TEIDAN: A Multilingual Multiparty Dialogue Corpus
Taiga Mori, Koji Inoue, Mikey Elmers, Divesh Lala, Tatsuya Kawahara
cs.CL · cs.HC
Quick verdict
TEIDAN: 69-session JA/EN triadic dialogue corpus, well documented and honestly caveated; data release unconfirmed and cross-language claims stay descriptive.
Summary
This ICMI Companion '26 resource paper presents TEIDAN, a multimodal corpus of spontaneous face-to-face three-party conversation, and offers its first standalone corpus-level description spanning Japanese and English. The Japanese portion (12 groups, 36 sessions, 24 acquainted lab members, six topics, 3:39:10) previously served only as task benchmarks; the new English portion (11 groups, 33 sessions, 33 strangers recruited via a staffing agency across 13 nationalities, three parallel topics, 6:07:46) brings the corpus to 69 sessions, 57 participants, and 9:46:57 of multimodal interaction (pin microphones, microphone array with DOA labels, cameras), transcribed as IPUs (≥200 ms pauses) in a CSJ-derived format tagging backchannels, fillers, self-interruptions, and vocalizations. Preliminary descriptive statistics report English pseudo-turns roughly 1.6× longer than Japanese ones (4.65 s vs 2.95 s; 2.08 vs 1.55 IPUs/turn), and conversation-analytic excerpts contrast long English opinion floors with compact Japanese listener contributions at non-TRP positions. The authors candidly flag the core confound — language entangled with participant relationship (acquainted vs strangers) and session format — and explicitly avoid inferential claims. Main limitations: dataset release is merely 'being considered' (no repository, DOI, or license); statistics are pooled means without per-group variance or sensitivity checks; and only three of six Japanese topics have English counterparts, with reworded prompts.
Review complete
Assessment at a glance
4
risk flags identified
● high 0
● medium 2
● low 2
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- TEIDAN is the first multilingual multimodal corpus supporting comparable analysis of spontaneous face-to-face triadic discussion, currently 69 sessions, 57 participants, and 9:46:57 across Japanese and English.
- The newly collected English portion (11 groups, 33 strangers, 13 nationalities, 3 parallel topics) extends the previously benchmark-only Japanese data into a cross-linguistic resource under a shared triadic protocol.
- Preliminary pooled statistics show English pseudo-turns are longer (mean 4.65 s vs 2.95 s) and contain more IPUs (2.08 vs 1.55) than Japanese ones.
- Qualitative analysis suggests Japanese participants build opinions incrementally with compact listener contributions at non-TRP positions, while English participants typically hold extended floors for opinion statements.
Strengths
First standalone corpus-level description of a bilingual triadic resource
Consolidates Japanese subsets previously used only as task benchmarks (refs [2],[6],[11]) with a new English extension into one documented resource: 69 sessions, 57 participants, 9:46:57 (Table 1), with per-session durations and IPU counts fully enumerated in Table 2 (12,384 JA + 17,370 EN IPUs).
Unusually candid confound disclosure
§2.2 and §5 explicitly state that the JA/EN contrast entangles language with participant relationship (acquainted lab members vs agency-recruited strangers) and session format; §4.1 twice disclaims inferential reading of pooled statistics — rare epistemic honesty for cross-cultural claims in this area.
Well-specified multimodal and transcription infrastructure
Quantitative-to-qualitative bridge
Careful ethics handling
Weaknesses
Corpus availability unconfirmed
Headline cross-linguistic contrast is confounded
Descriptive statistics without dispersion or robustness
'Parallel protocol' is only partial
Limited novel content and QA reporting
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
TEIDAN is a resource paper: the first standalone, corpus-level description of a multimodal corpus of spontaneous face-to-face three-party conversation in Japanese and English (ICMI Companion '26, 8 pages). The claims are appropriately modest: (i) TEIDAN spans 69 sessions, 57 unique participants, and 9:46:57 of interaction (Table 1); (ii) the newly collected English portion (11 groups, 33 sessions, 6:07:46) enables cross-linguistic comparison with the previously benchmark-only Japanese portion (12 groups, 36 sessions, 3:39:10, statistics reused from ref [11]); (iii) preliminary analyses illustrate uses for turn-taking, addressee recognition, next-speaker prediction, and multimodal grounding. No new method or model is claimed; the contribution is documentation plus demonstration.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 4 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is the TEIDAN corpus?
TEIDAN is a multimodal research corpus of spontaneous three-person face-to-face discussions, currently 69 sessions (36 Japanese, 33 English) with 57 participants and about 9 hours 47 minutes of audio-video. It was recorded with individual pin microphones, a microphone array, and cameras, and comes with IPU-based transcripts (speech segments bounded by pauses of at least 200 ms).
Is the TEIDAN dataset publicly available?
Not yet. The paper states the team is 'considering releasing the dataset' under conditions respecting participant consent and privacy, including per-participant opt-outs for video or audio+video redistribution; no repository, DOI, or license is specified at publication time.
What did the TEIDAN paper find about Japanese vs English conversation?
Descriptive comparisons suggest English pseudo-turns (speaker turns after filtering backchannels) average 4.65 s with 2.08 IPUs each, versus 2.95 s and 1.55 IPUs in Japanese. English sessions show long opinion-holding turns; Japanese excerpts show listeners making brief cooperative interventions before a speaker's turn is complete. The authors caution that language is confounded with participant relationship (acquainted lab members vs recruited strangers).
What was TEIDAN used for before this paper?
Subsets of the Japanese portion served as benchmarks for addressee recognition, next-speaker prediction, and voice activity projection in multi-party dialogue modeling. This paper is the first corpus-level description covering both the Japanese and the newly collected English portions.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figure 1 (recording layout, top and front views) is an image diagram; its content could not be assessed from the text extraction.
- The excerpt transcripts and English translations are author-provided; accuracy of the underlying Japanese audio transcription cannot be independently checked from text alone.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.