It is thought that, in conversation, we generally act and judge whether participants’ conversational behaviors are appropri-ate at that time; that is, we avoid interruption of talk in formal situations. However, which behaviors are appropriate inwhat kinds of situations has been not studied well. Therefore, in this study, we focus on the violation of turn-taking rules(i.e., overlap or interruption) and investigate which behaviors are appropriate in each situation through an experiment usinginteractive robots/agents that can regulate conversational behaviors. The results showed that violation in formal situationssignificantly lowered the sense of unity, but, on the other hand, the sense of unity in informal situations showed no signif-icant difference between contexts of violation and obeying of rules. Thus, the violation of conversational behaviors in aninformal situation may maintain a sense of unity and this may contribute to revealing the mechanism behind perception ofconversational behaviors.

We investigated the gain-loss effect of virtual/personal assistant character, which provides intelligent assistance tohumans, and focused on not only the first impression of using the assistance but also changing impression about the characterwhen used multiple times for assistance. The experiment used a fictive retrieval system (searching onomatopoeia), and thevirtual assistant character looked up for suitable words for the user (success), or failed to find the words (mistake). There werethree sessions, differing by the task of character’s mistake; two tasks were successes and one was a failure in each session. Theresults showed that the group of people who had low expectation from its first appearance, formed negative impressions afterthe final mistake, significantly. Consequently, final mistake influenced the formation of negative impression more than othermistakes, thus showing that the final mistake in multiple times of assistance was associated with loss effect.