Tracking Shifts in the Latent Meaning of “I” and Its Connection to Mental Health Outcomes in a Large Therapy Dataset

This project examines how the semantic meaning of first-person singular pronouns (e.g., “I,” “me,” “my”) changes over the course of psychotherapy and how these shifts relate to mental health outcomes. Using natural language processing and contextual word embeddings applied to a large corpus of psychotherapy transcripts, we track how clients’ self-referential language moves along two theoretically meaningful dimensions—good vs. bad and able vs. unable—across treatment. We test whether changes in these semantic representations predict reductions in depression, anxiety, and internalizing symptoms at both within-person and between-person levels, above and beyond known linguistic confounds (e.g., pronoun frequency, sentiment, word count). Analyses are preregistered on a holdout validation sample from a large digital mental health platform, with the goal of clarifying how evolving self-representation in language reflects psychological change during therapy.