Which event properties matter for which cognitive task?

Two important questions in the language sciences have been: What portion of event information is accessed and used during sentence processing? What portion of event information is relevant to the grammar of natural languages? An extensive amount of linguistic and psycholinguistic research over the past couple of decades has shown that quite a bit of event information is relevant to online sentence processing, but only a limited amount of information is relevant to the grammars of natural languages. One possible cause for this divergence is that grammar development and language comprehension are carried out by separate systems that are sensitive to different types of information. Another possible explanation is that grammar development and language comprehension are carried out by an integrated system, but the task demands of language learning and the task demands of language comprehension are different. In this chapter, we show that, grammatical systems that seem more “exotic” from the point of view of more well–known languages still make use of a limited set of properties and that these “exotic” languages still obey the same design constraints as more well–known systems; we then briefly report on some computational models of online reading experiments, which demonstrate quite clearly that a distinct and much larger kind of event knowledge is used by the human parser. We propose an explanation for this difference in the use of event knowledge: Grammars and parsers use different kinds of event knowledge because the tasks listeners and grammar learners must perform are quite distinct.