Dialogue System Characterisation by Back-channelling Patterns Extracted from Dialogue Corpus

Dialogue System Characterization by Back-channelling Patterns Extracted from Dialogue Corpus Masashi Inoue, Hiroshi Ueno Yamagata University 3-16, 4 Jyonan, Yonezawa mi@yz.yamagata-u.ac.jp, tmk56575@st.yamagata-u.ac.jp Abstract In this study, we describe the use of back-channelling patterns extracted from a dialogue corpus as a mean to characterising text-based dialogue systems. Our goal was to provide system users with the feeling that they are interacting with distinct individuals rather than artificially created characters. An analysis of the corpus revealed that substantial difference exists among speakers regarding the usage patterns of back-channelling. The patterns consist of back-channelling frequency, types, and expressions. They were used for system characterization. Implemented system characters were tested by asking users of the dialogue system to identify the source speakers in the corpus. Experimental results suggest that possibility of using back-channelling patterns alone to characterize the dialogue system in some cases even among the same age and gender groups. Keywords: keyword A, keyword B, keyword C 1. Introduction Text-based communication in conversational formats has become increasingly common among people because of the widespread use of messaging applications. Accordingly, a growing need exists for text-based dialogue systems that can conduct natural conversation beyond single command utterances. Dialogue systems may someday become uni- versal, which mean people will have to interact with dif- ferent systems in different situations and for different pur- poses. It will then be natural for dialogue systems to em- ploy different characters so that users can distinguish sys- tems and enjoy different types of interaction. Many studies have been conducted on generating charac- ters through vocalization (e.g., (Fujie et al., 2004).) Com- pared to characterization in which personal traits are added to spoken sound, relatively few studies have been con- ducted adding linguistic personal traits such as vocabulary and style. The addition of linguistic characters to dialogue systems is often based on tailored rules (Nass et al., 1995) instead of statistical information obtained from corpora. As an example of statistical dialogue system characterization, one study has adapted the Big Five personality model to verbal expressions (Mairesse and Walker, 2007). Later, both extroverted and introverted characters based on the model were implemented in a dialogue system (Andrews, 2012). When characters are defined based on categories such as age, gender and personality, the variety of characters are then limited: In the real world, two strongly extroverted females in their 20s may have unique qualities that distin- guish them, but the system can generate the same character to represent both of them when the system is created using categorical rules. Dialogue characterization can be achieved by modifying the utterances created by generic dialogue systems using a characterization module. However, a trade-off occurs be- tween the significance of characters in dialogue systems and the quality of modified utterances. When utterances are overly modified, they can be ungrammatical or awk- ward. To avoid such over-modification, we can use expres- sions that do not contain concrete semantic implications. An example of this type of expression is back-channelling (BC). BCs are “the short utterances produced by one …

Content truncated to 500 words