Login
Computer Science @ Rochester
Monday, December 08, 2003
11:00 AM
CSB 209
Matthew Stone
Rutgers U.
Semantics and Collaboration in Conversational Agents
Users expect spoken language interfaces to recognize their words and respond intelligently to their needs. It has proved suprisingly difficult to meet these expectations in a general way, because what we say and do in conversation so intimately reflects our information, our intentions, and our engagement with one another. Nevertheless, computer systems can hold up their end of a conversation, thanks in part to increasingly sophisticated models of meaning.

In this talk, I will use implemented prototypes to describe some of the characteristic features of meaning in conversational language for which there has been significant recent progress, including: - the signals and feedback that keep conversations going; - the collaborative context that informs language understanding and generation; - the coordination that allows one interlocutor to signal her expectations for the conversation, and another interlocutor to meet them.

An important common theme in these developments is a rejection of a static view of linguistic semantics as a predefined system of logical translation or formal truth-conditions, in favor of a collaborative view of meaning: As language users, we construct the meanings of our utterances, through the collaborative decisions we take as we draw on the rules of language to make ourselves understood.