I am not maintaining that what we primarily know is our own perceptions. This is largely a verbal question; but with the definition of knowledge given in Chapter VIII, it will be correct to say that from the first we know external objects, the question is not as to what are the objects we know, but rather as to how accurately we know them. Our non-inferential knowledge of an object cannot be more accurate than our reaction to it, since it is part of that reaction. And our reaction cannot be more accurate than the stimulus. But what on earth can you mean by the “accuracy” of a stimulus? I may be asked. I mean just the same as by the accuracy of a map or a set of statistics. I mean a certain kind of correspondence. One pattern is an accurate representation of another if every element of the one can be taken as the representative of just one element of the other, and the relations that make the one set into a pattern correspond with relations making the other set into a pattern. In this sense, writing can represent speech with a certain degree of accuracy; to every spoken word a written word corresponds, and to the time-order of the spoken words the space-order of the written words corresponds. But there are inflexions and tones of voice that cannot be represented in writing, except, to some extent, by musical notation. A gramophone record is a much more accurate representation of vocal sounds than any writing can be; but even the best gramophone record fails to be completely accurate. The impression made upon an observer is very analogous to a gramophone record or a photograph, but usually less accurate owing to the influence of the law of association, and the lack of delicacy in our senses. And whatever limitations there are to the accuracy of our impressions are limitations to the accuracy of our non-inferential knowledge of the external world.