Two Predictors of Psychotic Illness—Based on Speech
New research shows that meaningless speech can foretell a psychotic episode.
Posted Jul 30, 2019
A: I mean it seems like as much as we want to believe we matter in universe, I don't think there is really anything that suggests that we do.
B: Sometimes I get nervous and I get worried about time and I can't look at the clock.
C: Now, now I know how to be cool with people because it's like not talking is like, is like, you know how to be cool with people, it's like now I know how to do that.
D: Well, I think I do have strong feelings about politics.
Assuming these utterances are typical for the four speakers, the correct answer is C, followed by B. C's utterance has the lowest meaning density, whereas D's utterance has the highest. Low meaning density means using a lot of words to say very little, whereas high meaning density means using relatively few words to say a lot. People who use a lot of words but convey very little meaning are at significantly higher risk of developing a psychotic disorder, such as a psychotic disorder such as Schizophrenia or Bipolar I disorder.
These were the results of a study published on June 13, 2019, in NPJ Schizophrenia. The study, which was conducted by psychiatrists Neguine Rezaii, Elaine Walker, and Phillip Wolf of Emory University, tested whether a machine learning system trained to determine the meaning density and word choice in speech samples could help identify people who are at the brink of a psychotic episode.
Psychotic disorders, such as Schizophrenia, are characterized by experiences that aren’t based in reality. People might hear their own inner voices as alien in nature, or form sets of beliefs that are not backed by evidence. Hearing your inner voice as coming from somewhere outside of you is an example of a hallucination, whereas believing that, say, the CIA is tailing you is an example of a delusion. Hallucinations and delusions are key symptoms of psychotic disorders.
To test whether meaning density and word choice are reliable predictors of psychotic illness, the researchers trained a machine learning system to analyze the meaning density of language samples obtained from 30 participants. They found that the speech samples from the research subjects who later had a psychotic episode were significantly lower in meaning density compared to the control group, that is, the people at risk of psychosis used more words to convey a lot less compared to an average person.
After training the system to analyze the meaning density of speech samples, the researchers then tested its capacity to predict psychotic episodes in an additional 10 research subjects. The system was able to predict who would later develop psychosis with 80 percent accuracy.
The researchers also trained the machine learning system to identify words related to voices and sounds, such as "whisper," "chant," "sound," "loud," or "hear," in conversations among 30,000 contributors on Reddit. A high occurrence of these kinds of words is also thought to be a strong predictor of psychosis.
When the system used both meaning density and word choice to predict which of the 10 participants would develop psychosis, its accuracy rose to 90 percent.
These results are particularly promising for people who suffer from psychotic illness and their families, as the method can help identify when sufferers are on the verge of a psychotic episode and allow for early intervention and safety measures.