In my day job as a research psychologist, I study how human beings process language. I use many methods, one of which is to model semantics using a co-occurrence model of semantics. These computational models bootstrap word meaning by quantifying patterns of co-occurrence in a very large text corpus, using some fancy math that would be boring to explain here. The basic idea is this: the model can figure out that cats are more similar to dogs than they are to tractors because the word 'cat' occurs more often near the same words as the word 'dog' than near the same words as the word 'tractor'. Quantify that notion, and you have the model.
Recently I spent some time trying to construct a computational model of human judgments of a word's dangerousness and usefulness. My friend Lee Wurm has shown that people's behavior is measurably affected by a word's dangerousness and usefulness, suggesting that these dimensions are unconsciously assessed by people when they process words. He and I used my co-occurrence model to try to see if we could model these dimensions, with some success (the paper we wrote has been submitted for peer review).
One of the interesting side-effects of having a formal model of these dimensions is that we can compute quantitative estimates of dangerousness and usefulness for every word in the English language. Once we have done that, we can ask: What is the most dangerous, least useful thing there is?
The answer was kind of brilliant. According to our computer model, the quantifiably-proven most dangerous, least useful thing is [See if you can guess and then select the box to see.] anger!
[Image taken by me in about 1984 at the Acropolis in Greece.]
No comments:
Post a Comment