Conversation Entailment
1/1/2008 - Present
Given the increasing amount of conversation data (e.g., conversation scripts such as meeting scripts, court records, and online chatting), it becomes more and more important to develop technology that can automatically discover and infer knowledge from conversation, for example, knowledge about participants involved in the conversation. However knowledge discovery from conversation has received less attention compared to other data sources (e.g., free text and structured databases). Although conversation is in the form of language, it has unique characteristics compared to written text. The key distinctive features include turn-taking between participants, grounding between participants, and different linguistic phenomena of utterances (e.g., Utterances in conversation tend to be shorter, with disfluency, and sometimes incomplete or ungrammatical). Because of these distinctions, traditional approaches dealing with written text are not sufficient to derive knowledge from conversation. Special treatments that incorporate rich conversation context and address unique conversation behavior will be important. This project explores integration of conversation context and natural language processing in automated inference of knowledge about conversation participants. In particular, inspired by textual entailment, we formulate the inference problem as conversation entailment: given a conversation discourse and a hypothesis about a particular state of a conversation participant (e.g., profile, belief, etc.), the goal is to automatically decide whether the discourse entails the hypothesis.
Selected Papers:
- Towards Conversation Entailment: An Empirical Investigation. C. Zhang and J. Y. Chai. The 2010 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 756–766, MIT, MA, October, 2010. [PDF] [Bib]
- What do We Know about Conversation Participants: Experiments on Conversation Entailment. C. Zhang and J. Chai. The 10th Annual SIGDIAL Meeting on Discourse and Dialogue, pages 206–215, London, UK, September, 2009. [PDF] [Bib]
- The data sets used in these papers are available here.
