.V. TL: džalas-at li-tu-šaahid al-tilivizyuun sat-3sf to-3sf-watch the-T.V. “She sat to watch T.V.”
(L1 English: Group 2)
>an *ra>aa al-ħayawaan zoo (12) IL: >araad wanted.3sm to watched.3sm the-animal zoo >an ya-raa ħadiiqata al-ħayawaan TL: >araad wanted.3sm to 3sm-watch garden the-animal “He wanted to see the zoo.” (L1 English: Group 3) 2.5.2 Verbal agreement (narrative tasks). Participants of both L1 backgrounds also exhibited somewhat variable use of subject-verb agreement with relatively higher correct subject-verb agreement ratios than tense ratios (see Tables 2 and 3, Section 2.5.3). Incorrect use of subject-verb agreement was similar in the production of both groups, as almost all wrong agreement inflections used comprise third and first person singular. A few tokens exhibited the incorrect agreement inflection for second person singular feminine (in place of third singular feminine) and one token for first person plural in the entire data sets. In addition, a couple of tokens exhibited two incorrect agreement inflections at once on a single stem (as in *ya-ðhab-tu → ta-ðhab ‘she goes’). The 7
For the past tense narrative tasks, recall that the participants were asked to narrate the planned vacation activities (on a calendar) carried out by each character during their vacation which took place the preceding month. Thus, the starter, baµda riħlati-hi ‘after his trip…’ occurs within the past tense (i.e., After his trip, last month,…) context.
THE SPLIT-INFL HYPOTHESIS
143
observation here is that verbal agreement mismatches in the production of the participants of both L1 backgrounds exclusively comprise person and gender but not number. All participants seem to have already acquired the singular vs. plural agreement distinction. Examples (13)-(16) are representative samples of incorrect rule application of subject-verb agreement in the production of both groups. (13) IL: *hiya ðahab-tu she went-1s TL: hiya ðahab-at she went-3sf “She went.”
(L1 Japanese: Group 1)
(14) IL: *[huwa] ðahab-tu he went-1s TL: ðahaba went.3sm “He went.”
(L1 English: Group 2)
(15) IL: *hiya ya-sħuu 3sm-wake up she TL: hiya ta-sħuu she 3sf-wake up “She wakes up.”
(L1 Japanese: Group 2)
(16) IL: *huwa ta-qra>a al-kitaab 3sf-read the-book he TL: huwa ya-qra> al-kitaab he 3sm-read the-book “He reads the book.”
(L1 English: Group 1)
With respect to statistical differences in performance on subject-verb agreement, the data yielded findings that are somewhat different from those of tense. Two-way and one-way MANOVA and follow up tests revealed: a near effect for interaction between L1 and proficiency (F(2,50) = 2.91, p = .064) and a near proficiency effect for the English groups (F(2,24) = 2.74, p = .084) with respect to verbal agreement in the present vs. agreement in the past. In other words, while performance of the L1 English participants on subject-verb agreement in the present tense improved somewhat with proficiency (i.e., with more formal instruction), no such improvement was found in the L1 Japanese performance.
MOHAMMAD T. ALHAWARY
144
2.5.3 Tense and verbal agreement (narrative tasks). Most importantly for the purpose of the present study, a similar pattern of findings to that of verbal agreement emerges when we consider the interaction between tense and verbal agreement. Tables 2 and 3 below display the distribution of the data with respect to the participants’ performance on both tense and subject-verb agreement. Arabic L2 (English L1) Group1 (n=9) Group2 (n=9) Group3 (n=9) Arabic L2 (Japanese L1) Group1 (n=10) Group2 (n=10) Group3 (n=09)
Tense Correct/Total 154/170 157/204 150/192
% .91 .77 .78
Agreement Correct/Total 150/170 165/204 171/192
% .88 .81 .89
146/152 211/221 156/190
.96 .95 .82
135/152 183/221 169/190
.89 .83 .89
Table 2: Correct rule application of tense and agreement on past tense tasks Arabic L2 (English L1) Group1 (n=9) Group2 (n=9) Group3 (n=9) Arabic L2 (Japanese L1) Group1 (n=10) Group2 (n=10) Group3 (n=09)
Tense Correct/Total 97/144 134/187 111/195
% .67 .72 .57
Agreement Correct/Total 128/144 140/187 177/195
% .89 .75 .91
51/137 111/187 105/174
.37 .59 .60
117/137 168/187 147/174
.85 .90 .84
Table 3: Correct rule application of tense and agreement on present tense tasks
One way and two-way MANOVA tests revealed neither between nor within group differences with respect to the production of the past tense and verbal agreement in the past. However, the tests revealed a near significant interaction effect for L1 and proficiency (F(2,50) = 2.94, p = .064) and a near proficiency effect for the L1 English groups (F(2,24) = 2.74, p = .084) with respect to the development of the present tense verbal agreement. Thus, while the performance of the L1 Japanese groups showed no improvement along proficiency with respect to the development of the present tense or verbal agreement in the present, the performance of their L1 English counterparts showed marginal improvement with respect to the development of verbal agreement in the present tense. The performance of L1 English groups did not show improvement along proficiency with respect to the development of the
THE SPLIT-INFL HYPOTHESIS
145
present tense. Figure 1 illustrates these two marginal findings on the present tense and subject-verb agreement in the production of both L1 groups. Present Agreement
1.0
0.8
0.6 52
0.4 28
0.2
0.0 English L1
Japanese L1 .
Figure 1
2.5.4 Tense and verbal agreement (random task). The data gathered through the random task were included to control for artifacts that may have resulted from the nature of the narrative past and present tense tasks. The random tasks elicited production data of the present tense and verbal agreement through discrete picture description items. The participants needed simply to describe the individuals in the pictures and what they were doing (naturally in the present tense). Hence, the random task would allow for more immediate and straightforward contexts than the narrative tasks did. The narrative tasks, on the other hand, expected the participants to describe a series of events consistently within an artificial narrative time frame (present or past) from beginning to end without switching tense use, although some of them often did to varying degrees. Table 4 displays the distribution of the data from the random task with respect to the participants’ production of present tense and subject-verb agreement.
MOHAMMAD T. ALHAWARY
146 Arabic L2 (English L1) Group1(n=9) Group2(n=9) Group3(n=9) Arabic L2 (Japanese L1) Group1(n=10) Group2(n=10) Group3(n=09)
Tense Correct/Total 24/42 145/174 252/270
% .57 .83 .93
Agreement Correct/Total 26/42 141/174 237/270
% .62 .81 .88
124/141 212/226 154/168
.88 .94 .92
125/141 207/226 146/168
.89 .92 .87
Table 4: Correct rule application of tense and agreement on the random task
Present Random Agreement
1.0
0.8 12
0.6
54 52 52
0.4
0.2
0.0
4
4
English L1
Japanese L1 .
Figure 2
Two-way and one-way MANOVA and their follow up tests revealed the following findings: a significant L1 effect with respect to both (present) tense (F(1,50) = 9.38, p< .005) and verbal agreement (F(1,50) = 6.80, p< .013), a significant proficiency feature effect for tense (F(2,50) = 8.13, p< .002), and an interaction effect between L1 and proficiency for tense (F(2,50) = 7.26, p< .003) and another for agreement (F(2,50) = 4.52, p< .017). In addition, a significant effect for the feature tense (F(2,24) = 11.06, p<.001) and a near effect for the feature verbal agreement (F(2,24) = 3.99, p = .032) were found in the L1 English groups. No significant effects in the production of the L1 Japanese groups were revealed. Figure 2 illustrates the findings based on the
THE SPLIT-INFL HYPOTHESIS
147
participants’ production data of the present tense and subject-verb agreement on the random task. The findings show that the performance of the L1 English groups differed significantly from that of their L1 Japanese counterparts on both forms, tense and verbal agreement. In addition, while there was no significant difference in the performance of the L1 Japanese groups with respect to tense and verbal agreement, such a difference was detected in the production of the L1 English groups. The performance of the L1 English groups on (present) tense improved more significantly with proficiency (i.e., with more formal instruction) than on subject-verb agreement, with ratios of correct rule application of tense being significantly higher than those of subject-verb agreement. 3.
Discussion and conclusion With respect to the first research question two paths of development for tense and verbal agreement were found based on the narrative tasks. On one hand, the L1 Japanese participants seem to have taken the same developmental path for acquiring tense and subject-verb agreement; hence, no significant difference was found between the two target forms. The situation is dissimilar in the case of the L1 English participants. No difference was found between their performance on the past versus the present tense on the narrative tasks. However, a near significant difference was found between their development of verbal agreement in the present and the past tense, showing marginal improvement with proficiency in subject-verb agreement in the present than in the past tense. Another near significant difference was found between the development of verbal agreement and the present tense, showing marginal improvement with proficiency in the former but none in the latter. A different finding is obtained when we consider the results from the random task and exclude the results obtained from the narrative tasks, considering the less than significant effects obtained and the possible artifacts of the narrative tasks themselves (explaining perhaps in part the variability exhibited in the data). While performance on tense improved with proficiency more significantly so than on verbal agreement in the L1 English groups, no such difference was found in the performance of the L1 Japanese groups. Since the production of the two target forms by the Japanese groups did not exhibit a steady performance, the first research question can be answered negatively: Arabic L2ers, who are speakers of a language that exhibits tense but not verbal agreement, do not seem to acquire tense and verbal agreement on par with speakers of a language that exhibits both tense and verbal agreement marking. The English L1 participants seem to follow a different developmental route when learning tense than their Japanese counterparts. This conclusion is surprising. One would perhaps instead expect the Japanese participants to exhibit differences in the developmental paths of the
148
MOHAMMAD T. ALHAWARY
two target forms, since their L1 exhibits tense but not subject-verb agreement unlike their English counterparts whose L1 exhibits both forms (see Alhawary 2005). This finding con-comitantly answers the second research question pertaining to L1 transfer. Accordingly, there is no indication for L1 transfer, based on the production data of both target forms of the study (aside from the near significant effects obtained in the narrative tasks). Perhaps this is due in part to the nature of the target forms examined and the participants’ hypothesizing that both notions of subject-verb agreement and tense are conflated within the verbal agreement morphological features without paying specific attention to the nature of the verb form at the root-vocalic melody level in the past versus the present tense. More research needs to be conducted to explore this factor. The absence of any indication for L1 transfer may also be due to the impoverished morphological agreement paradigm in English. To investigate this line of research further, other studies need to be conducted with different typological constellations that include languages with rich morphological agreement paradigms. As for the third research question, the data of the present study seem to produce mixed evidence. The acquisition of the target forms by the Japanese participants seem to proceed along the same developmental path, hence assuming both Tense and Agr collapsed under one single maximal projection may be tenable so far as the L1 Japanese speakers are concerned. On the other hand, the acquisition of the target forms by the English participants does not seem to take place along the same developmental path or simultaneously at the same pace, with (present) tense showing steady improvement more so than subject-verb agreement. The L1 English data seem to provide evidence in support of the Split-INFL hypothesis (especially, (Guasti & Rizzi 2002)) and that Agreement (AGRP) and Tense (TP) may be independent maximal projections with agreement higher than tense, with the latter being more basic. The findings may be interpreted differently if we assume Speas’ (1995) analysis of the syntactic representation of null and non-null subject languages. Following Speas’ Economy of Projection Principle, languages that exhibit agreement features, no matter how impoverished, project an AGRP projection for agreement features to be checked, while in languages that lack agreement features (e.g., Japanese), there is no need for an AGRP; a TP instead is projected. The present findings provide an indirect support for L1 transfer. The L1 Japanese and L1 English participants may be each drawing on their own L1 syntactic representation, with a TP and AGRP projection. This line of reasoning would require further research including additional analysis of null subjects use in the participants as well as analysis of other L1 participants such as Spanish which exhibits both null subjects and verbal agreement.
THE SPLIT-INFL HYPOTHESIS
149
APPENDIX A A sample narrative elicitation task in the past tense of a male character. The vacation started from the 17th – 26th of last month (adapted from Burton & Maharag 1995).
REFERENCES REFERENCES
150
MOHAMMAD T. ALHAWARY
REFERENCES Abboud, Peter, Zaki Abdel-Malek, Najm Bezirgan, Wallace Erwin, Mounah Khouri, Ernest McCarus, Raji Rammuny & George Saad. 1983. Elementary Modern Standard Arabic Vol. 1. Cambridge: Cambridge University Press. ----------, Aman Attieh, Ernest McCarus & Raji Rammuny. 1997. Intermediate Modern Standard Arabic. Ann Arbor, MI: Center for Middle Eastern and North African Studies, University of Michigan. Alhawary, Mohammad T. 2005. “L2 acquisition of Arabic morphosyntactic features: Temporary or permanent impairment”. In Perspectives on Arabic linguistics XVII-XVIII, ed. Mohammad T. Alhawary and Elabbas Benmamoun, 273-312. Amsterdam & Philadelphia: John Benjamins. Anderson, Roger. 1978. “An Implicational Model for Second Language Research”. Language Learning 28.221-282. Badawi, Elsaid Muhammad, Muhammad H. Abdullatif & Mahmoud Al-Batal. 1992. Al-kitaab al-’asaasii Vol. 2. Tunisia: Al- Munazzama Al-‘arabiyya Li-ltarbiyya wa Al-thaqaafa wa Al-‘uluum. ---------- & Fathi Ali Yunis. 1983. Al-kitaab al-`asaasii Vol. 1. Tunisia: AlMunazzama Al-‘arabiyya li-ltarbiyya wa Al-thaqaafa wa Al-‘uluum. Bailey, Nathalie, Carolyn Madden & Stephen Krahen. 1974. “Is There a ‘Natural Sequence’ in Adult Second Language Learning?”. Language Learning 24.235-243. Brustad, Kristin, Mahmoud Al-Batal & Abbas Al-Tonsi. 1995a. Alif baa: Introduction to Arabic letters and sounds. Washington, D.C.: Georgetown University Press. ----------, Mahmoud Al-Batal & Abbas Al-Tonsi. 1995b. Al-kitaab fii ta‘allum al-‘arabiyya. Washington, D.C.: Georgetown University Press. Burton, Eric & Lois Maharag. 1995. Going places 1. London: Longman. Chomsky, Noam. 1991. “Some notes on economy of derivation and representation”. In Principles and parameters in comparative syntax, ed. Robert Freidin, 417-454. Cambridge, MA: MIT Press. ----------. 1993. “A minimalist program for linguistic theory”. In The view from Building 20, ed. Ken Hale and Samuel J. Keyser, 1-52. Cambridge, MA: MIT Press. ----------. 1995. The Minimalist Program. Cambridge, MA: MIT Press. ----------. 1998. “Minimalist Inquiries: The framework”. MIT Occasional Papers in Linguistics 15.1-14. ----------. 1999. “Derivation by Phase”. MIT Occasional Papers in Linguistics 18:1-43. ----------. 2001. “Beyond Explanatory Adequacy”. MIT Occasional Papers in Linguistics 20.1-28.
THE SPLIT-INFL HYPOTHESIS
151
Clahsen, Harald. 1990. “Constraints on Parameter Setting: A grammatical analysis of some acquisition stages in German child language”. Language Acquisition 1.361-391. ---------- & Martina Penke. 1992. “The acquisition of agreement morphology and its syntactic consequences: New evidence on German child language from the Simone-Corpus”. In The acquisition of verb placement, ed. Jürgen Meisel, 181-223. Dordrecht: Kluwer. ----------, Martina Penke & Teresa Parodi. 1993/1994. “Functional Categories in Early Child German”. Language Acquisition 3:4.395-429. Déprez, Viviane & Amy Pierce. 1993. “Negation and Functional Projections in Early Grammar”. Linguistic Inquiry 24.25-67. Eubank, Lynn. 1992. “Verb movement, agreement, and tense in L2 acquisition”. In The acquisition of verb placement, ed. Jürgen Meisel, 225244. Dordrecht: Kluwer. Griffin, William. 2003. “The Split-INFL Hypothesis and AgrsP in Universal Grammar”. In The role of agreement in natural language: TLS 5 Proceedings, ed. William E. Griffin, 13-24. Texas Linguistics Forum, 53. Guasti, Maria Teresa & Luigi Rizzi. 2002. “Agreement and tense as distinct syntactic positions: Evidence from acquisition”. In Functional structure in DP and IP: The cartography of syntactic structures Vol. 1, ed. Guglielmo Cinque, 167-194. Oxford: Oxford University Press. Krashen, Stephen. 1977. “Some issues relating to the Monitor Model”. In On TESOL ’77, ed. Douglas Brown, C. Yorio and R. Crymes, 144-158. Washington, D.C.: TESOL. Larsen-Freeman, Diane. 1975. “The Acquisition of Grammatical Morphemes by Adult ESL Learners”. TESOL Quarterly 9.409-430. Meisel, Jürgen. 1991. “Principles of Universal Grammar and strategies of language use: On some similarities and differences between first and second language acquisition”. In Point counterpoint: Universal Grammar in the second language, ed. Lynn Eubank, 231-276. Amsterdam & Philadelphia: John Benjamins. ----------. 1994. “Getting FAT: Finiteness, agreement and tense in early grammars”. In Bilingual first language acquisition: French and German grammatical development, ed. Jürgen Meisel, 89-129. Amsterdam & Philadelphia: John Benjamins. ----------, Harald Clahsen & Manfred Pienemann. 1981. “On Determining Developmental Stages in Natural and Second Language Acquisition”. Studies in Second Language Acquisition 3.109-135. ---------- & Natascha Müller. 1992. “Finiteness and verb placement in early child grammars: Evidence from simultaneous acquisition of French and German bilinguals”. In The acquisition of verb placement, ed. Jürgen Meisel, 109-138. Dordrecht: Kluwer.
152
MOHAMMAD T. ALHAWARY
Pierce, Amy. 1992. Language acquisition and syntactic theory: A comparative analysis of French and English child grammars. Dordrecht: Kluwer. Poeppel, David & Kenneth Wexler. 1993. “The Full Competence Hypothesis of Clause Structure in Early German”. Language 69.1-33. Pollock, Jean-Yves. 1989. “Verb Movement, Universal Grammar, and the Structure of IP”. Linguistic Inquiry 20.364-424. Prévost, Philippe & Lydia White. 2000. “Missing Surface Inflection or Impairment in Second Language Acquisition? Evidence from tense and agreement”. Second Language Research 16.103-133. Roeper, Tom. 2003. “Multiple grammars, feature-attraction, pied-piping, and the question: Is AGR inside TP?”. In (In)vulnerable domains in multilingualism, ed. Natascha Müller, 335–360. Amsterdam & Philadelphia: John Benjamins. Speas, Margaret. 1995. Economy, agreement, and the representation of null arguments. Ms., University of Massachusetts, Amherst. Verrips, Maaike & Jürgen Weissenborn. 1992. “Routes to verb placement in early German and French: The independence of finiteness and agreement”. In The acquisition of verb placement, ed. Jürgen Meisel, 283332. Dordrecht: Kluwer. Wexler, Ken. 1994. “Finiteness and head movement in early child grammars”. In Verb movement, ed. David Lightfoot and Norbert Hornstein, 305-350. Cambridge: Cambridge University Press. Zobl, Helmut & Juana Liceras. 1994. “Functional Categories and Acquisition Orders”. Language Learning 44.159-180.
LEXICAL PROCESSING IN TWO LANGUAGE VARIETIES AN EVENT-RELATED BRAIN POTENTIAL STUDY OF ARABIC NATIVE SPEAKERS∗
REEM KHAMIS-DAKWAR & KAREN FROUD Teachers College, Columbia University
1.
Background Arabic is a diglossic language, in which two varieties exist; “high” and “low”. The “high” variety refers to Modern Standard Arabic (MSA), which is used mainly in written texts and formal settings. The “low” variety refers to the different spoken dialects; these differ widely between Arab countries and communities, along geographical, religious and socio-economic lines (Ferguson 1959). Linguistic differences between Palestinian Colloquial Arabic (PCA) and MSA are manifest in all language domains, including morphological, phonological, syntactic and semantic (Rosenhouse 1997; Shlonsky 1997; Abu-Rabia et al. 2003; Saiegh-Haddad 2005). Nevertheless, there remains a question as to whether MSA and varieties of PCA are more appropriately regarded as distinct languages or as different dialects or registers of one language. This question is confounded by the extremely frequent use of codeswitching, defined as the alternation between two or more languages, by native speakers of Arabic. In this paper, we describe an attempt to delineate the nature of the underlying distinction between MSA and PCA, by examining neural responses to a highly constrained codeswitching situation. 1.1 Codeswitching in Arabic Diglossic switching between MSA and Arabic dialects has been suggested to constitute a “third” language or a “middle” variety by several researchers (Mitchell 1982; Mahmoud 1986). Others have looked for alternative descriptive approaches to account for MSA/dialect mixing on different linguistic levels (Badawi 1973; El-Hassan 1978; Holes 1986; Suleiman 1986; Owens & Bani-Yasin 1987; Eid 1990; Hary 1996). Often such models have taken the form of a continuum or scale between the two language varieties, ∗
We would like to thank the Neurocognition of Language Lab at Teachers College for help with various aspects of this study and all our participants. This research was funded in part by a Faculty Diversity Research Award provided by the Dean’s Office, Teachers College, Columbia University.
154
REEM KHAMIS-DAKWAR & KAREN FROUD
rather than trying to understand codeswitching in a diglossic context as a simple dichotomy between the “high” and “low” varieties. For example, Boussofara-Omar (1999, 2003) attempts to account for Arabic diglossic switching within the Matrix Language Frame (MLF) model and its derivatives, the 4-M and Abstract Frame models (Myers-Scotton 1993; Myers-Scotton & Jake 2001), arguing against the existence of a third language variety. She rather argues that any apparent “third/middle” language variety is a systematic and predictable outcome of MSA/dialect switching, governed by structural constraints. Boussofara-Omar’s study, based on a corpus analysis, is one of very few systematic investigations of diglossic codeswitching in Arabic, and no studies have used neurophysiological approaches to investigate diglossic switching. We will argue that neurophysiological investigations of this phenomenon provide a unique opportunity to determine whether MSA and dialectal varieties of Arabic are neurally represented as distinct or interrelated linguistic systems. 1.2 Previous ERP studies on codeswitching Studies of online language processing in Arabic using EEG have not investigated codeswitching. In part, this is due to the fact that most of the EEG studies of Arabic were conducted mainly by Boudelaa and colleagues, who have a research focus on lexical representations of morphological units in Arabic, utilizing mainly written MSA stimuli for examination of priming and repetition effects (Boudelaa & Marslen-Wilson 2000, 2004a, 2004b; Boudelaa & Gaskell 2002). Only one unpublished study has directly examined diglossic switching, and this utilized written single word stimuli to compare semantic priming and repetition effects within and between the two Arabic varieties (Ibrahim & Bentin, unpublished, cited in Eviatar & Ibrahim 2000). The event-related potential (ERP) method is a non-invasive technique for directly observing electrical potentials generated by the brain, which index aspects of cognitive and linguistic processing. Scalp sensors record electrical activity (the electroencephalogram, or EEG) across multiple trials in experimental conditions, and ERPs are then derived from the continuous EEG recording by averaging within and between subjects. ERP studies on codeswitching in other languages show that switching between languages is associated with an increased negativity (N200) at frontal sensor sites compared to non-switching conditions, but only when switching occurs into the second language (i.e., L1→L2 switch, but not L2→L1) (Jackson et al. 1999; Jackson et al. 2004). Moreover, codeswitching between languages is associated with an increased late positive complex (LPC, also referred to in different studies as the P600) between 350-700 milliseconds after codeswitching onset, compared to non-codeswitching conditions. This effect was exhibited in both L1 and L2 codeswitching (Meuter & Allport 1999; Moreno et al. 2002).
LEXICAL PROCESSING IN TWO LANGUAGE VARIETIES
155
Moreno et al. (2002) studied codeswitching between Spanish and English, using three written stimulus conditions: expected/control (e.g., She heard a knock at the door); lexical switch/unexpected (e.g., She heard a knock at the entrance), and codeswitched (e.g., She heard a knock at the puerta). They found a clear dissociation: lexical switch conditions showed a negativity around 400 milliseconds after the unexpected word was presented (N400), but no LPC effect. In codeswitched sentences, the N400 was attenuated and the LPC was significant. For Moreno et al., this suggests that lexical switching within the same language is processed similarly to semantic anomaly, with N400 indexing unexpectedness but not syntactic reanalysis. Codeswitching, on the other hand, requires a qualitatively distinct neural response, with LPC/P600 effects indexing syntactic reanalysis and a response to increased structural complexity. Our study builds on such results by examining sentences in MSA/PCA combinations, with and without codeswitching, and with and without semantic anomalies. We hypothesized that N400 responses to diglossically codeswitched sentences should only arise if the neural responses to such items were the same as those elicited by within-language anomaly or unexpectedness. N200, along with P600/LPC effects, by contrast, should arise if the switch between “high” and “low” varieties requires syntactic reanalysis, suggesting that the two varieties are really represented as distinct languages (in the same way that Spanish and English are considered distinct in the Moreno et al. study of bilingual codeswitching). Furthermore, N200 effects should only be apparent in the switch from PCA to MSA, and not vice versa, reflecting Jackson et al.’s (2004) finding that the early negativity was more pronounced in L1→L2 codeswitching. 2. Methods 2.1 Participants Five native speakers of Palestinian Colloquial Arabic (the Galilee dialect) participated in the present study voluntarily. Subjects were all right-handed, one female and four males, aged 25;10 to 36;2 (mean age 30;2), with no history of psychological or neurological disorder, and with normal hearing and normal or corrected-to-normal vision. They were recruited through word of mouth at Columbia University, New York, and all gave written informed consent prior to participation in the study. All participants learned MSA at school since first grade, in Arab schools in Israel. 2.2 Materials The stimuli included 234 sentences presented auditorily, because PCA is not usually written down. Sentences were digitally recorded at 44.1 kHz using Computerized Speech Lab (CSL – Kay Elemetrics Corp.) and all sound files
156
REEM KHAMIS-DAKWAR & KAREN FROUD
were standardized for intensity (volume) levels using RMS calculations with Matlab. The presentation of each sentence was paced so that words were presented every 1000 milliseconds; the stimuli were presented over loudspeakers, using E-prime stimulus presentation software. 117 of the sentences started in MSA, and 117 started in PCA. 78 control sentences provided an expected ending for the sentence, in the same language variety. 78 ended in the same language variety, but with an unexpected (semantically anomalous) word. 78 sentences ended with a word which made sense, but which was codeswitched (i.e., presented in MSA for sentences which started in PCA, and presented in PCA for sentences which started in MSA). Examples of each sentence type are given in (1)-(3) below. For the codeswitched condition, we used MSA-PCA semantic cognates with no phonemic correspondence (lexically different items) in order to minimize word form processing effects. (1) Control condition ween naam-at il- binet where slept-2sf the-girl PCA PCA PCA-PCA “Where did the girl sleep?” laa tukθir mina al-kalaam neg excess-2sm from the-talk MSA MSA MSA-MSA “Don’t speak too much!” (2) Semantic condition ween naam-at il- tawle where slept-2sf the-table PCA PCA PCA-PCA “Where did the table sleep?” laa tukθir mina al-malal neg excess-2sm from the-bore MSA MSA MSA-MSA “Don’t bore too much!” (3) Codeswitch condition wen naam-at il- qitta where slept-2sf the-cat PCA PCA PCA-MSA “Where did the cat sleep?”
LEXICAL PROCESSING IN TWO LANGUAGE VARIETIES
157
laa tukθir mina al-ħaki neg excess-2sm from the-talk MSA MSA MSA-PCA “Don’t speak too much!” 2.3 Procedure The participants were tested in a sound attenuated room and seated in a comfortable chair approximately 80 cm. in front of a monitor. 128-channel geodesic sensor nets (Tucker 1993) were placed on each participant’s head following measurement of the circumference and marking of the vertex. These nets are arrangements of electrodes, held in relative positions to each other with fine elastic. The electrodes are embedded in sponges which are soaked in a weak electrolyte solution (potassium chloride). Electrode impedances were maintained below 50 kΩ, and continuous EEG recordings were obtained via a high-impedance amplifier digitally sampling at 250Hz, with the reference electrode at the vertex. Stimuli were presented in a computer generated random order. All subjects were instructed to avoid, as much as possible, any head or eye movements, blinks or any other muscle activity and to remain as relaxed as they were able to. To maintain attention, participants were asked to respond to a question immediately following the presentation of each sentence: they were asked to press a button to indicate whether they thought the last word of each sentence was in the same language as the rest of the sentence, or in a different language. Subjects were told beforehand that we were focusing on MSA/PCA contrasts, and that they would be asked to distinguish between the two. 3. Results 3.1 Behavioral analysis Reaction time and error responses within the different conditions and across the subjects were examined. Mean reaction times for each subject were calculated, and any response that was slower or faster than 1 standard deviation from each individual mean was identified as a ‘timeout’ trial and removed from further analysis. Error trials were also discarded, because error recognition is known to have particular effects on ERP recordings (known as error-related negativities - e.g., Falkenstein et al. 1990). Simple effects on error rates and reaction times between conditions were examined using t-tests. Responses to the different conditions are shown in Figure 1.
158
REEM KHAMIS-DAKWAR & KAREN FROUD
Figure 1: Mean # of error responses and mean reaction times for four conditions (n = 5 subjects). * = statistically significant difference at p <0.05
The findings revealed no significant effects on reaction times, but there were significantly more error responses in the codeswitched conditions compared to the control sentences. Semantic anomaly had no significant effect on any behavioral measure. 3.2 EEG results EEG recordings were lowpass filtered at 40Hz, and segmented to epochs extending from 200 milliseconds pre- to 800 milliseconds post-onset of the target word. Epochs associated with the same trial types were averaged within subjects, and trials were rejected if they were contaminated by artifacts (e.g., eye blinks, movements) or if they were associated with a behavioral error response or timeout. Averaged data from all subjects were re-referenced offline to the average reference and then re-montaged so that readings from a subset of channels were entered into further analyses. For the early time window (100300 milliseconds post onset of the final word in each sentence), sensors over the left fronto-central quadrant were selected, based on findings of Jackson et al. (2004). For the later time window (300-750 milliseconds), the montage of the channels was based on previous reports from the literature demonstrating that both N400 and P600 are maximally visible over occipito-parietal sensors (Moreno et al. 2002); selected channels are identified in Figure 2. Averaged and montaged data from all subjects were grand averaged. ERPs were identified with respect to peak latency and amplitude, associated with required field topography (negativity for N200 and N400, positivity for P600) at the scalp. Point-to-point t-tests were conducted to establish intervals of statistical significance for wave separation.
LEXICAL PROCESSING IN TWO LANGUAGE VARIETIES
159
Figure 2: Left: montage of selected channels for N200 analyses. Right: montage of selected channels for N400 / P600 analyses.
3.2.1 Overall analyses. Grand average analysis of responses to sentence final words (control condition, semantic condition, and code-switch condition) was initially collapsed across MSA and PCA. This approach revealed a significant N200 effect in the early time window across the fronto-central quadrant, with brain responses to codeswitched sentences being significantly more negative than responses to other sentences (p < 0.05 from 192 to 220 milliseconds) (see Figure 3). In the later time window, N400 effects were shown for semantically anomalous sentences over the same sensors, between 312 and 384 milliseconds, but these effects were not significant. A significant P600 effect was observed in the codeswitched condition only (p < 0.05 from 292 milliseconds onwards, significant relative to both the control and semantic conditions). See Figure 4.
Figure 3: Grand average (N=5) ERPs to sentence final words (control condition, semantic condition and code-switch condition); waveforms generated from fronto-central sensors (time window: 100-400 milliseconds)
REEM KHAMIS-DAKWAR & KAREN FROUD
160
µV
Figure 4: Grand average (N=5) ERPs to sentence final words (control condition, semantic condition and code-switch condition); waveforms generated using N400/P600 sensors (time window 300-750 milliseconds).
Further analysis of sentence final words for each variety was conducted to examine the impact of language variety on the brain. We separated the three conditions (control, semantically anomalous, codeswitching) by language, and looked separately at MSA and PCA. 3.2.2 MSA analysis. Grand average analyses of the five subjects to MSA sentence final words (MSA control condition, MSA semantic condition, and MSA codeswitching condition) revealed no significant separation between neural responses to codeswitched and control sentences in the N200 time window across fronto-central sensors (see Figure 5). In the later time window, no N400 effect was observed to codeswitched sentences, although significant N400 effects were found for the semantically anomalous condition in MSA, relative to control sentences (p < 0.05, 348-368 milliseconds). Also, a significant P600 effect for MSA codeswitching relative to the semantically anomalous condition and the control condition was found (p < 0.05 from 608638 milliseconds). See Figure 6. The absence of an N200 in the MSA analyses is in accord with findings from earlier studies of codeswitching, which showed increased N200 amplitude when switching from a first to a second language, but not the reverse, suggesting that MSA→PCA switching is analogous to switching from L2 to L1. The evidence of a late positivity (P600) in response to codeswitched sentences corroborates the view that MSA and PCA are represented in the brain as distinct languages.
LEXICAL PROCESSING IN TWO LANGUAGE VARIETIES
161
Figure 5: Grand average (N=5) ERPs to MSA sentence-final words (MSA control condition, MSA semantic condition and MSA code-switching condition); waveforms generated from fronto-central sensors (time window: 100-400 milliseconds).
µV
Figure 6: Grand average (N=5) ERPs to MSA sentence final words (MSA control condition, MSA semantic condition and MSA code-switching condition); waveforms generated using N400/P600 sensors (time window 300-750 milliseconds).
3.2.3 PCA analysis. For the PCA sentences, the early negativity (N200) is only expected if PCA is neurally represented as an L1 in contrast with L2 MSA. Grandaveraged waveforms derived from the fronto-central sensors showed a highly significant negativity in the early time window for codeswitched sentences only (p < 0.05 between 192-260 milliseconds). See Figure 7.
REEM KHAMIS-DAKWAR & KAREN FROUD
162
Figure 7: Grand average (N=5) ERPs to final words for PCA control sentences, PCA semantically anomalous sentences, and PCA code switched sentences. Waveforms generated from fronto-central sensors (time window: 100-400 milliseconds).
N400 effects in the PCA analyses were not found to be reliably associated with semantic anomaly, with control sentences showing the greater negativity in the N400 time window (statistically significant at p < 0.05 between 440-468 milliseconds). No N400 in codeswitched sentences was found. A late positive complex (P600) was found in both semantically anomalous and codeswitched sentences, but the positivity was significant only in the codeswitched condition (p < 0.05 between 552-616 milliseconds). See Figure 8. These results provide further evidence that PCA is a separate linguistic system, at the level of neurological functioning, from MSA. Furthermore, for the five subjects who participated in our study, PCA appears to be processed like a first language, with MSA effects being consistent with second language representations.
µV
Figure 8: Grand average (N=5) ERPs to final words for PCA control sentences, PCA semantically anomalous sentences, and PCA code switched sentences. Waveforms generated using N400/P600 sensors (time window 300-750 milliseconds).
LEXICAL PROCESSING IN TWO LANGUAGE VARIETIES
4.
163
Discussion In this study, we examined the neural responses to sentences containing diglossic codeswitches from MSA to PCA and from PCA to MSA. These were contrasted with ERP responses to non-codeswitched sentences, with and without semantically anomalous final words. Diglossic switches between the two varieties in this experiment elicited the pattern of ERP responses predicted from previous studies investigating codeswitching between two different languages (Jackson et al. 1999; Meuter & Allport 1999; Moreno et al. 2002; Jackson et al. 2004). In codeswitched sentences, N200 and P600 effects were present, but N400 effects were not found. P600 effects were evident no matter the direction of the codeswitch, which runs counter to the notion that switching from one language variety to the other is simply a matter of a within language “lexical switch”. The N200 effects, by contrast, were variety-specific, being in evidence only for PCA → MSA switches. We suggest that this supports a view of PCA and MSA as represented in the brain as two separate languages. Moreover, the specificity of the conditions under which N200 was elicited suggests that PCA is processed as a first language, in contrast with MSA. We found significant differences in the amplitude of the N400 to semantically anomalous final words in MSA, but not in PCA, compared to the control sentences. This might be correlated with the fact that larger N400s are expected for less predictable words (Kutas & Hillyard, 1980). We suggest that MSA is less familiar than PCA to the Arabic speakers who participated in our experiment, based on differences in the amount of exposure and use of the two varieties. Recall that PCA is used for daily interactions and is acquired naturally, whereas MSA is formally taught at educational institutions and is mainly used in formal interactions and literacy activities. It is possible, given a larger number of subjects, to correlate the non-significant N400 effect in PCA (compared with MSA) to the familiarity level of both languages; we tentatively suggest that MSA sentences incorporate less familiar words generally, and therefore may elicit larger N400 effects. The interpretation of the N400 differences between the two language varieties remains open, however. The results confirm that the ERP methodology can provide us with a direct way of measuring the cognitive status and processing of the two language varieties. The results also suggest many further directions for this kind of work. This study should be extended to incorporate a larger sample of native speakers with dialectal and educational experiences which are as uniform as possible; findings should also be reduplicated in different dialects. The effects of combining diglossic codeswitching and semantic anomaly in the same sentences should be investigated. There are undoubtedly structural constraints on diglossic codeswitching which can be investigated using ERPs, for instance by controlling stimuli in terms of position in phrase effect, or morphological and bare form insertion effects. ERP responses to diglossic codeswitching
164
REEM KHAMIS-DAKWAR & KAREN FROUD
should also be compared with codeswitching to a different language (such as English), even in the same native speakers of Arabic. Studies of the Arabic diglossia utilizing neurophysiological tools are relatively new. Schooling in the Arab world is conducted orally in Colloquial Arabic. However, textbooks and written tests and assignments are given in MSA. Accordingly, native Arabic speaking children are required to accomplish a proper use and understanding of Classical Arabic/MSA, which is not based on their experience as native speakers of a specific dialect. A great deal of anecdotal evidence suggests that this imposes a substantial additional learning burden (e.g., Elgibali 1996; Khamis-Dakwar 2005). Neurophysiological approaches have the potential to break new ground in our understanding of the mental representation of diglossic variation, which in turn could provide a basis for new approaches to language teaching and acquisition research in diglossic communities. To summarize, our findings reveal that native Palestinian speakers process switching between the two language varieties similarly to switches between two different languages. Additionally, our data indicate that PCA is represented as a first language and MSA as a second language, for these participants. The study provides a preliminary basis for further comparative studies of similar situations where there is a mismatch between the language of instruction and the language of schooling. In particular, many urban African American children speak a language variety - African American Vernacular English (AAVE) - different from the language of the school (Delpit 1995; Wolfram & Schilling-Estes 1998). Neurophysiological investigations of the cognitive status of PCA, AAVE, and other vernacular language varieties, have the potential to provide evidence to enhance our understanding of cognitive representations of these socio-linguistic systems.
REFERENCES Abu-Rabia, Salim, David Share & Maysaloon Mansour. 2003. “Word Recognition and Basic Cognitive Processes among Reading-Disabled and Normal Readers in Arabic”. Reading and Writing 16.423-442. Badawi, Elsaid. 1973. Mustawayaat al-‘arabiyya al-mu‘aasflira fi Misflr [Levels of contemporary Arabic in Egypt]. Cairo: Dar Al-Ma‘aarif. Boudelaa, Sami & William Marslen-Wilson. 2000. “Non-concatenative Morphemes in Language Processing: Evidence from Modern Standard Arabic”. Proceedings of Workshop on Spoken Word Access Processes Vol. 1. Nijmegen, 23-26.
LEXICAL PROCESSING IN TWO LANGUAGE VARIETIES
165
---------- & Gareth Gaskell. 2002. “A Re-examination of the Default System for Arabic Plurals”. Language and Cognitive Processes 17:3.321-343. ---------- & William Marslen-Wilson. 2004a. “Abstract Morphemes and Lexical Representation: The CV-skeleton in Arabic”. Cognition 92:3. 271303. ---------- & William Marslen-Wilson. 2004b. “Allomorphic Variation in Arabic: Consequences for lexical processing and lexical architecture”. Brain and Language 90.106-116. Boussofara-Omar, Naima. 1999. Arabic diglossic switching in Tunisia: An application of Myers-Scotton’s MLF model. PhD diss., University of Texas, Austin. ----------. 2003. “Revisiting Arabic Diglossic Switching in Light of the MLF Model and its Sub-models: The 4-M Model and the Abstract Level Model”. Bilingualism: Language and Cognition 6:1.33-46. Delpit, Lisa. 1995. Other people’s children: Cultural conflict in the classroom. New York: New Press. Eid, Mushira. 1990. “Arabic linguistics: The current scene”. In Perspectives on Arabic linguistics I, ed. Mushira Eid, 3-37. Amsterdam & Philadelphia: John Benjamins. Elgibali, Alaa. 1996. “Beginning to understand Arabic”. In Understanding Arabic: Essays in contemporary Arabic linguistics in honor of El-Said Badawi, ed. Alaa Elgibali, 1-15. Cairo: AUC Press. El-Hassan, Shahir. 1978. “Variation in the Demonstrative System in Educated Spoken Arabic”. ARLI 9:1.32-57. Eviatar, Zohar & Raphiq Ibrahim. 2000. “Bilingual is as Bilingual Does: Metalinguistic abilities of Arabic-speaking children”. Applied Psycholinguistics 21:4.451-471. Falkenstein, M., J. Hohnsbein, J. Hoormann & L. Blanke. 1990. “Effects of errors in choice reaction tasks on the ERP under focused and divided attention”. In Psychological brain research, ed. C. Brunia, A. Gaillard, and A. Kok, 192-195. Amsterdam: Elsevier. Ferguson, Charles. 1959. “Diglossia”. Word 15.325-340. Hary, Benjamin. 1996. “The language continuum in Arabic multiglossia”. In Understanding Arabic: Essays in contemporary Arabic linguistics in honor of El-Said Badawi, ed. Alaa Elgibali, 69-90. Cairo: American University in Cairo Press. Holes, Clive. 1986. “The Social Motivation for Convergence in Three Arabic Dialects”. ITSL 61.27-67. Jackson, Georgina, Rachel Swainson, A. Mullin, Ross Cunnington & Stephen Jackson. 2004. “ERP Correlates of Receptive Language-Switching Tasks”. The Quarterly Journal of Experimental Psychology 57A:2.223-240. Jackson, Stephen, Georgina Jackson & Mark Roberts. 1999. “The Selection and Suppression of Action: Electrophysiological correlates of executive
166
REEM KHAMIS-DAKWAR & KAREN FROUD
control in humans”. Neuroreport 10.861-865. Khamis-Dakwar, Reem. 2005. “Children’s attitudes towards the diglossic situation in Arabic and its impact on learning”. In Languages, communities and education: A volume of graduate student research, ed. Zeena Zakharia and Tammy Arnstein, 75-86. New York: Teachers College, Columbia University. Kutas, Marta & Steven Hillyard. 1980. “Event Related Brain Potentials to Semantically Inappropriate and Surprisingly Large Words”. Biological Psychology 11.99-116. Mahmoud, Youssef. 1986. “Arabic after diglossia”. In The Fergusonian impact, ed. Joshua Fishman, 239-251. Berlin: Moutton de Gruyter. Meuter, Renata & Alan Allport. 1999. “Bilingual Language Switching in Naming: Asymmetrical costs of language selection”. Journal of Memory and Language 40.25-40. Mitchell, T. F. 1982. “More than a matter of ‘writing with the learned, pronouncing with the vulgar’: Some preliminary observations in the Arabic Koine”. In Standard languages: Spoken and written, ed. W. Hass, 123-156. Manchester: Manchester University Press. Moreno, Eva, Kara Federmeier & Marta Kutas. 2002. “Switching Languages, Switching Palabras (Words): An electrophysiological study of codeswitching”. Brain and Language 80.188-207. Myers-Scotton, Carol. 1993. Duelling languages: Grammatical structure in codeswitching. Oxford: Oxford University Press. ---------- & J. L. Jake. 2001. “Explaining aspects of code-switching and their implications”. In One mind, two languages: Bilingual language processing, ed. J. L. Nicol, 84-116. Malden, MA: Blackwell. Owens, Jonathan & Raslan Bani-Yasin. 1987. “The Lexical Bases of Variation in Jordanian Arabic”. LING 25.705-738. Rosenhouse, Y. 1997. “Morphological and Syntactical Acquisition of Spoken Arabic as Native Language: Preliminary findings”. Hilkat Lashon 27.7693. Saiegh-Haddad, Eleanor. 2005. “Correlates of Reading Fluency in Arabic Diglossic and Orthographic Factors”. Reading and Writing 18:6.559-582. Shlonsky, Ur. 1997. Clause Structure and Word Order in Hebrew and Arabic. New York: Oxford University Press. Suleiman, Salah. 1986. Jordanian Arabic between diglossia and bilingualism: Linguistic analysis. Amsterdam & Philadelphia: John Benjamins. Tucker, D. 1993. “Spatial Sampling of Head Electrical Fields: The geodesic sensor net”. Electroencephalography and Clinical Neurophysiology 87.145-163. Wolfram, Walt & Natalie Schilling-Estes. 1998. American English. Malden, MA: Blackwell.
IV
PRAGMATICS
A COGNITIVE APPROACH TO ANALYZING DEMONSTRATIVES IN TUNISIAN ARABIC∗
AMEL KHALFAOUI University of Minnesota
1.
Introduction Demonstratives have traditionally been analyzed as “pointing words” whose primary function is to indicate the relative spatial or temporal distance of a referent from the speech participants. Recent work, however, has moved away from this approach by giving alternative explanations for the use of demonstratives that focus on other cognitive and pragmatic meanings. Based on an analysis of spontaneous interactions of native Lao speakers collected on video, Enfield argues that the two Lao demonstrative determiners nii and nan do not encode a meaning of distance. He suggests instead that “pragmatic inference alone gives rise to the associations of nii with things ‘NEAR speaker’ and/or ‘here’, and nan with things ‘FAR from speaker’” (2003:115). Accordingly, Lao speakers choose between the two demonstratives based on their conceptions of “here-space”, which is determined by such factors as physical barriers, attention focus, and the addressee’s presumed access to information. Based on corpus analysis, Oh (2001) argues against the traditional claims made by a number of linguists (e.g., Halliday & Hasan 1976; Lyons 1977; and Greenbaum & Quirk 1990) that “the most basic use of this and that is as deictics (i.e., expressions used for pointing), the two being contrasted in terms of proximity to the speaker” (2001:124-5). Oh tested this claim against a corpus of written advertisements and showed that the most critical factor in determining the speaker’s choice among this, that, and it, which he analyzes as a demonstrative, is “focus” in the sense of calling the addressee’s attention to something for a particular purpose (the notion of imposed salience (Mulkern 2003)). For example, Oh found that this was used more frequently than that in central reference (i.e., when the referent is the advertised product or one of its parts or ingredients) when referring to entities already mentioned in discourse. He argues that this pattern constitutes compelling evidence that this is a high focus form, since it is used to direct the reader’s attention to the advertised ∗
This material is based upon work supported by the National Science Foundation under grant No. 0519890 (PI Jeanette Gundel). I would like to express my thanks to Professor Jeanette Gundel for her invaluable comments on previous drafts of this paper.
170
AMEL KHALFAOUI
product. Gundel et al. (1993) argue that the choice among demonstratives, like the choice among other determiners and pronouns, is determined by cognitive status, which is the assumed memory and attention status of the referent. This study aims to show how demonstratives in Tunisian Arabic (TA) are used to indicate cognitive status. The theoretical framework for this study is the Givenness Hierarchy (GH) proposed by Gundel et al. (1993). The GH is an implicational scale of six cognitive statuses ranked from the most restrictive, or in focus, to the least restrictive, or type identifiable (see Section 2). In a previous study (Khalfaoui 2004), I propose correlations between cognitive status and TA pronouns (personal and demonstrative pronouns) and determiners (the definite article and demonstrative determiners). The present study focuses on demonstratives by testing the hypotheses made about the correlations between cognitive status and single demonstrative forms, which occurred infrequently in the data used in the previous study. It also extends the analysis to phrases with double demonstrative determiners, one pre-nominal and one post-nominal (e.g., ha-l-ktab haða ‘this book’), which were not included in that study. The research methodology is also different from that of the earlier study. A written questionnaire that biases for the different statuses on the GH was developed, and participants were asked to choose, among a set of demonstrative forms, those they judged to be acceptable in the different contexts. This study also investigates whether the cognitive status associated with two demonstrative determiners within the same DP differs from the cognitive status associated with the individual forms when used separately. A review of the GH will be given in Section 2. The review is followed by a description of TA demonstrative pronouns and determiners in Section 3. Then, the methodology used for this study will be described in Section 4. Finally, results and discussion will be given in Section 5. 2.
The Givenness Hierarchy Gundel et al. (1993) propose an implicational scale of six cognitive statuses that are relevant in explaining the use of referring expressions (see Table 1). The cognitive statuses on the GH represent the assumptions speakers make about the memory and attention state of the addressee at the time of the utterance. Gundel et al. (1993) argue that different determiners and pronouns have these statuses as part of their conventional meaning. When speakers choose a particular form, they guide the addressees in restricting possible referents. For example, when speakers use the English indefinite article a, they are only expecting the speaker to identify the referent type. If they use the demonstrative determiner that, they restrict possible referents to the ones that are familiar to the addressee (i.e., already in memory).
DEMONSTRATIVES IN TUNISIAN ARABIC
UNIQUELY IDENTIFIABLE >
REFERENTIAL
IN FOCUS
ACTIVATED
FAMILIAR
>
>
>
it
that, this, this that N the N indefinite this N N Table 1: The Givenness Hierarchy (Gundel et al. 1993)
171 TYPE IDENTIFIABLE
> aN
The statuses on the GH are not mutually exclusive. Every status on the hierarchy entails all the lower statuses to the right. The GH thus predicts that a given form can be used for referents that meet the minimal required status, and it can also be used to encode higher statuses. For example, the English proximate demonstrative determiner this, which signals the status ACTIVATED, can be used to encode referents that are in focus, since IN FOCUS entails ACTIVATED. Forms signaled by the highest status, namely IN FOCUS, are the most restrictive, since their referents must meet the requirements of this status, while forms that signal the lowest status, TYPE IDENTIFIABLE, are the least restrictive, since they can be used to encode entities with any cognitive status on the hierarchy. The six cognitive statuses are briefly characterized below. TYPE IDENTIFIABLE: The addressee can access a representation of the object type described by the expression. This status is necessary and sufficient for the use of the English indefinite article a. The use of a in I met her in a restaurant is appropriate if the speaker assumes that the addressee knows the meaning of the word restaurant and can understand the type of thing the phrase a restaurant describes. REFERENTIAL: The speaker intends to refer to a particular object. The addressee needs to access an appropriate type of representation and s/he must retrieve an existing representation of the speaker’s intended referent or construct a new representation. The status REFERENTIAL is necessary and sufficient for the appropriate use of the indefinite this in Colloquial English. In (1), the speaker explicitly signals that she expects the hearer to construct a representation of the intended referent of the DP this guy by the time the sentence is processed. (1) I was coming to see you but this guy stopped me and asked me to give him directions to the children’s hospital. UNIQUELY IDENTIFIABLE: By using a form that signals this status, the speaker expects the addressee to identify the referent based on a previous mental representation in his/her memory or to construct a new unique representation, if enough descriptive content is encoded in the DP. In English, the status UNIQUELY IDENTIFIABLE is necessary and sufficient for the appropriate use of the definite article. For instance, the phrase the apartment
172
AMEL KHALFAOUI
next to mine in John lives in the apartment next to mine is acceptable even if the addressee has no previous knowledge of its referent, as it contains enough descriptive content that allows the addressee to form a unique representation of the apartment. FAMILIAR: The addressee can identify the intended referent because s/he already has a representation of it in memory (in long-term memory if it has not been recently mentioned or in short-term memory if it has). In English, the status FAMILIAR is necessary and sufficient for the appropriate use of the demonstrative determiner that. The phrase that apartment (next to mine) in John bought that apartment (next to mine) is appropriate only if the addressee already knows about the apartment the speaker is referring to. ACTIVATED: A referent is activated if it is represented in the addressee’s current short-term memory. The addressee can identify the referent, since it has been recently mentioned or because it is present in the extralinguistic context. In English, the status ACTIVATED is necessary and sufficient for the appropriate use of the demonstrative pronoun that, for the demonstrative pronoun and determiner this, and for stressed personal pronouns. The demonstrative pronoun that in that is a good one can be used appropriately to refer to an entity that both the speaker and the addressee can see or have just mentioned. IN FOCUS: A referent is in focus when it is in the short-term memory and also at the current center of attention. In English, this status is necessary for the appropriate use of unstressed pronominals. Gundel et al. indicate that membership in the IN-FOCUS set is partially determined by the syntactic position of the antecedent: “subjects and direct objects of matrix sentences are highly likely to bring a referent into focus” (1993:279). For instance, the unstressed personal pronoun it in (2) is felicitous because its referent is previously mentioned in the subject position of the previous sentence. (2) This apartment is huge. It must be very expensive. 3.
Tunisian Arabic demonstratives Tunisian Arabic has two demonstratives: the proximate haða and the nonproximate haðaka. They are non-reduced demonstratives which can function as determiners as well as pronouns1, and are marked for number and gender. When they function as determiners, they always occur post-nominally. TA has two reduced pre-nominal demonstratives: the proximate ha and the nonproximate hak. They function as determiners only and they do not carry number or gender features. This characteristic is not exclusive to TA. Other dialects such as Syrian Arabic (Cowell 1964), Kuwaiti and Moroccan Arabic 1
By “demonstrative determiner”, I refer to any demonstrative form that requires an NP complement. By “demonstrative pronoun”, I refer to any demonstrative form that functions as a DP by itself.
DEMONSTRATIVES IN TUNISIAN ARABIC
173
(Brustad 2000), Lebanese and Palestinian Arabic (Omar 1976) also have invariant pre-nominal demonstratives which function as determiners only. In this paper, the TA pre-nominal proximate and non-proximate demonstratives ha and hak will be glossed as “proximate1” and “non-proximate1”, while the post-nominal proximate and non-proximate demonstratives haða and haðak will be glossed as “proximate2” and “non-proximate2”, respectively. TA allows the using two demonstratives within the same DP (e.g., hak lktab haðaka ‘that book’). This characteristic is not common to all other Arabic dialects, but it is not exclusive to TA, as Syrian and Moroccan Arabic allow two demonstratives within the same DP (Cowell 1964; Youssi 1992). Like all other Arabic dialects, TA demonstrative determiners co-occur with the definite article, e.g., ha-l-ktab ‘this the book’, l-ktab haða ‘the book this’, except in possessive constructions as in ktab saara haðaka ‘that book of Sarah’s’. 3.1 Proximate demonstratives TA has one invariant pre-nominal proximate demonstrative determiner, namely ha, that does not carry number or gender features. In (3) and (4), for instance, ha remains invariant before the singular feminine NP l-tawla ‘the table’ and the singular masculine NP l-kursi ‘the chair’. It also remains invariant before the plural feminine NP l-twawil ‘the tables’, and the plural masculine NP l-krasi ‘the chairs’. Finally, ha can function only as a determiner, not as a pronoun, hence the ungrammaticality of (5). (3) ha-l-tawla /l-kursi prox.1-the-table /the-chair “this table/chair”
(4) ha l-twawil /l-krasi prox.1 the-tables /the-chairs “these tables/chairs”
(5) *n-ħib ha 1s-like prox.1 “I like this.” TA has a post-nominal proximate determiner, namely haða, that is marked for number and gender in the singular, and for number only in the plural. The proximate2 determiner haða agrees with the singular masculine NP l-bab ‘the door’ in (6), while the singular proximate2 determiner haði agrees with the singular feminine NP l-tawla ‘the table’ in (7). In (8), however, the proximate2 determiner haðum occurs with the plural masculine and feminine NPs, as it carries only number agreement features. (6) l-bab haða the-door prox.2.sm “this door”
(7) l-tawla haði the-table prox.2.sf “this table”
AMEL KHALFAOUI
174
(8) l-bnaat/l-wlaad haðum the-girls/the-boys prox.2.p “these girls/boys” The proximate2 haða can be used as a demonstrative determiner, as in (6)-(8) above, or as a pronoun, as in (9) and (10). (9) A: aš žib-t? what brought-2sm “What did you bring?” B: žib-t haði brought-1s prox.2.sf “I brought this (one).” (10)
n-ħib haðum 1s-like prox.2.p “I like these.”
3.2 Non-proximate demonstratives As shown in examples (11) and (12), TA has an invariant pre-nominal non-proximate demonstrative determiner, namely hak, that does not carry number or gender features. In (11) hak is used with the singular masculine lražil ‘the man’ and the singular feminine l-mra ‘the woman’, and in (12) it is used with the plural masculine NP l-ržaal ‘the men’ and the plural feminine NP l-nsa ‘the women’. Like the proximate1 ha, the non-proximate1 hak cannot be used as a pronoun, as shown in (13). (11) hak l-ražil /l-mra non-prox.1 the-man/the woman “that man/woman” (12) hak l-ržaal /l-nsa non-prox.1 the men/the women “those men/women” (13) *n-ħib hak 1s-like non-prox.1 “I like that.” The post-nominal non-proximate demonstrative carries number and gender agreement features in the singular, as in (14) and (15), and number agreement
DEMONSTRATIVES IN TUNISIAN ARABIC
175
features only in the plural, as in (16). The non-proximate2 haðaka can be used as a pronoun, as in (17). (14) l-wlad haðaka the-boy non-prox.2.sm “that boy” (15) l-bnaya haðika the-girl non-prox.2.sf “that girl” (16) l-wlaad/ l-bnaat haðukum the-boys/ the-girls non-prox.2.p “those boys/girls” (17) A: škun qal-li-k? who told.3sm-to-you.sf “Who told you?” B: haðaka prox.2.sm “that one” 3.3 Double demonstratives In addition to being used individually, the TA pre-nominal and postnominal proximate demonstrative determiners and the pre-nominal and postnominal non-proximates can co-occur within the same DP, as shown in (18) and (19). When they co-occur, both must be proximate or non-proximate, as shown by the ungrammaticality of (20) and (21). (18) ha l-ražil haða bahi prox.1 the-man prox.2.sm nice “This man is nice.” (19) hak l-ražil haðaka bahi non-prox.1 the-man non-prox.2.sm nice “That man is nice.” (20) *ža hak l-ražil came.3sm non-prox.1 the-man “This man came.”
haða prox.2.sm
176
AMEL KHALFAOUI
(21) *ža ha l-ražil haðaka came.3sm prox.1 the-man non-prox.2.sm “This man came.” 4.
Analysis In a previous study (Khalfaoui 2004), I proposed correlations between cognitive status and TA pronouns (personal and demonstrative pronouns) and determiners (the definite article and demonstrative determiners). The results show that the pre-nominal proximate ha, the post-nominal proximate haða, and the post-nominal non-proximate haðaka all require at least the status ACTIVATED, while the post-nominal non-proximate hak requires only the status FAMILIAR. The present study uses a different research methodology to further investigate demonstrative pronouns and determiners in TA by testing the hypotheses made in the previous study and by extending the analysis to double demonstrative determiners. In this study, 13 adult native speakers of TA, 6 females and 7 males, filled in a paper and pencil questionnaire written in TA using the Arabic script. Since the participants come from different educational backgrounds, no specialized or technical language was used; only everyday conversational language. The questionnaire consists of 11 sentences or short conversations. The context in each test item was biased toward a different cognitive status on the GH, adapting the coding guidelines developed by Gundel (2004). Each sentence or short conversation was preceded by an indication of the memory and attention state of the addressee with respect to the referent of the demonstrative expression. Of the 11 test items, two were biased toward the status IN FOCUS, three toward the status ACTIVATED, two toward the status FAMILIAR, two toward the status UNIQUELY IDENTIFIABLE, one toward the status REFERENTIAL, and one toward the status TYPE IDENTIFIABLE. The 11 DPs that follow each sentence or short conversation consist of four DPs with the four TA demonstrative determiners, two DPs with the two TA demonstrative pronouns, two DPs with double demonstratives, and three DPs that did not include demonstratives, which were used as fillers to distract the participants. The sentences and short conversations were presented in random order. An example from the questionnaire translated into English is provided in (22), where the context is biased toward the status FAMILIAR. (22) Context: Speaker B knows that somebody gave Sonia a gift, but she was not thinking about it [the gift] before Speaker A mentioned it. A: sonia µžib-ha ………… Sonia pleased.3sm-her ………………. “Sonia was pleased by …………..”
DEMONSTRATIVES IN TUNISIAN ARABIC
177
B: µraf-t-u baš yi-µžib-ha knew.1s-it fut. 3sm-please-her “I knew it was going to please her.” 1. hak l-kado haðaka ‘that gift’ 2. kadu-ha ‘her gift’ 3. ha-l-kado ‘this gift’ 4. l-kado mtaµha ‘her gift’ 5. l-kado haða ‘this gift’ 6. haðaka ‘that’
7. l-kado haðaka ‘that gift’ 8. ha-l-kado haða ‘this gift’ 9. l-kado ‘the gift’ 10. haða ‘this’ 11. hak l-kado ‘that gift’
When a form is chosen in a context biased toward a particular cognitive status, but not in contexts biased toward lower statuses, that status is taken as the one that is necessary for the appropriate use of that form. For example, if a form is placed in contexts biased toward at least FAMILIAR, but not at most UNIQUELY IDENTIFIABLE, REFERENTIAL or TYPE IDENTIFIABLE, the status FAMILIAR was determined to be the one that is necessary for the appropriate use of that form. 5.
Results and discussion Table 2 presents the results of the questionnaire testing the hypothesized correlations between the GH statuses and TA demonstrative pronouns and determiners. It shows the correlations between cognitive status and TA double demonstratives, which were not considered in the previous study. The table also shows the maximum number of tokens that could be chosen for each form. Since there are three sentences biased toward the status ACTIVATED, if a particular form is chosen by every participant, the maximum number of tokens for that form under the status ACTIVATED would be 39. IN FOCUS
ACTIVATED
FAMILIAR
Maximum possible 26 39 26 ha-NP 20 28 NP haða 12 32 haða 0 4 haðaka 1 9 NP haðaka 11 26 hak N 8 16 23 ha N haða 8 28 hak N haðaka 6 14 5 Table 2: Results of the questionnaire according to highest status of the referent
If a form is chosen for a context biased toward a particular status but not in contexts biased toward lower statuses, that status is determined to be necessary and sufficient for the appropriate use of that form. For instance, hak
178
AMEL KHALFAOUI
is taken to signal the status FAMILIAR, as the participants chose it when the referent was assumed to be at most FAMILIAR, but not when it was assumed to be at most UNIQUELY IDENTIFIABLE, REFERENTIAL or TYPE IDENTIFIABLE. All the forms considered in this study were placed in contexts biased toward the status FAMILIAR or higher but not in any of the lower statuses, indicating that the referent of a demonstrative has to be in memory. Table 2 also shows that the number of tokens chosen for the forms associated with the same cognitive status varies significantly. For example, the participants chose 4 tokens of the pronoun haða and 32 tokens of the determiner haða, although both forms are associated with the status ACTIVATED. It seems that other factors interact with the GH in restricting the choice among the different forms. 5.1 Proximate demonstratives The results of this study confirm the findings of the previous one, as the pre-nominal and the post-nominal determiners ha and haða, as well as the pronoun haða, require the status ACTIVATED, since they were not chosen when the context was biased toward any of the lower statuses. For example, in (23), the DP ha-l-ħis ‘this noise’, and the DP l-ħis haða ‘this noise’ were both chosen to refer to the noise the speaker and the addressee could hear at the speech event. The noise is assumed to be activated, since it is part of the immediate spatio-temporal context. In (24), the DP l-ħkaya haði ‘this story’ was chosen to refer to the fact that Leila changes her mind everyday. Its referent is assumed to be activated, since it refers to a fact that is inferable from the preceding sentence. (23) Context: Both speaker and addressee can hear a noise. qallaq-ni ha-l-ħis /l-ħis haða bothered.3sm-me prox.1-the-noise/the-noise prox.2.sm “This noise bothered me.” (24) Context: The addressee already knows that Leila changes her mind everyday. The speaker knows that the addressee already knows that about Leila. leila kul yum t-baddil ray-ha w-l-ħaq l-ħkaya Leila every day 3sf-change mind-her and-the-truth the-story haði mqallqit-ni prox.2.sf bothering-me “Leila changes her mind everyday, and this fact is bothering me.”
DEMONSTRATIVES IN TUNISIAN ARABIC
179
None of the participants chose the proximate demonstrative determiners and pronouns in (25), where the context was biased toward the status FAMILIAR. In this context, the addressee is assumed to know the referent of lkuntabli ‘the accountant’, but the accountant is not activated, since the addressee could not be assumed to be thinking about him before he was mentioned by the speaker. (25) Context: The addressee knows the accountant, but he was not thinking about him before the speaker mentioned him (the accountant). talb-ik *haða /*l-kuntabli haða /*ha-l-kuntabli called-you prox.2.sm/the-accountant prox.2.sm/prox.1-the-accountant “This accountant called you.” 5.2 Non-proximate demonstratives This study also confirms that the determiner hak requires the status FAMILIAR, as it was chosen in contexts biased towards the status FAMILIAR, but was not chosen in contexts biased toward lower statuses. For example, in (26) the referent of hak l-kado ‘that gift’ is assumed to be familiar, since both the speaker and addressee know that somebody gave Sonia a gift, but when Speaker A mentions it, the addressee is not thinking about it, and therefore it could not have been activated. (26) Context: Speaker B knows that somebody gave Sonia a gift, but she was not thinking about it [the gift] before Speaker A mentioned it. A: sonia µžib-ha hak l-kado Sonia pleased.3sm-her prox.1 the- gift “That gift pleased Sonia.” B: µraf-t-u baš yi-µžib-ha knew.1s-it fut. 3sm-please-her “I knew it was going to please her.” However, in (27) where the context is biased toward at most UNIQUELY hak was not chosen, indicating that hak requires familiarity; hence the unacceptability. In (27), the speaker is asking about a house that he has never seen. The addressee answers the question by giving information about the house as well as a kitchen in that house. The referent of l-kužina ‘the kitchen’ is assumed to be UNIQUELY IDENTIFIABLE, having acquired this status via bridging to the DP d-dar l-ždida ‘the new house’, which is already
IDENTIFIABLE,
AMEL KHALFAOUI
180
activated. However, it is not familiar, as the addressee cannot be assumed to have a representation of the kitchen in memory. (27) A: bahya d-dar l-ždida good the-house the-new “Is the new house a good one?” B: ih ama #hak l-kužina yes but #non-prox.1 the-kitchen “Yes, but that kitchen is small.”
zγira small
As in the previous study, the pronoun and post-nominal determiner haðaka signal the status ACTIVATED. Both were chosen in contexts biased toward the status ACTIVATED, but not in contexts biased toward lower statuses. For instance, the DP l-burtman haðaka ‘that apartment’ was chosen in (28), where its referent is assumed to be activated, since it is mentioned in the preceding sentence. None of the participants, however, chose haðaka to be placed in contexts biased toward being at most FAMILIAR, as in (29). (28) Context: Speaker B knows the apartment that A is talking about but was not thinking about it before Speaker A mentioned it. A: salwa naqlit min dar-ha l-qdima w-šrat burtman salwa moved.3sf from house-her the-old and-bought-3sf apartment “Salwa moved from her old house and bought an apartment.” B: ih fi-bal-i w-l-burtman haðaka huwa lli ħabb-it nišri-h yes in-mind-my and-the-apartment non-prox.2. he rel. liked.1s buy.1s-it “Yes, and it is that apartment that I wanted to buy.” (29) Context: Speaker B knows the accountant, but he was not thinking about him before Speaker A mentioned him (the accountant). A: talb-ik #l-kuntabli haðaka called.3sm-you #the-accountant non-prox.2 “That accountant called you.” B: waqtah “When?”
DEMONSTRATIVES IN TUNISIAN ARABIC
181
5.3 DPs with double demonstrative determiners One aim of this study is to determine whether TA DPs with two demonstrative determiners differ in cognitive status from those with single demonstratives. The results show that double proximate demonstratives DPs do not differ in cognitive status from DPs where either of the two demonstratives was used alone. Double proximate determiners require at least ACTIVATED. In (30) the double proximate demonstrative determiner was chosen with the NP l-burtman ‘the apartment’. The referent of the DP ha-lburtman haða is assumed to be ACTIVATED, because it is already mentioned in the previous sentence. None of the participants, however, chose the proximate double demonstrative in contexts biased toward the status at most FAMILIAR. For example, it is not acceptable in (31) where the context is biased toward the status FAMILIAR. (30) Context: Speaker B knows the apartment that Speaker A is talking about. A: salwa naql-it min dar-ha l-qdima w- šra-t burtman Salwa moved.3sf from house-her the-old and-bought-3sf apartment “Salwa moved from her old house and bought an apartment.” B: ih fi-bal-i w-ha-l-burtman haða huwa lli yes in-mind-my and-prox.1-the-apartment prox.2 he rel. ni-šri-h ana 1s.buy- it I “Yes, and it is this apartment that I wanted to buy.”
ħabb-it wanted-1s
(31) Context: Speaker B knows the accountant, but he was not thinking about him before Speaker A mentioned him (the accountant). A: talb-ik #ha-l-kuntabli haða called.3sm-you #prox.1-the-accountant prox.2 “That accountant called you.” B: waqtah “When?” With double non-proximate determiners, only familiarity is required, just like the examples where the non-proximate demonstrative determiner hak is used alone. The double non-proximate demonstrative determiner was not chosen by any of the participants in contexts biased toward a status lower than at most FAMILIAR. In (32), the double non-proximate determiner was chosen with the NP l-kado ‘the gift’. The context in (32) is biased toward the status
182
AMEL KHALFAOUI
FAMILIAR, as the addressee has a representation of the gift in mind, but he was not thinking about it before the speaker brought it up. None of the participants, however, chose the double non-proximate determiner in (33) where the context is biased toward the status UNIQUELY IDENTIFIABLE.
(32) Context: Speaker B knows that somebody gave Sonia a gift, but she was not thinking about it [the gift] before Speaker A mentioned it. A: sonya µžib-ha hak l-kado haðaka Sonia pleased.3sm-her non-prox.1 the-gift non-prox.2 “That gift pleased Sonia.” B: µraf-t-u baš yi-µžib-ha knew-1s-it fut. 3sm-please-her “I knew it was going to please her.” (33) Context: The addressee doesn’t know the grocer that the speaker is talking about. The speaker knows that the addressee doesn’t know the grocer. #ma-nu-ftur-š l-sbaħ fi-d-daar NEG-1s-eat-NEG the-morning in-the-house l-µattar haðaka nišri kaskrut kul yum nitµadda l-hak every day pass to-non-prox.1 the-grocer non-prox.2 buy sandwich “I don’t eat at home in the morning. Everyday I stop by that grocer and buy a sandwich.” 5.4 Distribution of the demonstrative forms according to cognitive status As shown in Table 2, all TA demonstratives correlate with at least the status FAMILIAR. A similar tendency is found in studies on other languages. For example, Gundel et al. (1993) propose correlations between the different statuses on the GH and pronouns and determiners in Mandarin Chinese, English, Japanese, Russian, and Spanish. With the exception of Chinese, where the distal demonstrative determiner nèi only requires the status UNIQUELY IDENTIFIABLE, all demonstrative forms in these languages require the status FAMILIAR or higher. Ariel (1988) argues that demonstrative phrases are “mid accessibility” markers, which refer to less accessible information than personal pronouns, and to higher accessible information than full definite DPs without a demonstrative determiner. She found that demonstratives refer to things that are previously mentioned in the text or in the physical surroundings. Botley and McEnery (2001) use a corpus of English to demonstrate that demonstratives have an accessibility status that is intermediate between that of pronouns and DPs with the definite article.
DEMONSTRATIVES IN TUNISIAN ARABIC
183
With the exception of the proximate demonstrative pronoun haða, all TA demonstrative forms were chosen in contexts higher than the required ones. This is expected, since every status on the Givenness Hierarchy entails all lower statuses. In (34), the double proximate demonstrative determiner in ha-lmra haði ‘this woman’, which requires the cognitive status ACTIVATED, was judged to be acceptable in a context biased for the higher status IN FOCUS. The referent of ha-l-mra haði ‘this woman’ is assumed to be in focus, since it was mentioned in a syntactic focus position in the immediately preceding sentence. (34) Context: The addressee knows the neighbor that the speaker is talking about. A: žarti hiyya lli µawnit-ni µla tanðiif neighbor-my she rel. helped.3sf-me on cleaning “It is my neighbor who helped me clean the house.”
l-daar the-house
B: l-ħaq nas mlaħ ha-l-mra haði the-truth people nice prox.1-the-woman prox.2 “This woman is really nice.” The number of tokens chosen for each form does not reach the optimal possible number. The results of the questionnaire demonstrate that even when certain forms meet the criteria for appropriate use in certain contexts, none of the participants placed them in those contexts. For example, none of the participants placed the proximate demonstrative pronoun haða in (35). (35) Context: Both speaker and addressee hear a noise. #qallaq-ni haða bothered.3sm-me prox.1 “This bothered me.” Although the GH explains the restrictions on the use of demonstratives in terms of cognitive status, there are other factors that further restrict the choice among demonstrative forms which are associated with the same cognitive status. It is beyond the scope of this paper to address those factors, but some restrictions on the use of the proximate demonstrative pronoun haða, and the choice among the post-nominal proximate and non-proximate determiners haða and haðaka are provided below.
184
AMEL KHALFAOUI
5.5 The proximate demonstrative pronoun Table 2 demonstrates that there are only four tokens of the proximate demonstrative pronoun haða that were chosen in contexts biased toward the status ACTIVATED. Moreover, it shows zero tokens in contexts biased toward the status IN FOCUS. This section tries to give a possible explanation using the examples in (36) and (37). (36) Context: Both speaker and addressee hear a noise. qallaq-ni ha-l-ħis /l-ħis haða /ha-l-ħis haða/ bothered.-me prox.1-the-noise/the-noise prox.2/prox.1-the-noise prox.2/ #haða prox.2 “This noise/this bothered me.” (37) Context: Speaker B does not know the doctor that Speaker A is talking about. A: ðahr-i maµatš yu-žaµ-ni back-my NEG 3sm-hurt-me “My back does not hurt me anymore.” tbib-i µta-ni dwa ðarba ðarba doctor-my gave.3sm-me medication efficient efficient “My doctor gave me very efficient medication.” haða/ B: aš ism-u ha-l-tbib /l-tbib haða /ha-l-tbib what name-his prox.1-the-doctor/the-doctor prox.2/prox.1-the-doctor prox2/ #haða #prox.2 “What is the name of this doctor/this?” As shown in (36) and (37), while the use of a full DP with a proximate demonstrative determiner is acceptable, the use of the proximate demonstrative pronoun haða is not. In (37), the demonstrative pronoun can refer to either the doctor or the medication, and in (36) the proximate demonstrative determiner can refer to the noise or to any other person or thing that is activated at this point. The preference for a full DP over a demonstrative pronoun in (36) and (37) may thus be due to the fact that the referent of the DP with the demonstrative determiner further restricts the number of possible referents because of the conceptual content of the noun.
DEMONSTRATIVES IN TUNISIAN ARABIC
185
5.6 Conditions on the use of haða and haðaka The proximate2 haða and the non-proximate2 haðaka require the status ACTIVATED. It follows that they should be interchangeable in such contexts. However, the results suggest that there are cases where the two forms can be interchangeable, as in (38), and others where only the proximate2 haða can be used, as in (39). If both forms require the same status, the question arises as to what factors further restrict the choice between them. One possibility to investigate in future research is that the difference may have to do with the speaker’s prior familiarity with the referent in the case of haðaka. (38) Context: The addressee knows the neighbor the speaker is talking about. A: žarti hiyya lli µawn-it-ni µla tanðiif d-dar neighbor-my she rel. helped-3sf-me on cleaning the-house “It is my neighbor who helped me clean the house.” B: l-ħaq nas mlaħ l-mra haði/haðika the-truth people nice the woman prox.2/non-prox.2 “This/that woman is really nice.” (39) Context: Both speaker and addressee hear a noise. qallaq-ni l-ħis haða /#l-ħis haðaka bothered.3sm-me the-noise prox.2/the-noise prox.2 “This noise is bothering me.” 6.
Conclusion The results of the present study are consistent with those in (Khalfaoui 2004): only the pre-nominal determiner hak correlates with the status FAMILIAR and the rest of the demonstratives, including pronouns, require the status ACTIVATED. Double demonstratives do not differ in cognitive status from single demonstratives. Other factors may influence the distribution of different demonstratives. One such factor involves ambiguity avoidance, which seems to influence the choice between a demonstrative pronoun and a DP with a demonstrative determiner. It seems that the noun in a DP with a demonstrative determiner further restricts the number of possible referents. The post-nominal determiners haða and haðaka, which signal the cognitive status ACTIVATED, can be interchangeable in some contexts, but not in others. In this case, it looks like previous familiarity might be relevant to the restriction on the use of haðaka. Grammaticality judgments and corpus analysis are needed to further investigate these and other factors that interact with the GH in restricting the distribution and interpretation of TA demonstrative forms.
186
AMEL KHALFAOUI
REFERENCES Ariel, Mira. 1988. “Referring and Accessibility”. Journal of Linguistics 24.6787. Botley, Simon & Tony McEnery. 2001. “Proximal and Distal Demonstratives: A corpus-based study”. Journal of English Linguistics 29.214-233. Brustad, Kristen. 2000. The syntax of Spoken Arabic: A comparative study of Moroccan, Egyptian, Syrian, and Kuwaiti dialects. Washington, D.C.: Georgetown University Press. Cowell, Mark. 1964. A reference grammar of Syrian Arabic. Washington, D.C.: Georgetown University Press. Enfield, N. J. 2003. “Demonstratives in Space and Interaction”. Language 79.82-152. Greenbaum, Sidney & Randolph Quirk. 1990. A student’s grammar of the English language. London: Longman. Gundel, Jeanette, Nancy Hedberg & Ron Zacharski. 1993. “Cognitive Status and the Form of Referring Expressions in Discourse”. Language 69.274307. ---------- & Ann Mulkern. 1998. “Quantity Implicatures in Reference Understanding”. Pragmatics and Cognition 6.21-45. Halliday, Michael & Ruqaiya Hasan. 1976. Cohesion in English. London: Edward Arnold. Holes, Clive. 1990. Gulf Arabic. London & New York: Routledge. Khalfaoui, Amel. 2004. Associations between cognitive status and the form of referring expressions in Tunisian Arabic: An application of the Givenness Hierarchy Theory. M.A. paper, University of Minnesota. Lyons, John. 1977. Semantics Vol. 2. New York: Cambridge University Press. Mulkern, Ann. 2003. Cognitive status, discourse salience and information structure: Evidence from Irish and Oromo. PhD diss., University of Minnesota. Oh, Sun-Young. 2001. “A Focus-Based Study of English Demonstrative Reference”. Journal of English Linguistics 29.124-148. Omar, Margaret. 1976. Levantine and Egyptian Arabic. Washington, D.C.: Foreign Service Institute. Youssi, Abderrahim. 1992. Grammaire et lexique de l’Arabe Marocain modern. Casablanca: Wallada.
A PROSODIC FEATURE THAT INVITES BACK-CHANNELS IN EGYPTIAN ARABIC∗
NIGEL WARD & YAFFA AL BAYYARI University of Texas at El Paso
1.
Back-channeling as a dialog skill To be a good listener you have to be able to show you are listening. In dialog, this includes the active display of attention, interest, understanding and/or willingness to let the other person continue. This is accomplished in part with back-channels, also known as “minimal responses” and “continuers”: the short utterances produced while the interlocutor has the turn. In English these are typically utterances such as uh-huh; in Egyptian Arabic, the most common back-channels are ah, mmm, laughter, tayyeb ‘fine’, and aywa ‘yes’. This raises the question of how listeners know when it is appropriate to produce a back-channel. Work in other languages, e.g., English and Japanese, (Yngve 1970; Ward & Tsukahara 2000; Fujie et al. 2005) suggests that back-channeling depends on speaker-related and listener-related factors. That is, the listener is free to produce back-channels based on his/her own understanding and intentions, but these back-channels are especially welcome at certain times in the dialog, determined by what the speaker is saying and how he/she is saying it. Further, these times are indicated in part by prosody: in several languages there is a prosodic cue that a speaker can use to indicate when he/she welcomes a back-channel from the listener. The present study was motivated by the desire to be able to teach Arabic back-channel skills to non-natives, specifically by extending an intelligent tutoring system, the Tactical Language Trainer (Johnson et al. 2005). The motivating problem is that a second language learner who lacks turn-taking skills, even if a master of the vocabulary and grammar, can easily appear uninterested, ill-informed, thoughtless, discourteous, passive, indecisive, untrusting, dull, pushy, or worse. Indeed, our earlier study of Japanese and ∗
We thank Thamar Solorio, Mustafa Mughazy, W. Lewis Johnson, John Amastae and the participants of the 20th Arabic Linguistics Symposium for discussion and assistance. This work was supported in part by DARPA and in part by the National Science Foundation under Grant No. 0415150.
188
NIGEL WARD & YAFFA AL BAYYARI
English back-channel behavior shows that not only do Japanese back-channel twice as often as Americans (Maynard 1989), but the interval between the prosodic cue from the speaker to the back-channel response by the listener was typically only half as long in Japanese (Ward & Tsukahara 2000). The potential for awkward intercultural interactions here is clear. Unfortunately the rules governing turn-taking are seldom taught to language learners, largely because they are not known. This has been the case for Arabic. This paper describes the initial identification of a prosodic cue in Egyptian Arabic that indicates to the interlocutor when back-channel feedback is especially welcome, and which makes it statistically more likely that the listener will indeed produce a back-channel in response. 2.
Prosody and turn-taking in Arabic Research findings relating to our question are found in two areas: turn-taking and prosody. Back-channeling is an aspect of turn-taking, i.e., the way speakers in dialog manage their interactions to allow smooth exchanges and minimize awkward silences and interruptions. In Arabic, the only work on this is that of Hafez (1991), which provides a useful taxonomy of the ways in which speakers manage turn-taking in Egyptian Arabic. Hafez further identifies lexical discourse markers which often accompany turn taking and turn yielding, however not those for back-channeling. In his brief discussion of back-channels, Hafez provides examples of their semantic and pragmatic functions; these appear to be similar to those seen in other languages. In particular, some back-channels do not display understanding, but merely attention. Hafez notes that back-channels can occur not only in “slots” (places where the interlocutor is momentarily silent) but also in “overlap” with the other speaker’s turn; again this is also seen in other languages. The second relevant body of work is that on the prosody of phrases, sentences and utterances. This reveals prosodic features which express pragmatic functions intimately related to turn-taking, including utterance type distinctions as well as expressions of completion and finality. Regarding the prosodic correlates of different sentence types, the basic facts are that statements typically have a falling final pitch contour (Eldin & Rajouani 1999; Kulk et al. 2005; Rifaat 2005), as do wh-questions (El-Hassan 1988; Kulk et al. 2005). Yes-no questions generally exhibit a pitch rise (El-Hassan 1988; Eldin & Rajouani 1999). Also of interest are the prosodic correlates of finality and completion. Rifaat (2005) observes that a pitch rise (a “[LH#]” tone) occurring “turn-medially” at a “phrase boundary” can indicate “non-finality” in Cairene Arabic. His example shows this occurring in the enumeration of a list, but a similar pattern can appear at phrase boundaries; this perhaps serves the same function as the “comma intonation” contour of English. Similarly, both Kulk et
BACK-CHANNELS IN EGYPTIAN ARABIC
189
al. (2005) and Corvetto (1982) observe that in Damascene Arabic non-final utterances have a level or rising contour, in contrast to the falling contour characteristic of final utterances. A related phenomenon is that described by Bergsträsser (1968) as a “Kadenz” (cadence), indicating semantic completion in Damascene Arabic. Cadence is a musical term, referring to a sequence of chords at the close of a piece of music, where the chords are typically sustained for a whole note, with a drop of two semitones seen between the two chords. This stands in contrast to the general pattern of pitch progressions in Arabic: where pitch glides are far more common than pitch jumps, and sustained level pitches are rare. Bergsträsser further observes that vowel lengthening can co-occur as part of a final Kadenz. Kulk et al. (2005) also observe vowel lengthening pre-pausally, although de Jong and Zawaydeh (1999) observe that vowel lengthening may be less common in Jordanian Arabic. At this level of description, there seems to be general agreement, perhaps surprisingly so, given that these findings reflect study of a wide variety of discourse types, dialects, speaking styles, elicitation methods, analysis methods, and theoretical frameworks. 3.
Corpus preparation There are many ways to approach the prosody of turn-taking. For this project we adopt a corpus-based analysis for two reasons. First, back-channels are intrinsically a dialog phenomenon and so they can only be observed in naturally occurring dialogs. Second, cause-effect relations at this time scale are not introspectable and so must be studied empirically. 3.1 The corpus The corpus used was the CallHome Corpus of Egyptian Arabic Speech (Canavan et al. 1997). This is a collection of “unscripted telephone conversations between native speakers of Egyptian Colloquial Arabic ... calls ... originated in North America and were placed to locations overseas (typically Egypt). Most participants called family members or close friends”. As a manageable-sized subset of the dialogs, we chose, pseudo-randomly, the first five dialogs from each of the three CDs. Specifically, this subset consists of the first 32 minutes of dialog AR_4023, the first 7 minutes of AR_4367, and the first 10 minutes of each of AR_4150, AR_4194, AR_4213, AR_4283, AR_4297, AR_4299, AR_4392, AR_4419, AR_4931, AR_4949, AR_4950, AR_4981, and AR_4985. This gave us 168 minutes. Later we realized that one of dialogs had severe line noise, but we decided against retroactively excluding it. These conversations were mostly between two people but a few contain parts where three or four people were active. Most speakers were adults, both women and men. Common topics included school, business matters, family
190
NIGEL WARD & YAFFA AL BAYYARI
health issues, financial transactions, babies, and general gossip and chit-chat. Dialog activities included openings and closings, telling family news, talking about work and school problems, invitations, persuasion, planning and making commitments, defending past actions, and joking, among others. Two speakers appeared to be not truly fluent in Arabic, and some English back-channels and back-channeling patterns were observed. 3.2 Identifying back-channels A native Arabic speaker (the second author) labeled all back-channels in this subset, giving 660 occurrences. Back-channels were labeled according to a standard definition: To count as an instance of back-channel feedback, an utterance had to meet three criteria: (D1) respond directly to the content of an utterance of the other (D2) be optional (D3) not require acknowledgement by the other Other definitions of back-channel exist, but in practice these all delimit roughly the same set of phenomena (Ward & Tsukahara 2000). Although designed for two-party dialogs, this definition also worked for the multi-speaker segments of the dialogs: for example, when one speaker made a statement and two listeners gave him back-channel feedback. In many cases the back-channel overlapped or interrupted the utterance of the main speaker; in the examples below this was salient in examples (10), (12), and (13). Back-channels were identified based on the audio only; transcriptions were added later, if at all. Back-channels in the corpus were fairly frequent, about four per minute. Although direct comparisons are not possible, this is similar to the frequencies reported for English and many other languages, more than in Chinese, and less than in Japanese (Maynard 1989; Clancy et al. 1996; Ward & Tsukahara 2000). However, back-channels were not evenly distributed. In particular, they were rare at the starts of the dialogs. This is probably for two reasons: first, most calls started with questions and answers (regarding, for example, family and health), and of course back-channels do not appear in such contexts, and second, at the beginning most speakers were somewhat uncomfortable, in part because they knew they were being recorded. There also seemed to be gender differences in back-channel use: women generally seemed to use fewer back-channels and their back-channels seemed to be shorter in duration than men’s. Large individual differences in interaction styles were also observed. For example, one dialog was completely lacking in back-channel behavior. Table 1 demonstrates the most frequent tokens used as back-channels. There was great variety: 115 different types. Note that this is a rough, broad
BACK-CHANNELS IN EGYPTIAN ARABIC
191
categorization, which obscures subtle phonetic variations and prosodic variations, although in fact such differences are likely to be pragmatically significant (Ward 2004, 2006). Token Gloss Occurrences ah yeah 234 mmm 147 laughter 76 tayyeb alright 28 aywa yes 18 aah yeah 16 yeah 16 haa/hah 14 ah ah yeah yeah 12 mmm-hum 11 maaši okay 11 yah 9 kuwayyis good 7 eeh yeah 7 other 54 Total 660 Table 1: Number of occurrences of various tokens as back-channels
3.3 Examples This subsection presents some examples of clear cases of back-channels (in italics), showing how they appear in context. The context is given in English and the back-channel itself in Arabic with glosses, unless the back-channel was an English word or a non-lexical utterance. (1) Speaker 1: Speaker 2: Speaker 1: (2) Speaker 1: Speaker 2: Speaker 1: Speaker 2: Speaker 1: (3) Speaker 1: Speaker 2: Speaker 1:
Last week I called your mom and she told me about your wedding mmm but I told her if it was any country other than Russia I’d go Dr. Malek gave me yesterday details of a cheap hotel for the reservation mmm-hum so I called and made the reservation yesterday kuwayyis awi “very good” and he welcomed me as he’ll be staying in the same hotel Yehya got married aah “yeah” and I was very glad to meet Amrawi at the wedding
192
NIGEL WARD & YAFFA AL BAYYARI
(4) Speaker 1: Speaker 2: Speaker 1: (5) Speaker 1: Speaker 2: Speaker 1:
They want people to speak in these languages and they record sample texts without saying who is speaking ah ah “yeah yeah” then they apply computer analysis to the data The tickets will cost us $2000-$2500 yaah, ah “oh! Yeah” this is just for the tickets because the baby needs a separate seat
3.4 Borderline examples Although identifying back-channels was unproblematic in the vast majority of cases, there were seventeen examples where classification was difficult. This is unavoidable, as there is no hard-and-fast distinction between back-channels and related phenomena. Although not particularly important for identifying prosodic cues to back-channels, here are some examples of these borderline cases. In several cases, a back-channel included a nuance of additional meaning. The example in (6) illustrates two back-channels which additionally convey a sense of agreement or acceptance of an offer. Similarly in (7) the back-channel also conveys a nuance of agreement: Speaker 2 is agreeing that Speaker 1’s plan is wise. Incidentally, Speaker 3’s initial agreement here is not labeled as a back-channel, since it continues on to be a full turn. Although back-channels are by definition optional, sometimes it seems that the speaker is expecting (although not requiring) a back-channel, as in (8). Although this is not apparent from the transcription, it is clear from the audio. (6) Speaker 1: Speaker 2: Speaker 1: Speaker 3: Speaker 2: Speaker 3:
I sent you some pictures and I will send a couple of our pictures in each letter aywa “yes” and don’t worry about your stuff, your brother will transfer them to our apartment tab kuwayyis awi “ok very good” ah ah “yeah yeah”
(7) Speaker 1: Speaker 2: Speaker 3:
we are going as employees not investors aywa aywa “right right” ħelw ħelw “good good”, and of course keep your status as is
(8) Speaker 1:
We heard you are coming to Canada so Faten will see her aunt, aren’t you? no no
Speaker 2:
BACK-CHANNELS IN EGYPTIAN ARABIC
Speaker 1: Speaker 2: Speaker 1: Speaker 2:
193
you are not intending to go? no no, we were just talking okay (laughs) it was not serious
Although back-channels by definition do not require acknowledgement by the speaker, sometimes they do incorporate a nuance of asking for more information. In example (9), the okay? is such a case. The response part of a conversational routine (also called an adjacency pair) sometimes falls into the back-channel category. In (10), ya rab ‘please God’ is a fixed response, but being not required in this context, was considered to be a back-channel. The same is true for the hah in (9). (9) Speaker 1: Speaker 2: Speaker 1: Speaker 2: Speaker 1:
ok, listen Ashraf hah most likely the Canada thing is okay okay? and it’ll be August next year to do the immigration
(10) Speaker 1: Speaker 2: Speaker 1: Speaker 2:
I was praying to God that she gets well exactly, leave her in God’s hands and weya rab “please God” we will pray for her too
3.5 Related phenomena To illustrate how back-channels both resemble and differ from other types of utterances, this subsection presents three of the six examples that initially seemed to be back-channels but were ultimately judged not to be. In example (11) below, Speaker 2 is promising that he would do his best to go to Canada; this goes beyond the normal function of a back-channel and clearly contributes new meaning to the dialog. In (12), Speaker 1 waits for an answer, so the mmm is in effect a required response; and therefore it cannot be considered a back-channel. Finally, in (13) ya reet ‘I hope so’ was said by a mother to her son. Although similar to a back-channel in terms of position, it conveys the clear meaning that she agrees and wants him to study. (11) Speaker 1: Speaker 2: Speaker 3:
please Ashraf do your best to come to Canada µinayya “you got it” because we are thinking that if everything is ok in Canada why don’t you come and live with us?
194
NIGEL WARD & YAFFA AL BAYYARI
(12) Speaker 1: Speaker 2: Speaker 1: (13) Speaker 1: Speaker 2: Speaker 1: 4.
the expenses for studying a year here would be about $2000, do you see? mmm so I get the-, by the way this call is being recorded I’m thinking of studying there and come back ya reet “I hope so” for a year or two until we settle
Analysis method Our aim was to find a prosodic feature that commonly appeared just before a back-channel by the interlocutor, and then determine whether this was in fact functioning as a cue. In order to do this, we chose to directly examine the speech signal itself, especially the F0 (pitch), rather than using any particular model or theoretical framework. Although there are numerous models and frameworks useful for the study of prosody, these have all been designed primarily for read speech and monolog phenomena. As such, these models are not necessarily adequate for describing the prosodic features involved in turn-taking. Working directly with the signal is advocated as the “direct method” by Shriberg and Stolcke (2004), and has the advantage that the relationship between prosodic features and the phenomena of interest can be discovered, at least in principle, without requiring any hand-labeling of intermediate features, such as sentence boundaries, pitch-accents, target tones, tunes, or turn yields. We also chose to analyze the data eclectically. Some practitioners of the direct method avoid any use of linguistic knowledge or hunches by the analyst, preferring to rely only on signal processing, statistical analysis and machine learning methods. However, such approaches often result in neural nets or decision trees which succeed in classifying the data at the price of being uninterpretably complex. This project, however, required both a simple qualitative description and a quantitative one. The qualitative description is needed so that the initial tutorial module can explain the desired behavior in a simple way that learners could grasp, and the quantitative description is needed so that drills can incorporate an automatic evaluation of the learner’s performance, and also so that the trainer’s non-player characters (animated agents) can model authentic Arabic turn-taking behavior in unscripted, real-time interactions with the learner. Thus, we used an integrated method for discovering the prosodic cues involved in back-channeling; this gave both qualitative and quantitative descriptions. This method uses both perceptually-based analysis and quantitative analysis, tightly integrated, for the formulation and testing of hypotheses. These analysis phases included simply listening, visually
BACK-CHANNELS IN EGYPTIAN ARABIC
195
inspecting graphical representations of the pitch and energy, and writing small programs to detect and evaluate various putative back-channel-inviting features. The process was iterative in that the perceptually based and quantitative phases were alternated. In a sense, the purpose of listening was to understand how to improve the quantitative description, and the purpose of improving the quantitative description was to direct attention to informative cases in the corpus. The most informative cases were generally those that a tentative version of the quantitative description did not handle correctly, either by failing to identify a back-channel cueing place where, in fact, there was a back-channel in the corpus (imperfect coverage), or by incorrectly identifying a back-channel cuing place where in fact no back-channel occurred (imperfect accuracy). Further description of the analysis method appears elsewhere (Ward & Al Bayyari 2006). 5.
Cues for back-channels Although our focus of attention was on the prosody, we did notice a few lexical phrases which also seem to cue back-channel responses. These phrases include waaxid baalak ‘are you paying attention’, šaayif ezzaay ‘see how’, boss ‘look’ and ba>ollak eeh ‘tell you what’. These instances explained only two or three per cent of all the back-channel occurrences in the corpus. The first, most salient, prosodic cue is a pitch upturn at the end of a phrase. This appears to be a relatively strong cue, functioning perhaps like ‘up talk’ in English. A second, rarer prosodic cue, only tentatively identified, is a low flat pitch associated with a lengthened vowel at a disfluency point. Finally, a sharp pitch down-slope seems to be the most frequent cue. The rest of this section describes this cue. 5.1 Pitch downdash as a cue for back-channels Back-channels are frequently preceded by a certain feature complex produced by the other speaker. The most distinctive feature of this feature complex is a sharply falling region of pitch: borrowing Bolinger’s terminology, we call this a “downdash” (Bolinger 1989). This fall is generally steady, and it is almost linear when viewed in log scale. The downdash is typically set off from adjacent pitch contours by sharp corners, that is, it does not smoothly transition from or to a different contour. A downdash seems to be generally ineffective as a cue for a back-channel if it is closely followed by a region of nearly level pitch. The downdash seems to be most effective as an invitation for a back-channel when it occurs after some substantial amount of discourse. It also seems most effective when followed by a pause, either immediately after or following another syllable or word. Downdashes often fall on syllables which
196
NIGEL WARD & YAFFA AL BAYYARI
are lengthened. In some instances the downdash is immediately followed by a region of saliently different speaking style, such as creaky voice, breathy voice, significantly less energy, or an in-breath, and in some cases this transition falls in the middle of a lengthened vowel. 6.
Examples Example (14), which is illustrated in detail in Figure 11, shows two cases of back-channeling.
Figure 1: Dialog fragment including downdash, upturn, and two back-channels
1
Audio for these examples is available at
http://www.cs.utep.edu/isg/members/yaffa/Arabic_BC.html
BACK-CHANNELS IN EGYPTIAN ARABIC
(14) Speaker 1: Speaker 2: Speaker 1: Speaker 2: Speaker 1:
197
most likely we are going to stay for a month and a week just to see how life is there okay and probably Uncle and Aunt are coming with us okay so it’d be a good chance for you to come visit us
Both cases are instances of okay: one occurs at 302 seconds, overlapping the other participant’s speech, and the other at 304 seconds, during a pause. The first one seems to be a response to the downdash found at 301.750 seconds. The second seems to be a response to the pitch upturn at 303.500 seconds. In the figures, each of the two strips includes two tracks and a timeline. In each strip the top track is one speaker and the bottom track the other. Each track includes a transcription, the signal, the pitch, and English translation. The example in (15), detailed in Figure 2, shows a turn yield, leading the other speaker to take the turn.
Figure 2: Dialog fragment including a Kadenz at a turn yield
198
NIGEL WARD & YAFFA AL BAYYARI
(15) Speaker 1: Speaker 2:
and she was talking, so I told her that you knew for a year ok, she’s getting married to a Russian guy, but by the time Maha knew, I had not told anybody
The phrase sana yaµni ‘a year I mean’ at about 54 seconds appears with each syllable at a successively lower pitch level. This is therefore a four-step Kadenz; two and three-step Kadenzes are also seen in the corpus. Although only the first syllable has a clearly flat pitch, perceptually this sounds like a downward staircase of pitches. The pitch just after 54 is so low that it is displayed overlapping the time marker. The pitch here is spotty because the pitch tracker behaves poorly for regions of creaky voice like this one. Example (16), detailed in Figure 3, demonstrates a pitch downdash that was not followed by a back-channel, even though, in the judgment of the second author, it would also have been appropriate if the listener had chosen to respond with a back-channel at this point.
Figure 3: Dialog fragment including a downdash but no back-channel
(16) Speaker 1:
Sarah, that little girl, is so cute she’s killing me; I told my mom if I were her I would go see her
BACK-CHANNELS IN EGYPTIAN ARABIC
199
In general, there is an element of choice, perhaps even of randomness in backchannel behavior. It seems that a listener typically responds with back-channel feedback at only some fraction of the opportunities given. Note that the downdash starts around second 399.250, and the rule, as discussed below, would predict a back-channel 800 milliseconds later, as shown by the oval. 7.
Quantitative description of the downdash feature complex This section gives a quantitative description of this feature complex, based on the features noted in Section 5.1, but leaving out those discussed in the last paragraph of that section since they are harder to quantify, and since we are not sure that they are consistently and distinctively part of the cue. Thus, the feature complex is deemed present whenever there is a time point that meets the following conditions: C1: The time point is part of an utterance that has lasted at least 1.8 seconds. C2: It is preceded by a downdash lasting at least 30 milliseconds. C3: The pitch in the downdash drops by at least 0.8% every 10 milliseconds. C4: It is followed by a pause (low energy region) within no more than 600 milliseconds and this pause lasts at least 150 milliseconds. C5: It is not followed by a flat pitch region before the pause, where a flat pitch region is one in which the pitch stays within 0.4% of the average pitch in that region for a period of at least 80 milliseconds. C6: It is not preceded by another back-channel prediction within 900 milliseconds. If this feature complex occurs, a back-channel response is likely to occur some 800 milliseconds later. This rule is the best of the many variants we have tried, where “best” means performs best according to the criteria described in Section 8. The specific parameters of this rule are tuned to optimize performance, but do not necessarily closely describe the most typical cases of back-channel cues. The typical rate of pitch drop is steeper, around 1.2% every 10 milliseconds, which is a semitone every 50 milliseconds. The pitch drop typically lasts much more than the 30 milliseconds required by the rule. The pause usually comes sooner than 600 milliseconds after the pitch drop. Condition C5 reflects the fact that the downdash functions in contrast with the Kadenz (downward pitch “staircase”) commonly found at turn ends. Although the Kadenz pattern typically has several regions of flat pitch, it seems that just a single region of flat pitch is enough to revoke the back-channel invitation associated with a downdash. Condition C6 means that a downdash occurring shortly after another downdash is treated as a reinforcement of the invitation to
200
NIGEL WARD & YAFFA AL BAYYARI
back-channel, rather than as an invitation for another back-channel. 8.
Evaluation As mentioned in Section 4, our analysis method relied on the use of the quantitative description to identify informative examples. Each putative rule, such as the one given above, was treated as a predictive rule. The task accomplished by such a rule is this: given some initial portion of one track of a two-person dialog, predict whether or not a back-channel is about to appear in the other track. This formulation makes it easy to identify false predictions, that is, places where the cue occurred but a back-channel did not follow, and missing predictions, where a back-channel occurred but no cue preceded it. The specific criterion for deciding whether a prediction was a success or failure was whether the predicted back-channel point occurred within half a second of the actual onset of a back-channel in the corpus; this window of opportunity was allowed since a back-channel can be produced slightly earlier or slightly later and have the same pragmatic effect (Ward & Tsukahara 2000). 8.1 Rule performance According to the F Measure, a way to combine accuracy and coverage into one unified performance measure, the rule given above is better than all the other rules we have considered so far. It accounts for 44% of the back-channels in our sub-corpus. One reason why this is less that 100% is that in Egyptian Arabic there also seem to be other prosodic features which cue back-channels, as noted above. Moreover, some back-channels seem to be produced in response to the semantic content of the speaker’s utterances, with prosody less important. The accuracy of the rule is 15%: it makes many predictions at places where a back-channel was not actually present. Since the accuracy expected by random predictions is 3.3%, this rule clearly has some predictive power. The reasons why the accuracy is less than 100% probably include inter-speaker differences (no single rule can be expected to model the behavior of all speakers in the corpus), the random element in back-channeling as noted above, and the limitations of the corpus. A more interesting reason is the inability of our current speech processing environment to handle pitch in regions of creaky voice, which are common in this corpus. 8.2 Utility The rule clearly has some validity, but its actual utility is not known: we do not yet know what level of performance would count as adequate, for example, to enable a computer system to be judged “as good as” (that is, an acceptable model of) a native speaker of Arabic in back-channeling ability, nor to permit a rule to be judged as “good enough” for pedagogical purposes.
BACK-CHANNELS IN EGYPTIAN ARABIC
201
9.
Open questions The first priority for future work is to refine this rule. It is undoubtedly not the best possible, and needs to be improved by further tuning the parameters or adding new parameters. Another obvious priority is an examination of how the occurrence of this prosodic cue and back-channels relates to various semantic, pragmatic, and interpersonal dimensions of dialog. The rest of this section mentions some other topics that should be addressed. 9.1 Experimental confirmation Although this paper has shown a correlation between the occurrence of this feature complex in one track and the subsequent appearance of a back-channel in the other track, the existence of a causal relationship remains unproven. Demonstrating this may require controlled perceptual experiments with synthesized speech. 9.2 Iraqi Arabic Given that many of the prosodic properties of Arabic seem to hold across dialects, it is natural to wonder whether this is also true of pitch downdash as a cue for back-channels. We have begun to examine this for Iraqi Arabic, using our own corpus of dialogs (Ward et al. 2006). Unlike the CallHome corpus, this corpus was recorded with the two speakers in the same room. This has two advantages for our purposes: First, since the recordings are not as band-limited as telephone conversations, it is possible to get more accurate and complete pitch estimates. Second, there is no significant line delay (unlike that probably present in some of the cell phone dialogs in CallHome). The best downdash-based predictive rule found so far achieves a coverage of 51% and an accuracy of 16% for this corpus. There are a number of small differences between the parameters for this rule and the rule for Egyptian Arabic presented above (Ward & Al Bayyari 2006). There is also one large difference: for the Iraqi corpus there is typically a delay of 300 milliseconds between the occurrence of the downdash and the appearance of the back-channel, but in the Egyptian corpus the typical delay is 800 milliseconds. This difference may be due to the absence of line delay or to the effects of simultaneous gestural cues to back-channels in the Iraqi data collection. 9.3 Other factors In the course of analysis we considered many features as possible back-channel cueing factors. In particular, we sought a correlation between vowel lengthening and a subsequent back-channel response, without much success. This may mean that lengthening is not an independent cue for back-channels. Indeed, the lengthening observed with pitch downdashes may
202
NIGEL WARD & YAFFA AL BAYYARI
be there merely to provide enough phonetic content to realize this pitch pattern. We have also examined the possible role of speaker gestures as an additional way to indicate when back-channels are welcome, using our Iraqi data. Contrary to expectation, no correlations were found. 9.4 Relation to other prosodic phenomena Although the role of pitch-downdash as a back-channel cue could not have been inferred from previous descriptions of Arabic prosody, it does not contradict them, since grammatical and semantic finality or completion are not closely correlated with interactional completion or invitations. In future reseacrh we would like to examine how the prosodic feature complex identified here relates to other prosodic phenomena, such as declination and lexical stress. It would also be interesting to attempt a unified description of the prosody of dialog acts, utterance types, finality, and turn-taking. It would also be interesting to examine the theoretical significance of this result: certainly the existence of a large difference in interactional import between pitch downdash and the Kadenz, although both are ways of reaching a low pitch, suggests that target-based models of prosody are inadequate for describing these phenomena. 9.5 Pedagogy Insofar as producing back-channels at appropriate times is part of being a good listener, this skill should be taught to learners of Arabic. In particular, they should be taught to recognize the downdash feature complex, and to understand how it contrasts with the Kadenz pattern. They also should be trained to respond quickly when they hear the downdash in conversation. As both of these abilities are quite different in nature from most language skills, the development of novel teaching methods may be required. We have developed a prototype 30 minute training sequence including exposure to dialogs from the corpora (rather than professionally recorded conversations, since people acting out dialogs typically follow different rules). It also includes practice in both the speaker role and the listener role, using software that provides feedback on learners’ attempts to produce the cue themselves and feedback on learners’ performance as they play the role of an attentive listener in response to one side of a pre-recorded dialog. Preliminary results are positive (see Escalante et al. 2007). There is also the question of when back-channeling skills should be taught. If taught early, there is the advantage that learners needing to interact with Arabic speakers can show polite attention through active listening, thereby increasing the chance that the speaker will produce more comprehensible utterances (Kraut et al. 1982), and the chance that the dialog will continue long
BACK-CHANNELS IN EGYPTIAN ARABIC
203
enough for the listener to understand. On the other hand, teaching back-channeling skills early has risks, given the current state of knowledge. We do not know whether or how Arabic speakers talking to non-natives change their use of back-channel cueing; perhaps it is abandoned in favor of a more crude (pause-based) form of turn-taking, or perhaps replaced with gestural cues, or perhaps the cues are exaggerated. We also do not know whether an early-stage learner whose skill at attentive listening exceeds other language skills will be perceived by Arabic speakers as polite or as phony. 9.6 Cross-cultural impressions Finally, we are interested in the ways in which the prosodic contours we have identified are interpreted by non-native speakers. To English speakers, some languages of the world can seem beautiful to listen to, but Arabic is probably not one of them. This is because certain features of Arabic have different roles, including some with negative connotations in English, such as pharyngeals, which may sound harsh to speakers of English, and the absence of de-accenting for given information (Hellmuth 2005), contributing to frequent pitch variation or wider pitch range, which may connote anger to speakers of English (Murray & Arnott 1993). Less obviously, some pitch contours used in questions have similar semantic functions but starkly different “attitudinal meanings” in English and Arabic (El-Hassan 1988). The prosodic features identified in this paper also have negative connotations in their roles in English dialog. The pitch downdash resembles the utterance-final pitch pattern used in American English to make an accusation or authoritative imperative (Bolinger 1986:208), and the downward staircase or Kadenz (or “terraced monotone”) resembles a final pitch pattern used in American English to express discouragement or resignation (Bolinger 1986:231, 1989:324). Since native English speakers listening to Arabic speech are likely to interpret these features as reflections of the speaker’s personality or attitude, the potential for intercultural misunderstandings is clear. It would be interesting to examine in detail these divergences in interpretation and their significance. We hope that a clear understanding of such differences could help prevent some misunderstandings in interactions between early-stage Arabic learners and Arabic speakers; or even counteract some false impressions of Arabic speakers that casual listeners may pick up from Arabic sound bites on radio or television.
204
NIGEL WARD & YAFFA AL BAYYARI
REFERENCES Bergsträsser, G. 1968. Zum arabischen dialekt von Damaskus [On Damascene Arabic]. Georg Olms Verlagbuchhandlung. Originally published in 1924 by Orient-Buchhandlung Heinz Lafaire, Hannover: Beiträge zur semitischen Philologie und Linguistik. Bolinger, Dwight. 1986. Intonation and its parts. Stanford: Stanford University Press. ----------. 1989. Intonation and its uses. Stanford: Stanford University Press. Canavan, Alexandra, George Zipperlen & David Graff. 1997. CALLHOME Egyptian Arabic Speech. Linguistic Data Consortium. LDC, University of Pennsylvania. Clancy, Patricia, Sandra A. Thompson, Ryoko Suzuki & Hongyin Tao. 1996. “The Conversational Use of Reactive Tokens in English, Japanese and Mandarin”. Journal of Pragmatics 26.355-387. Corvetto, Ines. 1982. “L’intonazione nell’arabo siriano” [The Intonation of Syrian Arabic]. Lingua e Stile 17.371-393. de Jong, Kenneth & Bushra Zawaydeh. 1999. “Stress, Duration, and Intonation in Arabic Word-Level Prosody”. Journal of Phonetics 27.3-22. Eldin, Nasser & A. Rajouani. 1999. “Analysis and Synthesis of Interrogative Intonation in Arabic”. Proceedings of the International Congress of the Phonetic Sciences. San Francisco, 1509-1512. El-Hassan, Shahir. 1988. “The Intonation of Questions in English and Arabic”. Studies in Contrastive Linguistics 22.97-108. Escalante, Rafael, Nigel Ward, Yaffa Al Bayyari & Thamar Solorio. 2007. Learning to show you’re listening: A back-channel trainer for Arabic. Paper presented at the Computer Assisted Language Instruction Consortium Symposium, May, in San Marcos, Texas. Fujie, Shinya, Kenta Fukushima & Tetsunori Kobayashi. 2005. “Back-channel Feedback Generation Using Linguistic and Nonlinguistic Information and its Application to Spoken Dialogue System”. Proceedings of the 9th European Conference on Speech Communication and Technology, Interspeech. Lisboa, 889-892.
BACK-CHANNELS IN EGYPTIAN ARABIC
205
Hafez, Ola. 1991. “Turn-taking in Egyptian Arabic: Spontaneous speech vs. drama dialogue”. Journal of Pragmatics 15.59-81. Hellmuth, Sam. 2005. “No de-accenting in (or of) phrases: Evidence from Arabic for cross-linguistic and cross-dialectal prosodic variation”. In Prosodies: With special reference to Iberian languages, ed. Sonia Frota, Marina Vigario and Maria Joao Freitas, 99-121. New York: Mouton de Gruyter. Johnson, Lewis, Carole Beal, Anna Fowles-Winler, Ursula Lauper, Stacy Marsella, Shrikanth Narayanan, Dimitra Papachristou, Andre Valente & Hannes Vilhjalmsson. 2005. Tactical language training system: An interim report. Paper presented at the Intelligent Tutoring Systems Conference, September, in Alagoas, Brazil. Kraut, Robert, Steven Lewis & Lawrence Swezey. 1982. “Listener Responsiveness and the Coordination of Conversation”. Journal of Personality and Social Psychology 43.718-731. Kulk, Friso, Cecilia Odé & Manfred Woidich. 2005. “The Intonation of Colloquial Damascene Arabic: A pilot study”. Proceedings of the Institute of Phonetic Sciences. Amsterdam, 15-20. Maynard, Senko. 1989. Japanese conversation. New Jersey: Ablex. Murray, Iain & John Arnott. 1993. “Toward the Simulation of Emotion in Synthetic Speech: A review of the literature on human vocal emotion”. Journal of the Acoustic Society of America 93.1097-1108. Rifaat, Khaled. 2005. “The structure of Arabic intonation: A preliminary investigation”. In Perspectives on Arabic linguistics XVII-XVIII, ed. Mohammad T. Alhawary and Elabbas Benmamoun, 49-67. Amsterdam & Philadelphia: John Benjamins. Shriberg, Elizabeth & Andreas Stolcke. 2004. “Direct Modeling of Prosody: An overview of applications in automatic speech processing”. Proceedings of the International Conference on Speech Prosody. Nara, 575-582. Ward, Nigel. 2004. “Pragmatic Functions of Prosodic Features in Non-lexical Utterances”. Proceedings of the International Conference on Speech Prosody. Nara, 325-328. ----------. 2006. “Non-lexical Conversational Sounds in American English”.
206
NIGEL WARD & YAFFA AL BAYYARI
Pragmatics and Cognition 14.113-184. ---------- & Wataru Tsukahara. 2000. “Prosodic Features which Cue Back-channel Feedback in English and Japanese”. Journal of Pragmatics 32.1177-1207. ----------, David Novick & Salamah Salamah. 2006. “The UTEP Corpus of Iraqi Arabic”. University of Texas at El Paso, Department of Computer Science. Technical Report UTEP-CS-06-02. ---------- & Yaffa Al Bayyari. 2006. A case study in the identification of prosodic cues to turn-taking: Back-channeling in Arabic. Paper presented at the Ninth International Conference on Spoken Language Processing, September, in Pittsburgh, Pennsylvania. Yngve, Victor. 1970. “On Getting a Word in Edgewise”. Proceedings of the Sixth Regional Meeting of the Chicago Linguistic Society. Chicago, 567-577.
V
COMPUTATIONAL LINGUISTICS
IMPLEMENTING AN OPEN SOURCE ARABIC RESOURCE GRAMMAR IN GF∗
ALI DADA & AARNE RANTA Chalmers University of Technology & Göteborg University
1. Introduction 1.1 Problem Language technology and software localization consume a significant share of many companies’ time and work. Translating an operating system or an application to different languages involves, in the traditional approach, translating out-of-context strings into different languages. This requires a language expert for each language, and still involves language-related problems due to the difficulty in translating such strings and taking care of morphological and syntactic variations at the same time. We illustrate this with an example. A mail reader application wants to display messages such as those in (1). (1) You have 1 new message You have 2 new messages You have 3 new messages You have 100 new messages If these are to be translated into Arabic, special morphological and syntactic considerations should be made, which include inflecting the word for message for number, as in (2). (2) 1 message 2 messages 3-10 messages 1-99 messages 100 messages
risaalatun risaalataani rasaa>ila risaalatan risaalatin
+Sing +Nom +Dual +Nom +Plur +Gen +Sing +Acc +Sing +Gen
The word messages is translated into three different forms, depending on number. Inflection also occurs in case, and count nouns are an extreme example. The case marking of singular and dual nouns is determined by their syntactic ∗
We thank Ken Beesley, Björn Bringert, Harald Hammarström, and Otakar Smrž for reading this paper and giving valuable comments and making corrections.
210
ALI DADA & AARNE RANTA
function (nominative in (2)). This is not the case for plurals, which assume the genitive case from three to ten, then accusative (singular) from eleven to nighty-nine, and genitive again for plurals that are multiples of hundred. This is not to mention noun-adjective agreement which should be taken care of when translating new messages into Arabic. The above details should not be the responsibility of the application programmer, and having translators do this work over and over again for each application can be costly and lead to repeated work and/or poor results. 1.2 Solution In this paper, we review an approach that addresses certain problems in language technology similar but not limited to the above case. This approach is based on developing libraries of natural language constructs and rules, which can be used by an application programmer who is not knowledgeable in a specific language. The core programming language is Grammatical Framework (GF) (Ranta 2004). The language library, called a resource grammar (Khegai & Ranta 2004), consists of rules that span the orthography, morphology, and syntax. This library can be reused in applications through an Application Programming Interface (API) by programmers who are unaware of the details of the specific language. Such a programmer uses a resource grammar assuming it will take care of morphological and syntactic rules. We elaborate on the approach, the programming language that implements it, and on Resource Grammars in Section 2.2. Most of the paper will describe the work done implementing a resource grammar for Arabic, so that Arabic language technology can benefit from and build on this approach, especially that the Resource Grammar itself is available as open source software (Ranta 2006). 1.3 Goals and contributions Implementing a resource grammar for Arabic in GF involves work on different linguistic levels, including the orthography, morphology, and syntax. The most important goal is to provide a resource of grammatical rules on all these linguistic levels, so that a language application writer can use as a library to produce grammatically correct Arabic sentences. Another goal is to provide a sample lexicon and lexical paradigms that the user or a lexicographer can use to easily augment the system. A demonstration application is also desirable to show in a graphical interface what the system can do. The demo can syntactically analyze a sentence showing its syntax tree, conjugate a sentence in tense, polarity, and word order, or give the inflection table of a word. One additional requirement is to keep the system, including the grammar files and the lexicon, as open source software, in order to ensure maximum feedback and to benefit the community.
ARABIC RESOURCE GRAMMAR IN GF
211
This work is not intended to be comprehensive or to cover all of the language in any sense, especially in the lexicon and the syntax. When we have coverage constraints, we should always attempt to concentrate on implementing the more common rules and paradigms. We provide a sample lexicon of words based on the Swadesh list (Hymes 1960), giving the details in Section 5.4. We implement all the parts of the Arabic morphology (inflection tables and paradigms) that are required for the 200 words in the Swadesh list. Because of the morphological variety of these words, our work turns out to cover many types of verbs and verb forms. We go even further than our test lexicon by describing the three nominal declensions (Retsö 1984). This enables us to handle any Arabic nominal form. The implementation, detailed in Section 5.2, serves as a strong starting point towards a complete morphological coverage. We write orthographic rules that correct the spelling of words generated by the morphological modules. This is described in Section 5.1. A minimal number of syntactic rules have been described to write basic grammatical sentences. These rules include sentence predication, noun modification, verb complementization, and noun determination. The complete listing and an example implementation are provided in Section 5.3. 2. Background 2.1 Modern Standard Arabic Modern Standard Arabic (MSA) has its own alphabet and a right-to-left script. This is an aspect that needs care and treatment when developing the system, so Unicode support is a must, and so is a clear, concise, and unambiguous way to include Arabic strings in files. Because of its script, Arabic has orthographic rules that need to be respected in the input or output to the system. An example is the spelling of the hamza (the glottal stop), an Arabic consonant that appears as distinct Unicode characters depending on a set of rules called the “hamza rules”. Another aspect of Arabic orthography is the optional writing of vowel diacritics, which require special attention regarding the input and output of the target system. Modern Standard Arabic is a Semitic language whose stem morphology is typically used as an example of non-linear morphological systems based on the concept of roots and patterns (Kiraz 1995). This morphotactic concept is very different from concatenative morphological systems where words are formed by concatenating different morphemes. In Arabic, stems are generally formed by applying a pattern of consonants and vowels, called CV patterns, to an abstract tri-consonantal root. For example, the root (ktb) is used to form words that have something to do with writing, including verbs (e.g., write and dictate), nouns (e.g., author, writing, desk, and library), and nominal adjectives (e.g., writer). A stem such as katab ‘wrote’ is formed by applying the CV pattern CaCaC to the ktb root. For this reason, Arabic stem morphology is usually classified as a
212
ALI DADA & AARNE RANTA
non-concatenative one, and this is an essential characteristic of Arabic that will need special treatment and an elegant representation. Arabic also has special syntactic features that should be addressed. The definiteness marker, along with some prepositions and pronouns do not appear as separate words, but are clitisized to stems. Prepositions require case agreement, so certain morpheme combinations are syntactically disallowed. Such complications should be handled in the level of the morphology or the syntax, depending on the simplicity and the logical appeal of the solution. 2.2 Grammatical Framework and the resource library Grammatical Framework is a special-purpose functional programming language for defining grammars of (formal or natural) languages. A common API and resource grammars for various natural languages accompany GF with a purpose similar to that of libraries in general programming languages: implementing pieces of code that can be reused by the application programmer. Resource grammars for GF have been developed for the following ten languages: Danish, English, Finnish, French, German, Italian, Norwegian, Russian, Spanish, and Swedish. GF makes a distinction between abstract and concrete syntax. The common API specifies a set of syntactic rules that are language independent (abstract syntax), and the resource grammar implements each rule according to the particular rules of the language (concrete syntax). The latter type of syntax involves word order, agreement, case inflection, etc. This distinction can abstract over language-dependent features and enables an application programmer to write sentences in a language only by describing their abstract syntax trees or by translating them from another language, preferably in a specific domain. The abstract representation would then act as interlingua. 3.
Example In this section, we take as an example translating a simple sentence from English into Arabic using our system. The Arabic resource grammar is independent of any other language, but we use a translation from English because it illustrates the simplicity of the idea. Consider the sentence in (3), whose abstract syntax tree is shown in Figure 1. (3) ya>kulu l-walad-u haaðihi t-tuffaaħ-at-a l-ħamraa>-a eat.3sm the-boy-nom. this.sf. the-apple-f.-acc. the-red.sf-acc. “The boy eats this red apple.”
ARABIC RESOURCE GRAMMAR IN GF
213
S ru NP VP ty ty Det CN V2 NP # # # ty the boy eat Det CN # ty this A CN # # red apple Figure 1: Abstract syntax tree for the example sentence
Our goal is that a user can get the Arabic concrete sentence from the abstract syntax description. This requires that each branch of the tree, which describes a syntactic rule, be implemented for Arabic, and that the lexicon and morphology provide the correct equivalents for the words in the leaves of the tree. For example, the syntactic rule that takes a determiner (Det) and a common noun (CN) to produce a noun phrase (NP) has the type signature in (4) according to the API. (4) DetCN : Det -> CN -> NP In GF notation, DetCN is the name of the rule, Det and CN are the types of the two arguments, and NP is the type of the value. For our example sentence, the Det can be haaðihi ‘this’, the CN at-tuffaaħa al- ħamraa> ‘the red apple’, so the NP becomes haaðihi at-tuffaaħa al-ħamraa> ‘this red apple’. The NP should be properly constructed; for example, the gender marking of the determiner should be derived from that of the noun. Since the noun tuffaaħa ‘apple’ is feminine, the inflected determiner should be in the feminine form (haaðihi ‘this.f’ and not ħaaða ‘this.m’). Similarly, the number and definiteness marking of the noun should follow that of the determiner, so we inflect tuffaaħa ħamraa> ‘red apple’ as a singular definite form because haaðihi ‘this.f’ is singular and definite. The implementation we give here is a simplified treatment that handles the example at hand, not the general case of noun determination. To be able to express all these rules, we will assign types to each word category representing its inherent features (e.g., gender for nouns and number for determiners). We also specify how the categories are inflected. For our three example categories, we define these types, as in (5).
214
(5) Det = { s n d CN = { s g h NP = { s n g p
ALI DADA & AARNE RANTA
: : : : : : : : : :
Species => Gender => Case => Str; Number; State } ; Number => State => Case => Str; Gender; Species } ; Case => Str ; Number; Gender; Person } ;
Each of the types is a GF record, where s is the string n is the number attribute (singular, dual, or plural) d is the state or definiteness (definite with -al, definite with a genitive construction >idaafa, or indefinite) g is the gender (masculine or feminine) h is the species or the natural gender (human or non-human) p is the person (1st, 2nd, or 3rd) For example, the type of CN shows that a noun in Arabic has an inherent gender and species, but is inflected for number, definiteness, and case. The => is the table operator in GF, so having three inputs to the CN table means that a CN is inflected in these three attributes. A simplified version of rule (1) can be implemented, as in (6). (6) DetCN det cn = { s = \\c => det.s ! cn.h ! cn.g ! c ++ cn.s ! det.n ! det.d ! c ; n = det.n; g = cn.g; p = P3 }; The rule accepts a case c from the sentence context using the table construction operator \\. This operator takes one argument c here because the NP inflection table inflects only in case. This case can be nominative, accusative, or genitive according to the grammatical function of the NP. The rule then says that the determiner string is chosen from its inflection table given the NP case and the gender and species of the CN. Similarly, the noun string is chosen using the NP case and the number and definiteness of the determiner. Both strings are concatenated to form the NP string. The NP then takes the number of the determiner and the gender of the noun; these will be used in other rules in the syntax. This is just a formal way to define what we descriptively
ARABIC RESOURCE GRAMMAR IN GF
215
explained before regarding the haaðihi at-tuffaaħa al-ħamraa> example. The user requires, in addition to the morphological and syntactic correctness, lexical entries for the words s/he will use. Thus, an Arabic application programmer should only need to make sure that the necessary lexical entries are available in the lexicon. If not, these entries can be added, as in (7), where facal, fuc~Alp, and >afcAl specify the CV patterns of the nouns in the singular and plural forms, regN, sdfN, and v1 specify the lexical paradigm names for regular nouns (those with broken plurals), nouns with sound feminine plurals, and Form I verbs. (7) child_N = regN "wld" "facal" ">afcAl" Masc Hum ; apple_N = sdfN "tfH" "fuc~Alp" Fem NoHum ; eat_V2 = v1 "?kl" a u ; The two entries after ?kl are the vowels associated with the perfect and imperfect verb forms. This way of specifying verbs in the lexicon is consistent with the way most Arabic dictionaries are written. Notice that lexical entries take the root consonants, and it is the duty of our system to generate all the correct stems, including proper hamza orthography. 4. System overview 4.1 Grammar files The system can be divided along three linguistic lines: the orthography, morphology, and the syntax. A sample lexicon file is also provided. 4.1.1 Orthography. Arabic has certain orthographical rules, such as the rules for writing the hamza, the occasional appearance of a šadda diacritic after the definite marker, and the extra alif, which is sometimes added for the accusative nunation. These and other rules are defined in a separate module called OrthoAra. 4.1.2 Morphology. This section spans different files to modularize the complex Arabic morphology. We first define the needed types in module ResAra, including the essential types for roots and patterns. We describe how words are formed from roots and patterns in the same module. We take into account here bi- and tri-consonantal roots and strong and weak patterns. Different functions in ResAra describe how words are formed from weak patterns (assimilated, hollow, and defective patterns). In PatternsAra we define all the CV patterns we internally need. We separate the actual morphology into two different parts, mostly for efficiency reasons. The first contains the basic definitions and inflection tables
216
ALI DADA & AARNE RANTA
of both the nominal and verbal morphology (located in ResAra) and the second contains the paradigms that will be used in the lexicon (ParadigmsAra). The first part is fully compiled, so the inflection tables are precomputed, but in the second we use the flag optimize = noexpand to inhibit precomputation which would otherwise take a long time. 4.1.3 Syntax. We provide a basic syntax for Arabic spread throughout several modules as required by the Resource Grammar API version 1.0. Each module is a phrase category module that concentrates on a particular phrase category (nouns, verbs, adjectives, sentences, etc.). A phrase category module tells how to construct phrases in that category (Ranta 2006). 4.1.4 Lexicon. Two lexical modules are available. StructuralAra is a list of function words (determiners, conjunctions, etc.) and LexiconAra is a list of basic content words (nouns, verbs, etc.), most of which are based on the words of the Swadesh list. This is where a lexicographer can look to see how lexical paradigms are used and how s/he can write new lexical entries. 4.2 Unicode infrastructure To accommodate the Arabic script, we have certain goals that should be realized by any infrastructure. Grammar writers should have the option to write or view the files in environments supporting only the ASCII character set. On the other hand, the end product should support Arabic Unicode characters, not just a Romanized transliteration. Accordingly, it is necessary to use a transliteration that can be unambiguously decoded into Arabic Unicode characters and vice versa, thus not losing any information. These requirements lead us to writing a script that performs a two-way conversion to and from the transliteration with minimal additional work. The available files are in Unicode but the examples we give here are in the transliteration which is based on the Buckwalter transliteration with a few changes. We note the following additions: ‘c’ for µayn, ‘L’ for al (otherwise represented as ‘Al’), ‘U’ to denote a long uu (otherwise represented as ‘uw’), and ‘I’ to denote a long ii (otherwise represented as ‘iy’). 5.
The grammar After laying the ground for our work and giving an overview of its components, we describe in this section the grammars and give more insight into the implementation. We follow the same order as Section 4.1, thus starting with the orthography, followed by the morphology and the syntax. We will also describe the lexicon and how it can be augmented.
ARABIC RESOURCE GRAMMAR IN GF
217
5.1 Orthography We use the implementation of the hamza rules as an example of orthographic rules. The hamza consonant in Arabic takes one of five shapes (each is a distinct Unicode character) depending on the adjacent vowels and its position in the word. In the transliteration we use, the different hamza carriers are in (8). (8) > < { & ’
on the alif under the alif on a tooth on waaw on the line
A word initial hamza is written on an alif. If followed by a or u (fatħa or damma), the hamza is written on top of the alif. If followed by i (kasra), the hamza is written under the alif. A word medial hamza follows the rule: the “stronger” vowel on both sides of the hamza in the hierarchical sequence (kasra-damma-fatħa-sukuun) determines the bearer in the parallel sequence yaa-waaw-alif-none. The bearer of a word final hamza follows the preceding vowel according to the parallel sequences mentioned above with the medial hamza (Badawi et al. 2004). We define an “abstract hamza” to be a hamza whose carrier is yet to be determined, and we denote it by ?. A hamza occurring in a morphological pattern is never abstract because its representation is determined by the pattern vowels, so an abstract hamza should be a root consonant. Since no two abstract hamzas co-occur in a tri-consonantal root, we expect to correct words with at most one abstract hamza. An implementation of rectifyHmz that takes a word with one possible occurrence of an abstract hamza, and provides an orthographically correct word, is in (9). We take advantage of GF’s regular expressions that give the expressive power shown in rectifyHmz. The two helping functions, bHmz and tHmz map directly from the hamza rules to GF code. The keyword oper (operation) is used to specify that rectifyHmz, bHmz, and tHmz are GF functions. (9) oper rectifyHmz: Str -> Str = \word -> case word of { "?" + v@("a"|"u") + tail => ">" + v + tail; "?" + v@("i") + tail => "<" + v + tail; head + v@("i"|"u"|"a"|"o") + "?" => head + v + (tHmz v); head+ "?" + tail => head + (bHmz (last head) (first tail)) + tail; _ => word };
218
ALI DADA & AARNE RANTA
-–hamza in middle of word (body) oper bHmz : Str -> Str -> Str = \d1,d2 -> case of { <"i",_> | <_,"i"> => "}"; <"u",_> | <_,"u"> => "&"; <"a",_> | <_,"a"> => ">"; _ => "’" }; -–hamza carrier sequence oper tHmz : Str -> Str = \d -> case d of { "i" => "}"; "u" => "&"; "a" => ">"; "o" => "’" }; In addition, we define a function to properly add the definite marker -al to the beginning of words. When this marker is added to a word, it may introduce a šadda diacritic which indicates consonant gemination. This happens with almost half of the consonants, called the “Solar Letters”, as opposed to the “Lunar Letters” that do not produce this effect. An additional orthographic rule that we implement is that of the accusative nunation. Nunation is an indefiniteness marker added as diacritics at the end of a noun or adjective. The diacritics of an indefinite accusative word appear on an additional alif, except if the last letter is a taa’ marbuuta, which functions as the feminine marker. 5.2 Morphology 5.2.1 Roots and patterns. As explained in Section 2.1, the Arabic morphotactics can be described in terms of tri-consonantal roots and CV patterns. Words are formed by applying roots to patterns in a process of interdigitization. With the help of the GF record types, we define types for roots and patterns, as in (10), where f, c, and l stand for the three root consonants, and h, m1, m2, and t stand for the head, middle, and tail of the patterns. (10) Root : Type = {f, c, l : Str} ; Pattern : Type = {h, m1, m2, t : Str} ; We also define functions to interdigitize roots and patterns. When the root ktb is applied to the pattern CaCaC, it forms the stem katab ‘wrote’, as in (11). (11) interdigitize : Root -> Pattern -> Str = \fcl, p -> p.h + fcl.f + p.m1 + fcl.c + p.m2 + fcl.l + p.t;
ARABIC RESOURCE GRAMMAR IN GF
219
All the patterns we internally require are defined in the module PatternsAra. Patterns are used by nouns, verbs, and adjectives, since all words pertaining to the Arabic open classes are formed by the same process and sometimes use common patterns. In the following sections, we describe the implementation of those three open classes. 5.2.2 Verbal morphology. Arabic verbs can be considered the core of its lexicon; most Arabic words are derived from verbs, including verbal nouns. Arabic verbs inflect for six morphological feature categories: gender (masculine and feminine), number (singular, dual, and plural), person (first, second, and third), tense (perfect, imperfect, and imperative), mood for the imperfect tense (indicative, subjunctive, and jussive), and voice (active and passive). This leaves the verb inflection table with 118 entries. We define all the required categories as types, which are parametrized when needed to reflect the exact structure of the inflection table. The code in (12) is from the ResAra module, and it shows the required definitions to reflect the verb inflection table. (12) param Number = Sg | Dl | Pl; Gender = Masc | Fem ; Mood = Ind | Cnj | Jus ; Voice = Act | Pas ; VForm = VPerf Voice PerGenNum | VImpf Mood Voice PerGenNum | VImp Gender Number ; PerGenNum = Per3 Gender Number | Per2 Gender Number | Per1 SgPl; SgPl = Sing | Plur; The GF table in (13) shows how the parameters presented above can be used, and it gives a feel for how the GF formalism parallels the way grammarians think of language. Our Arabic verb conjugation references were (El-Dahdah 1999) and (Ammar & Dichy 1999). The example we give in (13) is the table of suffixes of the Arabic perfect tense, which is common for the active and passive voice, and is one of the first tables given in an Arabic grammar book. This table is declared using the operator =>, and its type is from PerGenNum to Str. Tables can be selected by using the selection operator !, so katab + perfectSuffix ! (Per2 Masc Pl) ; would give katab + "otumo", where katab can be any perfect stem.
220
ALI DADA & AARNE RANTA
(13) perfectSuffix : PerGenNum => Str = table { Per3 Masc Sg => "a" ; kataba Per3 Masc Dl => "aA" ; katabA Per3 Masc Pl => "UA" ; katabuu Per3 Fem Sg => "ato" ; katabat Per3 Fem Dl => "ataA" ; katabataa Per3 Fem Pl => "ona" ; katabna Per2 Masc Sg => "ota" ; katabta Per2 _ Dl => "otumaA"; katabtumaa Per2 Masc Pl => "otumo" ; katabtum Per2 Fem Sg => "oti" ; katabtii Per2 Fem Pl => "otuna" ; katabtunna Per1 Sing => "otu" ; katabtu Per1 Plur => "onaA" katabnaa }; In our system, we only consider tri-consonant verbs leaving four-consonant variations for future work. Tri-consonantal verbs follow one of ten templates, numbered from I to X, where each lexical root can theoretically fit into any of these forms and produce new verbs with a certain nuance in meaning. In Arabic dictionaries, such as (Wehr 1980), the meaning of the base Form I is first given, and then derived stems follow, the following being a full listing: II faµµla, III faaµala, IV >afµala, V tafaµµla, VI tafaaµala, VII infaµala, VIII iftaµala, IX ifµalla, X istafµala. An example on lexical entries and semantic differences for a sample lexical root will follow below. Irregularities in verb forms and inflections occur for morpho-phonological reasons, depending on the root consonants. They can be due to the occurrence of one or a combination of the following: glides, hamzas, or šaddas. The system treats sound verbs (with none of these irregularities), assimilated verbs (with a glide as the first consonant), hollow verbs (with a glide as the second consonant), defective verbs (with a glide as the third consonant), geminate verbs (with second root consonant doubled), and verbs with a hamza in the root. We aim to simplify the work of lexicographers by mimicking the approach used in most Arabic dictionaries, so writing a lexicon for the system will involve providing the minimal amount of information needed to infer phonological irregularities and construct the whole inflection table. This information is the three root consonants, the verb forms (I to X), and the perfect and imperfect vowels for verb Form I. (14) below shows what information is expected from the lexicon writer, using the lexical root c-l-m> and some of its derived stems. (14) calima callama aclama tacallama istaclama
= v1 = v2 = v4 = v5 = v10
"clm" i a –-to know "clm" –-to teach "clm" –-to inform "clm" –-to learn/study "clm" –-to ask for information
ARABIC RESOURCE GRAMMAR IN GF
221
In our ParadigmsAra module, we have separate functions for each of the forms v1-v10, that take a three-consonant string, and the v1 form also takes two vowels. Checks are made to decide verb types. Separate functions are responsible for deriving the stems needed to form the inflection table. These functions, which are internal to the system and are not visible to the user, are called v1sound, v1hollow, v2sound, etc. The number of stems needed to generate the inflection table is five or ten, depending on the verb type. The table of sound verbs is fully calculated with five stems: katab, kutib, aktub, uktib, uktub, whereas that of hollow or geminate verbs requires ten: qaal, qul, qiil, qil, aquul, aqul, uqaal, uqal, qul2, quul. Macros that are common to several verb types take stem forms, and generate the whole table by making use of the common affixes used for tense and mood. We try to detect the similarities across verb types to abstract them away and end up with compact, yet readable code. 5.2.3 Nominal morphology. The current implementation of noun morphology was simplified after initial attempts to include several plurals per noun to distinguish plurals of plenty from plurals of paucity. This was especially difficult because nouns can take any number of plural forms up to five, with varying numbers of the different types (broken plurals of paucity and plenty, and sound plurals). The GF concrete syntax does not support lists which could help in writing a more comprehensive noun system. We give the nouns and adjective types with the parameters they depend on and their inherent features, as in (15). (15) param Number = Sg | Dl | Pl; State = Def | Indef | Const ; Case = Nom | Acc | Gen ; Gender = Masc | Fem ; Species = NoHum | Hum ; oper Noun : Type = { s : Number => State => Case => Str ; g : Gender ; h : Species }; oper Adj : Type = { s : Gender => Number => State => Case => Str }; Arabic nouns have inherent gender and species features, but they are also inflected for number, definteness, and case. The => table operator occurs three times in the formation of a noun string implying that the noun inflection table will have three inputs. The fact that a noun has two inherent attributes (gender and species) implies that this information will be input in the lexicon. Adjectives are inflected for gender, which is not one of their inherent features.
222
ALI DADA & AARNE RANTA
(16) below is a table called dec1sg from the Nominal module. This table defines the suffixes that attach to the singular or broken plural forms of the first (strong) declension of Arabic nominal forms. It calculates, given the state of the nominal form (definite or indefinite) and its case (nominative, genitive, and accusative), what its suffix will be. Note that N, F, and K are the nominative, accusative, and genitive nunation diacritics. (16) dec1sg : table { Indef => table { Nom => Acc => Gen => }; _ => table { Nom => Acc => Gen => } };
State => Case => Str =
"N"; "F"; "K"
"u"; "a"; "i"
Only words with the state “indefinite” take nunation, whereas the rest (al-definite or construct-state definites) take simple diacritics. Similarly, we define tables for the other two declensions and for the dual and sound masculine plurals. These tables are used inside other functions that specify how noun paradigms behave. One such function is sing which takes a nominal word (in singular or broken plural forms) of any declension and returns a table describing how the word inflects for definiteness and case. Note how the function tests the ending of the word to determine its declension and then uses the appropriate suffix table, as in (17). (17) sing : Str -> State => Case => Str = \word -> case word of { lemma + "I" => \\s,c => Al ! s + lemma + dec2sg ! s ! c ; lemma + ("A"|"Y") => \\s,c => Al ! s + word + dec3sg ! s ! c ; _ => \\s,c => Al ! s + word + dec1sg ! s ! c ; }; Al : State => Str = table { Def => "L"; _ => "" };
ARABIC RESOURCE GRAMMAR IN GF
223
5.3 Syntax The last part of the grammar we have implemented is a set of basic syntactic rules, one of which we gave in (5). We list all the API rules that we implement, along with their type signatures in (18). This list can be found in the abstract syntax of GF resource grammars. Our goal is to linearize the rules, both the categories and the functions, into Arabic according to the grammar. (18) cat Cl ; VP ; AP ; CN ; NP ; Pron ; Det ; QuantSg ; QuantPl ; Quant ; Num ; Ord ; V ; V2 ; A ; N ; fun PredVP UseV ComplV2 DetCN UsePron DetSg DetPl SgQuant PlQuant NoNum NoOrd DefArt IndefArt UseN PositA AdjCN
: : : : : : : : : : : : : : : :
NP -> VP -> Cl ; V -> VP ; V2 -> NP -> VP ; Det -> CN -> NP ; Pron -> NP ; QuantSg -> Ord -> Det ; QuantPl -> Num -> Ord -> Det ; Quant -> QuantSg ; Quant -> QuantPl ; Num ; Ord ; Quant ; Quant ; N -> CN ; A -> AP ; AP -> CN -> CN ;
We pick an example of linearization to demonsrate the expressive power of our formalism. The rule is that of verb predication, which can be used for the two rules UseV and ComplV2. For verb predication we need to decide the type of verb phrases (VP), so we give a simplified version of that in (19).
224
ALI DADA & AARNE RANTA
(19) param Tense = Pres | Past | Fut ; Polarity = Pos | Neg ; Order = Verbal | Nominal ; lincat VP = { s : Tense => Polarity => Order => Agr => Str ; comp : NP } ; Agr = PerGenNum ; A VP inflects for tense, polarity, word order, and PerGenNum. We keep the complement of the VP as a separate field because a verb and its object will be disjoint constituents in a VSO word order sentence. The operation in (20) takes a verb and constructs a verb phrase. Using the table construction operator \\, we take four arguments: t, p, o, and pgn, that stand for tense, polarity, word order, and person-gender-number agreement. Then the table is constructed, depending on the different combinations of tense and polarity. (20) oper predV : Verb -> VP = \v -> { s = \\t,p,o,pgn => let { pgn’ = case o of { Verbal => verbalAgr pgn; Nominal => pgn }; kataba = v.s ! ( VPerf Act pgn’ ); yaktubu = v.s ! ( VImpf Ind Act pgn’ ); yaktuba = v.s ! ( VImpf Cnj Act pgn’ ); yaktub = v.s ! ( VImpf Jus Act pgn’ ) } in case of { => yaktubu ; => "laA" ++ yaktubu ; <Past, Pos> => kataba ; <Past, Neg> => "lamo" ++ yaktub ; => "sawfa" ++ yaktubu ; => "lano" ++ yaktuba }; comp = { s = \\_ => [] } }; oper verbalAgr : PerGenNum -> PerGenNum = \pgn -> case pgn of { Per3 g _ => Per3 g Sg; _ => pgn };
ARABIC RESOURCE GRAMMAR IN GF
225
Word order in Arabic determines how the verb agrees with its subject. In nominal sentences, the verb shows full agreement, whereas in verbal sentences the verb does not agree in number. Instead, it inflects for the singular irrespective of the number of its subject. This rule is described in the case statement within the let ... in statement and in the verbalAgr helping function, as in (20). 5.4 Swadesh lexicon In this section, we provide an example lexicon in the LexiconAra module. The entry in (21) shows how a lexicographer enters verbs in the current system. The entries have the advantage of being identical to those in a dictionary, where the tri-consonantal root is given, along with the verb form (in Roman numerals instead of our Arabic numbering) and the perfect and imperfect vowels for the verbs in Form I. This way, only minimal knowledge of Arabic is needed to write the verb lexicon. An excerpt from the noun part of the lexicon is in (22). (21) bite_V blow_V breathe_V burn_V die_V fight_V give_V kill_V laugh_V play_V sing_V swell_V think_V
= = = = = = = = = = = = =
v1 v1 v5 v1 v1 v3 v4 v1 v1 v1 v2 v8 v2
"cD~" "nfx" "nfs" "Hrq" "mwt" "qtl" "cTy" "qtl" "DHk" "lcb" "gny" "nfx" "fkr"
a a ; a a ; ; a i i ; ; ;
a ; u ; i ; u ; u ; a ; a ;
(22) bird N bone N child N blood N father N day N fire N grass_N lake_N egg_N
= = = = = = = = = =
regN "Tyr" regN "cZm" regN "wld" regN "dm" regN "?b" regN "ywm" regN "nwr" regN "c$b" sdfN "bHr" sdfN" byD"
"facl" "faclp" "facal" "fac" "fac" "facol" "fAc" "fuclo" "fucaylp" "facolp"
"fucUl" "ficAl" ">afocAl" "ficA’" ">afocA’" ">af~Ac" "fIcAn" ">afcAl" Fem NoHum; Fem NoHum;
Masc NoHum; Fem NoHum; Masc Hum; Masc NoHum; Masc Hum; Masc NoHum; Fem NoHum; Fem NoHum;
226
ALI DADA & AARNE RANTA
The noun entries are not as elegantly represented as verb entries. This is, in part, due to the inherent features of nouns which should be provided, and partly because it is more difficult to deduce noun information, such as the particular broken plural pattern from the singular pattern. For nouns, we provide the above paradigm name, the two or three root consonants, the needed CV patterns, gender, and the species. Note that nouns with possible weak consonants (glides) in their roots, such as ywm ‘day’ and nwr ‘light’, use the same paradigm as their sound counterparts. It is the responsibility of the system to test for possible weaknesses after observing the root consonants and the patterns that combine with them. Finally, we give in (23) sample entries for adjectives, which are not as complex as those of nouns because of the regularity of Arabic adjectives. Most of the adjectives we used are regular, needing only a pattern and a root to be fully defined, whereas color adjectives do not even need a pattern because they all follow the same pattern. (23) bad_A big_A cold_A far_A good_A long_A smooth_A white_A red_A yellow_A
= = = = = = = = = =
regA regA regA regA regA regA regA clrA clrA clrA
"sy?" "kbr" "brd" "bcd" "jyd" "Twl" "ncm" "byD" "Hmr" "Sfr"
"fac~il"; "facIl"; "fAcil"; "facIl"; "fac~il"; "facIl"; "fAcil"; ; ; ;
6. Related work 6.1 Xerox Arabic morphological analyzer and generator A large scale implementation of the Arabic morphological system is the Xerox Arabic Morphological Analyzer and Generator (Beesley 1996, 1998, 2001; Beesley & Karttunen 2000). This system is developed using only the Xerox Finite State Technology tools (Beesley & Karttunen 2003), from which an Arabic Finite State Lexical Transducer is written. The system analyzes input in the form of fully, partially, or unvocalized words which provides the user with maximum flexibility. When vowels are included, the ambiguity decreases and the number of returned lexical hits may decrease, so the system takes advantage of any written diacritics. Along with the fully vocalized output words, the user is provided with an English gloss, the root, pattern, and tags indicating the part of speech, person, number, mood, and voice. The interdigitization process described earlier can be viewed as an intersection operation in finite-state calculus, so the ktb root and the CaCaC pattern can be intersected to form the katab stem. The system has a lexicon of roots
ARABIC RESOURCE GRAMMAR IN GF
227
where each root is hand-marked to show the patterns with which it combines. This and other lexicons are parsed and converted into lexc files, which are in turn compiled into a Lexicon Finite State Transducer (FST). Finite-state morphological rules, including those that handle all kinds of epentheses, deletions, and assimilations are compiled into a separate Rules FST. The Lexicon and Rules FSTs are then composed into one FST. 6.2 Buckwalter Arabic Morphological Analyzer Another notable computational model of the Arabic morphology is Tim Buckwalter’s Arabic Morphological Analyzer (Buckwalter 2002, 2004a,b). This analyzer parses Arabic words and gives all their possible morphological interpretations, each solution having a unique lemma ID. In addition to the lemma ID, the word constituents (prefixes, stems, suffixes) are returned, along with their part of speech and English glosses. Unlike the Xerox system, Buckwalter’s analyzer does not handle diacritics in the input. This analyzer was used in the morphological annotation of Arabic treebanks, including the Penn Arabic Treebank (Maamouri et al. 2004a,b, 2005a,b; Maamouri & Bies 2004) and the Prague Arabic Dependency Treebank (Hajič et al. 2004a,b). 6.3 More morphology treatments Another significant treatment of the Arabic morphology is given in (Kiraz 2001). Kiraz describes a multi-tape two-level formalism for the root-and-pattern morphology of Semitic languages. He argues that this model can account for Semitic morphology using high level notation and yet be computationally tractable (Kiraz 1994, 1995). Although Arabic morphology is typically considered non-concatenative, a concatenative approach is presented in (Cavalli-Sforza et al. 2000; Soudi et al. 2001). This system differentiates between the two operations of Arabic morphology: infixation and affixation. They argue that decoupling infixation and affixation reduces the number of rules required, giving as an example a treatment of hollow verbs. The lexicon in this system is lexeme-based, as compared to the root-and-pattern based lexicon in the Xerox system. Work on the latest Arabic morphology system is based on Functional Morphology (Forsberg & Ranta 2004), a methodology for building morphological systems in the Haskell programming language. Our treatment of Arabic morphology is similar to that of Functional Morphology. Both approaches use typed languages, making use of finite algebraic datatypes to define linguistic categories. Both languages are functional, so the approches use functions to realize linguistic abstractions. A large scale implementation of this approach is Huet’s Sanskrit dictionary and morphological system (Huet 2006a). Huet has built the Zen Computational Linguistics Toolkit based on his Sanskrit morphology implementation (Huet 2005, 2006b).
228
ALI DADA & AARNE RANTA
6.4 Syntax Work related to Arabic syntax is less frequent than that of morphology. We make note of the work in (Othman et al. 2003, 2004), where the authors describe a parser they wrote in Prolog to parse and disambiguate Arabic sentences. Shaalan (2005) builds on this work to develop a syntax-based grammar checker for Arabic called Arabic GramCheck. His system analyzes Arabic sentences and accepts only the grammatically correct ones, giving correction suggestions for the unacceptable results. 7.
Evaluation A decision like the place of introducing the orthographic rules can lead to the explosion of the GF grammars. We first introduced the hamza rules before the morphological rules, i.e., in the interdigitization function itself. The compiled morphological files were huge. After moving the hamza rules and placing them after the morphology, the size of the compiled file was reduced by a factor of 10. Another such rule was that of adding the definite article to nouns as described in Section 5.2.3. This rule was part of the morphological tables, and after moving it out and introducing it after the tables are calculated, the noun morphology compiled file size was reduced by a factor of 8. Other than these sudden explosions in GF grammars and the problems of representing nouns, GF has lots of merits and strengths when describing linguistic issues, such as inflection tables and grammar rules. The ability to use GF with a “different” language like Arabic was a prerequisite for the whole work. We note also the elegance in which a grammar formalism like GF enables us to represent linguistic rules and abstractions. We point to a downside in our system when compared to most other implementations, namely that we have not yet considered the analysis of Arabic texts, which are mostly unvocalized. This was solved in a simple way in Beesley’s FST system by writing rules that optionally map diacritics to the empty string. These rules are compiled into a transducer composed with the core FST. In this way, the system can recognize fully-vocalized, partially-vocalized, or non-vocalized texts. Beesley’s solution inspired us to think of a parallel approach to solve the problem in our system. It is yet to be seen if our solution will be as elegant and straightforward as in the Xerox system. 8.
Future work A resource grammar for a complex language like Arabic can be a very deep subject of work and study, in both main branches: morphology and syntax. Having the system code available as open source software will help getting comments and contributions from the audience. The following is a list of enhancements and completions that should be done to the system, which is still limited in language coverage.
ARABIC RESOURCE GRAMMAR IN GF
229
The implemented verb system can be augmented to include the remaining combinations of verb forms and phonological irregularities of three-consonant verbs that we did not cover. This needs minimum effort because the programmer should find the already implemented combination to be similar to the one s/he wants to add and copy, only changing the patterns to those required. Those patterns, if not already found in the PatternsAra module, should be added manually there. The next step to be taken in the verb system is to treat verbs with four consonant roots. This will take more effort because new types and paradigms should be added. For example, the new roots and patterns will differ from those already shown. A desired improvement is a computationally elegant way to specify several plurals for each noun, differentiating between plurals of the few and of the many. A more complete syntax with wider coverage should be written. The Arabic syntax is a complex one with several differences from Indo-European languages, and the example syntax developed covers only a small part of the language. The rest of the GF Resource Library API should be implemented for Arabic. This will lead us to tackle problems we have not solved yet, including that of clitisized words. We should come up with a way in order for our system to analyze the typically unvocalized Arabic texts. 9.
Conclusion We discuss in this paper the details of implementing an Arabic Resource Grammar in GF, and its implications in terms of grammar reuse. The implementation of the morphology and syntax is explained, along with examples from the different grammar modules: the orthography, morphology, and syntax. We show how the grammar formalism we use parallels the way grammarians and linguists think. The methodology we follow in building the morphological system is one of three implementations that use typed functional programming languages, describing morphology in terms of data types and abstracting code by using functions. The other two implementations are the Zen Toolkit and Functional Morphology.
REFERENCES Ammar, Sam & Joseph Dichy. 1999. Bescherelle les verbes arabes. Paris, France: Hatier. Badawi, Elsaid, M. G. Carter & Adrian Gully. 2004. Modern Written Arabic: A comprehensive grammar. New York: Routledge. Beesley, Kenneth. 1996. “Arabic Finite-State Morphological Analysis and Generation”. COLING 96. Copenhagen, 89–94.
230
ALI DADA & AARNE RANTA
----------. 1998. “Arabic Morphology Using Only Finite-State Operations”. Proceedings of the Workshop on Computational Approaches to Semitic Languages. Montreal, 50–57. ----------. 2001. “Finite-State Morphological Analysis and Generation of Arabic at Xerox Research: Status and plans in 2001”. Proceedings of the Workshop on Arabic Language Processing: Status and prospects. Toulouse, 1–8. ---------- & Lauri Karttunen. 2000. “Finite-State Non-Concatenative Morphotactics”. Proceedings of the Fifth Workshop of the ACL Special Interest Group in Computational Phonology. Luxembourg, 1-12. ---------- & Lauri Karttunen. 2003. Finite state morphology: Studies in computational linguistics. Stanford, CA: CSLI Publications. Buckwalter, Tim. 2002. Buckwalter Arabic Morphological Analyzer Version 1.0. LDC catalog number LDC2002L49. ----------. 2004a. “Issues in Arabic Orthography and Morphology Analysis”. Proceedings of the COLING 2004 Workshop on Computational Approaches to Arabic Script-based Languages. Geneva, 31-34. ----------. 2004b. Buckwalter Arabic Morphological Analyzer Version 2.0. LDC catalog number LDC2004L02. Cavalli-Sforza, Violetta, Abdelhadi Soudi & Teruko Mitamura. 2000 “Arabic Morphology Generation Using a Concatenative Strategy”. Proceedings of NAACL 2000. Seattle, 86-93. El-Dahdah, Antoine. 1999. A dictionary of Arabic verb conjugation. Beirut: Librairie du Liban. Forsberg, Markus & Aarne Ranta. 2004. “Functional Morphology”. Proceedings of the Ninth ACM SIGPLAN International Conference on Functional Programming. Snow Bird, Utah, 213-223. Hajič, Jan, Otakar Smrž, Petr Zemánek, Petr Pajas, Jan Šnaidauf, Emanuel Beška, Jakub Kráčmar & Kamila Hassanová. 2004a. Prague Arabic Dependency Treebank 1.0. LDC catalog number LDC2004T23. ----------, Otakar Smrž, Petr Zemánek, Jan Šnaidauf & Emanuel Beška. 2004b. “Prague Arabic Dependency Treebank: Development in data and tools”. NEMLAR International Conference on Arabic Language Resources and Tools, 110-117. Huet, Gérard. 2005. “A Functional Toolkit for Morphological and Phonological Processing: Application to a Sanskrit tagger”. Journal of Functional Programming 15.573–614. ----------. 2006a. Sanskrit Site. http://sanskrit.inria.fr/. ----------. 2006b. The Zen Computational Linguistics Toolkit. http://sanskrit.inria.fr/ZEN/. Hymes, D. H. 1960. “Lexicostatistics So Far”. Current Anthropology 1.3–44. Khegai, Janna & Aarne Ranta. 2004. “Building and Using a Russian Resource Grammar in GF”. Proceedings of Intelligent Text Processing and Computational Linguistics. Seoul, 38-41.
ARABIC RESOURCE GRAMMAR IN GF
231
Kiraz, George. 1994. “Multi-Tape Two-Level Morphology: A case study in Semitic non-linear morphology”. Proceedings of the 15th International Conference on Computational Linguistics Vol. 1. Kyoto, 180-186. ----------. 1995. Computational Analyses of Arabic Morphology. http://www.bell-labs.com/project/tts/gkiraz-arabmorph.ps. ----------. 2001. Computational nonlinear morphology with emphasis on Semitic languages. Cambridge: Cambridge University Press. Maamouri, Mohamed & Ann Bies. 2004. “Developing an Arabic Treebank: Methods, guidelines, procedures, and tools”. Proceedings of the COLING 2004 Workshop on Computational Approaches to Arabic Script-based Languages. Geneva, 2–9. ----------, Ann Bies, Tim Buckwalter & Hubert Jin. 2004a. Arabic Treebank: Part 2, Version 2.0. LDC catalog number LDC2004T02. ----------, Ann Bies, Tim Buckwalter & Wigdan Mekki. 2004b. “The Penn Arabic Treebank: Building a large-scale annotated Arabic corpus”. NEMLAR International Conference on Arabic Language Resources and Tools. 102–109. ----------, Ann Bies, Tim Buckwalter & Hubert Jin. 2005a. Arabic Treebank: Part 1 Version 3.0. LDC catalog number LDC2005T02. ----------, Ann Bies, Tim Buckwalter, Hubert Jin & Wigdan Mekki. 2005b. Arabic Treebank: Part 3 Version 2.0. LDC catalog number LDC2005T20. Othman, Eman, Khaled Shaalan & Ahmed Rafea. 2003. “A Chart Parser for Analyzing Modern Standard Arabic Sentence”. Proceedings of the MT Summit IX Workshop on Machine Translation for Semitic Languages: Issues and approaches. Lousiana, 37-44. ----------. 2004. “Towards Resolving Ambiguity in Understanding Arabic Sentence”. NEMLAR International Conference on Arabic Language Resources and Tools. 118–122. Ranta, Aarne. 2004. “Grammatical Framework: A type-theoretical grammar formalism”. Journal of Functional Programming 14.145–189. ----------. 2006. GF Resource Grammar Library v. 1.0. http://www. cs.chalmers.se/~aarne/GF/lib/resource-1.0/doc/. Retsö, Jan. 1984. “State, Determination and Definiteness in Arabic: A reconsideration”. Orientalia Suecana 33-35.341-346. Shaalan, Khaled. 2005. “Arabic GramCheck: A grammar checker for Arabic: Research articles”. Software - Practice and Experience 35.643–665. Soudi, Abdelhadi, Violetta Cavalli-Sforza & Abderrahim Jamari. 2001. “A Computational Lexeme-based Treatment of Arabic Morphology”. Proceedings of the Workshop on Arabic Language Processing: Status and prospects. Toulouse, 50–57. Wehr, Hans. 1980. A dictionary of Modern Written Arabic. Beirut, Lebanon: Librairie du Liban.
A LINK GRAMMAR PARSER FOR ARABIC
WARREN CASBEER, JON DEHDARI & DERYLE LONSDALE Brigham Young University
1.
Introduction A parsing system is critical to many natural language processing tasks. Applications such as information extraction, natural language generation, grammar checking, language tutoring, and machine translation use parsed input sentences as an input to subsequent processing. Developing an automatic parser for Arabic is nontrivial, as this language poses unique difficulties for traditional natural language processing approaches. Current Arabic parsers include a parser based on definite clause grammar (Al Daimi & Abdel-Amir 1994), a principle-based parser (using a GB framework) for tutoring (Weinberg et al. 1995), a lexical functional grammar for parsing (Abu-Arafah 1996), AraParse, which uses an Affix Grammar over a finite lattice (Ouersighni 2001), and a bottom-up chart parser using unification based grammar (Othman et al. 2003). Most parsers are able to handle many of the basic constructions of the language including sentences with transitive and intransitive verbs, copular sentences, idaafa constructs, word order variations in declarative sentences, wh- and yes/no questions, relative clauses, and embeddings. A number of problems still confront current Arabic parsers. Some difficult Arabic constructions for parsing include coordinating conjunctions (AbuArafah 1996; Ouersighni 2001), elliptical forms, negatives, anaphora, complex determiners, and agreement, as well as other dependencies within a sentence. Many current parsers are computationally costly or have limited coverage. Due to these issues, Arabic parsers for use in large-scale applications are not available (Ouersighni 2001). Ambiguity at any level (i.e., morphological, lexical, syntactic, semantic, and pragmatic) remains a particular issue that has not been treated to a large extent (Al Daimi 2001; Othman et al. 2003). Robustness is another area for potential improvement. Ouersighni (2001) comments on these issues: There is still no general language analyzer available for Arabic with sufficiently wide coverage to be sure that all expressions of the language are treatable with existing tools…Most systems select types of syntactic phenomena for treatment, with considerable lexical limitations. (Ouersighni 2001:1, emphasis original)
234
CASBEER, DEHDARI & LONSDALE
In this paper, we introduce a new parser for Arabic based on Link Grammar, a dependency-like grammar. This grammar has been implemented as a parser that is cost efficient, widely used, and freely available. In the following sections we discuss the Link Grammar parser and how we implemented it for Arabic syntactic processing. Link Grammar has been used to provide coverage for a variety of languages. Though originally developed for English, it has been applied to other languages, such as Persian (Dehdari & Lonsdale 2006) and Russian (Protasov 2005)1. The system is capable of resolving some of the issues discussed above, including lexical and syntactic ambiguity. Multiple parses are provided by the parser, as all possible structures are considered. These are presented in ranked order according to the constraints of the grammar. The parser displays robustness, since it can guess the categories of words not in the lexicon based on its knowledge of the syntactic environment. It can also skip unknown structures, providing the robustness necessary to focus on partial parsing when a complete parse is not feasible. This paper is organized as follows. First, preprocessing is described. This includes input formats, morphological decomposition, and necessary lexicons. Link Grammar is then introduced, and grammar development is described. To exemplify coverage in Arabic, sample parses are shown. Ambiguity resolution using the parser is then demonstrated, and an evaluation of the parser is given. Finally, potential applications and future work are offered. 2.
Preprocessing The syntactic analysis is aided by a number of preprocessing steps included in the system. First, the parser requires certain input formats for the input sentences in textual form. The system currently accepts Arabic-script UTF-8, CP 1256, or Latin-script Buckwalter transliterated input2, which then becomes transliterated if not already in this format. No information is gained or lost by transliterating the input. Additionally, a morphological engine is incorporated into the present system, as done in the Persian Link Grammar system. The present system sends the transliterated input text through the Buckwalter (2002) morphological analyzer for Arabic. Words from the input are segmented to constituent morphemes by the morphological engine prior to any subsequent syntactic analysis. Buckwalter’s system supports many Arabic morphological phenomena including feature affixes on nominals, possessive morphemes, direct object enclitics, and verbal form affixes. This system thus provides a tight coupling between morphology and syntax, providing a more detailed
1 2
http://sz.ru/parser/ http://www.qamus.org/transliteration.htm
A LINK GRAMMAR PARSER FOR ARABIC
235
description of Arabic morphological structure for the Link Grammar parser downstream. 3.
Link Grammar The Link Grammar parser provides a dependency-like method for parsing sentences. In this section, the basic principles of Link Grammar are first summarized from the foundational literature (Sleator & Temperley 1991). We then use simple examples to demonstrate how our Arabic grammar was developed. Afterwards, the parser is presented. 3.1 Basics of Link Grammar In Link Grammar, each word has links that must be established with other words in a sentence. Examples of link types include a subject link that attaches a subject to a verb, and an object link that combines a transitive verb with its object. Directionality and Relative Distance are the main principles upon which links in this grammar are defined and established. When two words have the same link type and their corresponding linking rules are in opposite directions (left, right), a link is established between them. Some words may have multiple links, some of which are more local to the word. According to the Principle of Distance, the sequence of linkage rules is equivalent to the sequence of application. Links must be established under the conditions of Planarity, Connectivity, Satisfaction, and Exclusion. Planarity states that links may not cross. Connectivity holds that all the words of the sequence must be connected. Satisfaction requires that links must satisfy the requirements of each individual word. Exclusion ensures that the same two words may only be connected by one link. The system uses a grammar file that contains a specification of words with their link requirements. For example, a simplified specification for the English word ‘frog’: frog: {A-} & {D-} & (S+ or O-); might list possible links and categories that a noun can initiate: adjective (A) and determiner (D) links to the left (-), and either a subject (S) link to the right (+) or an object link to the left (-). Optional items are embedded in curly braces, and precedence is indicated via parentheses. Further details in regards to the entries will be given later. The parser also uses a dictionary file that lists word classes accompanied by their linkage requirements. For this parser, we generated a sizeable lexicon of Arabic words from the Buckwalter (2002) distribution; we added link specifications for the various word classes thus enumerated. Functional word categories, on the other hand, were listed directly within the dictionary for the parser and their link specifications were hand-crafted.
236
CASBEER, DEHDARI & LONSDALE
3.2 Development of a Link Grammar Developing a Link Grammar involves identifying word classes for a language. These are then annotated by rules that establish links to, and thus relationships with, other word classes. The rules of the grammar can be complicated, involving multiple necessary links. To understand a linkage rule, consider the relationship between a subject and its verb. For a verbal sentence in Arabic, example (1) below shows linkage rules for a verb and a subject (since a transliterated script is used, sentences are read from left to right). (1) Word Link type Akl S+ eat.perf.ms
Word AHmd Ahmad
Link type S-
Both of these words have the same link type (S), which is a subject link. The verb Akl (أ, ‘to eat’) has a ‘+’ direction, which indicates establishment of a link towards the right. The proper noun AHmd can establish a link in the opposite direction. Since these words have the same link types with opposite directional requirements, a link may be established between them. A sample output of this is demonstrated in (2) below. (2) +----S----+ | | Akl AHmd A wide coverage Link Grammar for Arabic, based on such linkage rules, is now available to parse sequences of words in this language. The lexicon of words contains approximately 55,000 entries. The grammar currently contains about 50 rules, integrating approximately 40 parts of speech. Since the present system also segments words into morphemes, linkage rules may also define the relationships between these morphemes. For example, a verbal stem may have multiple morphemic links for tense, features, and direct object enclitics. These links show how each morpheme relates to the verb stem, showing direct relationships between morphemes. 3.3 A parser implementation The parser processes sentences based on the aforementioned linkage rules. It first reads in a Link Grammar dictionary that is separated into categories as described previously. Each word type is associated with certain linkage requirements used to parse word sequences to determine their grammaticality. The sentence along with its linked structure is then displayed. The parser may be run in command line mode or through a web interface3. Sentences may be 3
See http://linguistics.byu.edu/arabiclg.
A LINK GRAMMAR PARSER FOR ARABIC
237
processed individually or in batch mode using a file of sentences. A user enters a sentence according to the required input formats and the system parses it morphologically and syntactically according to the linkage rules. The linked structure is displayed on the screen. Consider example (3) below: (3)
Relationships between individual items are shown. The majority of items here are morphemes, including a tense marker, subject affixes, a stem, and a direct object suffix. To understand this example, it is necessary to begin from the stem $Ahd (ه, ‘to observe’), which is linked to four morphemes. First, phi-feature information is related to the stem by the prefixes t- and -wn, using the IVPRE (imperfective verb prefix) and IVSUFFSUBJ (imperfective verb suffix subject) link. These are the most local links to the verbal stem. These feature affixes do not inhibit morphemes that are further away from attaching directly to the stem. The tense marker s is related to the stem $Ahd by the AUXfut (auxiliary future) link. In addition, the VSUFFDO (verb suffix direct object) links hA (ه, ‘her’) to the stem in a similar fashion. This ability to establish multiple linkages is extended to words in a sentence as well. Such descriptive abilities are useful, since an in-depth morpho-syntactic analysis is possible with our framework. 4.
Results Currently, the parser can handle Arabic sentences containing a variety of constructions. So far the grammar assumes only VSO order, based on Modern Standard Arabic. This structure is handled well by the parser as expected due to the flexibility shown in describing Persian SOV order and English SVO order. The parser addresses morphological concerns, including verb formation and feature morphemes for nominals. In example (4) below, the noun hjm (‘attack’) is linked to its number feature (the plural -At) through the NSUFFFEMpl link. This link associates the noun with its suffix and hence its number feature information. Idaafa constructs, negation, aspect markers, clausal complements, relativized complementizers, comparatives/superlatives, copular sentences, conjunction, quantifiers and other structures have been
238
CASBEER, DEHDARI & LONSDALE
integrated. Sample parses (4) and (5) below demonstrate a variety of structures in Arabic. (4)
The verb rd (ر, ‘to respond’) in (4) has both a subject (S) link with jy$ ( , ‘army’) as well as a preposition (P) link to ElY ( , ‘to’). This is similar to the example given above with morphemes, showing that a word may link to other words that are not adjacent or contiguous to it. An adjective (A) link is also shown; this is a simple link that connects a noun with its adjective modifier. In addition, a quantifier object (QO) link is shown here as well. (5)
This example shows negation in the imperfect, directly linked to the root of the verb qdm (م, ‘to offer’). Note that this verb has six links (prefix feature, negation, subject, conjunction, object, and preposition). This illustrates how the root relates to both morphemes within the word it forms as well as to other words. Some of the link types in the above examples are followed by lower case letters (e.g., NEGi). In Link Grammar, these are termed “subscripts”. They provide more in-depth descriptions of link types. For example, using the subscript i in the link NEG indicates that the negation lm may only link to imperfect verbs, thus avoiding the ungrammatical linkage of this negation element to a perfect verb. The above example shows how idaafa constructions are handled by the GEN (genitive) link. The possessed object >hdAf (أهاف, ‘objectives’) is directly related to the possessor zyArp (زرة, ‘trip’) by this link. Longer idaafas could potentially be handled, though this has not been completely
A LINK GRAMMAR PARSER FOR ARABIC
239
investigated yet. Example (6) shows how the parser handles a nominal sentence with a long idaafa. (6)
xAld Abn Sdyq Emy +--------On-------+ +---GEN--+ +--GEN--+-NPOSS+ | | | | | xAld.N0 >bn.Nprop Sdyq.N0 Em.Ndu y.NSUFF
Nominal sentences with long idaafa constructions pose a complication for the system at the present. Instead of showing one long idaafa, the parser divides it up into two smaller idaafas. This is an issue that still needs to be addressed. Other issues that remain to be addressed include word order variations, topicalization, conditionals, and quotation embeddings. Quotation marks are currently discarded by the parser. Doing so could potentially be problematic for a parse of embedded citations, but so far has not been an issue. 5.
Ambiguity and scoring The identification of ambiguity in a parse is an issue that has received attention for a number of languages. However, it has not been developed for Arabic to a large extent. Ambiguity detection is possible using the Link Grammar system. The system is designed to report syntactic ambiguities, and it can also detect lexical ambiguity. As described above, all the structures that are possible, based on linkage rules, are reported by the system. These display ambiguities in meaning present in the sentences encountered. Rankings of these parses show which are the most feasible, according to a calculated cost vector. The cost vector includes four components: DIS, UNUSED, AND, and LEN. First, a user may enter a cost with a certain linkage by the use of square brackets; any number of sets may be used, and the more sets there are the greater the cost assigned to DIS. Any unused words (null links) in the parse are penalized, as reflected in UNUSED. The AND component applies to sentences with conjunctions. Linkages containing similar lengths of conjoined word-lists are preferred. The final component, LEN, prefers linkages which have the least total length of links. Any parse containing constraints will be listed later in the ranking. Thus, the system ranks lower-cost sentences above higher-cost ones. Example (7) shows a sententially ambiguous sentence. The input phrase is first shown, followed by the output from the system.
240
CASBEER, DEHDARI & LONSDALE
(7)
The parser displays the number of possible parses, followed by each of the parses identified by linkage numbers as determined by the cost vector ranking. After the linkage number, the sentence is displayed with its structure as defined by the linkage grammar rules. The grammar gives three possible linkages for this sentence. The first, which can be translated as ‘Ahmad Hussein changed’, is the highest ranked parse because it has the lowest cost vector (=1). The second linkage, ‘(He) changed Ahmad Hussein’, has a cost vector of 2, since subject links have lower penalized (DIS) cost than object links. The third linkage, ‘Ahmad changed Hussein’, has the highest cost vector (=3), because shorter links are preferred over longer ones. Even though there are other possible interpretations of the syntax of this sentence, they are not recognized by the parsing system at the moment. Three linkages are given by the parser, and these correspond to the three possibilities above. Other possible readings include ‘other than Ahmad Hussein’ (involving a preposition and its complement), as well as ‘Ahmad Hussein was changed’ (a passive), and ‘Change Ahmad Hussein’ (an imperative). The prepositioncomplement reading would generally be discarded by the system, since it does not form a complete sentence. The other two readings would get parsed in a fashion similar to the links in (7) above, although the verbal part-of-speech tags would differ. Since passive verb forms often appear similar to their nonpassive counterparts, we have not yet addressed them. The first possibility above shows how the system could identify lexical ambiguities. The string ‘gyr’ could be interpreted either as a verb or as a
A LINK GRAMMAR PARSER FOR ARABIC
241
preposition. At the moment, some parses from the system are not grammatical. An evaluation is needed to determine which rules are not restrictive enough. 6.
Discussion and results The present system can handle many basic constructions of Arabic. Basic declarative sentences, embeddings, copulars, noun constructs, relative clauses are built in. Aspect markers, adverbs, and quantifiers are other important components of Arabic that are included in the system; these do not appear to be mentioned in previous literature. The system can also handle some of the difficulties that were mentioned, including coordination, complex determiners, and negative forms. Certain issues are yet to be addressed in the present system. These include word order variations, numerical expressions, and questions. The system is capable of parsing long sentences, though better clause identification would be helpful. This could be implemented by identifying strong separators such as f( ) or w ( ) وand using the wall link that identifies sentence boundaries. The parser offers important advantages that need noting. Robustness is shown as words not in the lexicon are handled by the parser. It is able to guess the categories of such words based on surrounding structural information. In addition, it can skip ungrammatical sentence parts, thus avoiding large processing times. This allows it to identify ungrammatical portions of sentences, while still being able to show a parse for the grammatical sections. Ambiguity is another area that is addressed by the present system. The parser identifies and ranks ambiguous phrases and sentences through an exhaustive effort to produce all possible parses. Another important aspect of the parser is its speed; it is currently capable of processing about 500 sentences per minute using an average sentence length of 15 words. A file of multiple sentences may be parsed relatively quickly. 7.
Applications and future work The system described herein has the potential to be used in a variety of applications. Some of these will now be described. 7.1 Information extraction The Link Grammar framework is useful for the purposes of information extraction. It has been integrated with a cognitive modeling system to extract semantic information from English texts in the form of predicate logic. This could be applied to the present Link Grammar for Arabic to gather information from text.
242
CASBEER, DEHDARI & LONSDALE
7.2 Grammar checking One possible use is in word processors as a grammar checker. Issues dealing with grammar checking for Arabic have been discussed by Shaalan (2005). He termed his approach “syntax-based”, as opposed to statistics-based or rule-based checking methods. This approach requires a lexicon, a morphological analyzer, and a parser. When no parse succeeds on a given text, it is considered incorrect. This grammar checker was shown to be comparable with a commercially available one. Perhaps a Link Grammar implementation could be successful as well, as it would probably take a similar approach. Recently an English Link Grammar has been implemented as a grammar checker in AbiWord4, an open source word processor. Feature checking for agreement between words (e.g., phi-features between verbs and nouns) would need to be extended in the present system. This could be possible through the use of subscripts, which were introduced previously. 7.3 Morpho-syntactic information database Another potential use of the present system is in identifying and gathering interesting linguistic phenomena from large databases of written Arabic. Exporting the parses given from the Link Grammar parser to a Treebank formatted structure could potentially provide a more in-depth alternative to the current Arabic Treebank, since the current system provides descriptions of morpho-syntactic structures. Sleator and Temperley (1991) note that the grid of links output from the parser might be considered as a constituent framework. In the newest version of the parser, a system called “the phrase parser” enables a traditional constituent structure to be derived from a linkage. It does this by using a list of constituent types along with links that begin each one. These displays can take a variety of forms, including a tree data structure. It should be possible to evaluate Link Grammar in terms of transferal into a constituency structure. Sleator and Temperley tested the output from the English phrase parser with respect to the English Penn Treebank. On the complete text of the Penn Treebank, their parser correctly recognizes approximately 75% of the constituents5. We intend to more fully explore our Treebank formatted output with the use of this phrase parser. For Arabic, the links that begin constituent types would need reconsideration. Schneider (1998) discusses some other possibilities. Schneider considers each link type in terms of head and dependent. He lists the types, along with which side should be considered the head and which the dependent. Once a strict notion of a head for each type of link is found, conversion from a Link
4 5
http://www.abisource.com/ http://www.link.cs.cmu.edu/link/ph-explanation.html
A LINK GRAMMAR PARSER FOR ARABIC
243
Grammar to dependency or constituency grammars, or even to semantic frameworks, could be possible. 8.
Conclusions Link Grammar is capable of describing many distinct phenomena over a wide range of languages. An Arabic implementation of it has been developed for parsing. It has been shown that this parser is able to provide an in-depth morphosyntactic analysis of Arabic, as well as multiple parses of sentences that show ambiguity. Implementations for this parser could include use as a grammar checker or as a method for gathering linguistic phenomena from text corpora.
REFERENCES Abu-Arafah, Adnan. 1996. A grammar for the Arabic language suitable for machine parsing and automatic text generation. PhD diss., Illinois Institute of Technology. Al Daimi, Khalid & Maha Abdel-Amir. 1994. “The Syntactic Analysis of Arabic by Machine”. Computers and the Humanities 28.29-37. Buckwalter, Tim. 2002. Buckwalter Arabic Morphological Analyzer v. 1.0. Linguistics Data Consortium, University of Pennsylvania. LDC Catalog No.: LDC2002L49. Daimi, Kevin. 2001. “Identifying Syntactic Ambiguities in Single-Parse Arabic Sentence”. Computers and the Humanities 35.333-349. Dehdari, Jon & Deryle Lonsdale. 2006. A Link Grammar parser for Persian. Paper presented at the First International Conference on Aspects of Iranian Linguistics, June, in Leipzig, Germany. Maamouri, Mohamed, Ann Bies, Hubert Jin & Tim Buckwalter. 2003. Arabic Treebank: Part 1 v. 2.0. Linguistics Data Consortium, University of Pennsylvania. LDC Catalog No.: LDC2003T06. ---------- & Ann Bies. 2004. “Developing an Arabic Treebank: Methods, guidelines, procedures, and tools”. Proceedings of the COLING 2004 Workshop on Computational Approaches to Arabic Script-based Languages. Geneva, 2–9. Othman, Eman, Khaled Shaalan & Ahmed Rafea. 2003. “A Chart Parser for Analyzing Modern Standard Arabic Sentences”. Proceedings of the MT Summit IX Workshop on Machine Translation for Semitic Languages: Issues and Approaches. New Orleans, Louisiana.
244
CASBEER, DEHDARI & LONSDALE
Ouersighni, Riadh. 2001. “A major offshoot of the DIINAR-MBC project: AraParse, a morphosyntactic analyzer for unvowelled Arabic texts”. Proceedings of Arabic NLP Workshop at ACL/EACL. Toulouse. Protasov, Sergey. 2005. A Link Grammar for Russian. http://sz.ru/parser/Protasov-RussianLinkGrammar.pdf Schneider, Gerold. 1998. A linguistic comparison of constituency, dependency and Link Grammar. M.A. thesis, University of Zurich. Shaalan, Khaled. 2005. “Arabic GramCheck: A grammar checker for Arabic”. Software: Practice and Experience 35.643-665. Sleator, Daniel & David Temperley. 1991. “Parsing English with a Link Grammar”. Carnegie Mellon University Computer Science technical report CMU-CS-91-196. Weinberg, Amy, Joe Garman, Jeffery Martin & Paola Merlo. 1995. “A principle-based parser for foreign language tutoring in German and Arabic”. In Intelligent Language Tutors: Theory shaping technology, ed. Melissa Holland, Jonathan Kaplan and Michelle Sams, 23-44. Mahwah, NJ: Lawrence Erlbaum Associates.
INDEX OF SUBJECTS
acquisition L1 acquisition, 61, 65, 135-136 L2 acquisition, 135-149 agreement, 12 noun-adjective, 212 subject-verb, 55, 117, 135-148, 242 anaphora, 72, 77-78, 85 Bound Variable Anaphora, 89-92 Functor Anaphora Constraint, 85 Arabic Classical, 28 Egyptian, 71-85, 189-205 Iraqi, 203 Jordanian, 27, 87-104, 191 Kuwaiti, 174 Lebanese, 88, 90, 118, 174 Moroccan, 3-24, 129, 174 Palestinian, 28, 29, 105-129, 153164 Standard, 41, 51-65, 135-149, 160, 213-214 Syrian, 174, 191 Tunisian, 171-187 arity operations, 51-65 aspect, 109, 137, 139 back-channeling, 189-205 feedback, 190, 192, 201 Berber, 4, 9-10 Binding Conditions, 89, 92-96, 100 positive vs. negative, 94 Breton, 93 bundling, 54 case, 54, 211 accusative, 53 genitive, 212 nominative, 212 suffix, 4 causativization, 53-55, 60, 62, 64 clitics dative clitic l-, 114, 117 double clitics, 87, 91-94, 98-99 left-dislocation, 109-110, 117 possessive clitics, 121 special clitics, 107, 122, 128 codeswitching, 153-164 construct state, 72, 108, 115 decausativization, 53-55, 60, 62, 64 definite article, 7, 93 definiteness, 72, 74, 78, 82-86
demonstratives, 93 double demonstratives, 177, 182184 non-proximate, 176, 181 proximate, 175, 180 traditional analysis, 171 diglossia, 153, 163 diminutive, 5, 11-17, 26, 37, 56 discourse markers, 190 ellipsis, 98-99, 104 epithets, 87, 90, 93-94, 101-104 existential closure, 53 existential particle fii, 112 free state possessive nominals, 72 definiteness, 74, 82 distribution, 75, 83 possessivization, 76-77, 84 relativization, 75, 84 syntactic asymmetries of, 72, 79 word order, 74, 79-82 French, 54, 60, 72, 123 gender differences, 192 Givenness Hierarchy, 172-173, 179 Grammatical Framework, 211-231 Greek, 72 Hebrew, 26, 55, 57, 60, 72, 80 Hypocoristics, 15 Classical Arabic, 28 Hypocoristic Formation, 28-30 non-sound roots, 31-33 reduplicated, 42-47 sound symbolism, 11 types of, 28 interlanguage, 147-148 intonation, 107, 190 islands, 87-90, 99-102 weak vs. strong, 88, 90, 94 Japanese, 137, 189, 190, 192 Kadenz, 191 left-periphery, 109 Lexicon, 35 active lexicon, 52 erosion, 5 lexical root, 30, 32, 43 Lexicon Interface Guideline, 53-54
246
SUBJECT INDEX
Lexicon-Syntax Parameter, 52, 60 Link Grammar, 237-245 locality, 84-86 markedness, 27, 31, 35, 40 melodic overwriting, 39, 51, 57-58, 6063 Minimalist Program, 135, 139 Checking, 83 mora, 38, 59-61 bimoraic foot, 34, 46 bimoraic vowels, 30 intensive mora, 39 mora augmentation, 38, 40 moriac foot, 58 moraic syllable, 59 moraic trochee, 26, 34 ,36 trimoraic syllables, 36 morphological analyzer, 228-229 negation, 105-129 negative polarity items, 120 negative pronouns, 121 negative QP, 90-91 scope ambiguity, 78, 85 negative circumscription, 59 neuro-psycholinguistics, 153-164 noun deverbal nouns, 8-9, 12, 26, 40 intensive nouns, 38-39, 47 morphology, 223 foot structure, 34, 36 NP deletion, 87, 98, 100 orthography, 219 passivization, 51, 53, 55, 60-63 phonological processes deletion, 11, 59-60 gemination, 38-39, 41, 43, 54, 51, 59 lengthening, 60, 108, 191 shortening, 107 plural broken plural, 6-10, 16-17, 26, 36, 56-58 lexicalized plurals, 7 possessive nominals, 74-75, 84 possessivization, 72, 76, 78, 84 prosody, 190 pitch downdash, 197-201 prosodic circumscription, 58 prosodic hierarchy, 127 prosodic word, 109 quantifiers, 89, 92, 100
reciprocalization, 54, 62, 64 reconstruction, 87-90 without islands, 92 reflexivization, 54-55, 60 relativization, 72, 75, 78, 89, 84 resource grammar, 212-231 resyllabification, 108 resumption, 87, 101 apparent resumption, 87, 90, 9293, 101-104 properties of, 88 resumptive pronouns, 87-88, 90, 93, 101, 109 true resumption, 87, 90-92, 102104 roots, 25, 56-59, 213 bilateral, 33, 46-47 extraction, 64-65 hollow (glide medial), 27, 31-33, 46 input vs. output root, 27, 30-32, 42 order of root consonants, 56 primacy of, 26, 56 quadrilateral, 29, 34 weak (glide final) 31-32 saturation, 53 semantic anomaly, 158, 162-164 Split-INFL Hypothesis, 135 stress, 107 subject, 84, 92, 97, 103, 117 superiority effects, 84 Swadesh lexicon, 227 syllable adjacent syllables, 41 alignment within, 44 coda, 26, 39, 44-46 final long vowel, 32 onset, 26, 42, 44-45 reduplication, 46-47 size, 36 templates, 25-27, 31 ablaut templates, 11 historical change, 4, 17-18 MSA verbs, 55-56 tense, 109-111, 123-125, 135-149, 219 236 theta role, 52 agent, 54, 76, 84 cause, 53-54 experiencer, 53 external, 53 theme, 53-54, 76, 84 theta system, 52-53 turn-taking, 189-192, 196-205
SUBJECT INDEX
verb auxiliary verbs, 111, 123-125 causative, 6, 38, 45, 62-64 decausative verbs, 53-54, 62 demoninal verbs, 6 foot structure, 36-37 iambic verbs, 37 imperfect, 6, 55-58, 124 intensive, 38, 43, 45 morphology, 221 perfect, 55, 58 reciprocals, 54, 59 unergative, 54, 62
247
wh-clauses, 90 wh-island 87-90, 92-94 epithet within, 95 wh-movement, 90 wh-phrase, 84 wh-question, 84, 190 Word-Based Theory, 52, 56 word formation rules, 56 word order, 73, 89, 123, 227 possessive nominals, 72, 78-81 negation, 116 SVO order, 116-117 VSO order, 237
CURRENT ISSUES IN LINGUISTIC THEORY
E. F. K. Koerner, Editor
Zentrum für Allgemeine Sprachwissenschaft, Typologie und Universalienforschung, Berlin [email protected] Current Issues in Linguistic Theory (CILT) is a theory-oriented series which welcomes contributions from scholars who have significant proposals to make towards the advancement of our understanding of language, its structure, functioning and development. CILT has been established in order to provide a forum for the presentation and discussion of linguistic opinions of scholars who do not necessarily accept the prevailing mode of thought in linguistic science. It offers an outlet for meaningful contributions to the current linguistic debate, and furnishes the diversity of opinion which a healthy discipline must have. A complete list of titles in this series can be found on the publishers’ website, www.benjamins.com 293 Detges, Ulrich and Richard Waltereit (eds.): The Paradox of Grammatical Change. Perspectives from Romance. v, 254 pp. Expected January 2008 292 Nicolov, Nicolas, Kalina Bontcheva, Galia Angelova and Ruslan Mitkov (eds.): Recent Advances in Natural Language Processing IV. Selected papers from RANLP 2005. xii, 307 pp. Expected December 2007 291 Baauw, Sergio, Frank Drijkoningen and Manuela Pinto (eds.): Romance Languages and Linguistic Theory 2005. Selected papers from ‘Going Romance’, Utrecht, 8–10 December 2005. 2007. viii, 338 pp. 290 Mughazy, Mustafa A. (ed.): Perspectives on Arabic Linguistics XX. Papers from the twentieth annual symposium on Arabic linguistics, Kalamazoo, Michigan, March 2006. 2007. xii, 247 pp. 289 Benmamoun, Elabbas (ed.): Perspectives on Arabic Linguistics XIX. Papers from the nineteenth annual symposium on Arabic Linguistics, Urbana, Illinois, April 2005. xiv, 274 pp. + index. Expected December 2007 288 Toivonen, Ida and Diane Nelson (eds.): Saami Linguistics. 2007. viii, 321 pp. 287 Camacho, José, Nydia Flores-Ferrán, Liliana Sánchez, Viviane Déprez and María José Cabrera (eds.): Romance Linguistics 2006. Selected papers from the 36th Linguistic Symposium on Romance Languages (LSRL), New Brunswick, March-April 2006. 2007. viii, 340 pp. 286 Weijer, Jeroen van de and Erik Jan van der Torre (eds.): Voicing in Dutch. (De)voicing – phonology, phonetics, and psycholinguistics. 2007. x, 186 pp. 285 Sackmann, Robin (ed.): Explorations in Integrational Linguistics. Four essays on German, French, and Guaraní. ix, 217 pp. Expected January 2008 284 Salmons, Joseph C. and Shannon Dubenion-Smith (eds.): Historical Linguistics 2005. Selected papers from the 17th International Conference on Historical Linguistics, Madison, Wisconsin, 31 July - 5 August 2005. 2007. viii, 413 pp. 283 Lenker, Ursula and Anneli Meurman-Solin (eds.): Connectives in the History of English. 2007. viii, 318 pp. 282 Prieto, Pilar, Joan Mascaró and Maria-Josep Solé (eds.): Segmental and prosodic issues in Romance phonology. 2007. xvi, 262 pp. 281 Vermeerbergen, Myriam, Lorraine Leeson and Onno Crasborn (eds.): Simultaneity in Signed Languages. Form and function. 2007. viii, 360 pp. (incl. CD-Rom). 280 Hewson, John and Vit Bubenik: From Case to Adposition. The development of configurational syntax in Indo-European languages. 2006. xxx, 420 pp. 279 Nedergaard Thomsen, Ole (ed.): Competing Models of Linguistic Change. Evolution and beyond. 2006. vi, 344 pp. 278 Doetjes, Jenny and Paz González (eds.): Romance Languages and Linguistic Theory 2004. Selected papers from ‘Going Romance’, Leiden, 9–11 December 2004. 2006. viii, 320 pp. 277 Helasvuo, Marja-Liisa and Lyle Campbell (eds.): Grammar from the Human Perspective. Case, space and person in Finnish. 2006. x, 280 pp. 276 Montreuil, Jean-Pierre Y. (ed.): New Perspectives on Romance Linguistics. Vol. II: Phonetics, Phonology and Dialectology. Selected papers from the 35th Linguistic Symposium on Romance Languages (LSRL), Austin, Texas, February 2005. 2006. x, 213 pp. 275 Nishida, Chiyo and Jean-Pierre Y. Montreuil (eds.): New Perspectives on Romance Linguistics. Vol. I: Morphology, Syntax, Semantics, and Pragmatics. Selected papers from the 35th Linguistic Symposium on Romance Languages (LSRL), Austin, Texas, February 2005. 2006. xiv, 288 pp. 274 Gess, Randall S. and Deborah Arteaga (eds.): Historical Romance Linguistics. Retrospective and perspectives. 2006. viii, 393 pp.
273 Filppula, Markku, Juhani Klemola, Marjatta Palander and Esa Penttilä (eds.): Dialects Across Borders. Selected papers from the 11th International Conference on Methods in Dialectology (Methods XI), Joensuu, August 2002. 2005. xii, 291 pp. 272 Gess, Randall S. and Edward J. Rubin (eds.): Theoretical and Experimental Approaches to Romance Linguistics. Selected papers from the 34th Linguistic Symposium on Romance Languages (LSRL), Salt Lake City, March 2004. 2005. viii, 367 pp. 271 Branner, David Prager (ed.): The Chinese Rime Tables. Linguistic philosophy and historicalcomparative phonology. 2006. viii, 358 pp. 270 Geerts, Twan, Ivo van Ginneken and Haike Jacobs (eds.): Romance Languages and Linguistic Theory 2003. Selected papers from ‘Going Romance’ 2003, Nijmegen, 20–22 November. 2005. viii, 369 pp. 269 Hargus, Sharon and Keren Rice (eds.): Athabaskan Prosody. 2005. xii, 432 pp. 268 Cravens, Thomas D. (ed.): Variation and Reconstruction. 2006. viii, 223 pp. 267 Alhawary, Mohammad T. and Elabbas Benmamoun (eds.): Perspectives on Arabic Linguistics XVII–XVIII. Papers from the seventeenth and eighteenth annual symposia on Arabic linguistics. Volume XVII–XVIII: Alexandria, 2003 and Norman, Oklahoma 2004. 2005. xvi, 315 pp. 266 Boudelaa, Sami (ed.): Perspectives on Arabic Linguistics XVI. Papers from the sixteenth annual symposium on Arabic linguistics, Cambridge, March 2002. 2006. xii, 181 pp. 265 Cornips, Leonie and Karen P. Corrigan (eds.): Syntax and Variation. Reconciling the Biological and the Social. 2005. vi, 312 pp. 264 Dressler, Wolfgang U., Dieter Kastovsky, Oskar E. Pfeiffer and Franz Rainer (eds.): Morphology and its demarcations. Selected papers from the 11th Morphology meeting, Vienna, February 2004. With the assistance of Francesco Gardani and Markus A. Pöchtrager. 2005. xiv, 320 pp. 263 Branco, António, Tony McEnery and Ruslan Mitkov (eds.): Anaphora Processing. Linguistic, cognitive and computational modelling. 2005. x, 449 pp. 262 Vajda, Edward J. (ed.): Languages and Prehistory of Central Siberia. 2004. x, 275 pp. 261 Kay, Christian J. and Jeremy J. Smith (eds.): Categorization in the History of English. 2004. viii, 268 pp. 260 Nicolov, Nicolas, Kalina Bontcheva, Galia Angelova and Ruslan Mitkov (eds.): Recent Advances in Natural Language Processing III. Selected papers from RANLP 2003. 2004. xii, 402 pp. 259 Carr, Philip, Jacques Durand and Colin J. Ewen (eds.): Headhood, Elements, Specification and Contrastivity. Phonological papers in honour of John Anderson. 2005. xxviii, 405 pp. 258 Auger, Julie, J. Clancy Clements and Barbara Vance (eds.): Contemporary Approaches to Romance Linguistics. Selected Papers from the 33rd Linguistic Symposium on Romance Languages (LSRL), Bloomington, Indiana, April 2003. With the assistance of Rachel T. Anderson. 2004. viii, 404 pp. 257 Fortescue, Michael, Eva Skafte Jensen, Jens Erik Mogensen and Lene Schøsler (eds.): Historical Linguistics 2003. Selected papers from the 16th International Conference on Historical Linguistics, Copenhagen, 11–15 August 2003. 2005. x, 312 pp. 256 Bok-Bennema, Reineke, Bart Hollebrandse, Brigitte Kampers-Manhe and Petra Sleeman (eds.): Romance Languages and Linguistic Theory 2002. Selected papers from ‘Going Romance’, Groningen, 28–30 November 2002. 2004. viii, 273 pp. 255 Meulen, Alice ter and Werner Abraham (eds.): The Composition of Meaning. From lexeme to discourse. 2004. vi, 232 pp. 254 Baldi, Philip and Pietro U. Dini (eds.): Studies in Baltic and Indo-European Linguistics. In honor of William R. Schmalstieg. 2004. xlvi, 302 pp. 253 Caffarel, Alice, J.R. Martin and Christian M.I.M. Matthiessen (eds.): Language Typology. A functional perspective. 2004. xiv, 702 pp. 252 Kay, Christian J., Carole Hough and Irené Wotherspoon (eds.): New Perspectives on English Historical Linguistics. Selected papers from 12 ICEHL, Glasgow, 21–26 August 2002. Volume II: Lexis and Transmission. 2004. xii, 273 pp. 251 Kay, Christian J., Simon Horobin and Jeremy J. Smith (eds.): New Perspectives on English Historical Linguistics. Selected papers from 12 ICEHL, Glasgow, 21–26 August 2002. Volume I: Syntax and Morphology. 2004. x, 264 pp. 250 Jensen, John T.: Principles of Generative Phonology. An introduction. 2004. xii, 324 pp. 249 Bowern, Claire and Harold Koch (eds.): Australian Languages. Classification and the comparative method. 2004. xii, 377 pp. (incl. CD-Rom). 248 Weigand, Edda (ed.): Emotion in Dialogic Interaction. Advances in the complex. 2004. xii, 284 pp. 247 Parkinson, Dilworth B. and Samira Farwaneh (eds.): Perspectives on Arabic Linguistics XV. Papers from the Fifteenth Annual Symposium on Arabic Linguistics, Salt Lake City 2001. 2003. x, 214 pp. 246 Holisky, Dee Ann and Kevin Tuite (eds.): Current Trends in Caucasian, East European and Inner Asian Linguistics. Papers in honor of Howard I. Aronson. 2003. xxviii, 426 pp.
245 Quer, Josep, Jan Schroten, Mauro Scorretti, Petra Sleeman and Els Verheugd (eds.): Romance Languages and Linguistic Theory 2001. Selected papers from 'Going Romance', Amsterdam, 6–8 December 2001. 2003. viii, 355 pp. 244 Pérez-Leroux, Ana Teresa and Yves Roberge (eds.): Romance Linguistics. Theory and Acquisition. Selected papers from the 32nd Linguistic Symposium on Romance Languages (LSRL), Toronto, April 2002. 2003. viii, 388 pp. 243 Cuyckens, Hubert, Thomas Berg, René Dirven and Klaus-Uwe Panther (eds.): Motivation in Language. Studies in honor of Günter Radden. 2003. xxvi, 403 pp. 242 Seuren, Pieter A.M. and Gerard Kempen (eds.): Verb Constructions in German and Dutch. 2003. vi, 316 pp. 241 Lecarme, Jacqueline (ed.): Research in Afroasiatic Grammar II. Selected papers from the Fifth Conference on Afroasiatic Languages, Paris, 2000. 2003. viii, 550 pp. 240 Janse, Mark and Sijmen Tol (eds.): Language Death and Language Maintenance. Theoretical, practical and descriptive approaches. With the assistance of Vincent Hendriks. 2003. xviii, 244 pp. 239 Andersen, Henning (ed.): Language Contacts in Prehistory. Studies in Stratigraphy. Papers from the Workshop on Linguistic Stratigraphy and Prehistory at the Fifteenth International Conference on Historical Linguistics, Melbourne, 17 August 2001. 2003. viii, 292 pp. 238 Núñez-Cedeño, Rafael, Luis López and Richard Cameron (eds.): A Romance Perspective on Language Knowledge and Use. Selected papers from the 31st Linguistic Symposium on Romance Languages (LSRL), Chicago, 19–22 April 2001. 2003. xvi, 386 pp. 237 Blake, Barry J. and Kate Burridge (eds.): Historical Linguistics 2001. Selected papers from the 15th International Conference on Historical Linguistics, Melbourne, 13–17 August 2001. Editorial assistance Jo Taylor. 2003. x, 444 pp. 236 Simon-Vandenbergen, Anne-Marie, Miriam Taverniers and Louise J. Ravelli (eds.): Grammatical Metaphor. Views from systemic functional linguistics. 2003. vi, 453 pp. 235 Linn, Andrew R. and Nicola McLelland (eds.): Standardization. Studies from the Germanic languages. 2002. xii, 258 pp. 234 Weijer, Jeroen van de, Vincent J. van Heuven and Harry van der Hulst (eds.): The Phonological Spectrum. Volume II: Suprasegmental structure. 2003. x, 264 pp. 233 Weijer, Jeroen van de, Vincent J. van Heuven and Harry van der Hulst (eds.): The Phonological Spectrum. Volume I: Segmental structure. 2003. x, 308 pp. 232 Beyssade, Claire, Reineke Bok-Bennema, Frank Drijkoningen and Paola Monachesi (eds.): Romance Languages and Linguistic Theory 2000. Selected papers from ‘Going Romance’ 2000, Utrecht, 30 November–2 December. 2002. viii, 354 pp. 231 Cravens, Thomas D.: Comparative Historical Dialectology. Italo-Romance clues to Ibero-Romance sound change. 2002. xii, 163 pp. 230 Parkinson, Dilworth B. and Elabbas Benmamoun (eds.): Perspectives on Arabic Linguistics. Papers from the Annual Symposium on Arabic Linguistics. Volume XIII-XIV: Stanford, 1999 and Berkeley, California 2000. 2002. xiv, 250 pp. 229 Nevin, Bruce E. and Stephen B. Johnson (eds.): The Legacy of Zellig Harris. Language and information into the 21st century. Volume 2: Mathematics and computability of language. 2002. xx, 312 pp. 228 Nevin, Bruce E. (ed.): The Legacy of Zellig Harris. Language and information into the 21st century. Volume 1: Philosophy of science, syntax and semantics. 2002. xxxvi, 323 pp. 227 Fava, Elisabetta (ed.): Clinical Linguistics. Theory and applications in speech pathology and therapy. 2002. xxiv, 353 pp. 226 Levin, Saul: Semitic and Indo-European. Volume II: Comparative morphology, syntax and phonetics. 2002. xviii, 592 pp. 225 Shahin, Kimary N.: Postvelar Harmony. 2003. viii, 344 pp. 224 Fanego, Teresa, Belén Méndez-Naya and Elena Seoane (eds.): Sounds, Words, Texts and Change. Selected papers from 11 ICEHL, Santiago de Compostela, 7–11 September 2000. Volume 2. 2002. x, 310 pp. 223 Fanego, Teresa, Javier Pérez-Guerra and María José López-Couso (eds.): English Historical Syntax and Morphology. Selected papers from 11 ICEHL, Santiago de Compostela, 7–11 September 2000. Volume 1. 2002. x, 306 pp. 222 Herschensohn, Julia, Enrique Mallén and Karen Zagona (eds.): Features and Interfaces in Romance. Essays in honor of Heles Contreras. 2001. xiv, 302 pp. 221 D’hulst, Yves, Johan Rooryck and Jan Schroten (eds.): Romance Languages and Linguistic Theory 1999. Selected papers from ‘Going Romance’ 1999, Leiden, 9–11 December 1999. 2001. viii, 406 pp.
220 Satterfield, Teresa, Christina M. Tortora and Diana Cresti (eds.): Current Issues in Romance Languages. Selected papers from the 29th Linguistic Symposium on Romance Languages (LSRL), Ann Arbor, 8–11 April 1999. 2002. viii, 412 pp. 219 Andersen, Henning (ed.): Actualization. Linguistic Change in Progress. Papers from a workshop held at the 14th International Conference on Historical Linguistics, Vancouver, B.C., 14 August 1999. 2001. vii, 250 pp. 218 Bendjaballah, Sabrina, Wolfgang U. Dressler, Oskar E. Pfeiffer and Maria D. Voeikova (eds.): Morphology 2000. Selected papers from the 9th Morphology Meeting, Vienna, 24–28 February 2000. 2002. viii, 317 pp. 217 Wiltshire, Caroline R. and Joaquim Camps (eds.): Romance Phonology and Variation. Selected papers from the 30th Linguistic Symposium on Romance Languages, Gainesville, Florida, February 2000. 2002. xii, 238 pp. 216 Camps, Joaquim and Caroline R. Wiltshire (eds.): Romance Syntax, Semantics and L2 Acquisition. Selected papers from the 30th Linguistic Symposium on Romance Languages, Gainesville, Florida, February 2000. 2001. xii, 246 pp. 215 Brinton, Laurel J. (ed.): Historical Linguistics 1999. Selected papers from the 14th International Conference on Historical Linguistics, Vancouver, 9–13 August 1999. 2001. xii, 398 pp. 214 Weigand, Edda and Marcelo Dascal (eds.): Negotiation and Power in Dialogic Interaction. 2001. viii, 303 pp. 213 Sornicola, Rosanna, Erich Poppe and Ariel Shisha-Halevy (eds.): Stability, Variation and Change of Word-Order Patterns over Time. With the assistance of Paola Como. 2000. xxxii, 323 pp. 212 Repetti, Lori (ed.): Phonological Theory and the Dialects of Italy. 2000. x, 301 pp. 211 Elšík, Viktor and Yaron Matras (eds.): Grammatical Relations in Romani. The Noun Phrase. with a Foreword by Frans Plank (Universität Konstanz). 2000. x, 244 pp. 210 Dworkin, Steven N. and Dieter Wanner (eds.): New Approaches to Old Problems. Issues in Romance historical linguistics. 2000. xiv, 235 pp. 209 King, Ruth: The Lexical Basis of Grammatical Borrowing. A Prince Edward Island French case study. 2000. xvi, 241 pp. 208 Robinson, Orrin W.: Whose German? The ach/ich alternation and related phenomena in ‘standard’ and ‘colloquial’. 2001. xii, 178 pp. 207 Sanz, Montserrat: Events and Predication. A new approach to syntactic processing in English and Spanish. 2000. xiv, 219 pp. 206 Fawcett, Robin P.: A Theory of Syntax for Systemic Functional Linguistics. 2000. xxiv, 360 pp. 205 Dirven, René, Roslyn M. Frank and Cornelia Ilie (eds.): Language and Ideology. Volume 2: descriptive cognitive approaches. 2001. vi, 264 pp. 204 Dirven, René, Bruce Hawkins and Esra Sandikcioglu (eds.): Language and Ideology. Volume 1: theoretical cognitive approaches. 2001. vi, 301 pp. 203 Norrick, Neal R.: Conversational Narrative. Storytelling in everyday talk. 2000. xiv, 233 pp. 202 Lecarme, Jacqueline, Jean Lowenstamm and Ur Shlonsky (eds.): Research in Afroasiatic Grammar. Papers from the Third conference on Afroasiatic Languages, Sophia Antipolis, 1996. 2000. vi, 386 pp. 201 Dressler, Wolfgang U., Oskar E. Pfeiffer, Markus A. Pöchtrager and John R. Rennison (eds.): Morphological Analysis in Comparison. 2000. x, 261 pp. 200 Anttila, Raimo: Greek and Indo-European Etymology in Action. Proto-Indo-European *aǵ-. 2000. xii, 314 pp. 199 Pütz, Martin and Marjolijn H. Verspoor (eds.): Explorations in Linguistic Relativity. 2000. xvi, 369 pp. 198 Niemeier, Susanne and René Dirven (eds.): Evidence for Linguistic Relativity. 2000. xxii, 240 pp. 197 Coopmans, Peter, Martin Everaert and Jane Grimshaw (eds.): Lexical Specification and Insertion. 2000. xviii, 476 pp. 196 Hannahs, S.J. and Mike Davenport (eds.): Issues in Phonological Structure. Papers from an International Workshop. 1999. xii, 268 pp. 195 Herring, Susan C., Pieter van Reenen and Lene Schøsler (eds.): Textual Parameters in Older Languages. 2001. x, 448 pp. 194 Coleman, Julie and Christian J. Kay (eds.): Lexicology, Semantics and Lexicography. Selected papers from the Fourth G. L. Brook Symposium, Manchester, August 1998. 2000. xiv, 257 pp. 193 Klausenburger, Jurgen: Grammaticalization. Studies in Latin and Romance morphosyntax. 2000. xiv, 184 pp.
