Encoding Subjects: Evidence from Retrieval Interference

Realtime sentence comprehension is an incremental process, involving the rapid establishment of a syntactic representation and dependencies over that representation. This requires previously encountered linguistic material to be retrieved from memory. Following much recent work, we assume a content-addressable retrieval mechanism: constituent encodings are retrieved based on their inherent features (e.g. [+PL]), rather than relations between encodings, such as c-command (McElree et al., 2003). Recent research has shown that the outcome of this retrieval is not always sensitive to the syntactic representation and, so, not grammatically appropriate. For instance, during subject-verb attachment, subjects outside of the current clause make retrieval more difficult ((1); Van Dyke and Lewis (2003)). There are also cases where only the grammatically appropriate dependency is formed, arguing for structure-sensitive search (Sturt, 2003; Wagers and Phillips, 2009). However, within content-addressable retrieval, the properties of encodings that serve to guide retrieval can yield grammatical outcomes, without reference to structural information. Thus, if there is content-addressable retrieval, we need a theory of retrieval cues. These issues can be addressed by probing the extent to which the presence of subjects within the nominal domain makes subject-verb dependencies more difficult. Subjects in the nominal and clausal domain are similar along some dimensions: they c-command other DPs within their domain, and they are the DPs to which the external argument is linked. They are dissimilar in that they occur in different domains and are assigned Case by distinct heads. In particular, the prediction is that if cue-based search targets features these elements share, the parser will be unable to distinguish between subjects in the nominal and clausal domain. By comparing subject-verb attachment in the presence of nominal interveners, we are testing the set of retrieval cues the parser uses in the face of a number of choices.