Letter from Lucille Lang Day of Oakland in the February 2011 Harper’s:
The Method in It
In her article on prodromal psychosis ["Which Way Madness Lies," Report, December 2010], Rachel Aviv discusses an exam said to evaluate the risk of psychosis. The exam asks patients to describe “the similarities between an apple and a banana.” This is a very poor way of diagnosing the likelihood of mental illness.
Before I quote the rest of Day’s letter, think for a moment how you would answer this question.
Day goes on:
What went through my mind was: both have a sweet, fruity fragrance; both grow on trees; both can be eaten raw; both are used in cooking; and both appear in children’s lunch boxes. Then I read the correct response: “Both are fruit.” Apparently, one of the most common “wrong” answers is “Both have skin.”
I feel sorry for the people being judged on the basis of questions like this one. I interpreted the question as “How are these two fruits similar?” so I would never have answered correctly. Similarly, if someone asked me, “How is a cat like a dog,” I would not say, “Both are animals,” but would take that as a given and instead point out that both are house pets, have fur, and so forth. This is not indicative of mental illness.
The exam presumably is trying to get at a patient’s system of mental categories, in particular the degree to which the patient can easily access higher-level (more “abstract”) categories (like the taxon FRUIT). And since we can’t directly examine this system, we have to get at it through language — in this case, through an attempt to elicit the label fruit for this taxon by asking about “similarities”. The problem is that the testee somehow has to divine the intentions of the tester in asking the question, in order to avoid giving an answer that is “wrong” because it is more specific than the answer the tester is looking for.
There are more indirect ways of tapping category structure — by asking the testee to sort objects into groups or by asking the testee to pick out, from an array of objects (a banana, a carrot, and an apple, say) the object that doesn’t belong, that is not like the others — though these have problems of their own, and they certainly don’t avoid the mediation of language or the need for the testee to divine what sort of answer the tester is seeking.
The testee’s job is made more difficult by the fact that there is probably no context for this sort of question, no previous experience that the testee can bring to bear in performing this task — not even a sample item or two that could establish a model for the testee’s response.
(In an earlier discussion of test questions — “What is this question about?”, on Language Log, here — I started with the question “What color is a banana?” (addressed to a young child) and moved on to “What is the opposite of ___?” and “Which one of these things is not like the others?” and, in comments, “What is the next number in the series ___, ___, ___?” Answering such questions “correctly” involves mustering all sorts of cultural knowledge and also experience with the contexts in which people ask them such questions and the kinds of answers that are expected.)
Notice that the subjects of the similarities test are described as “patients”, so they’re probably used to being asked unaccountable, but somehow important, questions by professionals of various sorts. Still, the experience is rather weird and unnatural, and can be baffling. My man Jacques, during his many years of descent into neurological hell, was generally cooperative, but sometimes bridled at (say) being asked what month it was (“Why do you want to know?”) or being asked to count backwards from 100 by 7s (“Why should I?”), and often performed very badly (well short of his actual abilities) when put on the spot and obliged to reflect on what he was doing.