• Keine Ergebnisse gefunden

Challenges for a Theory of Agreement

Theoretical Background

2.3 Implementing Agreement in Syntactic Theory

2.3.1 Challenges for a Theory of Agreement

2.3.1.1 Feature conflicts

So far, the description of agreement patterns concentrated on the clear-cut cases.

A first complication was touched when discussing agreement with nouns which exhibit a feature conflict between a grammatical feature and the corresponding notional feature. Collective nouns were discussed as an example for plural agree-ment with a grammatically singular controller driven by notional plurality. The re-verse effect—notional singularity overriding grammatical plurality—also occurs as I will demonstrate below. Certain noun classes and types of noun phrases which are grammatically plural allow for or even require singular agreement.

In English, diseases like measles, mumps, rickets, shingles mostly take singu-lar agreement despite their grammatical plurality but also accept plural agreement (Quirk et al., 1985).33 Yet, these cases of singular agreement with plural nouns which can be attributed to notional concord have to be distinguished from singular agreement with singular nouns like linguistics which shows consistently singular

33The corresponding German names are mainly singular nouns (Ziegenpeter or Mumps

‘mumps’, Rachitis ‘rickets’, Gürtelrose ‘shingles’) and require singular agreement. The few ex-amples which are plural nouns (Masern ‘measles’ and Röteln ‘German measles’) do not allow singular agreement. Besides they have to be used with a definite determiner.

agreement. Instances like this can be easily handled by simply assuming that lin-guistics is grammatically singular despite the apparent plural marking. English has a whole bunch of such nouns often denoting sciences. A noun like news can be treated the same way when additionally assuming a zero-plural marking in order to account for the fact that news can occur with singular and plural verbs.

Overall, we can imagine three strategies to cope with feature conflicts. In the easiest case, grammatical number is specified lexically with probable ten-sion to either the morphological form (e.g., linguistics, scissors) or the semantics (e.g., people). More problematic are cases with apparent optionality even in the core agreement domains determined by semantic factors (e.g., collective nouns).

Here, we have to assume either a syntactic mechanism of semantic agreement or some way that ensures the availability of an appropriate grammatical feature value for each case. The latter can be achieved by assuming two homomorphous lexi-cal entries differing only in the corresponding feature specification or some pre-syntactic mechanism providing the appropriate grammatical feature specification in accordance with the perceived or intended conceptual feature value. Head-Driven Phrase Structure Grammar tackles the challenge by assuming two sets of agreement features—concord features and index features (Pollard and Sag, 1994;

Kathol, 1999; Wechsler and Zlati´c, 2003). The corresponding features usually have identical values, but the values can well diverge, e.g., in the case of collec-tive nouns. In addition, three types of agreement are distinguished: concord, index agreement and semantic/pragmatic agreement. This gives reasonable flexibility to cope with the variation observed in the cases discussed above. Agreement with collective nouns and alike received less attention in syntactic theory in the Chom-skyan tradition, but see Den Dikken (2001). In contrast, the problem of conjoined NPs which will be discussed in turn gained substantial attention.

2.3.1.2 Conjoined Noun Phrases

For agreement with conjoined NPs we can distinguish two strategies: resolution and agreement with just one of the conjuncts, usually the closest. In languages allowing both options, the choice of one or the other depends on the syntac-tic configuration—Arabic, for instance, allows the closest conjunct strategy only when the verb precedes the subject NP (Aoun et al., 1994)—and semantic prop-erties, e.g., animacy (Levin, 2001) or the semantic relation of the conjuncts (Dal-rymple and Nikolaeva, 2006).

The strategy of resolution computes a feature value for the conjoined NP based on the feature values of the conjuncts or notional properties of the referent(s).

(48) Peter and Mary are married.

The resolution rule for the conjunction of two singular NPs gives usually plural or

dual in languages which have a dual. Singular agreement, however, is a possibility when the referent is a single entity.

(49) Romeo and Juliet is played on Monday.

(50) Trade and investment is better than aid.

The Romeo and Juliet example might be trivial since the conjoined NP is the title of the play and does not refer to two individuals. The emergence of singular agreement in (50) is less transparent, though trade and investment still can be considered a name (of a strategy) as witnessed by the absence of a determiner.

Overall, number resolution is fairly transparent and results in notional concord.

The situation is slightly more complex for person and gender. Accordingly, we find considerable variation (cf. Corbett, 2006). Since gender and person are not the main concern of the present dissertation I ignore these cases here. I only want to mention a specific resolution strategy, which is, however, only limitedly available—insertion of a resumptive pronoun. Two examples from German are given below.

‘You and Peter, you are a dream pair.’

(52) Ich,

The second strategy–single conjunct agreement34—makes the verb (or some other agreement target) agree with the conjunct that is closer to it. Though German exhibits single conjunct agreement only to a very limited extent (Johannessen, 1996), I will illustrate the strategy with the help of a German example.

(53) a. Aber

‘But to the left were the inner Alster and the white light ads.’

(example from Johannessen (1996))

34Another term is ‘partial agreement’. But this term is also used for agreement in only a subset of features, e.g., gender but not number with postverbal nouns in Arabic (Aoun et al., 1994).

Resolution of the conjoined NP die Binnenalster und die weißen Lichtreklamen (‘inner Alster and the white light ads’) results in plural agreement. Single con-junct agreement results in singular agreement when the singular concon-junct die Bin-nenalster is taken as controller or plural agreement when the plural conjunct die weißen Lichtreklamen. The data show that plural agreement is available regard-less of the order of the two conjuncts whereas singular agreement is only available when the singular conjunct is closer to the verb. Hence single conjunct agreement means agreement with the closest conjunct in German. The alternative—furthest conjunct agreement seems to be rare cross-linguistically (cf. Corbett, 2006, and references therein). Single conjunct agreement occurs in a variety of languages including such unrelated ones like various varieties of Arabic (Aoun et al., 1994, 1999; Aoun and Benmamoun, 1999; Soltan, 2006; Badecker, 2007), Irish (Mc-Closkey, 1986) and Serbian/Croatian (Wechsler and Zlati´c, 2003) (for more lan-guages see Badecker, 2007; Corbett, 2006 and Dalrymple and Nikolaeva, 2006).

For experimental data on English and Lebanese Arabic see Lorimor (2007); for German see Hemforth and Konieczny (2003).

Agreement with conjoined NPs and in particular single conjunct agreement has been extensively discussed in the literature. Various suggestions have been made as to how to derive the observed patterns syntactically, e.g., coordination of clauses rather than NPs and subsequent deletion (e.g., Aoun et al., 1994, 1999 but see also Munn, 1999), late insertion of the second conjunct (Soltan, 2006) or multiple Agree in the sense of Chomsky (2000) and deferring the decision to the morphological component (van Koppen, 2006). For an Optimality Theoretic account see Badecker (2007).

2.3.1.3 Defective and Absent Controllers

A descriptive agreement rule for subject–verb agreement in number following the format in (2.3) could be stated as in (2.3.1.3).

(54) (i) If the subject NP is specified for singular then the verb has to be spec-ified for singular, too.

(ii) If the subject NP is specified for plural then the verb has to be speci-fied for plural, too.

A rule like (2.3.1.3) runs into problems when the controller is not specified for any number value or even absent. For illustration consider the German examples below.

(55) Irren err.INF

ist is

menschlich.

human

‘To err is human’

(56) Dass

‘It disappoints me that Peter did not show up.’

(57) Gestern

‘It was much laughed yesterday.’

Note that in all examples above the verb occurs with singular marking.35 Hence, we can characterize singular as the default and re-state (2.3.1.3) as in (2.3.1.3).

(58) If the subject NP is specified for plural then the verb has to be specified for plural, too and for singular otherwise.

Finally, there are cases lacking a subject NP but nevertheless showing agreement as in the Spanish example below.

(59) a. He

Note that (59b) does not show some sort of default agreement and is thus dif-ferent from the subject lacking impersonal passive in (57). Sentences like (59b) are usually assumed to involve a subject pronoun (pro) which is only apparently absent because it happens to be phonologically zero. Sentences like (57) can be subsumed under the assumption of pro but differ from (59b) in that the subject is not referential (for discussion see Rizzi, 1986).

A final issue concerns the question what happens when the feature systems of controller and target differ in the range of possible values. Corbett (2000) discusses inter alia Modern Hebrew, which distinguishes three number values for nouns (singular, plural and dual) but only two for verbs (singular and plural). A dual subject NP occurs with a plural verb. Unless we want to assume that the verb has dual forms which are all homophonous with the corresponding plural forms, we need some mechanism that either translates dual into plural or ensures compatibility of dual and plural.

35As pointed out Koster (1978), the coordination of two clauses in subject function also results in singular agreement, just as in (56).