The study aims to identify linguistic markers that indicate the therapeutic alliance between a patient and clinician in psychotherapy sessions. To do so, the study used natural language processing (NLP) to quantify first-person pronoun usage and non-fluency language markers, which may be relevant to the communicative and emotional aspects of the therapeutic relationship. The results suggest that therapists' first-person pronoun usage frequency and patients' speech transition marking relaxed interaction style could be potential metrics of alliance. Behavioral data from an economic game that measures social exchange also indicated that therapists' first-person pronoun usage may influence alliance ratings through their diminished trusting behavior towards therapists.
The therapeutic alliance is challenging to assess due to the difficulty in finding objective markers that underlie patients' subjective experience of closeness with clinicians. Currently, gold standard assessments of alliance rely on either self-reports or human observers' qualitative coding of treatment interactions, which are subjective, labor-intensive, and time-consuming. Efforts have been made to assess alliance using machine learning, including patient-clinician language use, head and body movements, facial expressions, respiration rate, heart rate variability, and brain activities, but many of these approaches fail to provide interpretability.
The study aimed to address this gap by identifying personal pronoun usage and non-fluency as communicative function markers of both patients and therapists from single-session transcripts and regressing these features on post-session alliance scores rated by the subjects. Overall, the study shows that communicative language features in patient-therapist dialogues could be markers of alliance and highlights the potential of machine learning to improve clinical outcomes through the timely identification of negative session-level alliance.