Androids that replicate humans in form also need to replicate them in behaviour to achieve a high level of believability or lifelikeness. We explore the minimal social cues that can induce in people the human tendency for social acceptance, or ethopoeia, toward artifacts, including androids. It has been observed that people exhibit a strong tendency to adjust to each other, through a number of speech and language features in human–human conversational interactions, to obtain communication efficiency and emotional engagement. We investigate in this paper the phenomena related to prosodic alignment in human–computer interactions, with particular focus on human–computer alignment of speech characteristics. We found that people exhibit unidirectional and spontaneous short-term alignment of loudness and response latency in their speech in response to computer-generated speech. We believe this phenomenon of prosodic alignment provides one of the key components for building social acceptance of androids.

An adaptive speech recognizer is a key function in the design of a robust spoken dialogue system. Our research focuses on the human tendency of prosodic alignment to one’s conversational partners. A spoken dialogue system might be able to exploit this human tendency to implicitly influence people to manage their speech at the prosodic level in order to accommodate its recognition capabilities. Consequently, this would decrease recognition errors. Prosodic alignment in human-computer interaction has been studied as part of the problems of personality alignment in the context of animated conversational characters. The present study examines human prosodic alignment tendency at more micro level, and explores whether people’s speech amplitude and pause length align to those of computer generated voices within a dialogue exchange. We found that people exhibit spontaneous short-term alignment of speech prosody to the slight prosodic changes in a computer’s voice within a session, even without the help of animated conversational characters.

This paper examines whether humans perceive additional meaning in computer utterances with different voice prosody and change their decisions based on such interaction if the prosodic differences carry informational significance. We conduct a route selection experiment in which participants were asked to find a route to a goal in a 3-D maze generated by computer graphics. The maze system occasionally outputs a confirmation in response to the participant’s choice of a route at a junction. The prosodic characteristics of the confirmation utterances reflect whether the route selected is right for reaching the goal in a normal alley or wrong, leading to a dead end. In this experiment, participants are able to recognize the prosodic meaning and successfully make a decision that leads to a route through the maze. This result suggests that participants can perceive the differences in pitch range of the confirmation voice, and used the pitch range in reselecting their route. Furthermore, wide pitch range confirmation voice indicates a dead end resulted in a higher reselection ratio. This suggests that participants gave a negative interpretation to wider pitch range responses.

This paper reports the results of a study on the prosodic marking of information status, i.e., whether information is ‘new ’ or ‘given’ in the discourse, in Tokyo Japanese noun phrases (NPs). Previous investigations have indicated that such discoursal information is reflected in accent distribution in Dutch NPs, and in degree of accentuation in Italian NPs. An acoustic analysis of Japanese data, in which accent is lexically determined, reveals that information status, both for accented and unaccented words, is mainly reflected in pitch range, with higher range used to signal new information, and lower range used to signal given information. In addition to the effect of discourse, there is ‘downstep’, which lowers the pitch of words following an accented word inside a prosodic phrase. A perception study shows that this pitch range variation has limited cue value for listeners about the information status of particular words. 1.

We discuss the design of a life-like agent interface that considers the social aspects in human-agent interaction. The underlying hypothesis of the design is that human social behaviors toward life-like agents are on a par with those toward humans. Humans tend to sympathize with and follow other humans based on their affiliation needs. Therefore, it should be with life-like agents. We have designed and incorporated life-like guide agents into an exhibition guide system based on the hypothesis. The system was used at the 12/sup th/ ATR Open House, where we had the opportunity to study the behaviors of its users. We observed that the behaviors of the users were directed toward maintaining their good social relationships with their life-like agents, thereby supporting our hypothesis. The results indicated the importance of incorporating an agent's capability to induce the effect of human affiliation needs in a system design.

Speech corpora with prosodic labels are getting more and more important not only for speech synthesis but also for discourse modeling. A widely used labeling system for Japanese prosody, J-ToBI, however, is insufficient for applications like discourse modeling and it even lacks an accurate method for automatic labeling. In this paper, we propose an automatic labeling method for J-ToBI style description of tonal events in Japanese speech, aiming at applying it to a general-purpose labeling of Japanese prosody. The proposed method takes into account the linguistic constraints on the tone structure, which improves the accuracy of automatic labeling. We achieve a fairly good performance in a preliminary experiment using a read speech corpus.