[Ordinary people accept universals like dog or triangle because they seem self-evident. Inasmuch as they are unaware of the transcendent or metaphysical quality in universals they are unconscious mystics. Bertrand Russell, on the other hand, feels obliged to justify them, or at least refute the denial of them, on purely logical grounds. Whether he is entirely successful would require more training in logic than most people possess, and even logicians with different philosophical committments might differ about the degree of his success. It is probably safe to say, however, that in this case, as in many others, the logical approach enormously complicates and often obscures what had previously seemed obvious and clear. We do not deny that the logician’s attempt to clarify a thing logically may be valuable or sometimes necessary. But for most people, a mystical acceptance of many mysterious things—such as the mind, or free will, or equality—seems a wiser recommendation than a risky injunction to enter the jungles of logic from which they may not emerge with their common sense mystical convictions intact.]
Assuming that we can get rid of all universals except similarity, it remains to be considered whether similarity itself could be explained away. We will consider this in the simplest possible case. Two patches of red (not necessarily of exactly the same shade) are similar, and so are two instances of the word “red.” Let us suppose that we are being shown a number of coloured discs and asked to name their colours say in a test for colour-blindness. We are shown two red discs in succession, and each time we say “red.” We have been saying that, in the primary language, similar stimuli produce similar reactions; our theory of meaning has been based on this. In our case, the two discs are similar, and the two utterances of the word “red” are similar. Are we saying the same thing about the discs and about the utterances when we say the discs are similar and when we say the utterances are similar? or are we only saying similar things? In the former case, similarity is a true universal; in the latter case, not. The difficulty, in the latter case, is the endless regress; but are we sure that this difficulty is insuperable? We shall say, if we adopt this alternative: if A and B are perceived to be similar, and C and D are also perceived to be similar, that means that AB is a whole of a certain kind and CD is a whole of the same kind; i.e., since we do not want to define the kind by a universal, AB and CD are similar wholes. I do not see how we are to avoid an endless regress of the vicious kind if we attempt to explain similarity in this way.
I conclude, therefore, though with hesitation, that there are universals, and not merely general words. Similarity, at least, will have to be admitted; and in that case it seems hardly worth while to adopt elaborate devices for the exclusion of other universals.
It should be observed that the above argument only proves the necessity of the word “similar,” not of the word “similarity.”
Some propositions containing the word “similarity” can be replaced by equivalent propositions containing the word “similar,” while others cannot. These latter need not be admitted. Suppose, for example, I say “similarity exists.” If “exists” means what it does when I say “the President of the United States exists,” my statement is nonsense. What I can mean may, to begin with, be expressed in the statement: “there are occurrences which require for their verbal description sentences of the form ‘a is similar to b’.” But this linguistic fact seems to imply a fact about the occurrences described, namely the sort of fact that is asserted when I say “a is similar to b.” When I say “similarity exists,” it is this fact about the world, not a fact about language, that I mean to assert. The word “yellow” is necessary because there are yellow things; the word “similar” is necessary because there are pairs of similar things. And the similarity of two things is as truly a non-linguistic fact as the yellowness of one thing.
We have arrived, in this chapter, at a result which has been, in a sense, the goal of all our discussions. The result I have in mind is this: that complete metaphysical agnosticism is not compatible with the maintenance of linguistic propositions. Some modern philosophers hold that we know much about language, but nothing about anything else. This view forgets that language is an empirical phenomenon like another, and that a man who is metaphysically agnostic must deny that he knows when he uses a word. For my part, I believe that, partly by means of the study of syntax, we can arrive at considerable knowledge concerning the structure of the world.
Bertrand Russell (from An Inquiry into Meaning and Truth, 1950)