Date: Oct 10, 2018 3pm-4:30pm
Location: Dunton Tower: Room 2203
Speaker: Raj Singh
Title: Parsing preferences, processing complexity, and scalar implicatures
Abstract:
Scalar implicatures are inferences that enrich the basic meaning of a sentence. For example, a listener who hears the sentence Sam ate some of the cookies will often conclude that Sam didn’t eat all of the cookies, even though the basic meaning of the sentence says nothing about whether Sam ate all of the cookies.
It is commonly assumed that implicature computation involves strictly more operations than those involved in parsing and understanding the basic meaning of the sentence. Thus, one might expect that a sentence with its implicature would be harder to process and less preferred than the sentence without its implicature. Work in experimental psycholinguistics over the past fifteen years has shown that this is only sometimes true.
This puzzle is the focus of our talk. Why is it that the extra work involved in implicature computation only sometimes reveals itself in processing tasks? We summarize the evidence that shows just when the extra work is costly, and we review various complexity metrics that could plausibly be recruited to explain the empirical findings. These metrics include syntactic complexity, semantic strength, and automata-theoretic complexity measures. We reject all of these in favour of conversational principles that relate parsing strategies with relations between questions and answers.