• Keine Ergebnisse gefunden

The Feature Selection and Feature Copying Model

Agreement in Sentence Production

3.4 Modeling the Computation of Agreement in Sentence Productionin Sentence Production

3.4.5 The Feature Selection and Feature Copying Model

TheFEATURE SELECTION AND FEATURE COPYING MODEL proposed by Franck et al. (2008) combines interactive and modular strategies and brings together cur-rent psycholinguistic models of language production and recent models in theoret-ical syntax (e.g., Chomsky, 1995). In particular, the model adopts the theorettheoret-ical distinction between interpretable and non-interpretable features. Crucially, one and the same feature type (e.g., number and gender) can be interpretable on one category (e.g., on the noun), but uninterpretable on another category (e.g., on the verb).

The computation of agreement is assumed to involve two steps, which how-ever differ from the marking and morphing distinction proposed by Bock and her colleagues. First, nominal features are selected from the lexicon based on mes-sage features. This selection process is called FEATURE SELECTION. Next, the selected features are transmitted to the agreement target. This transfer process is called FEATURE COPYING.

Feature selection is the interactive component in the model and takes place in the functional lexicon. Extending the DP-hypothesis (Abney, 1987), number is assumed to be head within the functional layer of nominal expressions (Ritter, 1991, 1993a).21 The number head hosts number affixes which are attached to the corresponding noun during the course of derivation. Number features can be se-lected from the functional lexicon in two different ways: directly from the lexicon as part of the lemma specifications or indirectly on the basis of the message level numerosity. This process of feature selection is guided by activation and feed-back. Conceptual and morphophonological factors influence feature selection and thereby agreement. This view is in accordance with research on lexical selection.

If agreement features are stored in the lexicon just like common lexical entries, feature selection should behave just like lemma selection which has been shown

21For similar assumptions regarding the representation of gender see Ritter, 1993b; Picallo, 1991.

to be sensitive to phonological influences (Dell, 1986; Rapp and Goldrick, 2000).

Feature selection is subject to some kind of competition between features. Con-ceptual and morphophonological information contributes to the activation level of the competing entries. The stronger a syntactic feature is associated to a mor-phophonological marker the more likely it is to be selected because it gets higher activation feedback from the morphophonological level. A feature that is only weakly associated to such a marker is less likely to be selected.

Feature copying which takes place within the syntax represents the modu-lar component of the model. The model is oriented at recent developments in generative syntax and in particular the minimalist program (Chomsky, 1995 and subsequent work). Franck and colleagues maintain the early minimalist concept of spec–head agreement and combine it with the more recent Agree mechanism (Chomsky, 2000). In their view, agreement involves the operation Agree and/or spec–head agreement. Agree establishes a connection between the subject NP in its base position and AgrS a position dedicated to subject agreement. The relation between the subject NP and AgrS becomes local when the subject moves to the specifier position of AgrS. In this situation, agreement is established in a spec–

head relation. The need to get valued forces the verb to move to AgrS. Feature copying conceives this valuation process as a feature transfer from the subject NP to the verb mediated via the functional head AgrS. Franck et al. (2006) have shown that the computation of verb agreement is more at risk of disruption when ensured by Agree only. The feature selection and feature copying assumes that the computation of agreement is controlled by syntactic factors only. In contrast to feature selection, feature copying is encapsulated from notional and phonological information. It is a purely syntactic process determined by syntactic constraints such as hierarchical relations and in particular c-command.

In this model, attraction errors arise when the number feature of the distractor is erroneously copied to the verb. Structural intervention is assumed to be a pre-requisite for this sort of interference. This assumption is in accordance with the findings discussed in section 3.3.2 which show that attraction is sensitive to syn-tactic factors. A more fine grained analysis of the type of structural intervention is given in Franck et al. (2006). This study has shown that attraction is stronger when the distractor intervenes both linearly in terms of precedence and hierarchically in terms of c-command. Furthermore, attraction was more likely in configurations where agreement was ensured by Agree only without further support from spec–

head agreement.

In summary, the feature selection and feature copying account integrates in-teractivity and modularity. Inin-teractivity is posited for lexical selection including feature selection while modularity is claimed for syntactic processing including feature copying. As a result, feature selection from lexicon is sensitive to various sources of information whereas feature copying is solely determined by syntactic

factors. The model focuses on syntactic factors for agreement, and, respectively attraction and seems to underestimate semantic and morphophonological factors.

Part of the observed semantic and morphophonological effects can be attributed to the interaction assumed during feature selection. In particular, the model fails to explain the finding that ambiguity of case marking promotes attraction. Fur-thermore, it does not offer an explanation for the usually observed singular–plural asymmetry. Though this might be explained in terms of markedness by stating that marked elements or feature values are more likely to interfere.

who Clark think are in the garden.

(Kimball and Aissen, 1971)

4

Agreement in Sentence