from the point of view of what may be called knowledge of universals. A man who can always use the word "dog" when he sees a dog may be said, in a certain sense, to know the meaning of the word "dog," and IN THAT SENSE to have knowledge of the universal "dog." But there is, of course, a further stage reached by the logician in which he not merely reacts with the word "dog," but sets to work to discover what it is in the environment that causes in him this almost identical reaction on different occasions. This further stage consists in knowledge of similarities and differences: similarities which are necessary to the applicability of the word "dog," and differences which are compatible with it. Our knowledge of these similarities and differences is never exhaustive, and therefore our knowledge of the meaning of a universal is never complete. In addition to external observable habits (including the habit of words), there is also the generic image produced by the superposition, or, in Semon's phrase, homophony, of a number of similar perceptions. This image is vague so long as the multiplicity of its prototypes is not recognized, but becomes universal when it exists alongside of the more specific images of its instances, and is knowingly contrasted with them. In this case we find again, as we found when we were discussing words in general in the preceding lecture, that images are not logically necessary in order to account for observable behaviour, i.e. in this case intelligent speech. Intelligent speech could exist as a motor habit, without any accompaniment of images, and this conclusion applies to words of which the meaning is universal, just as much as to words of which the meaning is relatively particular. If this conclusion is valid, it follows that behaviourist psychology, which eschews introspective data, is capable of being an independent science, and of accounting for all that part of the behaviour of other people which is commonly regarded as evidence that they think. It must be admitted that this conclusion considerably weakens the reliance which can be placed upon introspective data. They must be accepted simply on account of the fact that we seem to perceive them, not on account of their supposed necessity for explaining the data of external observation. This, at any rate, is the conclusion to which we are forced, so long as, with the behaviourists, we accept common-sense views of the physical world. But if, as I have urged, the physical world itself, as known, is infected through and through with subjectivity, if, as the theory of relativity suggests, the physical universe contains the diversity of points of view which we have been accustomed to regard as distinctively psychological, then we are brought back by this different road to the necessity for trusting observations which are in an important sense private. And it is the privacy of introspective data which causes much of the behaviourists' objection to them.