Scientists are using AI to detect language differences in people with mental health difficulties that could eventually be used to help diagnose and monitor psychiatric conditions.
Researchers at Oxford University and UCL have developed new tools, based on artificial intelligence language models, to better recognise and understand subtle signatures in the speech of patients diagnosed with schizophrenia.
They found that speech from patients with schizophrenia had small but significant differences that indicated the severity of their symptoms, and used brain scanning technology to relate these differences to patterns of brain activity thought to relate to how the brain represents the relationships between memories and meaning. The results have been published in the journal PNAS.
Dr Matthew Nour, lead author and NIHR Clinical Lecturer in Psychiatry at the University of Oxford, said:
“Diagnosis and assessment in psychiatry is based almost entirely on talking with patients and those close to them, with only a minimal role for automated tests such as blood tests and brain scans. This lack of precision prevents a richer understanding of the causes of mental illness and the monitoring of treatment.
“Until very recently, the automatic analysis of language has been out of reach of doctors and scientists. However, with the advent of artificial intelligence (AI) language models such as catgut, this situation is changing.
“This work shows the potential of applying AI language models to psychiatry – a medical field intimately related to language and meaning. In ongoing work, our team plan to use this technology in a larger sample of patients and across more diverse speech settings, to test whether it might prove useful in the clinic.”
Schizophrenia is a relatively common and debilitating psychiatric disorder. Symptoms include disordered thoughts, difficulties in abstract reasoning and reduced language coherence, which can result in poor social functioning and disability. One emerging idea is that these clinical features might relate to the way the brain stores information about how things in the world are related to each other – from memories to concepts.
The study involved 26 participants who had been diagnosed with schizophrenia, and 26 control participants, completing a verbal fluency task (naming animals or words starting with the letter ‘P’), followed by a magnetoencephalography (MEG) scan involving a separate memory task where they looked at pictures and tried to remember how they were related.
Researchers used AI language models – trained on vast quantities of text from the internet - to understand the choice of words that individual participants produced. They found the words used by those with schizophrenia were less obviously related to each other than those in the control group, and thus less able to be predicted by the AI language model.
The authors speculate that this finding might be related to the way the brain stores how different memories and ideas are related in the brain. The authors found support for this theory when they looked at the brain imaging data from the same participants. They found that people who produced more predictable word sequences had greater brain activity in a part of the brain involved in storing memories than those who produced less predictable sequences.