Workshop of Human-Agent Interaction 2026 (HAI 2026) in Osaka, Japan.
As artificial intelligence becomes deeply integrated into our daily lives, designing truly inclusive Human-Agent Interaction (HAI) is an urgent societal imperative. Historically, HAI research has been heavily skewed towards hearing-centric spoken dialogue systems, often marginalizing Deaf and hard-of-hearing users.
The SignHAI 2026 workshop challenges this paradigm by shifting the research focus from isolated sign language recognition to dynamic, multimodal sign language dialogue. We emphasize the critical role of non-verbal conversational dynamics, specifically turn-taking, in facilitating seamless human-agent interactions.
By bridging HAI researchers, multimodal AI experts, and the Deaf community, SignHAI 2026 aims to establish new technological baselines, provide crucial benchmark resources, and set a strategic roadmap for the next generation of inclusive multimodal dialogue foundation models.
[Accessibility Note] To ensure an inclusive environment, we are planning to provide online International Sign (IS) interpretation during the workshop.
Invited Talk information is updated (October 16th, 2026)
The Challenge Track dataset is now available, and the track description has been updated (September 29th, 2026)
Information on presentation submission is updated (September 22nd, 2026)
Our workshop has been accepted to HAI 2026 and the website has been opened !! (July 28th, 2026)
We invite presentation submissions for our workshop across the two tracks outlined below. Presenters are requested to submit a title and an abstract, which will be reviewed by the organizers. Accepted titles and abstracts will be posted on the workshop website (here); please note that no formal proceedings or papers will be published (non-archival).
Challenge Track (Semi-Competition)
This track invites participants to propose and explore tasks using our newly released signing activity annotations for the Japanese Sign Language Colloquial Corpus (JSL corpus). Possible tasks include signing activity detection, future signing activity prediction (projection), and interactional timing analysis. The annotations are available in the dataset repository. We also provide an example implementation of projection, including pretrained models and code for pose extraction, preprocessing, inference, and evaluation. Participants must obtain the original JSL corpus separately through the official application procedure and comply with the corpus’s terms of use.
General Track
We welcome research papers, position papers, and late-breaking results covering a broader range of topics, including sign language translation, non-verbal behaviors in visual dialogue, and inclusive agent design.
Topics of interest include, but are not limited to:
Inclusive Human-Agent Interaction (HAI)
Sign language dialogue and multimodal dialogue foundation models
Non-verbal behaviors in visual dialogue (e.g., eye gaze, preparatory phases of articulation)
Computational modeling and prediction of turn-taking in sign language
Sign language recognition and translation
AI accessibility for the Deaf and hard-of-hearing communities
Evaluation metrics and ethical considerations for inclusive agent design
Presentation Submission October 14th, 2026
Notification October 18th, 2026
Workshop Day November 16th, 2026 (afternoon)
※ We are going to maek the notification to make it earlier than the early bird registration of HAI 2026 main conference.
We invite presentation submissions for our workshop. Presenters are requested to submit a title and abstract, which will be reviewed by the organizers. Accepted titles and abstracts will be posted on our website; no any papers or other materials will be included in formal proceedings.
To submit your presentation, please use the form below:
https://forms.gle/2R33tDUyUGHAxLqx5
Please also note that all presenters need to register for the HAI main conference.
Tsubasa Uchida (NHK Science and Technology Research Laboratory)
NHK has been developing sign language avatar technologies to expand the availability of sign language services. Our earlier systems generated Japanese Sign Language (JSL) avatar videos from predefined sentence templates and have been applied to practical services for weather information and live sports commentary, as well as to field trials for transportation information. Building on these systems, our current research focuses on a translation-based approach that automatically generates avatar videos from Japanese breaking-news text. The proposed end-to-end pipeline translates Japanese text into JSL glosses, assigns auxiliary motion-control labels to the glosses, and then blends and renders the corresponding JSL motions. The assigned labels specify spatial placement, facial expressions, and mouthing, enabling the avatar to reproduce context-dependent grammatical features of JSL. This talk will review NHK’s development and practical use of template-based systems and introduce our latest translation-based system. It will present findings from a technical evaluation of system performance and an evaluation of the generated JSL avatar videos involving Deaf and hard-of-hearing sign language users. These findings can contribute to expanding access to information in sign language through avatar technology.
References:
NHK STRL Annual Report 2025: Sign language CG production technologies
NHK STRL Open House 2026: Expanding Sign Language CG Content Using AI
◼ Biography
Tsubasa Uchida is a researcher at NHK Science and Technology Research Laboratories, where he develops sign language avatar systems and accessibility technologies for broadcasting. He received his M.E. degree in Media and Telecommunications Engineering from Ibaraki University in 2011 and joined NHK in the same year. After working at the NHK Sendai Broadcasting Station, he was transferred to the research institute in 2014. His research interests include automatic sign language generation, avatar animation, and accessible information delivery. From May to November 2023, he was a visiting researcher at the Institute of German Sign Language and Communication of the Deaf at the University of Hamburg in Germany.
Our workshop is tentatively scheduled for the afternoon of 16th November.
We have accepted following presentations:
Challenge Track:
Beyond Speech: Evaluating Voice Activity Projection for Sign Language Turn-Taking Detection
Ninlawat Phuangchoke (Asian Institute of Technology), Mayumi Bono (National Institute of Informatics)
Detecting Signing Activity and Sentence-Level Segmentation in JSLCorpus
Ozge Mercanoglu Sincan (University of Surrey), Sobhan Asasi (University of Surrey), Richard Bowden (University of Surrey)
Signing Activity Annotations for the JSL Corpus
Takao Obi (Institute of Science Tokyo), Kasidit Sarnrak (Institute of Science Tokyo), Kamolpat Thananopavarn (Institute of Science Tokyo), Koji Inoue (Kyoto University), Kotaro Funakoshi (Institute of Science Tokyo)
General Track:
Two-part Motion Diffusion Model for Generating 3D Sign Motion of Avatars
Kohei Hakozaki (NHK Foundation), Tomoya Murakami (Japan Broadcasting Corporation), Tsubasa Uchida (Japan Broadcasting Corporation), Masaru Miyazaki (Japan Broadcasting Corporation)
Pose-based pre-annotation of mouth actions in sign language conversations under annotator disagreement
Biplav Sharma Regmi (Asian Institute of Technology), Tomohiro Okada (National Institute of Informatics), Mayumi Bono (National Institute of Informatics), Shin'ichi Satoh (National Institute of Informatics)
Visualizing Turn-End Prediction to Support Hard-of-Hearing Participation in Casual Conversation: A Wizard-of-Oz Study of Triadic Conversation
Hina Ishihama (Tsukuba University of Technology), Kohei Kikuchi (Tsukuba University of Technology), Chiemi Watanabe (Tsukuba University of Technology)
Deaf-led Data Creation for Sign Language Conversation Analysis and AI: Practices and Challenges of a Deaf-led Framework
Naoki Asada (National Institute of Informatics), Tomohiro Okada (National Institute of Informatics), Mayumi Bono (National Institute of Informatics)
Turn-Taking Across Modalities: Temporal Coordination in Speech-to-Finger Braille Interpretation and Implications for Inclusive Dialogue AI
Ryosaku Makino (Musashi University), Haruka Abe (University of Tsukuba), Kanato Ochiai (National Institute for Japanese Language and Linguistics), Mayumi Bono (National Institute of Informatics), Satoshi Fukushima (The University of Tokyo)
Hard-Negative Evaluation for Sign Language Video–Text Matching
Marshall Thomas (University of Surrey)
Each presentation is going to be assigned at least 15 minutes including Q&A (10 min. for presentation and 5 min. for Q&A).
We will make the program soon.
Takao Obi (Institute of Science Tokyo)
Kazuhiro Nakadai (Institute of Science Tokyo)
Yutaka Osugi (Tsukuba University of Technology)
Mayumi Bono (National Institute of Informatics)
This workshop is supported by JST ASPIRE (Grant number: JPMJAP25B3) and JST PRESTO (Grant number: JPMJPR24I4)