NaturalTurn: a method to segment speech into psychologically meaningful conversational turns

Area of Science:

• Computational Linguistics
• Social Sciences
• Cognitive Science

Background:

• Analyzing conversational data is crucial for social, cognitive, and computational sciences.
• Existing methods struggle to accurately segment speech-to-text transcripts into conversational turns, hindering research.
• Scalable methods are needed to process large, complex conversational datasets.

Purpose of the Study:

• To introduce NaturalTurn, an algorithm for accurate conversational turn segmentation.
• To differentiate primary speaker turns from secondary listener utterances (e.g., backchannels).
• To enable deeper analysis of conversational dynamics and their relationship with social outcomes.

Main Methods:

• Developed the NaturalTurn algorithm to distinguish primary and secondary conversational utterances.
• Applied NaturalTurn to a large conversation corpus.
• Compared NaturalTurn's performance against a baseline segmentation model.

Main Results:

• NaturalTurn demonstrated higher accuracy in capturing conversational turns compared to the baseline.
• The algorithm produced turn durations and gaps consistent with empirical literature.
• NaturalTurn revealed stronger linguistic alignment and uncovered hidden links between turn-taking and affective outcomes.

Conclusions:

• NaturalTurn offers a pragmatic advancement in processing conversational transcripts.
• This method facilitates research linking turn-taking dynamics to significant social interaction outcomes.
• Enables researchers to better understand the structure and impact of human conversation.