Annotation of Addressing Behavior in Multi-Party Conversation

This paper reports an attempt to annotate addressing behaviors in everyday conversation to collect both verbal and nonverbal resources related to addressing. Although several earlier studies have discussed behaviors that can be used as addressing devices, no study has surveyed actual conversations to determine the means of addressing used. Currently, we are developing an annotation framework for addressing behaviors, which we refer to as multi-dimensional annotation. This paper introduces the current version, including the basic idea, labels, and results of the pilot annotations of naturally occurring conversation data. The distribution of tags indicates the characteristics of both the participants and conversations. Additionally, some cases found through annotation suggest that simply detecting movement and body direction may not fully capture address information.