Patterns of speech may be telltales of particular symptoms
DIAGNOSING MENTAL illness is difficult. Giving broad names such as “schizophrenia” and “bipolar disorder” to particular sets of symptoms helps psychiatrists and patients discuss and treat what is going on, but many traits are symptomatic of more than one such named condition, giving plenty of scope for mislabelling. Moreover, the specialised interviews required to detect the presence of particular traits are time-consuming and require specific training to conduct.
A shortcut to reliable psychiatric diagnoses would therefore be desirable. Justin Baker, a psychiatrist at Harvard Medical School, and Louis-Philippe Morency, a computer scientist at Carnegie Mellon University, in Pittsburgh, think they have one. As they told the International Conference on Multimodal Interaction in Boulder, Colorado, on October 19th, they believe they can extract a lot of relevant information from patients’ speech patterns.
Rather than tackle all mental illness at once, Dr Baker and Dr Morency focused on psychosis. This is the experience by a patient of hallucinations or delusions—in other words of a “reality” at variance with the general consensus of other people. Psychosis is particularly symptomatic of schizophrenia, but is also common in bipolar disorder. And it is, in turn, capable of division into traits of its own, such as impulsive hostility, emotional withdrawal, conceptual disorganisation and delusions of grandeur. These can be used to refine an initial diagnosis of a broader condition.
To test the idea that patterns of word use might help diagnose such traits, Dr Baker and Dr Morency let their computers loose on transcripts of 53 interviews with 28 psychotic patients at the McLean Hospital in Belmont, Massachusetts. These were people who had been studied thoroughly by conventional psychiatric techniques. They had therefore had their specific psychotic traits classified. And patterns the two researchers did indeed find.
For example, patients who commonly used power-related words like “important”, “superiority” and “exploit” generally turned out to have the traits of delusions and grandiosity. Conversely, there was an inverse correlation between patients’ use of words relating to time and space, such as “yesterday”, “lately” and “nearby”, and the trait of poor reality monitoring. Those who rarely or never used these words showed significant detachment from reality.
In addition to these specific correlations, the overall severity of a patient’s psychosis, regardless of the detailed pattern of traits, appeared to correlate with his use of emotionally loaded words. The absence of positive words in a transcript, and a preponderance of negative ones, such as “gloomy”, “dark” and “sadly”, was characteristic of those whose psychotic disorder was severe.
Nor was frequency of word use the only signal that Dr Baker and Dr Morency’s computers picked up. They also noticed patterns in phenomena called sentence repair and language perplexity.
Everyone repairs sentences from time to time during conversations, saying things like “John likes, I mean, loves Mary.” But constant repairing is rare in the mentally healthy. Dr Baker and Dr Morency, however, found it common in patients who had the psychotic traits of apathy, avolition and defensiveness.
Language perplexity is a measure of how easy it is, partway through a sentence, to guess what is coming next. The harder it is to do this for a given individual’s speech, the more perplexing is his language. And the more perplexing a psychotic patient’s language is, Dr Baker and Dr Morency discovered, the more likely it is that his particular traits include excitement and conceptual disorganisation.
The study Dr Baker and Dr Morency have carried out is small, and so will need confirmation using larger groups of patients. But if that confirmation comes, it will give psychiatrists a new diagnostic tool. And it is a tool that might eventually be applied to areas other than psychosis. Dementia and Parkinson’s disease, too, are thought to shape speech in subtle ways as they begin to develop. The Baker-Morency approach might thus permit earlier diagnosis of these conditions as well.