Self-Other Dynamics (SOD): A Transtheoretical Coding Manual

Building on both modern and classical psychotherapy theories, I propose the Self-Other Dynamics (SOD) approach. This transtheoretical approach integrates key elements of patients' self-states, therapists' interventions, and their dyadic interactions, with the goal of identifying adaptive and maladaptive dynamics associated with psychotherapy outcomes. The vast number of self-states, the dynamic contexts in which they may emerge, and interventions that therapists may utilize to address them have presented a formidable challenge that research methodologies have struggled to tackle until recently. Recent advancements in artificial intelligence (AI), particularly the development of large language models (LLMs), offer substantial promise in addressing this challenge. The goal of this project is to leverage LLM capabilities to automatically detect a patient's self-state and recommend the most appropriate intervention to enhance their well-being. To achieve this goal, I developed the SOD coding scheme, which integrates micro-level processes of patients, therapists, and dyads that are considered central according to a wide range of therapeutic approaches. Using this scheme, patient-therapist dialogues will be annotated speech turn by speech turn. Initially, a subset of the data will be manually annotated using the SOD coding scheme. Subsequently, LLMs will be trained to automatically annotate a larger dataset, enabling the scaling up of categories and providing clinical insights into adaptive and maladaptive self-other dynamics. This paper describes the work in progress, beginning with the background of the SOD theoretical approach, followed by detailed guidelines for the coding scheme. Preliminary experiments indicate that LLMs can reliably annotate psychotherapy data at a level comparable to human reliability.