Linguistic Shifts Around Phase Transitions in Psychotherapy: A Computational Exploration

Abstract

Introduction

The use of language as an objective indicator of mental illness has gained increasing recognition [1]. In psychotherapy, linguistic shifts often reflect key changes in a person’s mental state. Our research examines how linguistic features from clients’ diaries during inpatient psychiatric care correspond to phase transitions (PTs)—points of significant shifts in mood, insight, or symptom severity [2]. Despite progress in time-series analysis of quantitative psychotherapy data, the content of qualitative sources remains largely underexplored. We address this gap by analyzing the content-specific nature of PTs.

Methods

The dataset comprised Therapy Process Questionnaire (TPQ) results and textual diary entries from 27 clients. The TPQ provided time series data annotated with clinically identified PTs. Text data were processed using the Linguistic Inquiry and Word Count (LIWC) tool, resulting in 117 extracted features related to psychological and linguistic categories. To detect changes in these features within 3-, 5-, and 7-day windows around each client's PTs, we used the Wilcoxon Signed-Rank Test with a higher threshold (p < 0.15) appropriate for exploratory analysis. We then aggregated feature values across clients and applied the Wilcoxon Rank-Sum Test (p < 0.05) to assess overall pre- and post-transition differences.

Results

Significant changes were found across the 5- and 7-day windows. At 5 days, the most frequent individual shifts involved prepositions, word count, verbs and pronouns, while the 7-day window showed changes in word count, sentiment, and verbs. Aggregated analysis revealed common characteristics, including decreases in impersonal pronouns, achievement-, perception- and feeling-related terms, alongside increases in positive emotion, and past-focus. Unique to the 5-day window were reductions in affect, insight, analytical thinking, and future-focus. The 7-day window alone showed declines in power- and work-related terms, cognitive processes, drives, and sadness, with increases in negative emotion, anger, auxiliary verbs, and time-related terms.

Our findings confirm that linguistic patterns—particularly changes in content—shift around phase transitions in a measurable way and offer valuable insights into underlying psychological dynamics. It is worth noting that these patterns reflect mixed trajectories, as clients experienced either improvement or deterioration in psychological state. This likely accounts for the simultaneous increases in both positive and negative emotion features in the aggregated results. Future work should aim to develop clinically relevant interventions by linking specific content-related patterns to distinct mental states.

References

[1] M. Spruit, S. Verkleij, K. de Schepper, and F. Scheepers, “Exploring language markers of mental health in psychiatric stories,“ Applied Sciences, vol. 12, no. 4, p. 2179, 2022.

[2] G. K. Schiepek et al., “Psychotherapy is chaotic—(not only) in a computational world,“ Frontiers in Psychology, vol. 8, p. 379, 2017.