Skip to main content

Transforming the understanding
and treatment of mental illnesses.

Celebrating 75 Years! Learn More >>

Computational Methods Identify Psychosis Symptoms in Spoken Language

Research Highlight

Changes in how a person speaks can indicate disturbances in thought and communication, which are core symptoms of schizophrenia and other psychotic disorders. These language changes can emerge before more observable psychosis symptoms but be difficult for clinicians to recognize, which can delay diagnosis and reduce the quality of care. A study supported in part by the National Institute of Mental Health aimed to identify speech patterns associated with psychosis to provide clinicians with an easy-to-collect, scalable, and objective marker for earlier identification of psychotic disorders like schizophrenia.

The research team led by Justin T. Baker, M.D., Ph.D. , Jeffrey M. Girard, Ph.D. , and Alexandria K. Vail  examined data from 38 adults with a psychotic disorder who had been admitted to the hospital for inpatient treatment. All participants were interviewed one or more times for a total of 78 sessions. At each session, a doctor involved in the study administered a recorded semistructured clinical interview that was used as the source of the language analyses and a standard measure of positive (e.g., delusions, hallucinations, feelings of grandiosity) and negative (e.g., emotional withdrawal, difficulty thinking, lack of motivation) psychosis symptoms.

The researchers then used data-driven computational methods to analyze language in the interviews. They calculated 14 measures reflecting three aspects of spoken language:

  • Lexical measures, which characterize the frequency of words from specific categories
  • Sentence coherence measures, which characterize the predictability of word sequences
  • Disfluency measures, which characterize the frequency of involuntary speech disruptions, including the use of fillers such as “umm” or “I mean,” repeats of words or short phrases, and midsentence word changes

The researchers found several significant associations between these language measures and psychosis. Some language measures were related to positive symptoms only, some to negative symptoms only, and some to both types of symptoms. For example, when controlling for negative symptoms, participants with a greater number of positive symptoms used more words about perceptual processes (such as “look” or “heard”) and relativity (such as “here” or “until”) but fewer words in other categories (i.e., achievement, cognitive processes, risk, and reward) and fewer speech fillers.

Conversely, negative symptoms were uniquely associated with using more speech fillers and fewer relativity words. Expressing more negative emotion words tracked with both positive and negative symptoms, suggesting that a negative emotional bias may be a common feature of psychosis. The researchers draw connections between the language patterns and disturbances in thought and behavior that could be used to distinguish people with a psychotic disorder.

The researchers acknowledge that these findings are preliminary. Although models predicting clinicians’ scores of psychotic symptoms from participants’ spoken language were promising, they needed more data to be clinically useful. However, the interview process was feasible for hospital staff and patients and produced enough data for computational analysis. This finding suggests that language measures might successfully predict psychosis with a larger sample.

Future studies are also needed to examine such measures with more diverse samples and over time to fully understand the relationship between spoken language and psychosis symptoms. In addition, there are likely other verbal and nonverbal signals not measured in this study that are relevant to the assessment of psychosis. Nonetheless, this study provides promising evidence that language measures can be automatically generated from clinical interviews and effectively applied to the standardized identification of psychosis.

Reference

Girard, J. M., Vail, A. K., Liebenthal, E., Brown, K., Kilciksiz, C. M., Pennant, L., Liebson, E., Öngür, D., Morency, L.-P., & Baker, J. T. (2022). Computational analysis of spoken language in acute psychosis and mania. Schizophrenia Research, 245, 97–115. https://doi.org/10.1016/j.schres.2021.06.040 

Grants

MH116925 , MH096951