Contribution Tracking Contribution Tracking: Models and Skills for Collaborative Language Use Under Uncertainty

This article explores the representations and reasoning by which dialogue systems can work collaboratively and creatively to communicate successfully with interlocutors in situations of transient uncertainty. In these situations, dialogue is shaped by the hierarchy of activities underway in the interaction, by the dynamics of linguistic context, and by the array of possibilities that interlocutors take as open ambiguities. Only by integrating information of all these types can a dialogue agent interpret incremental contributions from interlocutors during periods of uncertainty, or formulate its own context-dependent utterances to help pinpoint the context and resolve ambiguities. An empirical study with an implemented human-computer dialogue system in a referential communication task shows how these skills of contribution tracking enable new, more flexible, more robust, and more coherent strategies for interactive language use.