Embodied semiotics aims to explicate the cognitive and social processes by which an actor's bodily movement becomes a meaningful "signal" when understood and responded to by interactant in everyday situation. The authors have been annotating and analyzing the "Miraikan SC Corpus," a multimodal corpus of interactions between science communicators (SCs) and visitors at the National Museum of Emerging Science and Innovation (Miraikan), with a focus on such characteristic body movements as moving and pointing. However, research on the aspects of how actors' bodies interact with physical objects in the material environment has been insufficient. Then, we designed and experimentally recorded the "Furoshiki Wrapping Corpus", in which learner's understanding of an instructor's physical action with objects can be observed through the learner's bodily motions. This paper outlines the design of the corpus, the procedure for sharing and releasing the data, and the characteristics of the recorded data.

Early detection of psychological distress is important for preventing serious mental health issues. This includes situations where individuals hide their stress from others or are not aware of their own condition. Traditional methods mainly rely on physiological signals such as skin temperature and heart rate, which require contact-based sensors. In this study, we propose a method for detecting concealed nervousness in spoken dialogue using facial and acoustic features obtained from non-contact devices such as cameras and microphones. We labeled the data based on self-reported nervousness levels and trained a machine learning model. The combined use of facial expressions and speech features achieved an accuracy of 0.74. We also examined individual differences in how nervousness appears. Some people tend to show tension in facial expressions, while others show it in vocal tone. Statistical analysis showed that relaxed facial muscles reduce the visibility of nervousness in expressions, and facing the conversation partner makes nervousness more detectable in speech.

This paper proposes the Externalized Cognitive Loop (ECL) model as a framework for understanding how syntactic meaning-making can be reactivated through dialog with generative AI.The model was developed based on qualitative analysis of interactive logs between a human subject and ChatGPT, focusing on situations where narrative capability had collapsed due to the breakdown of internal syntactic structures.In conventional natural language processing, semantic alignment is often prioritized, while the recursive integrity of the speaker's cognitive structure tends to be overlooked.The ECL model shifts the focus to the preservation and reactivation of syntactic structures that enable externalized, recoverable forms of questioning, especially in cognitively fragile or emotionally disrupted contexts. The proposed model consists of five recursive stages, from pre-syntactic expression to full reentry into structured dialog.It is evaluated through detailed narrative logs and situated within a broader theoretical framework that includes the SISS (Structured Intellectual Safe Space) and SRI (Structural Responsiveness Index) models. The findings suggest that generative AI, when designed or engaged to maintain syntactic openness rather than impose semantic closure, can serve as a catalyst for reinitiating discourse in cases where internal language loops have failed. This has implications for both dialog system design and the understanding of linguistic resilience in human-machine interaction.

This study investigates how amusement park staff manage dual involvements (Raymond & Lerner, 2014) in service encounters with child visitors, focusing on their simultaneous engagement in ride operation and conversation. Staff are required to conduct service encounters under several constraints. For example, when operating a Ferris wheel, they must guide and assist visitors while handling the ride's operation at specific timing intervals. Additionally, in interactions with children, staff often initiate and expand sequences of talk to entertain and engage them. Using interaction analysis, this study examines how staff adjust their conduct under such constraints. The analysis reveals that staff deal with dual involvements by inserting and extending conversations within time-limited operation sequences, while simultaneously adjusting their bodily orientation to remain engaged in both ride operation and interaction with children.

This study analyzes the characteristics of a child's directive speech acts from the perspective of decontextualization. The analysis targets family conversation data from the "CEJC-Child." A previous study has shown that various rhetorical functions are used in parents' utterances classified under the label "Directive Speech Act" in dialogue act information. This study focused on the utterances of a kindergarten child that has a younger brother and found that the main directive expressions were primarily low-level decontextualization expressions based on the child's own desires at that time and place. On the other hand, it was also found that there were expressions with different degrees of decontextualization that were conscious of being an older sibling.

In recent years, there has been growing interest in developing dialogue robots capable of building close relationships with humans. Prior research in human interpersonal communication suggests that sharing negative opinions about others can play a critical role in fostering interpersonal closeness. Based on this insight, we developed a dialogue robot that shares negative opinions about third parties with users, and investigated whether such behavior would promote closeness in human-robot interaction. Contrary to our hypothesis, the results of a dialogue experiment revealed no significant effect of the robot's negative opinion sharing on relationship formation. We discuss a possible reason for this discrepancy: negative statements made by robots about others may be perceived as less ethically acceptable than when made by humans. These findings offer important insights for the design of dialogue robots aimed at fostering close relationships with users.

The purpose of this study is to realize a dialogue system that dynamically selects and shares information that is estimated to be of high interest to the user from among a number of target person's information. By integrating our interest estimation model and RAG framework, we have developed a LINE chatbot that shares information about target person that may be interesting to the user according to the content of the dialogue. As a result of the dialogue experiment, the proposed chatbot could increase the user's interest in the target person and improve the user's willingness to build their relationship.

This study examines action repair in the context of temae movements in the Japanese tea ceremony, analyzing how practitioners return to normative procedures following disfluencies or errors in movement. Although temae follows highly formalized protocols, interruptions and mistakes can occur during actual practice, and a sequence of behaviors aimed at correcting these disruptions is often observed.In this study, such corrective processes are defined as action repair, and concrete examples were extracted from video data recorded by the 1st author. In this study, such corrective processes are defined as action repair, and concrete examples were extracted from video data recorded by the 1st author. Drawing on the concept of repair from conversation analysis (Schegloff et al., 1977) and the Repair Interval Model (Nakatani & Hirschberg, 1993), we analyzed the tendencies and characteristics of repair behaviors based on Hirose's research on microslips. As a result, we identified repair forms unique to the tea ceremony, as well as cases where normative actions and repair processes co-occur, providing insights into the nature of natural, embodied repair.

This study examines how signers elicit responses from recipients at a possible completion point of turn constitutional unit (TCU) during a telling in Japanese Sign Language (JSL) conversations. The analysis focuses on the recipients' responses at possible completion of TCUs and how the signers' behaviors differ depending on whether such responses, such as nods, are displayed before the TCU reaches its completion. The findings reveal that when the recipient has already responded before the possible end of the TCU, the signer proceeds directly to the next TCU. In contrast, when no response has been given, the signer tends to hold the final handshape of the TCU and only releases it to begin the next TCU after the recipient displays a response. This suggests that in JSL, the holding of a handshape at the end of a TCU can serve as a means for prompting a recipient's response.

In recent years, the collaborative use of multiple large language models (LLMs) has attracted increasing attention for its potential to enhance problem-solving capabilities and enable high-fidelity social simulations. While these efforts have yielded valuable insights, a unified understanding of the underlying phenomena requires clarifying the mathematical principles that govern the collective behavior of LLMs.In this study, we systematically examine the dynamics of LLM agent populations through a framework grounded in statistical mechanics. Focusing on consensus formation via dialogue between agents holding opposing views, we uncover phase transitions induced by variations in network structure and agent personality. Moreover, we introduce a method for decomposing pairwise agent dynamics into interpretable components that capture individual personality traits.

Realizing joint action based on Theory of Mind (ToM) is an important step for enabling large language models (LLMs) to collaborate effectively with humans. However, autonomously generating action plans that take into account the perspectives and beliefs of others remains a challenging problem for LLMs. In this study, we aim to achieve autonomous joint action with embodied LLM agents equipped with ToM, and propose a decentralized sequential decision-making method, ToM-JAC (Joint Action Control with Theory of Mind), which plans and adapts actions online. We demonstrate the effectiveness of the proposed method through two-agent joint action tasks in a Minecraft environment.