Dynamical Cognitive Modeling of Syntactic Processing and Eye Movement Control in Reading

Integrating eye-movement control and sentence processing would mark an important step forward for mathematical models of natural language processing. We present an integrated approach by combining the SWIFT model of eye-movement control (Engbert et al., 2005) with key components of the LV05 (Lewis & Vasishth, 2005) parser. The integrated generative model can reproduce reading time patterns that have been explained in terms of similarity-based interference in the psycholinguistic literature. A crucial problem for such complex models is parameter estimation. We build upon recent advances on successful parameter identification in dynamical models, investigate likelihood profiles for single parameters, and present pilot results on MCMC sampling within a Bayesian framework of parameter inference.