Colloquium by Rui Chaves (Buffalo): Two sources of probabilistic gradience in island effects

Abstract:  In this talk I present linguistic and experimental evidence suggesting that the graded acceptability of certain filler-gap dependencies is contingent on at least two independently motivated factors. One factor is more structural in nature. As originally suggested by Hofmeister et al. 2013, if a gap is syntactically, semantically, or pragmatically highly unlikely in a particular sentence, then it is less likely for it to be postulated and less acceptable. But in ideal conditions the parser's  expectations can be altered, and such gaps can be made more likely and acceptable. I will show just how extreme this adaptation can be. The second factor is more semantic in nature.  In a nutshell, referents are extractable to the degree that they are deemed important to the proposition expressed by the utterance, drawing inspiration from insights by Erteschik-Shir (1973,1981),  Kuno (1987) and Van Valin (2005, 288). In particular, I provide evidence that Deane (1991) may have been essentially on the right track when proposing that the conventionalized world knowledge that the main verb evokes plays a key role in determining whether a referent is extractable. The idea that conventionalized general knowledge – sometimes referred to as a "frame", "script" or "cognitive model" – guides the perception and interpretation of the world around us has long permeated various branches of cognitive science, including psychology, linguistics, and artificial intelligence. In this talk I discuss experimental evidence suggesting that frames also play a role in explaining the gradience of certain long-distance dependency phenomena.