Making Meaning of Psychotherapy using Natural Language Processing Restricted; Files Only

Abstract

Decades of research support the effectiveness of psychotherapy for creating lasting relief from symptoms, improving interpersonal relationships, and boosting overall wellbeing. Despite extensive research that psychotherapy works, little is known about its mechanisms. This dearth of knowledge is largely due to the nature of psychotherapy data, which consists of a dialogue between two individuals. Direct assessment of psychotherapy content via hand-coding of relevant features in psychotherapy dialogue is labor intensive, expensive, and difficult to implement at scale. Natural language processing (NLP), a subfield of computer science, offers promising methods for analyzing psychotherapy dialogue. In Study 1, we used a large psychotherapy corpus to extract clinically relevant features from dialogue in a data-driven manner, characterized previously unexplored within-session temporal dynamics, and probed relationships between client speech and presenting concern. We found support for the use of neural topic modeling to generate fine-grained emotions and symptoms directly from text and found associations between affective features of dialogue and presenting concerns. In Study 2, we collected psychotherapy dialogue and self-report measures of symptoms, session ratings, and therapeutic alliance from a psychotherapy training clinic. We found diverging within-client and between-client associations with symptoms and linguistic features: negative emotions were associated with symptom fluctuations at the within-client level, while lower positive emotions were associated with symptoms at the between-client level. Anger emerged as an important dialogue feature in several ways: higher client anger scores were associated with better session ratings, while higher therapist anger scores were associated with better client-rated alliance on and next-session symptom improvements. We note therapist anger was not directed at clients, but typically contained expressions of anger on their behalf or the validation of anger. Lastly, in Study 3, we offer an initial case study highlighting the potential of large language models for use in psychotherapy process research. We adapted an observational coding scheme of multi-theoretical therapist interventions and generated annotations on psychotherapy transcripts using GPT-4. Findings highlight the use of NLP to study psychotherapy dialogue to describe psychotherapy in practice, identify linguistic markers of symptoms, and isolate potential mechanisms of therapy.

Table of Contents

General Introduction........................................................................................................................1 Study 1

Introduction..........................................................................................................................7 Methods................................................................................................................................9 Results................................................................................................................................13 Discussion..........................................................................................................................23

Study 2 Introduction........................................................................................................................28 Methods..............................................................................................................................29 Results................................................................................................................................38 Discussion..........................................................................................................................59

Case Study: Annotating Therapist Interventions with a Large Language Model Introduction........................................................................................................................63 Methods..............................................................................................................................65 Results................................................................................................................................69 Discussion..........................................................................................................................78

General Discussion........................................................................................................................81 References......................................................................................................................................87 Appendix......................................................................................................................................104 

About this Dissertation

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Laney Graduate School
Department	Psychology
Subfield / Discipline	Clinical Psychology
Degree	Ph.D.
Submission	Dissertation
Language	English
Research Field	Psychology, Counseling Psychology, Clinical
Palabra Clave	NLP psychotherapy
Committee Chair / Thesis Advisor	Michael T. Treadway, Emory University
Committee Members	Linda Craighead, Emory University Phillip Wolff, Emory University Jinho D. Choi, Emory University Christian Webb, McLean Hospital/Harvard Medical School

Última modificación

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	File download under embargo until 21 August 2026	2024-07-25 21:54:45 -0400	File download under embargo until 21 August 2026

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions