Identification of Personal Information Shared in Chat-Oriented Dialogue

Identification of Personal Information Shared in Chat-Oriented Dialogue Sarah Fillwock∗, David Traum† ∗Michigan State University 428 S. Shaw Lane, East Lansing, MI 48824 fillwoc2@msu.edu †Institute for Creative Technologies, University of Southern California 12015 Waterfront Dr, Playa Vista, CA 90094 traum@ict.usc.edu Abstract We present an analysis of how personal information is shared in chat-oriented dialogue. We develop an annotation scheme, including entity-types, attributes, and values, that can be used to annotate the presence and type of personal information in these dialogues. A collection of attribute types is identified from the annotation of three chat-oriented dialogue corpora and a taxonomy of personal information pertinent to chat-oriented dialogue is presented. We examine similarities and differences in the frequency of specific attributes in the three corpora and observe that there is much overlap between the attribute types which are shared between dialogue participants in these different settings. The work presented here suggests that there is a common set of attribute types that frequently occur within chat-oriented dialogue in general. This resource can be used in the development of chat-oriented dialogue systems by providing common topics that a dialogue system should be able to talk about. Keywords: dialogue, personal information, attributes, knowledge base, chat-oriented, user model 1. Introduction Spoken dialogue has been studied from the perspectives of many different disciplines, with a primary focus on ana- lyzing the turn-taking procedure of dialogue, the effects of social context, and the underlying structure of sequences of dialogue (Eggins and Slade, 1997). Within this body of work, spoken dialogue is often divided into two major categories: task-oriented dialogues and chat-oriented di- alogues. Although task-oriented dialogues have received the most attention historically, interest in chat-oriented di- alogue has been on the rise in recent decades. One major factor in this increase of interest has been the popularity of artificial conversational dialogue agents. As defined by Eggins and Slade (1997), chat-oriented di- alogue - or casual conversation as they name it - is char- acterized by its topic flexibility, the informal nature of the exchange, and the fact that the participants are not trying to accomplish any particular functional task through their dia- logue. Aiming to overcome the seeming open-domain gen- erality of chat-oriented dialogues, much of the effort has been placed into identifying the sub-types of chat-oriented dialogue and characterizing the general structure of these sub-types (Eggins and Slade, 1997; Slade and Gardner, 1993; de Silva Joyce and Slade, 2000). Although there has been much work on defining and an- alyzing the characteristic structure of chat-oriented dia- logues, there has been less work on exploring the content of these dialogues in terms of their characteristic topics. An understanding of the topics that are commonly focused on during chat-oriented dialogues would provide both greater insight into the function of chat-oriented dialogues on an inter-personal level and also allow for interesting analyses based on the popular topics of conversation. One promising area of inquiry is the sharing of personal information be- tween the conversational participants.It has been observed that people frequently …

Content truncated to 500 words