A Casual Conversation System Using Modality and Word Associations Retrieved from the Web

Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing, pages 382–390, Honolulu, October 2008. c⃝2008 Association for Computational Linguistics A Casual Conversation System Using Modality and Word Associations Retrieved from the Web Shinsuke Higuchi Rafal Rzepka Graduate School of Information Science and Technology Hokkaido University , Sapporo Japan 060-0814 {shin h,kabura,araki}@media.eng.hokudai.ac.jp Kenji Araki Abstract In this paper we present a textual dialogue system that uses word associations retrieved from the Web to create propositions. We also show experiment results for the role of modal- ity generation. The proposed system automat- ically extracts sets of words related to a con- versation topic set freely by a user. After the extraction process, it generates an utterance, adds a modality and verifies the semantic re- liability of the proposed sentence. We evalu- ate word associations extracted form the Web, and the results of adding modality. Over 80% of the extracted word associations were evalu- ated as correct. Adding modality improved the system significantly for all evaluation criteria. We also show how our system can be used as a simple and expandable platform for almost any kind of experiment with human-computer textual conversation in Japanese. Two exam- ples with affect analysis and humor generation are given. 1 Introduction Many task-oriented dialogue systems (Liu et al., 2003; Reitter et al., 2006) have been developped. Research on non-task-oriented dialogue systems like casual conversation dialogue systems (”chatbots”) is on the other hand not very common, perhaps due to the many amateurs who try to build naturally talking systems using sometimes very clever, but rather un- scientific methods although there are systems with chatting abilities as (Bickmore and Cassell, 2001) but concentrate on applying strategies to casual con- versation rather than their automatic generation of those conversations. However, we believe that the main reason is that an unrestricted domain is dispro- portionately difficult compared to the possible use such a system could have. It is for example very hard to predict the contents and topics of user utterances, and therefore it is almost impossible to prepare con- versational scenarios. Furthermore, scenarios need more or less specific goals to be useful. However in our opinion, sooner or later non-task-oriented di- alogue systems will have to be combined with task oriented systems and used after recognizing that the user’s utterance does not belong to a given task. This would lead to more natural interfaces for e.g. infor- mation kiosks or automatic guides placed in public places where anyone can talk to them about anything (Gustafson and Bell, 2000; Kopp et al., 2005) re- gardless of the role the developers intended. For this reason we have also started implementing emotive- ness recognition and joke generation modules that are presented later in the paper. Well-known examples of non-task-oriented dia- logue systems are ELIZA (Weizenbaum, 1966) and A.L.I.C.E 1, though the former was built to parody a Rogerian therapist which can be regarded as a task. Both systems and their countless imitators 2 use a lot of rules coded by …

Content truncated to 500 words