研究人与数字代理在多人对话中的语言同步，发现互动受限且因群体而异。
Speech Entrainment in Multi-Party Conversations with a Digital Agent
- 通过真实多人对话数据，分析人类与数字代理的语言同步机制。
- 人类对身边对话者有局部同步，但整体和与代理的同步较弱。
- 家庭与成人组表现不同，揭示群体差异对互动的影响。
已有研究表明，对话中的个体倾向于调整自身说话风格以匹配对方。然而，以往研究多聚焦于人与人之间的二元互动。本文首次探究了由人类群体与数字代理共同参与的多方对话中的语言同步现象。该场景有助于理解非人类参与者如何影响对话的短期与长期动态。为此，我们收集并分析了一个独特数据集，包含成人组与亲子家庭对话会话，从而比较不同群体中同步效应的差异。我们考察了多种基于知识与模型的同步特征，发现尽管个体能与他人局部同步，但整体层面以及与数字代理的同步仍较为有限，且受群体类型影响显著。
原文摘要 · Abstract (English)
It has been widely observed that individuals engaged in conversation tend to adapt their speaking style to more closely match the other interlocutors. However, most prior work has focused on dyadic interactions among humans. In this paper, we investigate entrainment effects in a novel setting: a multi-party interaction between groups of humans and a digital agent. This scenario offers important insights into how the participation of non-human actors modulates both short-time and temporally resolved conversational dynamics. To address these questions, we collect and analyze a unique dataset that consists of both adult and family (parent/children) sessions, enabling us to examine how entrainment manifests differently across these cohorts. We consider a range of knowledge-driven and model-based entrainment features and find that, while individuals locally entrain with other humans, global entrainment, and entrainment with the agent remains limited and cohort-dependent.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。