Models of incremental interpretation

Edward Stabler, UCLA

Fluent speech is typically understood "incrementally", word-by-word, with syntactic and other ambiguities resolved at the earliest possible point (as evidenced by recent studies from Tanenhaus et al., for example). This is often taken as favoring "constraint-based" over "modular" processing theories. But there really is no "constraint-based" alternative to "modular" syntax that can account for performance data. And second, it is a mistake to think that traditional modular approaches have trouble allowing the incremental, word-by-word influence of discourse and other factors in language recognition. The latter point is demonstrated here by exhibiting a traditional, modular model in which discourse and non-linguistic factors are used incrementally.

After briefly sketching a range of well-understood parsing models for minimalist syntax, this paper will show how one variety of these allows incremental recognition, allowing assessment of the current parse at a very fine grain, so that non-syntactic (e.g. discourse, background) factors can select among the (modular, syntactically defined) options "immediately." The proposed models have a very simple structure and exhibit performance apparently compatible with recent psycholinguistic studies, but they require an ability to interpret partial structures, and a departure from some prominent 'minimalist' proposals about semantics (from Heim&Kratzer, von Stechow, and others).