What to do when a linguist says she's expecting: Coherence and coreference expectations in sentence processing
In order for a discourse to make sense, not only must the individual utterances that comprise the discourse be well-formed but the set of utterances as a whole must come together in a coherent way. Given this, comprehenders of a discourse can reasonably expect that upcoming linguistic material will not appear arbitrarily but rather will relate in meaningful ways to previous utterances. Those underlying discourse relations can be signalled overtly, but in many cases they must be inferred given the content of a set of utterances and the comprehender's estimation of the speaker's intended message. My work considers the extent to which natural language interpretation relies on expectations about the kinds of discourse relations that will be inferred to hold between current and upcoming material. If discourse comprehension is expectation-driven, what characterizes those expectations, what are they conditioned on, and what phenomena do they influence? In this talk, I consider two types of discourse-level dependencies: the relationships that can be inferred to hold between pairs of sentences (coherence) and the relationships between subsequent mentions of the same entity (coreference). I discuss recent work and open questions regarding comprehenders' expectations for these two types of discourse-level dependencies and the methods available for measuring such expectations. I speculate about the types of cues that guide expectations about upcoming discourse structure. Lastly, I take a step back and question the assumption that a highly predictable discourse is what comprehenders are really expecting.