FIXATION-RELATED BRAIN IMAGING
OF WORD-BY-WORD READING
This is a newer project that combines computational linguistics and cognitive neuroscience methods to investigate individual differences in word-by-word grammar processing. We compare two grammar models, one with no syntactic constituent structure (a.k.a. n-gram) and one with hierarchical structure that encodes syntactic constituents and phrasal dependencies (a.k.a. CFG). The amount of processing effort required to integrate each word into each of these models can be indexed by the surprisal measure (Hale, 2001). Crucially, the two grammars differ in their surprisal measures at points of integration complexity (e.g., when a retrieval is needed to combine a verb with a distal subject, or to retrieve the distal antecedent of a pronoun.)
Preliminary evidence from both ERP and fMRI show differences in activation for the CFG vs. ngram grammar models, providing a means of diagnosing the type of model an individual is constructing while reading each word. Collection of neurological data is on-going.
Analyses of syntactic and semantic word-by-word processing with eye-tracking data shows interactions of CFG surprisal and semantic richness and reading ability. This suggests that surprisal can be a useful tool for discriminating skill-related differences in representation quality.