Future doctors may ask us to say more than “Ahhh.” Several groups of neuroscientists, psychiatrists and computer scientists are now investigating the extent to which patients' language use can provide diagnostic clues—before a single laboratory test is run. Increased computing power and new methods to measure the relation between behavior and brain activity have advanced such efforts. And although tests based on the spoken word may not be as accurate as gene sequencing or MRI scans, for diseases lacking clear biological indicators, language mining could help fill the gap.


Psychiatrists at Columbia University interviewed 34 young adults at risk for psychosis, a common sign of schizophrenia that includes delusions and hallucinations. Two and a half years later five of the subjects had developed psychosis, and the remaining 29 remained free of the disorder. A specially designed algorithm combed the initial interviews collectively to look for language features that distinguished the two groups and found that psychosis correlated with shorter sentences, loss of flow in meaning from one sentence to the next and less frequent use of the words “that,” “what” and “which.” When later tested on each individual interview, the computer program predicted who did and who did not develop psychosis with 100 percent accuracy. The results were recently published in Schizophrenia, and a second round of testing with another group of at-risk subjects is now under way.

Parkinson's Disease

Twenty-seven subjects in a study at Favaloro University in Argentina listened to recorded sentences containing verbs associated with specific hand shapes (such as “applaud” or “punch”). As soon as they understood the sentence, participants pressed a button while keeping both hands in either a flat or clenched-fist position. Healthy subjects responded more quickly when the verb and hand shape were compatible (flat for “applaud,” clenched fist for “punch”) compared with when they were incompatible. Subjects with early-stage Parkinson's disease, however, showed no difference in their reaction times. Such disconnects could serve as an early sign of the disease, before the onset of severe problems. Now the researchers are conducting a similar study with subjects who currently are healthy but carry a genetic mutation associated with Parkinson's.

Amyotrophic Lateral Sclerosis (ALS)

ALS is typically characterized as a movement disorder, which can lead some patients to speak unclearly because of weak muscles. A new study led by Sharon Ash at the University of Pennsylvania shows that the disease may also disrupt grammar usage. Forty-five subjects were asked to narrate the events in a series of 24 pictures, using their own words. ALS patients produced more incomplete sentences (“And he's angry 'cause it—”), more missing determiners (“Owl flew around”) and more errors in verb tense (“And the deer push him off a cliff”), compared with healthy controls. MRI scans revealed that people who made more grammatical errors also showed more deterioration of brain regions associated with language, suggesting that grammar analysis may be a relatively simple way to assess disease onset and severity. In an ongoing follow-up study, Ash and her colleagues are analyzing patients' shorter utterances, prompted by a single picture.