Complement Coercion: Distinguishing Between Type-Shifting and Pragmatic Inferencing.

Although Complement Coercion has been systematically associated with computational cost, there remains a serious confound in the experimental evidence built up in previous studies. The confound arises from the fact that lexico-semantic differences within the set of verbs assumed to involve coercion have not been taken into consideration. From among the set of verbs that have been reported to exhibit complement coercion effects we identified two clear semantic classes - aspectual verbs and psychological verbs. We hypothesize that the semantic difference between the two should result in differing processing profiles. Aspectual predicates (begin) trigger coercion and processing cost while psychological predicates (enjoy) do not. Evidence from an eye-tracking experiment supports our hypothesis. Coercion costs are restricted to aspectual predicates while no such effects are found with psychological predicates. These findings have implications for how these two kinds of predicates might be lexically encoded as well as for whether the observed interpolation of eventive meaning can be attributed to type-shifting (e.g. McElree et al., 2001) or to pragmatic-inferential processes (e.g. De Almeida, 2004).