Bram Vandekerckhove is a researcher at the university of Antwerp
Exemplar-based generalization at the interface between syntax and semantics
Language processing theories often assume that linguistic abstractions are stored in long-term memory as fixed and stable linguistic categories, such as phonological and semantic features, parts of speech, thematic roles, etc., and that it is by classifying language input into these stable categories that language users can generalize linguistic patterns. An alternative to category-based theories is exemplar-based language processing. In this approach, generalization of linguistic patterns is achieved by matching linguistic items to concrete representations of linguistic experience that are stored in long-term memory. In this approach, abstraction arises because features that a target exemplar and the memory exemplars have in common are identified and distinguishing features tuned out, resulting in the online summarization of the exemplars into temporary, local categories.
In this talk, I will show that exemplar-based extrapolation, and in particular the aspect of online abstraction, has considerable explanatory value for processing accounts of two linguistic problems that lie at the interface of syntax and semantics, namely thematic fit and adjective order. I will present evidence from both computational psycholinguistics and neurolinguistics.