New papers probe LLM turn-taking and social norm adaptation in group chats

Arxiv CS.CL (NLP/language models)

Four papers examine LLMs' abilities for addressee detection, turn-change prediction, next speaker prediction, and endpoint detection in multiparty conversations. Proposals include a multimodal voice activity projection framework and a benchmark for local social norm adaptation.

LLMturn-takingmultiparty-conversation

See 3 more sources

Next-Turn: Duration-Aware Streaming Endpoint Detection via Time-to-Next-Speech-Onset Prediction1 day agoTristan Tsoi, Jiajun Deng, Yingke Zhu, Huu Quyen Dang, Tianxiang Cao, Nikita Kuzmin, Tao Zhong, Simon Lui

MuVAP: Multimodal Multiparty Voice Activity Projection for Turn-taking Prediction in the Wild2 days agoHaotian Qi, Gabriel Skantze

LoSoNA: A Benchmark for Local Social Norm Adaptation in Group Conversations3 days agoMateusz Winiarek, Maksymilian Bilski, Mateusz Jacniacki

Arxiv CS.CL (NLP/language models)·Ryo FukudaTakatomo KanoSiddhant AroraMarc DelcroixNaohiro TawaraAtsunori OgawaYuya ChibaAtsushi AndoWilliam ChenShinji Watanabe

Jun 17, 4:00 AM