The inference that has to be made has two steps, one physical and one psychological. The physical inference is of the sort we considered a moment ago, in which we pass from a sensation to a physical occurrence. We hear noises, and think they proceed from the policeman’s body. We see moving shapes, and interpret them as physical motions of his lips. This inference, as we saw earlier, is in part justified by testimony; yet now we find that it has to be made before we can have reason to believe that there is any such thing as testimony. And this inference is certainly sometimes mistaken. Lunatics hear voices which other people do not hear; instead of crediting them with abnormally acute hearing, we lock them up. But if we sometimes hear sentences which have not proceeded from a body, why should this not always be the case? Perhaps our imagination has conjured up all the things that we think others have said to us. But this is part of the general problem of inferring physical objects from sensations, which, difficult as it is, is not the most difficult part of the logical puzzles concerning testimony. The most difficult part is the inference from the policeman’s body to his mind. I do not mean any special insult to policemen; I would say the same of politicians and even of philosophers. The inference to the policeman’s mind certainly _may_ be wrong. It is clear that a maker of wax-works could make a life-like policeman and put a gramophone inside him, which would cause him periodically to tell visitors the way to the most interesting part of the exhibition at the entrance to which he would stand. They would have just the sort of evidence of his being alive that is found convincing in the case of other policemen. Descartes believed that animals have no minds, but are merely complicated automata. Eighteenth-century materialists extended this doctrine to men. But I am not now concerned with materialism; my problem is a different one. Even a materialist must admit that, when he talks, he means to convey something, that is to say, he uses words as signs, not as mere noises. It may be difficult to decide exactly what is meant by this statement, but it is clear that it means something, and that it is true of one’s own remarks. The question is: Are we sure that it is true of the remarks we hear, as well as of those we make? Or are the remarks we hear perhaps just like other noises, merely meaningless disturbances of the air? The chief argument against this is analogy: the remarks we hear are so like those we make that we think they must have similar causes. But although we cannot dispense with analogy as a form of inference, it is by no means demonstrative, and not infrequently leads us astray. We are therefore left, once more, with a _prima facie_ reason for uncertainty and doubt.