Towards Conversation Entailment: An Empirical Investigation

Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, pages 756–766, MIT, Massachusetts, USA, 9-11 October 2010. c⃝2010 Association for Computational Linguistics Towards Conversation Entailment: An Empirical Investigation Chen Zhang Joyce Y. Chai Department of Computer Science and Engineering Michigan State University East Lansing, MI 48824, USA {zhangch6, jchai}@cse.msu.edu Abstract While a significant amount of research has been devoted to textual entailment, automated entailment from conversational scripts has re- ceived less attention. To address this limi- tation, this paper investigates the problem of conversation entailment: automated inference of hypotheses from conversation scripts. We examine two levels of semantic representa- tions: a basic representation based on syntac- tic parsing from conversation utterances and an augmented representation taking into con- sideration of conversation structures. For each of these levels, we further explore two ways of capturing long distance relations between lan- guage constituents: implicit modeling based on the length of distance and explicit mod- eling based on actual patterns of relations. Our empirical findings have shown that the augmented representation with conversation structures is important, which achieves the best performance when combined with ex- plicit modeling of long distance relations. 1 Introduction Textual entailment has received increasing attention in recent years (Dagan et al., 2005; Bar-Haim et al., 2006; Giampiccolo et al., 2007; Giampiccolo et al., 2008; Bentivogli et al., 2009). Given a segment from a textual document, the task of textual entailment is to automatically determine whether a given hypoth- esis can be entailed from the segment. The capa- bility of such kind of inference can benefit many text-based applications such as information extrac- tion and question answering. Textual entailment has mainly focused on infer- ence from written text in monologue. Recent years also observed an increasing amount of conversa- tional data such as conversation scripts of meetings, call center records, court proceedings, as well as on- line chatting. Although conversation is a form of language, it is different from monologue text with several unique characteristics. The key distinctive features include turn-taking between participants, grounding between participants, different linguistic phenomena of utterances, and conversation impli- catures. Traditional approaches dealing with tex- tual entailment were not designed to handle these unique conversation behaviors and thus to support automated entailment from conversation scripts. Example 1: Conversation Segment: B: My mother also was very very independent. She had her own, still had her own little house and still driving her own car, A: Yeah. B: at age eighty-three. Hypothesis: (1) B’s mother is eighty-three. (2) B is eighty-three. To address this limitation, our previous work (Zhang and Chai, 2009) has initiated an investigation on the problem of conversation en- tailment. The problem was formulated as follows: given a conversation discourse D and a hypothesis H concerning its participant, the goal was to identify whether D entails H. For instance, as in Example 1, the first hypothesis can be entailed from the 756 conversation segment while the second hypothesis cannot. While our previous work has provided some interesting preliminary observations, it mostly …

Content truncated to 500 words