Dialogue Structure Annotation for Multi-Floor Interaction

Dialogue Structure Annotation for Multi-Floor Interaction David Traum1, Cassidy Henry2, Stephanie Lukin2, Ron Artstein1 Felix Gervitz3, Kimberly A. Pollard2, Claire Bonial2, Su Lei1, Clare R. Voss2, Matthew Marge2, Cory J. Hayes2, Susan G. Hill2 1USC Institute for Creative Technologies, 2U.S. Army Research Laboratory, 3Tufts University Playa Vista, CA 90094, Adelphi, MD 20783, Medford, MA 02155 traum@ict.usc.edu Abstract We present an annotation scheme for meso-level dialogue structure, specifically designed for multi-floor dialogue. The scheme includes a transaction unit that clusters utterances from multiple participants and floors into units according to realization of an initiator’s intent, and relations between individual utterances within the unit. We apply this scheme to annotate a corpus of multi-floor human-robot interaction dialogues. We examine the patterns of structure observed in these dialogues and present inter-annotator statistics and relative frequencies of types of relations and transaction units. Finally, some example applications of these annotations are introduced. Keywords: dialogue structure annotation, human-robot interaction, multiparty dialogue 1. Introduction We present an annotation scheme for meso-level dialogue structure (Traum and Nakatani, 1999), specifically de- signed for multi-floor dialogue. The scheme includes both a transaction unit for clustering utterances from multiple participants and floors that contribute to realization of an initiating participant’s intent, and relations between indi- vidual utterances within the unit. While there are stan- dard annotation schemes for both dialogue acts (Bunt et al., 2012) and discourse relations (Prasad and Bunt, 2015), these schemes do not fully address the issues of dialogue structure. Of particular interest to us, and not previously ad- dressed in other schemes, are cases in which the units and relations span across multiple conversational floors. Dia- logues can be characterized by distinct information states (Traum and Larsson, 2003). These include sets of par- ticipants, participant roles (e.g. active, ratified participant vs. overhearer), turn-taking or floor-holding, expectation of how many participants will make substantial contributions at a time (Edelsky, 1981), and other factors. Often distinct dialogues with different information states are going on at the same time. There are a number of ways in which such dialogues can be related to each other, including: • having the same purpose but distinct participants, e.g., teams competing in a trivia contest to come up with the answer first. • co-located such that participants in one can observe and possibly comment on the other, such as groups of people sitting at different tables at a restaurant. • having one or more (but not all) participants in com- mon, where such participants are multicommunicating (Reinsch et al., 2008), e.g., someone in a meeting is texting with one or more people outside the meeting. In the multicommunicating case, the multiple dialogues that a multicommunicator is part of might involve com- pletely separate topics or be more closely related, such that satisfaction of the goals of one depends on actions in the other. For example, a question arising in a meet- ing might be conveyed and answered over the text channel. We use the term multi-floor dialogue to refer to cases in which …

Content truncated to 500 words