We analyze over 20,000 transcripts of real behavioral treatment sessions to develop a measure for patient-therapist synchrony.  We achieve this by using large language models, which enable extracting intricate features without the need for human annotation. These tools, which have shown a promising ability to capture both the subtle and core aspects of conversations, allow us to (1) conduct ecological research with big data and (2) develop novel semantic measures that capture the dynamics and content in free-speech therapeutic conversations. To validate our measures, we assess their correlation with objective metrics and other semantic features such as the number and duration of sessions, patient’s expressions of positivity and negativity, phrases of agreement, and word count. Our goal for this project is to establish a framework for conversation analysis in psychotherapy and other fields, introducing new metrics that advance the ecological research of psychotherapy.

Supervisor: Ariel Goldstein