Using Large Language Models to Understand Thought Disorder and Predict Psychosis Open Access

Abstract

Thought disorder (TD) is defined as any internal, cognitive disturbance affecting the organization, control, processing, or expression of thought. It is a key characteristic of schizophrenia and psychosis spectrum disorders. Two kinds of TD are widely recognized in schizophrenia and are typically measured through patients’ language. One is reflected in local language disorganization, caused by unusual word choices and sequencing. The other concerns non-local disruptions in the flow of ideas across multiple sentences, implying conceptual disorganization. Prior work on TD using computational methods has focused on local disruptions. Here we propose a novel approach to the detection and measurement non-local thought disorganization. We used a new class of artificial intelligence called large language models (LLMs) to examine disorganization across sentences by measuring its ability to predict sentences subjects spoke during their interviews. The model was provided with patient narratives of individuals at clinical high-risk (CHR) for psychosis. The narratives were provided to the model either intact or with the sentence order shuffled. If people’s narratives lack global organization, then shuffled narratives should be as effective as intact narratives in at facilitating the prediction of sentences. As expected, intact narratives were more effective than shuffled narratives in facilitating sentence prediction in CHR individuals who did not convert to psychosis (CHR-) than in individuals that converted to psychosis (CHR+). Also as predicted, the LLM was less able to predict individual words in CHR+ than in individuals CHR-. These findings support and expand upon work suggesting that conversion to psychosis is signaled by both local and non-local disruptions in thought.

Table of Contents

Background

Methods

Participants

Measures and Protocol

Procedures

Design

Results

Discussion

References

Supplementary materials:

Large Language Models (LLMs):

Subsetting CHR- :

About this Master's Thesis

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	M.A.
Submission	Master's Thesis
Language	English
Research Field	Psychology, Clinical Psychology, Cognitive
Keyword	large language models thought disorder artificial intelligence psychosis
Committee Chair / Thesis Advisor	Phillip Wolff, Emory University

Last modified

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Using Large Language Models to Understand Thought Disorder and Predict Psychosis ()	2023-07-17 11:18:58 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions