Grammatical Agreement Research Paper

(1) French                   a.         La^fsg fille^fsg est^3sg belle^fsg.

‘The girl is beautiful’

1. Les ^pl garcons ^mpl sont^3pl beaux^mpl.

‘The boys are handsome’

(2) French                    a.         Je^1sg les^3pl ai^1sg ues^fpl.

‘I have seen them’

1. Tu^2sg l’^3sg a^2sg vu^msg.

‘You have seen him’

(3) Finnish                   a.         Taina loysi minun ^1sg_GEN kirja-ni^{1sg Poss}.

‘Taina found my book’

1. Taina loysi heidan^3pl_GEN kirja-nsa^3Poss.

‘Taina found their book’

(4) German                  a.         Ein kleines^nsg Madchenⁿ saß ^3sg

auf der Treppe, und es^3nsg lachte.

‘A little girl sat on the stairs,

and she laughed’

1. Ein kleiner^msg Junge ^msg stand^3sg

daneben, und er ^3msg heulte.

‘A little boy stood beside, and he

cried’

In (1), the definite article agrees with the following noun, and both the copula verb and the predicative adjective agree with the whole subject phrase. If any item of (1a) is exchanged with the corresponding item of (1b), the sentence becomes ungrammatical (e.g., *la fille est beaux). Example (2) shows two agreement relations interwined: the auxiliary agrees with the subject pronoun, while the participle agrees with the object clitic. Sentence (3) is an example of a possessed noun agreeing with the possessor in the genitive. In (4) it is the pronoun in the second clause that agrees with the antecedent noun phrase in the first clause. Notice that Madchen ‘girl’ is a neuter noun, triggering a neuter pronoun even though it refers to a female being; in contrast, the English pronouns in the translation are selected solely on semantic grounds.

Grammatical agreement can be seen as an interplay of morphological, syntactic and semantic aspects (Anderson 1988, Lapointe 1988, Corbett 1994, 1995, 1998, Pollard and Sag 1994). Morphosyntactically, agreement is a relation between syntactic constituents in virtue of the fact that the word forms they consist of bear similar information, either inherently or by means of morphological affixation (inflection). Semantically, agreement serves to keep record of discourse referents: only constituents that relate to the same referent may overtly agree with each other (Lehmann 1982). The sentences in (1) are about one referent (a single person or a group of persons), while those in (2) are about two referents in a seeing-relation.

Generally, agreement relations are found among the following elements:

(a) NP(or DP-) internally: determiner and noun, attribute and noun, possessor and possessed noun;

(b) a predicate and its arguments: verb–subject, verb–object, preposition–object, predicative noun/adjective–argument; and

(c) a pronoun or anaphor and its antecedent.

The categories that may appear in these relations are gender (e.g., feminine f, masculine m, animate, or neuter n), number (e.g., dual, plural pl, and singular sg), and person (1st, 2nd, and 3rd, or inclusive 1+2).

Nouns and pronouns (or pronominal affixes/clitics) function as the controller (or source) of an agreement relation in virtue of their features. Nouns are classified inherently for gender (or noun class), and sometimes also for number (cf. scissors, trousers, brethren), whereas pronouns are classified inherently for person and number, and sometimes also for gender. In some instances, this inherent classification is semantically based: as a default, female beings are represented by feminine nouns, but male beings by masculine nouns; aggregates are represented by plural nouns; addressees are represented by second person pronouns, but speakers by first person pronouns. In most other cases feature assignment is purely grammatical (consider the distinct genders in French lune^f, soleil^m vs. German Mond^m, Sonne^f, ‘moon,’ ‘sun’) and may even override the default (cf. German Madchenⁿ ‘girl’) (Corbett 1991).

The grammatical elements that agree with the noun or the pronoun are said to be the agreement targets (or controllees); in (1), the article la is the target of fille, which is f, and both the copula est and the adjective belle are targets of the DP la fille, which is fsg. The agreement features of controller and target do not have to be identical, but they must be compatible with each other, as in la fille^fsg and est^3sg. In order for a DP to be the controller of an agreement relation, the features of the noun and the other DP-internal elements have to be unified. For short, φ-features index the referential elements of a discourse (pronouns and DPs), as well as the categories that predicate of these elements.

In the wide sense, one speaks about ‘agreement’ if, e.g., the verb bears information about arguments. Sometimes this information is clearly pronominal, sometimes it is less than that. In Italian, ho^1sg vista^fsg is possible alongside with io^1sg l’^3fsg ho^1sg vista ^fsg , so that one can conclude, the information is pronominal. In German, one cannot drop the pronouns; habe^1sg gesehen (‘have seen’) is clearly incomplete. Forms such as ich^1sg habe^1sg, du^2sg hast^2sg ‘I have, you have’, however, are redundant because the information of the subject pronoun is doubled by the agreement morphology. Redundancy is one of the criteria for canonical agreement offered by Corbett (2003) (where he discusses 19 criteria); in that sense, agreement in German is more canonical than agreement in Italian. (Corbett (2003) discusses 19 criteria for canonical agreement.)

Siewierska (1999) distinguishes three types: „(i) Anaphoric agreement markers are in complementary distribution with free nominal or pronominal arguments; (ii) ambiguous agreement markers occur obligatorily both in the presence and absence of free nominal or pronominal arguments; (iii) grammatical agreement markers are obligatory, like ambiguous markers, but, unlike ambiguous or anaphoric markers, must necessarily be accompanied by overt nominal or pronominal arguments.“ Accordingly, Italian shows ambiguous agreement, while German shows grammatical agreement.

If the verb bears information about just one argument, it is usually about the nominative argument (the subject in nominative-accusative patterns, or the object in ergative-absolutive (=nominative) patterns). That the Italian participle can bear information about the object, has ultimately to do with the fact that these participles also occur in passive constructions.

The notion of ‘agreement’ is sometimes used in a still broader sense, including other types of correspondences as well, for instance, case concord in (5), and sortal correspondence in (6).

(5)                               a.         Latin: DP-internal case concord

Illa^fsg-m_ACC femina^fsg-m_ACC bella^fsg-m_ACC videbam.

‘I saw that beautiful lady

1. Finnish: DP-internal case concord

tuo-n_GEN nariseva-n_GEN tuoli-n_GEN alla

that squeaky chair under

‘under that squeaky chair’

1. German: Case concord between the predicative noun and its argument

Man nannte ihn_ACC einen Idioten_ACC; er_NOM wurde ein Idiot_NOM genannt.

‘He was called an idiot’

(6) The horse neighs; the dog barks; the cock crows.

These relations, however, differ in many respects from agreement described above, and, therefore, are not included here. Case reflects the grammatical role of a constituent (subject or object), rather than its referential status, whereas sortal correspondence is based on much finer semantic distinctions than φ-features can offer.

Several theories have been proposed to account for agreement phenomena. In a first systematic attempt, the constituents introduced by a phrase structure rule, such as S → DP + VP, are indexed with the same set of φ-feature values:

(7)                               a.         S → DP^φi + VP^φi (subject+verb phrase)

1. DP^φj → D^φj + NP^φj (determiner+noun phrase)

This account does not allow the instantiation of different (although still unifying) feature values, as in the French example (1), in which one constituent is fsg and another one is 3sg, and it does not account for anapher-antecedent agreement because these elements do not originate from a common syntactic node. On the other hand, this account is not specific enough because it would describe case concord with the same means. Moroever, the claim that φ-feature values distribute along with phrase structure rules seems rather arbitrary.

Notice that (7) is symmetric between controller and target. In contrast, General Phrase Structure Grammar proposed an asymmetric analysis: controller elements can inherit a feature structure AGR lexically, whereas target elements (mostly the functors) receive such a structure only by morphological affixation; the distribution of AGR-features within a clause is then checked by a general control-agreement principle (Gazdar et al. 1985). Again, this theory is both too general and too restricted with respect to agreement.

In the Chomskyan framework of Principles and Parameter Theory (PPT), later replaced by Minimalist Syntax (Chomsky 1995), the bearer of agreement features are functional categories represented by independent syntactic projections (Laka 1988, Pollock 1989, Chomsky 1991). Consider the schematic structure in (8). This structure describes subject-verb agreement when the AGR-node is identified with AGRS, which is specified for certain φ-feature values, as well as for case features (such as NOM). In order to get marked for subject agreement morphologically, the verb stem must move to AGRS, while the subject DP moves into SPEC.AGRS, where it receives nominative case (both movements leave a trace, which is coindexed in (8)). Similarly, AGRO is responsible for object-verb agreement as well as accusative case, and AGRPOSS is responsible for possessor-noun agreement as well as genitive case. Regardless of various modifications in detail, the basic idea remained the same: there is one syntactic configuration that determines both case and agreement. Agreement, as well as case, are checked between AGR and SPEC.AGR (and both must be licensed by a particular argument role of the verb or noun).

(8)                               AGRP

                                     ∕ \

 SPEC.AGR AGR’

 Dpi                 ∕ \

 AGR VP

 ∕ \ ∕ \

 AGR V_j t_i … t_j

 φ+case

 features

This configuration deals with agreement as an additional structure superimposed on argument structure (such as subject-verb, object-verb, or possessor-noun). It is unclear what additional structure would account for agreement with attributes, adverbs, secondary predicates, or dependent objects. More recently, Chomsky (2000, 2004) proposed to dispense with additional AGR structures and syntactic Spec-Head relations. Agreement is now assumed as a relation between a probing head and a target goal which is in the probe’s c-command domain. Note that the terminology has been shifted: now the controller (say, a subject NP) is called target, and the controllee (the verb) is called probe. TENSE rather than arbitrary _AGRS is the category responsible for subject agreement, and little v (which c-commands VP) for object agreement. The probing head bears unvalued φ-features that must find their values within the respective domain.

For most syntacticians, agreement is a local phenomenon and part of narrow syntax. The crosssentential anapher-antecedent relation certainly is non-local and cannot be captured syntactically. There are more instances of non-local (‘long-distance’) agreement between matrix and embedded clauses where it is argued that they can be handled locally (see below). In view of these more complex phenomena, Bobaljik and Wurmbrand (2005) come to the conclusion that „agreement is with an (LF) phasemate“ (where a phase is something like a clause and LF is ‘logical form’), thus, for these authors agreement ist postsyntactic (see also Bobaljik 2008), a position that is not so far from lexically-based accounts.

Another problem for syntactic accounts comes from the assumption that agreement and caseassignment are the two sides of the same coin. If a verb agrees with just one argument, more than often it is the nominative argument: in a nominative-accusative language the verb agrees with the subject, and in an ergative-nominative (‘absolutive‘) language with the object. Hindi is compatible with that correlation even if it is more complex in showing two types of differential realizations. Verbs select ergative for the subject only in the perfective, and accusative for the object only if that is definite or human-specific. Thus, a transitive verb of Hindi can show _ERG-ACC or _ERG-NOM in the perfective, but _NOM-ACC or _NOM-NOM in the imperfective. The agreement rule is ‘Agree with the highest nominative’ (which is the subject in the imperfect but the object in the _ERG-NOM pattern); there is no agreement in the _ERG-ACC option. Thus, Hindi very clearly exemplifies a narrow correlation between case and agreement.

But other Indo-Aryan languages (with millions of speakers) behave totally different, even if they have similar case-patterns. The Nepali verb agrees with the ergative subject in an _ERG-NOM pattern (9a, Deo and Sharma 2006), while in the opposite, the Kutchi-Gujarati perfective verb agrees with the focused accusative object in a _NOM-ACC pattern (9b, Patel 2007), i.e., contrary to what is expected, the verb agrees here with the marked rather than the unmarked argument in terms of case. There obviously is no case-agreement correlation in general. Wunderlich (2012) proposes an analysis where case and agreement, based on the same argument ranking, are determined by different constraint-rankings (such as *_ERG ≫ _ERG! or _ERG! ≫*_ERG; _AGR(S) ≫ _AGR(O) or *_AGR(S)/context ≫ _AGR(S) ≫ _AGR(O), etc.). Nepali combines _ERG! ≫*_ERG with _AGR(S) ≫ _AGR(O), while Kutchi-Gujarati combines *_ERG ≫ _ERG! with *_AGR(S)/+perf ≫ _AGR(S) ≫ _AGR(O). (_ERG! = realize ergative; *_ERG = do not realize ergative, etc.)

(9)        Agreement with case-marked arguments

1. Mai-le mero                luga:                dho-en.            Nepali

I-ERG              my                   clothes._NOM      wash-_PERF.1sg

‘I washed my clothes’

1. Reena chokra-ne         mar-ya.                                   Kutchi-Gujarati

Reena.f._NOM      boys-_ACC                     hit-_PERF.mpl

‘Reena hit the boys’

The relation between controller and controllee is asymmetric, as well as the (reversed) relation between probing head and goal. Yet, the agreement relation itself is symmetric: agreement holds if the pieces of information on the controller and on the head are compatible. Thus, a unification-based approach might be promising.

Following the idea that φ-features keep record in what the individual referents are predicated of, agreement is seen as a mechanism that rests on the normal composition procedures; there is no need for a special syntax of agreement. In this spirit, agreement was analyzed as a syntax-semantics interface-phenomenon in Head-driven Phrase Structure Grammar (HPSG; Pollard and Sag 1994) as well as in Lexical Decomposition Grammar (LDG; Wunderlich 1994, 1997). In HPSG, all grammatical information of lexical items or more complex forms is encoded by means of feature structure; the φ-features that are shared in an agreement relation (regardless of whether they are lexically inherent or morphologically introduced) form a substructure INDEX (as part of CONTENT), and when a clause is composed, the relevant INDEX structures are unified.

In LDG, all morphological and phrasal composition is determined by the Theta Structure of lexical items, constituted as a list of lambda abstractors related to the Semantic Form (a partial semantic representation). In this theory, φ-features are associated with (indexed to) individual variables under a binder, such as the λ-abstractor or the Russellian iota operator ‘ι’; in the process of composition, these indexes are unified at the common binder of the variables in question. This is illustrated in (9), representing the composition of Latin illa femina bella ‘that beautiful woman’ from three individual entries; unification is represented by the sign ‘∪’.

Some of the elements may be underspecified for f or sg, but any other specification would lead to ungrammaticality (such as *illa femina bellus^msg). Notice that the indexed information becomes relevant only when semantic composition takes place; therefore this account indeed locates agreement at the syntax-semantics interface. Index unification itself is symmetric; all asymmetries result from the syntactic categories. ∪

This approach can easily be extended to all other kinds of agreement, taking into account the specific mode of composition. If one forms the assertion illa femina bella modesta est ‘that beautiful woman is modest’, one predicates ‘be modest’ (λu^fsg _MODEST(u)) of the result of (10), and if one continues with the pronoun ea^fsg ‘she’, illa femina bella is a possible antecedent (referring to the same person) because the φ-features are compatible.

In the Italian example (11), the subject agrees locally with the auxiliary, and non-locally with the participle embedded in a dependent infinitive clause. The agreement facts thus indicate that ‘have fear’ and ‘be arrived too late’ predicate of the same individual referent. If one assumes that ha paura is a subject control verb, the identity of arguments follows from the lexical entry of control verbs; by functional composition, the argument variable of ‘be arrived too late’ then fuses with the argument variable of ha paura, and the φ-features must unify, as shown in (11c).

Another language well-known for long-distance agreement is Hindi (Mahajan 1989, Wunderlich 1994, Boeckx 2004, Bhatt 2005, Chandra 2007). An imperfective control verb such as caah-t-‘want’ agrees (together with a possible auxiliary) with the nominative subject (12a), while the perfective variant agrees with the embedded nominative object (12b) because ergative on the subject blocks agreement with the subject. There are two further possibilities, namely that no agreement takes place in the perfect and the verbs get the default ending msg (12c), or that the embedded verb agrees with the object in the imperfect (12d, not available in all dialects). The respective acceptability rates are 95%, 88%, 78%, and 67% according to a study by Bornkessel-Schlesewsky 2008).

(12)     LDA with Hindi control verbs

1. Raam^msg roTii^fsg                           khaanaa^msg       caahtaa^msg        thaa^msg.

Ram                 bread               eat._INF               want._IMPF          be._PAST

‘Ram habitually wanted to eat bread’

1. Raam ne roTii^fsg             khaanii^fsg          caahii^f              thii^fsg.               LDA

Ram _ERG           bread               eat._INF               want._PERF          be._PAST

‘Ram had wanted to eat bread’

1. Niinaa^fsg ne roTii^fsg                           khaanaa^msg       caahaa^msg         thaa^msg.

Ram     _ERG       bread               eat._INF               want._PERF                   be._PAST

‘Ram had wanted to eat bread’

1. Raam^msg roTii^fsg                           khaanii^fsg          caahtaa^msg                thaa^msg.

Ram                 bread               eat._INF               want._IMPF          be._PAST

‘Ram habitually wanted to eat bread’

Under what condition the option b or c is selected is controversially discussed. According to Chandra (2007), LDA is possible only with specific objects, while Bhatt (2005) states that the embedded object can get scope over the matrix verb only when LDA takes place. Thus, ‘Naim wanted to read every book’ with no agreement only allows the reading want > ∀, while the LDA version b allows both readings want > ∀ and ∀ > want (with the latter being preferred). Bobaljik and Wurmbrand (2003) observed a similar pattern in Itelmen, see (16c) below.

This suggests that LDA is possible by reconstruction, where a verb complex is formed and the argument structure of the lowest verb is inherited to the result. The analysis of (12b) is sketched in (13). The suffix -ii in khaan-ii instantiates the feature complex fsg, which may be associated with either argument (the association with the subject ‘u’ being the default), whereas the perfect form caah-ii instantiates f, which cannot be associated with the argument role assigned ergative. In this context, only the second alternative of khaanii can survive because of unification.

Other types of verbs in Hindi that make LDA possible by forming verb complexes are raising-tosubject verbs like ‘seem’ in (14a), raising-to-object verbs like ‘see’ in (14b), light verbs like ‘let’ in (14c), and modals with a lexically marked subject like ‘should’ in (14d). Here, again, ergative/accusative blocks subject agreement, and a verb complex is formed.

(14)     More LDA verbs in Hindi

1. Raam ne roTii^fsg              khaayii^f                       lagtii^f               thii^fsg.

Ram     _ERG       bread               eat._PERF                         seem._IMPF          be._PAST

‘Ram seemed to have eaten bread’

1. Raam ne billii^fsg              aatii^f                 dekhii^f             thii^fsg.

Ram     _ERG       cat                    come._IMPF                  see._PERF            be._PAST

‘Ram had seen a cat coming’

1. Nadia-ne sarosh-ko         gaaRii^fsg           chalaa-ne         d-ii^fsg

Nadia-_ERG                 Sarosh-_ACC        car-_F.SG             drive-_GER.OBL     let-_PERF.F

‘Nadia let Sarosh drive car.’

1. Raam-ko davaaii^fsg          khaa                lenii^f                chaahiye          thii^fsg.

Ram-_ACC           medicine          eat                   take._INF             should              be._PAST

Ram should have taken medicine

In principle, LDA is recursive: the following sentences, although not usual, are grammatical (Shravan Vasishth p.c.).

(15)                             a.         Raam ne          larki_fsg              aatii^f                 dekhnii^fsg                  chaahii^f

Ram-_ERG                     girl c                ome._IMPF           see._INF                            want._PERF

‘Ram wanted to see a girl coming.’

1. Raam-ne har kitaab^fsg      parhnii^fsg          chaahii^f            lagtii^f                hai

Ram-_ERG           every book       read._INF            want._PERF          seem._IMPF                   be._PRES

‘Ram seemed to have wanted to read every book.’

LDA is found in many more languages. In Basque, the auxiliary (which encodes up to three arguments) agrees with the number of the embedded object (16a, Etxeparre 2006). In Itelmen, a Chukotko-Kamchatkan language, the matrix verb agrees with person. number of the direct or indirect embedded object (16b,c, Bobaljik and Wurmbrand 2003, 2005); (16c) shows the same scopal reading as found in Hindi. Note that the matrix verbs here are control verbs, i.e. restructuring verbs forming a verb complex.

(16)                             LDA with restructuring verbs in Basque (a) and Itelmen (b,c)

1. [liburu-ak itzultze-ko]      eskatu              dizkidate

book-pl._NOM     give-back         asked               have.3pl_ERG.1s_DAT.3pl_NOM

‘They asked me to give back the books’

1. əntxa-βum=nIn kəma-nk nənJ č i            zəl-es.

forget-1sg=3.CL me-DAT        fish                  INTERJ give-INF

‘He forgot to give me (a) fish’

1. t’-əntxa-ce?n [ miλ okno?-n   sop-es ]

1sg-forget-3pl.OBJ [ all window-pl close-INF ]

‘I forgot to close all the windows’ (∀ > forget)

Another group of LDA-languages includes the NE Caucasian Daghestan languages. In Tsez, the matrix verb agrees with the gender class of the embedded object (17a); Polinsky and Potsdam (2001) state the following conditions: “(i) The embedded clause must be in the absolutive position, (ii) the trigger must be the absolutive argument of the embedded clause, (iii) the absolutive NP must be the topic of the embedded clause.” In Khwarshi, the matrix verb likewise agrees with the gender class of the embedded object, but under slightly different conditions: the embedded object must be either topic (17b) or focus (17c), and it must be moved into initial position (Khalilova 2007). Note that ‘know’ is not a restructuring verb because the embedded clause is finite. Just to illustrate what LF-‘phasemate’-ship (Bobaljik and Wurmbrand 2003) here means, (17a) might be paraphrased as ‘the mother knows of the bread (=topic) that the boy ate’.

(17) LDA with non-restructuring verbs in Tsez (a) and Khwarshi (b,c)

1. eni-r [už –aa             magalu             b-aac’ -ru-λi]    b-iy-xo

mother-dat       boy-erg            bread.III.NOM III-eat-PASTPARTIC-NOML                                                                                                                         III-know-PRES

‘The mother knows that the boy ate the bread.’

1. Zihe-n uža-l                b-iq’-še            b-iti-xx-u.

cow(III)-TOP   boy.OBL-LAT III-know-PRES III-divide-CAUS-PASTPARTIC

‘The boy knows that the cow was stolen’

1. K?aba            zihe                  b-ot’uq’q’-u     b-iq’-še            uža-l.

black               cow(III)           III-come-PASTPARTIC          III-know-PRES boy.OBL-LAT

‘The boy knows that the black cow has come’

Despite theoretical differences, all accounts of LDA are agreed that apparent non-local agreement is in fact local. Either the matrix verb and the embedded verb form a verb complex in which subject and embedded object are clausemates (clause union in Hindi, Basque, Itelmen etc.), or the embedded object is moved into a peripheral position (topic or focus) in which it is accessible to the next higher domain (in the spirit of Chomsky’s 2000 edge condition) – which is the case in the Daghestan languages (17).

Since LDA in the sense of (16) always is lexically triggered by the matrix verb (note that Kutchi Gujarati has just one lexical item of this sort, namely par ‘have to’, according to Patel 2007), lexically-based accounts like HPSG and LDG seem especially suitable for the analysis of these constructions: there is something in these lexical items that makes them able to ‘see into’ a dependent clause. Furthermore, discourse categories like topic and focus (to be dealt with in discourse-semantics) seem to play an important role for connecting matrix and dependent clause. In contrast, syntactic accounts assuming that agreement is mediated by extra-syntactic AGR heads (like Mahajan 1989) come in trouble to analyze LDA; therefore, the syntactic minimalism in Chomsky (2000) was a major step for approximating syntactic theory to lexical frameworks.

Most languages have the possibility of ‘no agreement’ in using a default verb form that otherwise could agree, such as neuter, 3rd person or masculine singular. In these cases, the feature complex introduced by the verb form is not associated with an argument variable. Since in German only nominative forms can agree with the verb, agreement is possible in (18a), but is blocked in (18b), so that only the default form friert^∅, lacking an index value, can appear.

(18)                             a.         Ich^1sg_NOM          friere^1sg                                                                       ‘I am cold’

1. Mich^1sg_ACC *friere^1sg / *friert^3sg/ friert^∅ ‘I am cold’

Not all languages use agreement. Clearly, a language such as Chinese with nearly no morphology cannot have agreement, but also Malayalam (a Dravidian language spoken in Kerala, SW India), which shows rather rich derivational morphology and also morphological case, nevertheless does not have any agreement. Moreover, the extent to which agreement is used, differs from language to language. Not only may the number of distinctions in the gender, number, and person categories, as well as the richness of inflectional affix inventories vary, it also depends on language-specific grounds in which contexts agreement is obligatory, optional, or even forbidden.

Hungarian marks plural on nouns, adjectives, and verbs. However, adjectives only agree with a noun when they are used predicatively (19a,b). Surprisingly, in the context of a numeral, which inherently contains the concept of plurality, the noun must not be marked for plural (19c). Furthermore, a predicative adjective or verb only agrees with a subject phrase which is overtly marked for plural, see (19d,e).

(19)                             Hungarian plural agreement

1. gyors hajók^pl/ *gyorsak_pl        hajók_pl

‘fast ships’

1. Ezek^pl a hajók^pl gyorsak^pl/ *gyors^sg.

‘These ships are fast’

1. öt hajó^sg/ *öt hajók^pl

‘five ships’

1. Az öt hajó^sg gyors^sg/ *gyorsak^pl.

‘The five ships are fast’

1. Az öt nagynéni^sg sört iszik^3sg/ *isznak^3pl.

DET five aunt beer drinks

‘The five aunts drink beer’

Since Hungarian has the plural forms gyorsak ‘fast’ and hajók ‘ships’, these forms must somehow be blocked in (19a,c). Ortmann (1999) proposed that the agreement facts in (19) follow from the interaction of economy and expressivity principles. More specifically, if one assumes the economy constraint ‘Avoid multiple expression of plurality in NPs’ to rank above the expressivity requirement ‘Mark plural on the noun!’, both gyors hajó and gyorsak hajók would violate one of these constraints, and gyors hajók turns out to be the optimal form in a context in which more than one ship is meant. In contrast, the singular form öt hajó is preferred because the numeral öt ‘five’ already expresses plurality; in this case, it is better to violate the lower-ranked constraint rather than the higher-ranked one. The tableau illustrates that the winner exhibits the fewest violations of higher-ranked constraints.

If plural is unmarked on the noun, the noun phrase is sg and thus incompatible with a plural verb form, as shown in (19e).

As we have seen, Hungarian (unlike English and German) avoids NP-internal agreement. There are also languages that avoid subject-verb agreement, even if they have the suitable affixes. Kurdish, for instance, marks plural only on the verb, see (21). In contrast, Georgian marks plural on the verb only if the subject refers to animate beings, see (20).

(20)                             Kurdish

Mirov /*mirovan_pl                                       hat-in_3pl.

Man                                        come.past-3pl

‘The men came’

(21) Georgian

1. Burt-eb^pl-i goravs³ / *        goraven^3pl.anim.

ball-pl-NOM    roll

‘The balls are rolling’

3. Knut-eb^pl-i goraven^3pl.anim /*goravs³.

kitten-pl-NOM roll

‘The kittens are rolling’

These few examples show that the implementation of agreement considerably varies from language to language.

Bibliography:

1. Anderson S R 1998 Inflection. In: Hammond M, Noonen M (eds.) Theoretical Morphology. Approaches in Modern Linguistics. Academic Press, San Diego, CA, pp. 23‒43
2. Baker, M C 2008 The Syntax of Agreement and Concord. Cambridge University Press
3. Bhatt R 2005 Long-distance agreement in Hindi-Urdu. Natural Language and Linguistic Theory 23: 757-807
4. Bobaljik J D, Wurmbrand S 2003 Long distance object agreement, restructuring and antireconstruction. In: Kadowaki M, Kawahara S (eds.) Proceedings of NELS 33. GLSA, Amherst, pp. 67-86.
5. Bobaljik J D, Wurmbrand S 2005 The domain of agreement. Natural Language and Linguistic Theory 23: 809-865
6. Bobaljik J D 2008 Where’s Phi? Agreement as a post-syntactic operation. In: Harbour D, Adger D, Béjar S (eds) Phi-Theory: Phi features across interfaces and modules. Oxford University Press, Oxford, pp. 295-328
7. Boeckx C 2004 Long-distance agreement in Hindi: Some theoretical implications. Studia Linguistica 58: 23-36.
8. Boeckx C 2006 Agreement Systems. John Benjamins, Amsterdam
9. Bornkessel-Schlesewsky I et al. 2008. Bridging the gap between processing preferences and typological distributions: Initial evidence from the online comprehension of control constructions in
10. In: Richards M, Malchukov A L (eds.) Scales (=Linguistische ArbeitsBerichte 86), pp. 397-436. Institut für Linguistik, University of Leipzig
11. Chandra P 2007 (Dis)Agree: Agreement and movement reconsidered. Ph.D dissertation, Univ. of Maryland, College Park
12. Chomsky N 1991 Some notes on economy of derivation and representation. In Freidin R (ed.) Principles and Parameters in Comparative Grammar. Cambridge, MA: MIT Press.
13. Chomsky N 1995 The Minimalist Program. MIT Press, Cambridge, MA
14. Chomsky N 2000 Minimalist inquiries: The framework. In: Martin R, Michaels D, Uriagereka J (eds) Step by Step: Essays on Minimalist Syntax in honor of Howard Lasnik. MIT Press, Cambridge, MA, pp. 89–155.
15. Chomsky N 2004 Beyond explanatory adequacy. In: Belletti A (ed.) Structures and Beyond. The Cartography of Syntactic Structures. Oxford University Press, Oxford, vol. 3, pp. 104-131 11
16. Corbett G G 1991 Gender. Cambridge University Press, Cambridge, UK
17. Corbett G G 1994 Agreement. In R. E. Asher (Ed.), Encyclopedia of Language and Linguistics Pergamon Press, Oxford, UK, Vol. 1 pp. 54‒60
18. Corbett G G 1995 Agreement. In: Jacobs J, von Stechow A, Sternefeld W, Vennemann T (eds.) Syntax: An International Handbook of Contemporary Research. De Gruyter, Berlin, Vol. 2, pp. 1235‒1244
19. Corbett G G 1998 Morphology and agreement. In: Spencer A, Zwicky A M (Eds.) The Handbook of Morphology. Blackwell, Oxford, UK, pp. 191‒205
20. Corbett G G 2003 Agreement: the range of the phenomenon and the principles of the Surrey Database of Agreement. Transactions of the Philological Society 101: 155–202
21. Corbett G G (ed.) 2003 Agreement: A Typological Perspective. Special issue of Transactions of the Philological Society 101: 149-370
22. Corbett G G 2006 Agreement. Cambridge University Press, Cambridge
23. D’Alessandro R, Fischer S, Hrafnbjargarson G H (eds) 2008 Agreement Restrictions. De Gruyter, Berlin
24. Deo A, Sharma D 2006 Typological variation in the ergative morphology of Indo-Aryan languages. Linguistic Typology 10: 369-418.
25. Etxepare, R 2006 Number long-distance agreement in (substandard) Basque. ASJU=International Journal of Basque Linguistics and Philology 40: 303-350
26. Gazdar G, Klein E, Pullum G, Sag I 1985 Generalized Phase Structure Grammar. Blackwell, Oxford, UK
27. Khalilova, Z 2007 Long-distance agreement in Khwarshi. In: Kokkonides, M (ed.) Proceedings of LingO. Faculty of Linguistics, University of Oxford, pp. 116-123
28. Laka I 1988 Configurational heads in inflectional morphology: The structure of the inflected forms in Basque. International Journal of Basque Linguistics and Philology 22: 343‒365
29. Lapointe S G 1988 Toward a unified theory of agreement. In: Barlow M, Ferguson C A (eds.) Agreement in Natural Language: Approaches, Theories and Descriptions. CSLI, Stanford, CA, pp. 67‒87
30. Lehmann C 1982 Universal and typological aspects of agreement. In Seiler H, Stachowiak F J (eds.) Apprehension: Das sprachliche Erfassen von Gegenständen. Narr, Tübingen, Germany, vol. 2 pp. 201‒267
31. Mahajan, Anoop K. 1989. Agreement and Agreement Phrases. MIT Working Papers in Linguistics 10, 217-252. Cambridge, Mass.
32. Ortmann A 2000 Where plural refuses to agree: Feature unification and morphological economy. Acta Linguistica Hungarica, 47: 249‒288
33. Patel, Pritty. 2007. Split ergativity and the nature of agreement. University College London.
34. Polinsky M, Potsdam E 2001 Long-distance agreement and topic in Tsez. Natural Language and Linguistic Theory 19: 583-646
35. Pollard C, Sag, I A 1994 Head-driven Phrase Structure Grammar. The University of Chicago Press, Chicago
36. Pollock J-Y 1989 Verb movement, Universal Grammar, and the structure of IP. Linguistic Inquiry 20: 365‒424.
37. Siewierska, A 1999 From anaphoric pronoun to grammatical agreement marker: why objects don’t make it. In: Corbett G G (ed.) Agreement. Special issue of Folia Linguistica 33(2):225-51
38. Wunderlich D 1994 Towards a lexicon-based theory of agreement. Theoretical Linguistics 20: 1‒35
39. Wunderlich D 1997 Cause and the structure of verbs. Linguistic Inquiry 28: 27‒68
40. Wunderlich D 2012 Case and agreement variation in Indo-Aryan. Ms. Leipzig, Dec. 5, 2012 (http://www.zas.gwz-berlin.de/wunderlich.html)