Search papers, labs, and topics across Lattice.
This paper introduces TIDES, a longitudinal bilingual dataset that captures the complexities of multi-party social dynamics by tracking 12 university project teams over a semester, resulting in 75,971 utterances in English and Korean. The dataset's socio-structural annotations enable the modeling of team evolution, and experiments reveal that fine-tuning on TIDES significantly enhances next-speaker prediction performance compared to a bigram baseline. Despite these improvements, human evaluations indicate that higher prediction accuracy does not correlate with more natural or coherent utterances, highlighting a critical area for future research in multi-party conversation modeling.
Fine-tuning on TIDES boosts next-speaker prediction accuracy by 13.8 percentage points, yet paradoxically leads to less preferred utterances in human evaluations.
Group conversations are fundamental to human collaboration, yet standard large language models (LLMs) still struggle with the complexities of multi-party interaction. This challenge persists in part because existing group conversation datasets are often limited to short-term lab settings with contrived tasks, failing to capture the long-term social dynamics of real-world teams. To bridge this gap, we introduce TIDES, a high-resolution longitudinal dataset tracking 12 university project teams over a full semester. Comprising 75,971 utterances in both English and Korean from in-person meetings, TIDES provides a naturalistic record of teams working on self-managed projects. Our socio-structural annotations-covering interaction types, emergent roles, and development stages-allow for modeling of team evolution over months. Experiments show that fine-tuning on TIDES improves next-speaker prediction by 13.8 percentage points over a bigram baseline (64.53%) and yields performance comparable to strong proprietary zero-shot models. The model also comes within 2.1 percentage points of the published state of the art on the AMI Meeting Corpus while using approximately 42% less training data. However, human evaluations suggest that better next-speaker prediction does not necessarily yield more natural or coherent utterances, as fine-tuned models were generally less preferred than vanilla models. This potential mismatch motivates further study of how structural modeling can support natural multi-party generation.
