Introduction to the Issue on Computational Models of Natural Language

Cognitive scientists might be forgiven for thinking that computational psycholinguistics ended somewhere around 1982. By that point, Augmented Transition Networks had come and gone, grammar-writing was out of vogue, and the study of garden-path sentences was doing quite well on its own. Computational linguists had a string of “other” relationships: first with logic, then statistics. Connectionism ushered in a whole new outlook on modeling. The computational linguistics and cognitive science communities were indeed largely estranged. However by 2010, the old lovers were ready to get back together. Connectionism had grown up; some of its controversial leading ideas now held jobs in big, probabilistic models of language. In addition, computational linguistics had accustomed itself to a notion of evaluation surprisingly close to the “task” in experimental psychology. It was time for the first workshop on Cognitive Modeling and Computational Linguistics (CMCL). This took place at the Association for Computational Linguistics meeting, and it has continued every year since then. This issue contains a selection of articles that, in large part, originated at CMCL. They are unified in both substance and methodology. Methodologically, they make use of computational ideas, techniques, and resources that often afford quite detailed predictions about cognition. They bridge different levels of linguistic analysis, working out the implications of proposals that were formerly trapped inside disciplinary niches. Their scientific common ground is harder to see; specific topics range from millisecond-scale eye movements all the way up to typological universals. However, certain themes do recur, as Fig. 1 illustrates.