Reading between the lines: Combining pause dynamics and semantic coherence for automated assessment of thought disorder.

Clinical Snapshot

Key Findings

• P Value: Not reported in abstract

• Effect Size: Late fusion (pause + semantic) model: mean Spearman ρ = 0.455 vs. semantic-only model: mean Spearman ρ = 0.413; absolute improvement Δρ = 0.042

• Primary Outcome: Spearman correlation between SVR model-predicted scores and clinician-rated FTD severity, averaged across three datasets

• Nnt Or Sensitivity: No sensitivity/specificity reported (continuous regression outcome). Pause-only models described as 'robustly predicting' FTD severity across datasets; late fusion consistently outperforms both unimodal approaches. Magnitude of correlation (ρ ≈ 0.45) indicates moderate predictive accuracy, explaining approximately 20% of variance in clinical FTD ratings.

• Confidence Interval: Not reported

Clinical Application

The framework requires ASR infrastructure and computational processing pipelines, which are increasingly accessible but not yet embedded in routine clinical workflows. Naturalistic diary recording (as in the AVH dataset) is particularly scalable for remote monitoring. However, ASR accuracy varies with audio quality, accent, and speech pathology, and the pipeline requires validation across diverse recording environments before clinical deployment. In Australia, formal thought disorder assessment is conducted within public mental health services, early psychosis programmes (e.g., EPPIC, headspace), and private psychiatric practice, typically using clinician-administered scales. The RACGP and RANZCP do not currently endorse automated speech analysis tools for FTD assessment. No TGA-listed or PBS-subsidised digital health tools for automated thought disorder assessment currently exist in Australia. The framework's scalability is relevant to Australia's geographically dispersed population and telehealth-expanded services post-COVID-19, but prospective validation in Australian cohorts — including Aboriginal and Torres Strait Islander populations and culturally and linguistically diverse communities — would be essential before any clinical adoption. Data sovereignty and privacy considerations under the Australian Privacy Act 1988 would apply to audio diary data collection. Adults with schizophrenia spectrum disorders or at-risk mental states undergoing clinical assessment of thought disorder severity; potentially applicable in early psychosis services, inpatient psychiatry, and community mental health settings where scalable, objective FTD monitoring is sought

Abstract

Formal thought disorder (FTD), a hallmark of schizophrenia spectrum disorders, manifests as incoherent speech and poses challenges for clinical assessment. Traditional clinical rating scales, though validated, are resource-intensive and lack scalability. Automated speech recognition (ASR) allows for objective quantification of linguistic and temporal features of speech, offering scalable alternatives. Furthermore, ASR-derived utterance timestamps provide access to pause dynamics, which are thought to reflect the cognitive processes underlying speech production. Yet, their added value beyond semantic measures remains insufficiently explored. In this study, we evaluated a scalable multimodal framework that integrates pause features with semantic coherence metrics across three datasets: naturalistic self-recorded diaries (AVH, n = 140 participants), structured picture descriptions (TOPSY, n = 72 participants), and dream narratives (PsyCL, n = 43 participants). Pause-related features were evaluated alongside established coherence measures using support vector regression (SVR) to predict clinical FTD scores. Models using pause features alone robustly predict manually rated FTD severity consistently across datasets. Integrating pause features with semantic coherence metrics enhanced predictive performance compared to coherence-only models, with late fusion yielding the most robust and consistent gains in all three datasets. On average across datasets, Spearman correlation increased from ρ = 0.413 for semantic-only models to ρ = 0.455 with late fusion. The performance gains from semantic and pause features integration held consistently across all contexts, though the nature of the most informative pause patterns was dataset-dependent. These findings suggest that both pause dynamics and semantic coherence reflect complementary aspects of thought disorganization. Our integration framework provides a scalable approach for refining the assessment of disorganized speech to advance automated speech analysis in psychosis.

References

1. 1.Chen, F., Xu, W., Li, C., Pakhomov, S., Cohen, A., Bhola, S., Yin, S., Tang, S. X., Mackinley, M., Palaniyappan, L., Ben-Zeev, D., & Cohen, T. (2026). Reading between the lines: Combining pause dynamics and semantic coherence for automated assessment of thought disorder. Neuropsychologia. https://doi.org/10.1016/j.neuropsychologia.2026.109473