RI: SMALL: COLLABORATIVE REFERENCE IN OPEN DOMAINS

Project Details

Description

Reference, the ability to single out objects in the world usinglinguistic expressions, is a fundamental part of communication.Cognitive scientists have showed that people collaborate on reference,for example by negotiating about how to describe things. This projectlays the groundwork for computer systems that can do the same.Eventual applications range from talking robots that can communicatewith people in real physical environments, to tutoring anddecision-support systems that can give comprehensible explanations ofspecialized concepts. These applications are open domains, in thesense that one interlocutor, or both, may have no simple descriptorfor an unfamiliar object. So the system has to have flexiblestrategies for describing things in words and inference mechanismsthat acknowledge the possibility of a misunderstanding by eithersystem or user.This project undertakes exploratory work on the communicativestrategies and inference mechanisms required for collaborativereference in open domains. During the academic year, PI Stone workswith a computer science student to extend a prototype system withadditional methods for collaborating under uncertainty, drawing on thestudent's ongoing research in cognitive modeling, planning andlearning. During the summer, Stone works with a three-memberinterdisciplinary team doing corpus analysis, grammar development andintegration to characterize generic strategies for describing shape,using rough descriptions and part-whole relationships. The resultsare an open-source prototype with more general collaborative referenceabilities along with specific hypotheses, informed by publishablelinguistic and conceptual analyses, about how these abilities mightaffect the system's success in referring to unfamiliar objects.
StatusFinished
Effective start/end date8/15/107/31/12

Funding

  • National Science Foundation (National Science Foundation (NSF))

Fingerprint Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.