The 2026 ACII Dyadic Conversations (DaiKon) Workshop & Challenge
The 2026 ACII Dyadic Conversations (ACII-DaiKon) Workshop & Challenge aims to set a new benchmark in the field of interpersonal affect and social dynamics during dyadic conversations. As advancements in conversational affect modeling have accelerated, existing benchmarks often focus primarily on individual speakers, failing to adequately capture the intricate, coupled dynamics between conversational partners. This workshop seeks to fill that gap by introducing a comprehensive framework that emphasizes the complexities of directional influence, timing coordination, and rapport development.
Overview of the Workshop
The ACII-DaiKon Workshop will feature three coordinated sub-challenges that utilize a shared dataset known as the Hume-DaiKon dataset. This invaluable resource consists of:
- 945 dyadic conversations
- A total of 743.4 hours of audiovisual data
- Data collected under naturalistic conditions across five different languages
Each sub-challenge is designed to address specific facets of dyadic interaction:
- Directional Interpersonal Influence Prediction: This sub-challenge focuses on understanding how one partner’s affective state may influence the other’s during conversation.
- Turn-Taking Prediction: Participants will predict who will speak next and the timing of that speech, encompassing both next-speaker identification and time-to-next-speech estimates.
- Rapport Trajectory Prediction: This challenge aims to model how rapport between conversation partners evolves throughout their interaction.
Benchmark Features and Evaluation Metrics
The ACII-DaiKon benchmark supports various modeling approaches, including multimodal analysis, temporal reasoning, and cross-context generalization. To facilitate rigorous evaluations, the benchmark includes fixed train/validation/test splits and standardized metrics. Evaluation will employ several metrics tailored to each sub-challenge, such as:
- Concordance Correlation Coefficient (CCC)
- Pearson correlation
- Macro-F1 score
- Mean Absolute Error (MAE)
Baseline experiments have provided initial performance benchmarks, revealing promising yet challenging results. For instance, the best test results for each sub-challenge include:
- 0.40 CCC and 0.50 Pearson for influence prediction
- 0.66 Macro-F1 and 1.50 seconds MAE for turn-taking prediction
- 0.68 CCC and 0.70 Pearson for rapport trajectory modeling
These results indicate that while existing methods can capture broad dyadic patterns, there remains significant difficulty in robustly modeling directional dependence and long-term interpersonal dynamics.
Importance of the Workshop
The ACII-DaiKon Workshop serves as a vital platform for fostering collaboration and discourse among researchers across various disciplines. Participants will engage in discussions about:
- Data validity and its implications for modeling
- Evaluation protocols that can enhance comparability
- Culturally aware modeling techniques for dyadic interactions
This workshop not only aims to advance the field of affective computing but also to pave the way for future research that can lead to more nuanced and accurate representations of human conversational dynamics.
Related AI Insights
- DataClaw: Benchmark for Exploratory Real-World Data Analysis
- TechCrunch Disrupt 2026: Key M&A Insights for Startups
- Foundation-Model Agents in Industrial Automation: Capabilities & Challenges
- DRLU-Based Semantics for Quantitative Bipolar Argumentation
- Ethos Secures $22.75M for Voice-Enabled Expert Network
- Efficient Temporal Datalog for Real-Time Event Recognition
- Anthropic’s Claude AI Agents Now Feature Creative ‘Dreaming’
- Cost-Effective Vision-Language Models for Pet Detection on AWS
- Apple Settles $250M Lawsuit Over Siri AI Delays
- 5 Easy Tips to Make Zorin OS Faster & More Efficient
