PLUS: A neurocomputational model for the Processing of Linguistic Utterances based on the Unification-Space architecture

Principle Investigator:

Prof.dr. G.A.M. Kempen, Universiteit Leiden, Faculteit der Sociale Wetenschappen

Abstract

Recent years have seen an increasing number of neurophysiological and neuroimaging studies on aspects of language processing. Unfortunately, these studies often lack an explicit account of the complex representational and processing issues that play a role in the computation of syntactic and semantic structures during language perception and production. On the other hand, there are several explicit computational models that do specify the representational levels and predict behavioral data. However, these models fail to take into explicit consideration the growing body of data from neuroimaging (in particular BOLD (fMRI)responses) and neurophysiology (ERPs). This situation is unfortunate, since explicit computational models can guide the search for the underlying neural architecture, and neural data may constrain the space of theoretical options for computational models.

This research proposal aims at bringing together these two levels of research: to develop a computationally explicit model of sentence processing that accounts not only for behavioral data but for recent ERP and BOLD effects as well. In designing and implementing such a model, we take the Unification-Space model of human sentence processing (Kempen & Vosse, 2000) as starting point. Although this model embodies some elementary principles of 'brain-style computing' and holds promises for a satisfactory explanation of neuroscientific data (Hagoort, 2003), it does not qualify as a real neurocomputational model. In order to achieve this aim, a next-generation model is required that is more solidly based on recent ideas concerning brain-style computing. The development of this new model will go hand in hand with testing its predictions in ERP and fMRI experiments that will inform its further development.