Category Mistakes


Richard Marshall interviews Ofra Magidor in 3:AM Magazine:

3:AM: You say it’s important for linguistics, computer science – how so?

OM: In the case of linguistics, it is fairly obvious why category mistakes are important: one of the central tasks of linguistics explaining why some sentences are fine and others are infelicitous. In fact, category mistakes are a particularly interesting case, because a plausible argument can be made for explaining their oddness in terms of each of syntax, semantics, and pragmatics – so this is a good phenomenon to explore for anyone who is interested in the distinction between these three realms of language. This is probably why in the late 1960s category mistakes played a key role in one of the central disputes in the foundations of linguistics – that between interpretative semanticists (who claimed that syntax is autonomous of semantics) and generative semanticists (who rejected the sharp divide between these two realms).

I should also note there was a period in the 1960s when there was quite a lot of discussion of category mistakes happening in a parallel in linguistics and in philosophy, but there was practically no interaction at all between the two fields on this topic (they even used different terms – in linguistics authors usually refer to category mistakes as ‘selectional violations’). One thing I tried to do in the book was to bring together these two parallel debates. I’d like to think that these days there is much more co-operation between linguists and philosophers of language so this kind of divide is less likely to happen.

Moving to computer science: one straightforward way in which category mistakes are relevant is because of the field of computational linguistics. Suppose for example that you have an automatic translator which is given the sentence ‘John hit the ball’. If the translator looks up the word ‘ball’ in a dictionary, it will encounter (at least) two meanings: a spherical object that is used in games, and a formal gathering for dancing. It is obvious that the most natural interpretation of the sentence used the former meaning, and one way to see that is to note that if ‘ball’ were interpreted in the ‘dance’ sense, the sentence would be a category mistake. So being able to recognize category mistakes can help the automatic translator reach the correct interpretation.

But there is also a more general way in which the topic is relevant to computer science: computer programs use variables of various types which are assigned values – and it is very common to encounter cases where the value is of the wrong type for the variable. So there is an issue about how the program is going to deal with this kind of type mismatch which is in some ways parallel to the question of how natural languages deal with category mistakes.

More here.