Computational models of human sentence processing: am model comparison approach

• Applicant: Professor Dr. Shravan Vasishth
• Subject Area: General and Comparative Linguistics, Experimental Linguistics, Typology, Non-European Languages
• Term: from 2007 to 2011
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 38541091

This project will be carried out in collaboration with partners in two universities in theUnited States (Michigan State University, and University of Michigan). The first research goal is to evaluate in a uniform computational environment the performance of several existing computational models of sentence processing using data from (1) 21 eyetracking and self-paced reading experiments with German and English that have already been completed by Vasishth and colleagues, and (2) three relatively large eyetracking databases, the Potsdam Sentence Corpora and the Paragraph Corpus, developed by Kliegl and colleagues. The models in question are Dependency Locality Theory (DLT), Surprisal, Entropy Reduction, cue-based parsing and three interference theories. The 21 experiments were specifically designed to address the predictions of some of these theories and focus on two major open questions in sentence processing: the effects of locality and interference on online sentence parsing. We evaluate the relative quantitative fits of the models (including additive combinations of models), taking into account the free parameters that each model utilizes. To our knowledge, such a comparison of several prominent computational models has never been carried out in psycholinguistics and will provide a first clear picture of their relative merits. In addition, the empirical component will deliver new experimental results that will evaluate the differing quantitative predictions of the models, and two new eyetracking corpora that will provide in a single dataset a wide range of theoretically important phenomena in sentence processing relating to locality and interference. This will serve as a comprehensive test set for evaluating the theories under consideration and as a database for future research.

• DFG Programme: Research Grants
• Participating Person: Professor Reinhold Kliegl, Ph.D.