Dataset for “SpecialTime: Automatically Detecting Dialogue Acts from Speech to Support Parent-Child Interaction Therapy”

Parent-child interaction therapy (PCIT) helps parents improve the quality of interaction with children who have behavior problems. The therapy trains parents to use effective dialogue acts when interacting with their children. Besides weekly coaching by therapists, the therapy relies on deliberate practice of skills by parents in their homes. We developed SpecialTime, a system that provides parents engaged in PCIT with automatic, real-time feedback on their dialogue act use. To enable automatic classification of spoken dialogue acts in parent-child interaction, we created a dataset of 6,022 parent dialogue acts, annotated by experts with dialogue act labels that therapists use to code parent speech. We developed an algorithm that classifies the dialogue acts into 8 classes with an overall accuracy of 78 %.

亲子互动疗法（Parent-child interaction therapy, PCIT）可帮助家长改善与存在行为问题儿童的互动质量。该疗法会训练家长在与子女互动时运用有效的对话行为（dialogue act）。除了治疗师提供的每周督导外，该疗法还依赖家长在家中对所学技能进行刻意练习。我们研发了名为SpecialTime的系统，可为参与亲子互动疗法的家长提供针对其对话行为使用情况的自动化实时反馈。为实现亲子互动场景中口语对话行为的自动分类，我们构建了包含6022条家长对话行为的数据集，该数据集由专家使用治疗师用于标注家长言语的对话行为标签（dialogue act label）进行注释。我们研发了一种可将对话行为划分为8个类别的算法，其总体准确率可达78%。