Jean-Pierre Koenig: Why is linking simple?

Since Fillmore (1968), research on the interface between syntax and semantics has shown that the semantic properties of participants in situations described by verbs play an important role in the syntax of basic clauses. But, as Pinker (1989) already noted, a relatively small number of semantic properties seems to influence the syntax of natural languages. The question is why.

In the first part of this talk, I outline a linking theory, i.e. a theory of the mapping between participant roles and argument-structure positions/grammatical functions. That theory is based on the research my collaborators and I have conducted. The upshot of that research is that, indeed, a limited range of semantic properties are relevant to linking. I show that constraints on the maximal structural complexity of verb meaning and the distinction between semantic arguments and semantic adjuncts limit the number of participant roles that matter to linking. I then show that some information about event structure is systematically ignored for selection of direct arguments (modal structure and non-profiled co-events). I also suggest that linking is further simplified if we distinguish between grammatical constraints on linking and meta-grammatical generalizations about linking constraints. For the former, only generalizations over natural situation types matter.

In the second part of my talk I present research that we conducted that demonstrates that the range of semantic properties that matter for sentence processing is much larger. Our research suggests that  the processing of instrument phrases is influenced not only by how predictable an instrument is (e.g., how predictable is spear as an instrument of jabbing), but by how similar in semantic space that instrument is to other instruments that could have occurred instead (i.e., how similar spear is to other sharp pointed instruments like sword, machete, and so on). That kind of fine-grained situational knowledge, typical of sentence processing is simply not relevant to syntax. I then explore two possible hypotheses for that difference between the range of semantic properties that matter to syntax and those that matter to sentence processing. One, strongly hinted at by Pinker, is that, the human syntactic abilities are innately attuned to a small set of semantic elements. The other, which our research group has put forth, is that the observed differences follow from the different demands of language acquisition (grammar formation) and language comprehension.