It is one of the merits of Max Black’s Companion to the Tractatus that he emphasizes the continuity of Wittgenstein’s philosophical development by frequent quotation from his later writings. In this particular case the retractation does not amount to a complete abandonment of the earlier point of view. Of course, it puts the original theory of elementary propositions in jeopardy, because it was through their logical independence that elementary propositions were specified in the Tractatus. But whatever its effect on his view about the terminus of analysis, it leaves his view about the process of analysis—the usable part of the original theory—unchanged.
There is also another important idea which survives the change. Wittgenstein still maintains that the essential nature of things cannot be described in factual propositions, but can only be revealed in the a priori connections between factual propositions. The difference is that in the Tractatus what was supposed to be revealed in this way was only the most general structure of reality, which is reflected in the truth-functional structure of all factual discourse. The doctrine was that reality consists essentially of simple things, and that this basic structure is revealed in any a priori connection between a pair of factual propositions. For it was supposed that any such connection could be reduced to a tautology or a contradiction, and that the existence of this tautology or contradiction made only a very general demand on reality: all that was required was that reality should consist of simple things. But in the Bemerkungen he allows that it is not always possible to explain an a priori connection between two factual propositions by reducing it to a tautology or contradiction. Sometimes this explanation will work, but his new view was that some a priori connections depend on words which cannot be analyzed in this way, and that in such cases a truth-functional reduction to tautology or contradiction is not possible. When he finds an a priori connection of this latter kind, he can still say that it is a feature of the logical grammar of the words, but in such a case what the logical grammar of the words reveals about the structure of reality will be something very specific, rather than its general truth-functional framework. For example, the structure of the spectrum of colors, or the structure of the range of sounds would be revealed in this way. In fact, Wittgenstein’s new view was that there are many specific features of the structure of reality which are reflected in the logical grammar of the words which we use to divide and classify what we experience.
At this point it is important to observe what made the earlier theory of necessity succeed in those cases in which it did succeed. There are two kinds of case. First, there are a priori propositions which depend on definitions, and in such cases we can see how the definition confers necessary truth on the proposition, and we can often see that the definition is merely a convention which need not have been adopted. For instance, the a priori proposition “All planets move round the sun” owes its necessary truth to the definition of the word “planet” as “a heavenly body in orbit around the sun.” We can see how this definition works, and we can see that it need never have been adopted. Therefore, in order to explain this kind of a priori proposition, we do not have to appeal to the nature of things.
Secondly, there are the truths of logic, and in their case too the explanation of necessary truth avoids any direct appeal to the nature of things. There is, of course, an indirect appeal, because tautologies reveal the general structure of reality. But there is no direct appeal, because tautologies are not theories with factual sense. So in both these types of case the appeal to logical grammar is successful because it eliminates any direct appeal to the nature of things. But when Wittgenstein extended the scope of this explanation of a priori propositions by appealing to the logical grammar of descriptive words which, according to the earlier theory ought to be definable, but which in fact are not definable, it was not so clear whether it was successful. Can incompatibilities of colors be explained by the observation that they depend on the logical grammar of color-words? It is not even clear how this explanation was meant to be taken. When Moore heard him lecturing on this point in Cambridge in 1933, he was puzzled. He wondered what the ultimate source of the necessity was supposed to be in such cases. For example, was the proposition, “Nothing can be blue and yellow simultaneously” supposed to owe its necessary truth in the last resort to the logical grammar of the words, or to the nature of the things to which we apply the words?
Now when Wittgenstein wrote the Tractatus, he believed that, though the propositions of logic lack factual sense, they reveal the general structure of reality. So when he changed his theory of elementary propositions, and allowed that they might contain incompatible words, it would be natural to assume that his new view was that the logical grammar of those words revealed specific features of the structure of reality. But the way in which his ideas about necessary truth actually developed was more complex than this. For in the early 1930s his work on the foundations of mathematics was leading him toward a more conventionalist theory of necessary truth. He was beginning to think that all necessary truth depends on conventions which need not have been adopted.
This tendency to conventionalism complicates the development of his ideas. One line of development has already been described: instead of claiming to find a single basic pattern in reality—the pattern of simple things arranged in various ways—he came to think that the logical grammar of unanalyzable words reveals a multitude of more specific patterns in reality. But at the same time his shift toward conventionalism was changing the very meaning of the thesis that the structure of reality exhibits this pattern, or that pattern. He was moving away from the view that the structure of reality is a rigid framework, fixed in advance, and toward the view that it is a projection of human thought. In his theory of necessity conventionalism was replacing objectivism, and, in his account of the way in which we divide and classify what we experience, nominalism was replacing realism. So, although he located the source of specific necessary truths in the nature of things, he was beginning to take an entirely different view of this location. According to his new view, it would be nearer the truth to regard it as our creation. This shift toward anthropocentric theories is one of the two most important developments in his later philosophy.
The other important development in his later philosophy is also connected with the theory of elementary propositions. In the period between the two wars the elementary propositions of the Tractatus were frequently identified with sense-datum propositions. Though this identification was not part of the theory itself, it is a possible application of it, and many analytic philosophers in this period made it almost automatically. Now it is often supposed that the sense-datum language could stand by itself as the firm substructure on which the remainder of empirical language could be built. The idea was not that this was the order of infant learning, but that the system of empirical language could be rearranged in this way. The point of the rearrangement, according to philosophers as diverse in viewpoint as Russell and C.I. Lewis, was that it would set the stage for the combat with Cartesian doubt. Although the sense-datum language was not learned first, it could have been learned first, and, if it had been, we would have started with propositions about our own sensory experiences which we could not doubt. For example, it was supposed that however doubtful I might be about the actual shape of a particular rock, I could not have any doubts about my visual impression of its shape. So each of us would train himself to use propositions about his own sensory experiences in the privacy of his mind, and then in one way or another we would all fight our way back to some sort of knowledge of the external world and establish communication with one another. Wittgenstein took this traditional picture of the relationship between minds and the physical world and turned it upside down. He argued that the so-called sense-datum language is, and must be, dependent on the language in which we describe the external world.
He developed this attack on the idea of a private language in his first posthumously published book, Philosophical Investigations, and it has had a profound effect on the analytic philosophy of the last fifteen years. There are more essays devoted to it in the excellent collection edited by George Pitcher than to any other single topic. Wittgenstein’s polemic would not have been so important if it had been developed only against theories of perception which assume that the sense-datum language could function independently of the language in which we describe the external world. What makes his arguments so important is that they traverse the whole philosophy of mind. They seek to prove that it would be equally impossible for emotions or desires or intentions to be reported in a private language. Public criteria are needed across the whole field of mental phenomena. This is a general challenge to the Cartesian assumption that our knowledge of the contents of our own minds is independent of our knowledge of the external world—an assumption which had dominated western philosophy for three centuries. The challenge is based on an idea which is neither original nor recondite: mental phenomena cannot be dissociated from their observable manifestations in behavior. But the way in which Wittgenstein develops this idea is elaborate and subtle. The obvious advantage of behaviorism is that it explains how one person can know so much of what goes on in the mind of another, and how we can communicate with one another about such things. Wittgenstein tries to keep this advantage without denying, as some behaviorists have done, that there is anything more to mental phenomena than observable behavior. But while he avoids this paradoxical denial, he takes care not to be misled by the patch-work of physical imagery which the mind has to borrow in order to form a view of itself. This delicate balance between three points is difficult to maintain, and not everyone agrees that he succeeded in maintaining it.
It is particularly important that the concept of intention is among those which are treated in this way. For this concept must lie somewhere near the center of any theory of meaning. At this point it is possible to trace two connections between Wittgenstein’s earlier and later theories of language, an obvious connection, and a deeper one. The obvious connection is that it is a mistake to suppose that the language of sense-datum propositions, with which elementary propositions were often identified by other philosophers, might stand alone. The deeper connection is that at a certain point in the development of his account of elementary propositions in the Tractatus, Wittgenstein stopped, when he might have gone on. He stopped at the correlation of words with simple things. If he had gone on to ask how such correlations could be made, he would have said more about people, and their intentions, and the rules which are associated with their intentions. A word is not brought to life by mere juxtaposition with the thing which it designates. Any attempt to say what more is needed inevitably brings in the philosophy of mind, which scarcely makes an appearance in the lunar landscape of the Tractatus.