This project will analyze the linguistic and extra-linguistic factors that influence the interpretations of questions and the answers to them. Linguistic factors include surface-level cues such as a speaker's choice of lexical items and the syntactic structure of the question. Extra-linguistic factors include the discourse context and the speaker's goals in asking the question. The project will address theoretical proposals regarding the division of labor between syntax, semantics, and pragmatics in question-answer dynamics, and will provide empirical data that bear on debates concerning what licenses certain kinds of answers to certain kinds of questions. The proposed research seeks to integrate corpus, psycholinguistic and developmental methods, and computational modeling in a novel way. The project will provide valuable opportunities for undergraduate students, from experimental design and running to the presentation of results in scholarly venues. The researchers will disseminate these results through academic conferences and publications, and engage in community outreach events. Both the existing connections between the researchers and colleagues in cognitive and computer science, and the proposed implementation of these results in a computational model ensure that the results of the research will be applicable to work in dialogue research, concerning the design of intelligent conversational agents for AI.
While previous work on questions has focused primarily on the nature of exhaustive answers, what licenses non-exhaustive (or mention-some) answers, which are frequent in natural dialogue, has gone under-investigated. Thus there is a need to systematically investigate the factors that determine whether and when a question is interpreted non-exhaustively. This research uses a three-pronged interdisciplinary approach, combining tools from experimental linguistics, developmental psychology, and computer science, gathering empirical data from natural language corpora and psycholinguistic and developmental studies, and capturing the findings in a computational model. Corpus data inform us about what speakers actually do. Adult psycholinguistic studies inform us about what is allowed and assessed as likely or unlikely. Developmental studies inform us about the trade-off between semantics and pragmatics, and the role of the input. A computational model aims to capture these data in a probabilistic framework. Together, these interconnected research approaches represent an integrated and holistic study of the linguistic and cognitive underpinnings of questions and questioning.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.