Mental models and nonmonotonic reasoning

Johnson-Laird & Byrne (J-L & B) are equivocal concerning the scope of mental-model theory. On the one hand, they are careful to note that mental models are aimed primarily at explaining deduction, although commonsense inference is not deductive in character. On the other hand, they contend that mental-model theory solves the problem of nonmonotonic reasoning, which is not deductive and is characteristic of commonsense inference. This equivocation requires clarification: An account of deductive reasoning casts light on a fascinating if rather arcane human ability; an account of nonmonotonic inference in general would be little short of a theory of thinking. It is not clear, therefore, exactly what J-L & B see as the domain of the mental-model account. I shall argue that mental-model theory does not in fact address the problem of understanding commonsense nonmonotonic reasoning, still less provide a solution to it. Everyday, commonsense reasoning may be conceived of as a species of inference to the best explanation: It involves inferring from given information to what best explains and is best explained by that information (Fodor 1983). Such inference is nonmonotonic, because the addition of new information can invalidate what were previously plausible conclusions. So, for example, the plausible inference from hearing the sound of purring behind the door to the conclusion that the cat is trapped in the cellar is immediately overridden if I catch sight of the cat in the garden. The premise on which the inference is based, the purring, need not be withdrawn, although another explanation for this fact may be sought. By contrast, in monotonic reasoning, the conclusion of a valid argument can only be challenged if one of its premises is false.