Subject-Verb Agreement in Modern Standard Arabic an LFG Implementation in the Xerox Language Engineering Environment

Subject-Verb Agreement in Modern Standard Arabic An LFG Implementation in the Xerox Language Engineering Environment

Frederick Hoyt University of Texas at Austin

Abstract MSA is a pro-drop language, meaning that verbal agreement morphemes can be interpreted as “pro- This report descibes a grammar for mod- nouns.” The first and second person agreement elling the morphosyntax of verbal agree- forms are always pronominal in this sense, as are ment in Modern Standard Arabic (MSA). all plural agreement forms. This means that an in- The grammar is implemented in the Xe- dependent NP corresponding to the subject can only rox Language Engineering (XLE) envi- appear with pronominal agreement if the NP is inter- ronment, which is based on the Lexical preted as a discourse function, in particular as a FO- Functional Grammar framework. MSA CUS rather than as the subject. However, in the 3rd- has a complex verbal agreement system person-singular, independent subject NPs can also which defies being modelled in terms of be used with the verb, in which case the agreement constraints on a single formal representa- marking simply “matches” some subset of the sub- tion. Instead, the complexities of agree- ject’s AGR-features (Fassi-Fehri, 1988). ment in MSA are best decribed as two agreement systems, each of which re- 1.1 Agreement and word order quires its own constraints; the interaction If a sentence contains a singleton subject NP, how of these produces the complexities in the the verb is marked for agreement depends on the data. The XLE architecture allows dif- word order of the subject relative to the verb. In VS ferent grammatical representations to be order the verb agrees with the subject only in gender used, each of which is subject to differ- and is marked in the singular, whether the subject is ent constraints. Use of XLE therefore al- singular (1) or plural (2). Plural marking on the verb lows the agreement facts in MSA to be is only acceptable if the NP is interpreted with con- modelled in a compactly modular gram- trastive focus, and hence in LFG terms as a discourse mar which generates a low number of am- function rather than as a SUBJ (3): biguities in parsing. (1) kætaba l-waladu l-wæ:gıba.ˇ wrote.prf.3MS the-boy the-paper “The boy wrote the paper.” 1 Verbal Agreement Marking in MSA (2) kætaba l-awlæ:du l-wæ:gıba.ˇ Arabic verbs are richly inflected and show marking wrote.3MS the-boys.MP the-paper for number (singular, plural, dual), gender (mascu- “The boys wrote the paper.” line, feminine), and person (1st, 2nd, 3rd), as well as (3) kætabu l-awlæ:du l-wæ:gıbaˇ ( wa-lE: wrote.3MP the-boys.MP the-assignment and-not for mood (which will not be addressed here). I refer ıl-bænæ:tu ). to number, gender, and person features collectively the-girls.FS as agreement- or AGR-features. “The BOYS wrote the homework (and not the girls).” In SV word order the verb agrees with the subject 1.2 Agreement with Conjoined Subjects NP in gender and number. If the subject is singular, Agreement marking is further complicated with sub- the verb is marked as singular (4); if the subject is jects consisting of conjoined NPs. In VS word order, plural, the verb must be marked as plural (5); singu- the verb agrees only with the first conjunct (16). As lar marking is unacceptable (6): before, if the verb has non-singular agreement mark- (4) Pal-waladu kætaba l-wæ:gıba.ˇ ing then the conjoined NP must interpreted as a FO- the-boy wrote.prf.3MS the-paper CUS rather than as a subject: “The boy wrote the paper.” : (5) Pal-awlæ:du kætabu l-wæ:gıba.ˇ (16) kætabat ıl-bıntu wa-l-waladu l-wæ gıba.ˇ the-boys.MP wrote.3MP the-paper wrote.3FS the-girl.FS and-the-boy.MS the-paper “The boys wrote the paper.” “The girl and the boy wrote the paper.” (6) * Pal-awlæ:du kætaba l-wæ:gıba.ˇ (17) kætabæ: ıl-bıntu wa-l-waladu the-boys.MP wrote.3MS the-paper wrote.3DL the-girl.FS and-the-boy.MS : E: P “Same.” l-wæ gıbaˇ ( wa-l ana ). the-assignment and-not I In sentences with an initial auxiliary verb both the “The GIRL AND THE BOY wrote the paper (and VS and the SV rules apply. In Aux-S-V word order, not me).” the auxiliary agrees only in gender while the main As before, the verb agrees with a plural first conjunct verb agrees in both gender and number: only in gender: : : (7) kæ nat ıl-bıntu taktubu l-wæ gıba.ˇ : : was.3FS the-girl.FS write.3FS the-paper (18) kætaba l-awlæ du wa-l-banæ tu wrote.3MS the-boys.MS and-the-girls.FS “The girl was writing the paper.” l-wæ:gıba.ˇ (8) kæ:nat ıl-bænæ:tu yaktUbna l-wæ:gıba.ˇ the-paper was.3FS the-girl.FP write.3FP the-paper “The boys and the girls wrote the paper.” “The girls were writing the assingment.” (19) kætabat ıl-banæ:tu wa-l-awæ:du l-wæ:gıba.ˇ If the subject precedes the auxiliary, then both verbs wrote.3FS the-boys.MS and-the-girls.FP the-paper agree with it in both gender and number: “The girls and the boys wrote the paper.” (9) Pal-bıntu kæ:nat taktubu l-wæ:gıba.ˇ In sentences with SV word order, the verb agrees the-girl.FS was.3FS write.3FS the-paper with the whole conjoined subject rather than just one “Same.” of its conjuncts. The gender of the whole conjoined (10) Pal-bænæ:tu kUnna yaktUbna l-wæ:gıba.ˇ the-girls.FP were.3FP write.3FP the-paper NP is masculine (20) unless both conjuncts are fem- “Same.” inine nouns (22): (11) * Pal-bænæ:tu kæ:nEt yaktUbna l-wæ:gıba.ˇ (20) Pal-bıntu wa-l-waladu kætabæ: l-wæ:gıba.ˇ the-girls.FP be.3FS write.3FP the-paper the-girls.FS and-the-boy.MS wrote.3MD the-paper “Same.” “The girl and the boy wrote the paper.” P : : The agreement patterns in VS vs. SV word order are (21) * al-bıntu wa-l-waladu kætabatæ l-wæ gıba.ˇ the-girls.FS and-the-boy.MS wrote.3FD the-paper summarized in the following diagrams: “The girl and the boy wrote the paper.” gen (22) Pal-bıntu wa-PUmmu-hæ kætabatæ: the-girl.FS and-mother.FS-nom-cl3FS wrote.3FD (12) Verb Subject (13) Subject Verb l-wæ:gıba.ˇ the-paper gen+num “The girl and her mother wrote the paper.” gen (23) * Pal-bıntu wa-PUmmu-hæ kætabæ: the-girl.FS and-mother.FS-nom-cl3FS wrote.3MD l-wæ:gıba.ˇ (14) Aux Subject Main-Verb the-paper “The girl and her mother wrote the paper.” gen+num (15) Subject Aux Main-Verb Likewise, the number value of the whole conjoined NP will be plural unless both conjuncts are singular, gen+num gen+num in which case the whole NP is dual (26): gen (24) l-banæ:tu wa-l-awæ:du kætabu the-girls.FS and-the-boys.MS wrote.3MP l-wæ:gıba.ˇ (32) Main-Verb [ NP1 & NP2 ] the-paper “The girls and the boys wrote the paper.” (33) [NP1 & NP2 ] Main-Verb (25) Pal-bænæ:tu wa-l-awlæ:du kætabæ: the-girls.FP and-the-boys.MP wrote.3MD gen+num l-wæ:gıba.ˇ | gen{z } the-paper

“Same.” (34) Aux [ NP1 & NP2 ] Main-Verb (26) Pal-bıntu wa-l-waladu kætabæ: l-wæ:gıba.ˇ the-girl.FS and-the-boy.MS wrote.3MD the-paper gen+num “The girl and the boy wrote the paper.” | {z } (35) [NP1 & NP2 ] Aux Main-Verb (27) * Pal-bıntu wa-l-waladu kætabu l-wæ:gıba.ˇ the-girl.FS and-the-boy.MS wrote.3MP the-paper “Same.” | {z } gen+num gen+num In Aux-S-V order, the auxiliary agrees with the ﬁrst 1.3 SV Agreement and Anaphoric Agreement conjunct, while the main verb agrees with the whole The agreement marking patterns seen in the SV conjoined NP. This creates striking mismatches in word orders are identical to patterns in agreement the agreement marking between the auxiliary in the between anaphoric pronouns and their antecedents. main verb. For example, in (28, 29) the auxiliary is In particular, anaphoric pronouns agree with their in the feminine singular, agreeing with the ﬁrst con- antecendents in gender, person, and number: junct while the main verb is in the masculine dual : U (28) or plural (29), dependening on whether the two (36) ˇgiræ n-na, raPaytu-h m yašrabu:na neighbors.MP-cl1P saw.1S-cl3MP drink.3MP conjoined nouns are singular or plural: l-hæšiš. the-hashish (28) kænæt ıl-bıntu wa-l-waladu yaktUbæ: was.3FS the-girl.FS and-the-boy.MS write.3DL “Our neighbors, I saw them smoking hashish.” l-wæ:gıba.ˇ (37) * ˇgiræ:n-na, raPaytu-hu yašrabu:na the-paper neighbors.MP-cl1P saw.1S-cl3MS drink.3MP “The girl and the boy were writing the paper.” l-hæšiš. the-hashish (29) kæ:næt ıl-banæ:tu wa-l-awlæ:du was.3FS the-girls.3FS and-the-boys.MP “Same.” yaktubu:na l-wæ:gıba.ˇ write.3MP the-paper If the antecedent is a conjoined NP, the same gender “The girls and the boys were writing the assing- and number resolution rules apply as in agreement ment.” marking with pre-verbal conjoined subjects:

In S-Aux-V word order, the auxiliary and main verb (38) Pabu:-y wa-PUmm-i, hUmmæ: mın agree with the conjoined NP in number and gender: father.MS-cl1S and-mother.FS-cl1S pro.3MD from asli Qarabi:. ˙ (30) ıl-bıntu wa-l-waladu kænæ: yaktUbæ: origin.MS-gen Arab.MS-gen the-girl.FS and-the-boy.MS was.3FS write.3DL “My father and my mother are of Arab origin.” :ˇ l-wæ gıba. P : PU the-paper (39) * abu -y wa- mm-i, hu mın father.MS-cl1S and-mother.FS-cl1S pro.3MS from “The girl and the boy were writing the paper.” asli Qarabi:. ˙ (31) * ıl-banæ:tu wa-l-awlæ:du kæ:næt origin.MS-gen Arab.MS-gen the-girls.3FS and-the-boys.MP was.3FS “Same.” yaktubu:na l-wæ:gıba.ˇ write.3MP the-paper This suggests that the dependency between pre- “The girls and the boys were writing the assing- verbal subjects and the agreement marking on the ment.” verb is similar to the dependency between an The agreement patterns for conjoined subject NPs anaphoric pronoun and its antecedent, and therefore are summarized in the following schemata: that it is a semantic dependency. 1.4 Discussion b. Constrains person, gender, and number. The data reviewed above suggest the following de- For example, the sentence in (44) has the syntactic scriptive generalizations for verbal agreement mark- structure in (45): ing in Standard Arabic: (44) kætabat ıl-bænæ:tu l-wæ:gıba.ˇ (40) Standard Arabic has two kinds of agreement: wrote.3FS the-girls.3FS the-paper a. Grammatical agreement: agreement marking “The girls wrote the paper.” constraining the form of the subject of the verb. Sf1 (45) ↑=↓ b. Semantic agreement: agreement marking constraining the interpretation of the subject. (41) Agreement type is determined by word order: a. Grammatical agreement obtains in VS order or Vf2 NPf3 NPf4 ↑=↓ ↑SUBJ ↓ ↑OBJ ↓ any verb stem - subject pair; ( )= ( )= b. Semantic agreement obtains in SV word order as kætabat ıl-bænæ:tu ıl-wæ:ˇgıba well as in antecedent-pronoun binding.

PRED ‘girl’ These pose two problems for a syntactic theory GEN fem SUBJ f3 NUM plur AGR which views the agreement relation as a simple mat-   PERS 3rd   " DEF + # ter of feature unification between two sets of fea- 1 2  PRED  ↑SUBJ ↑OBJ   f ,f  ’write<( )( )>’  tures, one specified on the verb and the other on    PRED ‘paper’  the subject NP. The first problem is that the agree-  GEN masc   OBJ f4 NUM plur  AGR ment marking for a given verb stem seems to license  PERS    3rd   nouns in one position more restrictively than it does  " DEF + #      in others. The second problem is that two verb stems REL x-is-group-of-girls can occur in the same clause and agree with the same OP the-unique-x ARG1 GEN fem subject but show different agreement markings.   AGR NUM plur   PERS 3rd     2 Analysis and Implementation  REL x-is-a-paper   OP the-unique-x     ARG2 GEN masc  In LFG terms, grammatical agreement is represented   AGR NUM sing    PERS 3rd  as a set of constraints on the different projections of    REL ARG WRITES ARG    1- - 2  a sentence. The solution I propose uses two pro-   jections from C-structure: the standard f-structure 2.1 Agreement with singleton subject NPs as well as an additional s-structure projection (for Agreement in VS word order is expressed with f- “semantic structure”) which is projected from f- structure constraints. These are implemented as lex- structure (Wechsler & Zlatic, 2003): ical templates which are included in the verb lexi- C-structure cal entries. Verb stems marked in the 1st- and 3nd- persons as well as all plural verbs have “pronominal agreement marking,” meaning that they are lexically F-structure S-structure (Grammatical Information) (Semantic Information) specified with a lexical template which introduces a pronominal subject at f-structure: Grammatical agreement is represented as a set of constraints on the f-structure of the sentence, while (46) ProAgr = (^ SUBJ PRED)= 'pro' (^ SUBJ PRON)= pers semantic agreement is represented as constraints on (s::^ ARG1 AGR) = (^SUBJ AGR) its s-structure: (s::^ ARG1 REL) = 'pro'. (42) Grammatical Agreement: Constraints on F-structures This constraint optionally introduces an s-structure a. Verbal agreement marking imposes constraints pronoun and identifies the AGR-features of the f- on the subject function of the clause; b. Constrains only the subject’s gender feature. structure SUBJ function with the AGR-feature of the (43) Semantic Agreement: Constraints on Interpretation s-structure argument. For example, the sentence in a. Verbal agreement marking constrains the inter- (47) has the f-structure in (48) and the s-structure in pretation of the subject; (49): (47) kætabu l-wæ:gıba.ˇ The first conjunct specified semantic agreement con- wrote.3MP the-assignment straints as above, while the second conjunct specifies “They [masc] wrote the paper.” grammatical agreement: (48) f-structure: "katabu il-waajiba" (55) GramAgr( _G ) = (^SUBJ AGR GEN) =c _G

PRED 'write<[0-SUBJ:pro], [3:assignment]>' @(PersAgr 3rd ). PRED 'pro' SUBJ AGR GEN masc, NUM plur, PERS 3rd CASE nom, PRON pers This template introduces a constraining equation PRED 'assignment' 0 OBJ AGR ANIM -, GEN masc, NUM sing, PERS 3rd 1 which requires the value of the f-structure SUBJ 11 3 CASE acc, DEF + 17 ASPECT Perfect gender feature to match that of the argument of the (49) s-structure: template. For this constraint to be satisfied, an independent NP will have to provide a gender feature AGR GEN masc, NUM plur, PERS 3rd ARG1 REL 'pro' 0 with the appropriate value. The only other mecha- AGR GEN masc, NUM sing, PERS 3rd 1 ARG2 2 12 3 OP 'the-unique-x', REL 'x-is-an-assignment' 18 REL 'ARG1-write-ARG2' nism in the grammar which can introduce an independent subject NP is the following rule: The agreement marking on the verb introduces a (56) S --> [V: ^=! ] “pro” subject specified with 3rd-masculine-plural [([NP: (^SUBJ)=!]), AGR-features. This pronominal subject projects ([NP: (ÔBJ)=!]), ([NP: (^FOCUS)=! an s-structure ARG1 function with which it shares (^GF) = %f2 identical AGR-features. The pronominal feature [s::%f2 = s::(^FOCUS)]])] specification precludes the addition of an indepen- [([VP: (^COMP)=!])]. dent subject NP because such an NP would fail to The interaction of the constraining equation above satisfy the coherency condition. Instead, the inde- and this rule guarantees that grammatical agreement pendent NP must be interpreted as a FOCUS: will occur between the subject and verb only in VS (50) kætabu l-awlæ:du l-wæ:gıba.ˇ word order. For example, the sentence in (57) has wrote.3MP the-boys.MP the-paper the f- and s-structures in (57a-b), in which the agree- “The BOYS wrote the paper.” ment marking on the verb introduces no information (51) f-structure: into the representation because it only constrains the "katabu il-awlaadu il-waajiba" gender feature of the subject: PRED 'write<[0-SUBJ:pro], [5:assignment]>' PRED 'pro' SUBJ AGR GEN masc, NUM plur, PERS 3rd (57) kætabat ıl-bænæ:tu l-wæ:gıba.ˇ CASE nom, PRON pers wrote.3F the-girls.FS the-paper PRED 'assignment' OBJ AGR ANIM -, GEN masc, NUM sing, PERS 3rd 5 CASE acc, DEF + “The girls wrote the paper.” PRED 'boy' 0 FOCUS AGR ANIM +, GEN masc, NUM plur, PERS 3rd (58) f-structure: 1 19 3 CASE nom, DEF + "katabat il-banaatu il-waajiba" 25 ASPECT Perfect PRED 'write<[3:girl], [5:assignment]>' PRED 'girl' (52) s-structure: SUBJ AGR ANIM +, GEN fem, NUM plur, PERS 3rd 3 CASE nom, DEF + PRED 'assignment' 0 ARG1 3 AGR GEN masc, NUM plur, PERS 3rd OBJ AGR ANIM -, GEN masc, NUM sing, PERS 3rd 2 OP 'the-unique-x', REL 'x-are-boys' 1 19 5 CASE acc, DEF + 0 AGR GEN masc, NUM sing, PERS 3rd 25 ASPECT Perfect 1 ARG2 4 20 5 OP 'the-unique-x', REL 'x-is-an-assignment' 25 REL 'ARG1-write-ARG2' (59) s-structure:

AGR GEN fem, NUM plur, PERS 3rd 2 ARG1 0 3rd-person-singular verbs have a special disjunctive 3 1 OP 'the-unique-x', REL 'x-are-girls' 18 AGR GEN masc, NUM sing, PERS 3rd lexical template for agreement features: 24 ARG2 4 25 5 OP 'the-unique-x', REL 'x-is-an-assignment' 26 REL 'ARG1-write-ARG2' (53) 3rdSingSubj( _G ) = {[@ProAgr @(3rdSing _G )] Agreement in SV word order is expressed with s- | @(GramAgr _G )}. structure constraints which identify the pre-verbal NP with a pronominal subject introduced by the For example, the perfect stem kætaba “(he) wrote” @ProAgr template. S-structure identiﬁcation entails would include the following: that AGR-features of the preverbal subject match (54) kataba V * @(TransPerf write) those of the pronominal subject. This enforces full @(3rdSingSubj masc). agreement in gender and number. (60) IP --> [NP: (^ TOPIC) = !] auxiliary and the subject of the complement. As be- I': ^=!. fore, s-structure identity implies identity of AGR- (61) I' --> [e: features, capturing full agreement between the sub- (^GF) = %f1 [s::%f1 = s::(^TOPIC)] ject and the verb. For example, the sentence in (68) (^TOPIC CASE) = Nom] has the f- and s-structures in (68a-b): [S: ^=!]. (68) kæ:na l-awlæ:du yaktUbu:na l-wæ:gıba.ˇ GF = {SUBJ|OBJ}. was.3MS the-boys.MP write.3MP the-paper “The boys were writing the paper.” According to these rules, NPs in the preverbal po- (69) f-structure: "kaana il-awlaadu yaktubuuna il-waajiba"

sition are TOPICS rather than subjects — in other PRED 'be<[4:write]>[3:boy]' PRED 'boy' words, SV or S-Aux-V word order should really SUBJ AGR ANIM +, GEN masc, NUM plur, PERS 3rd 3 CASE nom, DEF + be called Top-V or Top-Aux-V word-order. The PRED 'write<[4-SUBJ:pro], [7:assignment]>' PRED 'pro' TOPIC is identified in s-structure with the subject SUBJ AGR GEN masc, NUM plur, PERS 3rd CASE nom, PRON pers COMP of the clause. For example, the sentence in (62) has PRED 'assignment' OBJ AGR ANIM -, GEN masc, NUM sing, PERS 3rd 0 4 7 CASE acc, DEF + the f- and s-structure representations in (62a-b): 1 5 26 25 ASPECT Imperfect, CASE Acc 29 ASPECT Perfect (62) Pal-awlæ:du kætabu l-wæ:gıba.ˇ (70) s-structure: the-boys.MP wrote.3MP the-paper “The boys wrote the paper.” 4 AGR GEN masc, NUM plur, PERS 3rd 5 ARG1 2 25 3 OP 'the-unique-x', REL 'x-are-boys' (63) f-structure: 0 AGR GEN masc, NUM sing, PERS 3rd 1 ARG2 6 "il-awlaadu katabu il-waajiba" 26 7 OP 'the-unique-x', REL 'x-is-an-assignment' 27 REL 'ARG1-write-ARG2' PRED 'write<[2-SUBJ:pro], [5:assignment]>' PRED 'pro' SUBJ AGR GEN masc, NUM plur, PERS 3rd Agreement in S-Aux-V word involves two CASE nom, PRON pers PRED 'assignment' anaphoric control equations. The pre-verbal OBJ AGR ANIM -, GEN masc, NUM sing, PERS 3rd 2 5 CASE acc, DEF + 3 PRED 'boy' subject is interpreted as a TOPIC function, the 17 TOPIC AGR ANIM +, GEN masc, NUM plur, PERS 3rd 24 25 1 CASE nom, DEF + s-structure projection of which is identified with the 26 ASPECT Perfect s-structure projection of the subjects of the auxiliary (64) s-structure: and the main verb: (71) Pal-awlæ:du kæ:nu yaktUbu:na l-wæ:gıba.ˇ AGR GEN masc, NUM plur, PERS 3rd 2 ARG1 0 3 1 OP 'the-unique-x', REL 'x-are-boys' the-boys.MP were.3MP write.3MP the-paper 18 AGR GEN masc, NUM sing, PERS 3rd 24 ARG2 4 “The boys were writing the paper.” 25 5 OP 'the-unique-x', REL 'x-is-an-assignment' REL 'ARG1-write-ARG2' 26 (72) f-structure: In clauses with auxiliaries, agreement between the "il-awlaadu kaanu yaktubuuna il-waajiba" PRED 'be<[4:write]>[2-SUBJ:pro]' PRED 'pro' subject and main verb is represented as an anaphoric SUBJ AGR GEN masc, NUM plur, PERS 3rd PRON pers control relation. Auxiliary verbs select a comple- PRED 'boy' TOPIC AGR ANIM +, GEN masc, NUM plur, PERS 3rd ment, and then assert s-structure identity between 1 CASE nom, DEF + PRED 'write<[4-SUBJ:pro], [7:assignment]>' their subjects and the subjects of their complements: PRED 'pro' SUBJ AGR [2-SUBJ-AGR] CASE nom, PRON pers COMP kaana V @(RaisePerf be ) 2 PRED 'assignment' (65) * 3 OBJ AGR ANIM -, GEN masc, NUM sing, PERS 3rd 24 4 CASE acc, DEF + @(3rdSingSubj Masc ) 27 5 7 28 23 ASPECT Imperfect, CASE Acc @KHS. 29 ASPECT Perfect (66) RaisePerf(_PRED) = @(RaiseVerb _PRED) (73) s-structure: @Perf. (67) RaiseVerb(_PRED) = 4 AGR GEN masc, NUM plur, PERS 3rd 5 ARG1 0 (^PRED) = '_PRED<(^COMP)>(^SUBJ)' 23 1 OP 'the-unique-x', REL 'x-are-boys' 2 AGR GEN masc, NUM sing, PERS 3rd s::(^SUBJ) = s::(^COMP SUBJ) 3 ARG2 6 24 7 OP 'the-unique-x', REL 'x-is-an-assignment' s::^ = s::(^COMP). 25 REL 'ARG1-write-ARG2' 26 27 This is similar to the annotation used in the I’-rule, except that instead of asserting s-structure identity 2.2 Agreement with conjoined NPs between the topic and subject of one clause, it as- Turning to conjoined subjects, in VS word order serts s-structure identity between the subject of the with a conjoined subject NP the first conjunct is the “head” of the NP while the conjunction and second [(ÂGR NUM) ~= Sing]}] conjunct form a constituent which is adjoined to the | [(s::^ AGR NUM) = Dual first conjunct: ((ADJ ^) AGR NUM) =c Sing (ÂGR NUM) =c Sing]}. (74) NP --> [NP:^=! s::! = s::^] [ConjP: ! $ (ÂDJ) As an example, the conjoined NP in (79) has the s::! = s::^]. grammatical structure in structure in (80-82): (75) ConjP --> [Conj: ^=!] (79) Pal-bænæ:tu wa-l-awlæ:du [NP: !=^]. the-girls.FP and-the-boys.MP “the girls and the boys” The whole conjoined NP will inherit the f-structure (80) f-structure: features of its first conjunct, so that in VS word or- "il-banaatu wa- il-awlaadu" der, the verb agrees with the first conjunct. The s- PRED 'girl' PRED 'boy' 5 AGR ANIM +, GEN masc, NUM plur, PERS 3rd structure relation (REL) feature for the conjoined ADJ 2 3 CONJ PRED 'and' NP is represented as a predicate which takes the two 11 CASE nom, DEF + 1 AGR ANIM +, GEN fem, NUM plur, PERS 3rd conjuncts and returns a set containing them. REL 12 CASE nom, DEF + features for NP s-projections are determined by (76): (81) s-structure: 5 AGR GEN masc, NUM plur, PERS 3rd 2 3 0 OP 'the-unique-x', REL 'x-are-girls' (76) RelCalc = CONJ 11 4 OP 'the-unique-x', REL 'x-are-boys' {[s::! = s::^ 1 12 REL 'the-set-of-all-x-in-CONJ' ~(ADJ^) ~(ÂDJ)] (82) c-structure:

| CS 1: NP:12 [s::! $ (s::^ CONJ) (s::^ REL) = 'all-x-in-CONJ' NP:1 ConjP:11 (ADJ^)] il-banaatu:0 Conj:3 NP:5 | wa-:2 il-awlaadu:4 [s::! $ (s::^ CONJ) (^ADJ)]}. Reversing the order of the conjuncts changes the c- The s-structure gender and number features of the structure (86) and f-structure (84), but produces an conjoined NP are resolved according to pragmatic s-structure (85) identical to the one in (81): principles which err in favor of masculine gender (83) Pal-bænæ:tu wa-l-awlæ:du and plural number. A conjoined NP is masculine the-girls.FP and-the-boys.MP unless both conjuncts are feminine, and plural un- “the girls and the boys” less both conjuncts are singular. This is represented (84) f-structure: by the following lexical templates: "il-awlaadu wa- il-banaatu" PRED 'boy' PRED 'girl' 5 AGR ANIM +, GEN fem, NUM plur, PERS 3rd (77) GenCalc = ADJ 2 3 CONJ PRED 'and' {[(^AGR GEN) = (s::^ AGR GEN)] 11 CASE nom, DEF +

| 1 AGR ANIM +, GEN masc, NUM plur, PERS 3rd [(s::^ AGR GEN) = Masc 12 CASE nom, DEF + (ÂGR GEN) ~= ((ADJ ^) AGR GEN) (85) s-structure: (ADJ^)] 5 AGR GEN masc, NUM plur, PERS 3rd 2 3 0 OP 'the-unique-x', REL 'x-are-boys' | CONJ 11 4 OP 'the-unique-x', REL 'x-are-girls' 1 [(s::^ AGR GEN) = Masc 12 REL 'the-set-of-all-x-in-CONJ' (ÂDJ)]}. (86) c-structure: (78) NumCalc = {[(s::^ AGR NUM) = (ÂGR NUM) CS 1: NP:12

~(ADJ ^) NP:1 ConjP:11 ~(ÂDJ)] | il-awlaadu:0 Conj:3 NP:5 [(s::^ AGR NUM) ~= Sing wa-:2 il-banaatu:4 (ÂDJ)] | Agreement with conjoined NPs in VS word order [(s::^ AGR NUM) = Plur {[((ADJ ^) AGR NUM) ~= Sing] works according to the same constraints as described | above for singleton NPs. Because the first conjunct defines the f-structure AGR-feature of the whole ships, leaving unification to take place at s-structure. conjoined NP, grammatical agreement for conjoined To put it differently, s-structure is a grammatical rep- NPs in VS word order works in exactly the same resentation, rather than simply a place-holder for a fashion as it does for singleton NPs. In the interest semantic representation. Sentences which are equiv- of space, examples will not be provided. alent in terms of the number and AGR-values of In SV order agreement is as before except that s- their noun phrases but which vary in terms of word structure AGR features for the subject are calculated order produce identical s-structures, even though according to (77) and (78). This means that (87) they produce distinct f- and c-structures. For exam- and (88) will have an identical f-structure for their ple, the sentences (91-94) each have different agree- s-nodes (89) and will share the s-structure in (90): ment markings for the left-most verb stem and ac- (87) Pal-alwæ:du wa-l-bænæ:tu kæ:nu cordingly have distinct f- and c-structures. However, the-boys.MP and-the-girls.FP were.3MP they share the s-structure in (95): yaktUbu:na l-wæ:gıba.ˇ write.3MP the-paper (91) kætabat ıl-bænæ:tu l-wæ:gıba.ˇ wrote.3FS the-girls.FP the-paper “The boys and the girls were writing the paper.” “The girls wrote the paper.” (88) Pal-bænæ:tu wa-l-alwæ:du kæ:nu the-girls.FP and-the-boys.MP were.3MP (92) Pal-bænæ:tu kætabna l-wæ:gıba.ˇ yaktUbu:na l-wæ:gıba.ˇ the-girl.FP wrote.3FP the-paper write.3MP the-paper “Same.” “The girls and the boys were writing the paper.” (93) kæ:nat ıl-bænæ:tu yaktUbna l-wæ:gıba.ˇ (89) f-structure: was.3FS the-girl.FP write.3FP the-paper PRED 'be<[8:write]>[6-SUBJ:pro]' PRED 'pro' “The girls were writing the paper.” SUBJ AGR GEN masc, NUM plur, PERS 3rd PRON pers (94) Pal-bænæ:tu kUnna yaktUbna l-wæ:gıba.ˇ PRED 'write<[8-SUBJ:pro], [11:assignment]>' the-girls.FS were.3FP write.3FP the-paper PRED 'pro' SUBJ AGR [6-SUBJ-AGR] CASE nom, PRON pers “Same.” COMP PRED 'assignment' OBJ AGR ANIM -, GEN masc, NUM sing, PERS 3rd 8 (95) s-structure: 11 CASE acc, DEF + 6 9 AGR GEN fem, NUM plur, PERS 3rd 36 ASPECT Imperfect, CASE Acc 2 ARG1 0 7 1 OP 'the-unique-x', REL 'x-are-girls' 37 ASPECT Perfect 3 18 AGR GEN masc, NUM sing, PERS 3rd 24 ARG2 4 25 5 OP 'the-unique-x', REL 'x-is-an-assignment' (90) s-structure: 26 REL 'ARG1-write-ARG2' 5 AGR GEN masc, NUM plur, PERS 3rd 8 2 3 0 OP 'the-unique-x', REL 'x-are-girls' 9 ARG1 CONJ 36 23 4 OP 'the-unique-x', REL 'x-are-boys' 6 1 The conclusion is that f-structure and s-structure 7 24 REL 'the-set-of-all-x-in-CONJ' 37 AGR GEN masc, NUM sing, PERS 3rd 38 ARG2 10 are mutually constraining, just as f-structure and c- 39 11 OP 'the-unique-x', REL 'x-is-an-assignment' 40 REL 'ARG1-write-ARG2' structure are. This uniformity of s-structure representations might provide a means to extract propositional information References from parsed text in spite of the variability in word order and agreement marking. Stephen Wechsler & Larisa Zlatic. 2003. The Many Faces of Agreement. CSLI (Stanford). 3 Summary and Conclusion Abdelkader Fassi-Fehri. 1988. Agreement in Arabic, Binding and Coherence. in Agreement in Natural Lan- The grammar developed here captures the agree- guage: Approaches, Theories, Description. C. Fergu- ment facts described in Section 1, accepting the ac- son & M. Barlow. Prentice-Hall, Englewood Cliffs, ceptable sentences and excluding the unacceptable N.J. pp.107-158. ones. Furthermore, it does so while generating, at the most, four 4 ambiguous parses for test sentences, and in most cases generating only one parse. This re- sult is based on a test-suite containing all the logical permutations of subject-verb relationships, includ- ing plural and conjoined nouns. The analysis suggests that MSA uses f-structure more as a means to constrain c-structural relation-