A Classification of Dialogue Actions in Tutorial Dialogue

Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 73–80 Manchester, August 2008 A Classification of Dialogue Actions in Tutorial Dialogue Mark Buckley and Magdalena Wolska Dept. of Computational Linguistics Saarland University 66041 Saarbr¨ucken, Germany {buckley|magda}@coli.uni-sb.de Abstract In this paper we present a taxonomy of di- alogue moves which describe the actions that students and tutors perform in tutorial dialogue. We are motivated by the need for a categorisation of such actions in order to develop computational models for tutorial dialogue. As such, we build both on exist- ing work on dialogue move categorisation for tutorial dialogue as well as dialogue taxonomies for general dialogue. Our tax- onomy has been prepared by analysing a corpus of tutorial dialogues on mathemati- cal theorem proving. We also detail an an- notation experiment in which we apply the taxonomy and discuss idiosyncrasies in the data which influence the decisions in the dialogue move classification. 1 Introduction The field of Intelligent Tutoring Systems has seen recent developments moving towards adding natu- ral language capabilities to computer-based tutor- ing (Graesser et al., 1999; Zinn, 2004; Litman and Silliman, 2004), motivated by empirical investiga- tions which point to the effectiveness of human tu- tors (Bloom, 1984; Moore, 1993; Graesser et al., 1995). However, to be able to interact with a stu- dent through the medium of natural language dia- logue, the system must have a model of how such tutorial dialogues can progress and what utterances are licenced. In order to develop such a model of dialogue, we need to understand and describe the “actions” performed with words, i.e. speech acts (Austin, 1955) or dialogue moves. This in- volves identifying and categorising the functions c⃝ 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported li- cense (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. that utterances may have in dialogue and their re- lationships to each other. Researchers in conversation and dialogue the- ory have proposed various general categorisa- tions of dialogue moves. DIT++ (Bunt, 2000) is an example of a comprehensive multidimen- sional taxonomy of dialogue acts for informa- tion dialogues based on DAMSL (Allen and Core, 1997), a general-purpose extensible taxonomy pro- posed as a standard for dialogue annotation. The DAMSL dialogue act taxonomy characterises utter- ances along four dimensions which correspond to four levels of functions utterances may have. The forward looking function describes the utterance’s effect on the following interaction, the backward looking function, its relation to previous dialogue, the communicative status describes the compre- hensibility or interpretability of the utterance, and the information level characterises the content of the utterance. Tsovaltzi and Karagjosova (2004) proposed an extension of the DAMSL classification based on an analysis of tutorial dialogue corpora. The pro- posed taxonomy adds a Task dimension which concentrates on tutor actions in the dialogue.1 Building on this work, we propose a further ex- tension to this taxonomy inspired by our analysis from the point of view of task-related goals. The classification we present (i) includes modifications of the DAMSL categorisation …

Content truncated to 500 words