Early derivational grammars defined agreement as an asymmetric relation in which the controller is the element from which grammatical information originates and the target the element that inherits such information. This controller-target asymmetry is a key aspect of feature-copying models of agreement such as that developed within the recent Minimalist Program (Chomsky 1995, 2000, 2001). Central to this approach is the assumption that features are expressed as a bundle on a single position in the syntactic tree (Tense, or T), and are uniformly dealt with by Agree, the operation that is responsible for checking and copying feature values from the controller to the target (see Figure 1). Chomsky (1995:308 ff) asserts that feature-type matching is a pre-requisite for the performance of Agree, and thus a mismatch would impede the copying of feature values from controller to target and therefore the correct realization of the dependency. The controller-to-target directionality of the copying is determined by the asymmetry in feature values existing between the specifications on the controller and those on the target: while controllers enter the process already endowed with feature values (e.g. 1st, 2nd or 3rd for person; singular or plural for number), targets do not, hence the need for the copying operation (see Figure 1). This asymmetry has interpretive consequences: not only is the controller the source from which the copying process originates, it is also the element that carries visible or interpretable information to the conceptual system responsible for assigning an interpretation to the dependency in a subsequent computational step. Conversely, agreement information is not interpretable on the target, which inherits the controllers’ information to fill its “empty” person and number specifications.
Agree operates within narrowly syntactic boundaries, as unvalued features need to receive a value before the syntactic representation is passed on for subsequent semantic-pragmatic analysis. This means that the computation of agreement relations takes place during the syntactic build-up of the sentence, independently of the thematic and semantic-pragmatic information of the arguments involved.
This approach is derivational in the sense that the creation (or derivation) of a well-formed linguistic expression goes through specific steps in which operations like Agree are applied. Crucially, purely syntactic computational steps precede semantic-pragmatic analysis.
1.1.1. Why a Purely Syntactic Analysis Cannot Work: Unagreement
There are exceptions to the systematic covariance characterizing agreement realization. Across languages, patterns are found in which controller and target do not systematically co-vary, but the well-formedness and the acceptability of the sentence are, however, preserved. As an example, take the British English pattern in (2b), compared to (2a), probably one of the most frequently used examples of agreement mismatch.
|(2)||a.||The faculty3.sgis3.sg meeting tomorrow|
|b.||The faculty3.sgare3.pl meeting tomorrow|
The 3rd person singular subject faculty can be followed by a 3rd person plural verb, and any British English speaker would find this combination perfectly grammatical. Like other collective nouns, faculty is formally singular but its referent can be identified in a plurality of individuals: as such, it can trigger semantic agreement on the verb, as opposed to the purely syntactic agreement of (2a).
The Spanish agreement system presents an interesting mismatch that targets the realization of subject-verb agreement and on which we will rest our proposal. Beside standard patterns like the one in (3) below, a person mismatch between a plural subject and verb is allowed: example (4) below illustrates what is known as Unagreement (Hurtado 1985; Jaeggli 1986, among others). Here, the presence of a person mismatch between a 3rd person plural subject and a 1st person plural verb does not prevent the sentence from being grammatical, as opposed to the outright ungrammaticality of (5). In (4), grammaticality is ensured by superimposing the verbal person value onto the nominal one, thus shifting the interpretation of the subject from a 3rd person plural to a 1st person plural one.
|(3)||a.||Los turistas3.plvisitaron3.pl un castillo muy bonito|
|b.||The tourists visited a very nice castle|
|(4)||a.||Los turistas3.plvisitamos1.pl un castillo muy bonito|
|b.|| We tourists visited a very nice castle|
|(5)||a.||*El turista3.sgvisitaste2.sg un castillo muy bonito|
|b.||*The tourist (you)visited a very nice castle|
Throughout the article, we will use the term “mismatch” to refer to a grammatical mismatch like Unagreement, while “anomaly” or “violation” will refer to an ungrammatical mismatch. Constructions in which full agreement between subject and verb is present will be referred to as standard or canonical agreement.
In his extensive description of agreement mismatches across languages, Corbett (2006:172) defines agreement patterns like Unagreement as “informative mismatches” that provide information not available elsewhere in the sentence, in this case information concerning the participants in the speech act. More precisely, in (4) it is verbal inflection that makes it clear that the speaker is part of the group of tourists. The type of mismatch illustrated in (4) is thus clearly different from the one in (2b): the verb superimposes its person feature on the subject and covertly shifts the interpretation of the nominal. Unagreement is thus yet another instance of how an exclusively syntactic analysis of agreement is unable to account for the variability of agreement realization across languages.
Unagreement poses relevant challenges to a standard Minimalist analysis in mainly two directions. Firstly, it seems to question the assumption that agreement is a uni-directional process taking place from the controller to the target: the 1st person plural specification of the verb clearly cannot result from a copying operation proceeding from the subject. Secondly, the 1st person plural interpretation assigned to the dependency is not compatible with the standard assumption that person is interpretable on the subject argument: in Unagreement it is the verb that provides the relevant person information to interpret the dependency. To rescue this analysis, one may possibly argue that in Spanish, plural lexical forms have a double person specification in the lexicon: one for 3rd person and one for 1st person. In Unagreement patterns, the latter, but not the former specification would be selected, thus allowing the performance of regular copying processes. Straightforward as it may sound, this explanation faces, however, two problems. First, the derivation should contain two lexical entries for los turistas, one specified for 3rd and one for 1st person, but this would add redundancy to the computation, something that any design wishes to avoid. Second, two lexical entries should be presupposed also for lexical forms in positions other than the subject, for example in object position. This appears not to be possible in Spanish, as 3rd person NPs in object position are never given a 1st person plural interpretation. This fact confirms that the shift in interpretation is not driven by an inherent property of Spanish NPs, but by the verb’s inflectional morphology, suggesting that Agree can operate in a reverse fashion here (we will come back to reverse Agree in Section 2).
A second limitation of the standard Minimalist analysis of agreement is to be found in the way features are represented in syntax. A feature bundle reduces computational load during feature checking, as it can be accessed uniformly by the system, but at the same time it obscures two important facts. Namely:
The cross-linguistic variation of agreement realization
: across languages agreement can manifest itself differently, depending, for example, on the position of the subject, a phenomenon known as partial agreement. A case in point is Arabic: preverbal subjects trigger full agreement on the verb, but postverbal ones only in gender, thus suggesting a differential access to features by the mechanisms driving agreement (Shlonsky 1989
The inherent interpretive differences among person and number
, which speak against a single-cluster representation of features, with recent psycholinguistic research also not supporting this view (see Section 3
below). While number features express the mere numerosity of the subject argument (e.g. a single entity vs. a plurality), person features express the status of the subject with respect to the participants in the speech act. Thus, 1st person expresses identity with (or inclusion) of the speaker, 2nd person identity with (or inclusion) of the addressee, 3rd person exclusion of both speaker and addressee.
A partial solution to the second point raised here seems to be offered by Cartography (cf. Cinque and Rizzi 2008; Shlonsky 2010 for reviews), a research program recently developed within the Principle & Parameters (Chomsky 1981) framework. Cartography proposes a decomposition of the agreement field into a series of independent syntactic heads, according to the heuristic principle “one morphosyntactic property – one feature – one head” (Cinque and Rizzi 2008:50). This analysis attributes agreement features the key role of “atoms” of syntactic computations and can thus capture the underlying differences existing among them. Under this approach, the syntactic tree of Figure 1 is “unpacked” to obtain a much richer and articulated representation (see Figure 2), in which independent projections are responsible for person, number and gender agreement. In this case, the grammar accesses agreement features separately: person, number and gender agreement thereby results from the establishment of distinct Agree relations.
Figure 2. Cartographic decomposition of the agreement field (based on Shlonsky 1989), in which each agreement feature has an independent representation in syntax: person is indicated by the Person Phrase (PersonP), number by the Number Phrase (NumberP) and gender by the Gender Phrase (GenderP).
Download figure to PowerPoint