Incorporating Extra-Linguistic Information into Reference Resolution in Collaborative Task Dialogue

Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pages 1259–1267, Uppsala, Sweden, 11-16 July 2010. c⃝2010 Association for Computational Linguistics Incorporating Extra-linguistic Information into Reference Resolution in Collaborative Task Dialogue Ryu Iida Shumpei Kobayashi Takenobu Tokunaga Tokyo Institute of Technology 2-12-1, ˆOokayama, Meguro, Tokyo 152-8552, Japan {ryu-i,skobayashi,take}@cl.cs.titech.ac.jp Abstract This paper proposes an approach to ref- erence resolution in situated dialogues by exploiting extra-linguistic information. Recently, investigations of referential be- haviours involved in situations in the real world have received increasing attention by researchers (Di Eugenio et al., 2000; Byron, 2005; van Deemter, 2007; Spanger et al., 2009). In order to create an accurate reference resolution model, we need to handle extra-linguistic information as well as textual information examined by exist- ing approaches (Soon et al., 2001; Ng and Cardie, 2002, etc.). In this paper, we incor- porate extra-linguistic information into an existing corpus-based reference resolution model, and investigate its effects on refer- ence resolution problems within a corpus of Japanese dialogues. The results demon- strate that our proposed model achieves an accuracy of 79.0% for this task. 1 Introduction The task of identifying reference relations includ- ing anaphora and coreferences within texts has re- ceived a great deal of attention in natural language processing, from both theoretical and empirical perspectives. Recently, research trends for refer- ence resolution have drastically shifted from hand- crafted rule-based approaches to corpus-based ap- proaches, due predominately to the growing suc- cess of machine learning algorithms (such as Sup- port Vector Machines (Vapnik, 1998)); many re- searchers have examined ways for introducing var- ious linguistic clues into machine learning-based models (Ge et al., 1998; Soon et al., 2001; Ng and Cardie, 2002; Yang et al., 2003; Iida et al., 2005; Yang et al., 2005; Yang et al., 2008; Poon and Domingos, 2008, etc.). Research has contin- ued to progress each year, focusing on tackling the problem as it is represented in the annotated data sets provided by the Message Understanding Con- ference (MUC)1 and the Automatic Content Ex- traction (ACE)2. In these data sets, coreference re- lations are defined as a limited version of a typ- ical coreference; this generally means that only the relations where expressions refer to the same named entities are addressed, because it makes the coreference resolution task more information extraction-oriented. In other words, the corefer- ence task as defined by MUC and ACE is geared toward only identifying coreference relations an- chored to an entity within the text. In contrast to this research trend, investigations of referential behaviour in real world situations have continued to gain interest in the language generation community (Di Eugenio et al., 2000; Byron, 2005; van Deemter, 2007; Foster et al., 2008; Spanger et al., 2009), aiming at applica- tions such as human-robot interaction. Spanger et al. (2009) for example constructed a corpus by recording dialogues of two participants collabo- ratively solving the Tangram puzzle. The corpus includes extra-lingustic information synchronised with utterances (such as operations on the puzzle pieces). They analysed the …

Content truncated to 500 words