8. I shall now examine briefly a heterodox suggestion by, for example, Ayer{1} to the effect that discourse about sense data is, so to speak, another language, a language contrived by the epistemologist, for the situations which the plain man describes by means of such locutions as "Now the book looks green to me" and "There seems to be a red and triangular object over there." The core of this suggestion is the idea that the vocabulary of sense data embodies no increase in the content of descriptive discourse, as over and against the plain man's language of physical objects in Space and Time, and the properties they have and appear to have. For it holds that sentences of the form

X presents S with a f sense datum

are simply stipulated to have the same force as sentences of the form

X looks f to S.

Thus "The tomato presents S with a bulgy red sense-datum" would be the contrived counterpart of "The tomato looks red and bulgy to S" and would mean exactly what the latter means for the simple reason that it was stipulated to do so.

    As an aid to explicating this suggestion, I am going to make use of a certain picture. I am going to start with the idea of a code, and I am going to enrich this notion until the codes I am talking about are no longer mere codes. Whether one wants to call these "enriched codes" codes at all is a matter which I shall not attempt to decide.

    Now a code, in the sense in which I shall use the term, is a system of symbols each of which represents a complete sentence. Thus, as we initially view the situation, there are two characteristic features of a code: (1) Each code symbol is a unit; the parts of a code symbol are not themselves code symbols. (2) Such logical relations as obtain among code symbols are completely parasitical; they derive entirely from logical relations among the sentences they represent. Indeed, to speak about logical relations among code symbols is a way of talking which is introduced in terms of the logical relations among the sentences they represent. Thus, if "O" stands for "Everybody on board is sick" and "D" for "Somebody on board is sick", then "D" would follow from "O" in the sense that the sentence represented by "D" follows from the sentence represented by "O."

    Let me begin to modify this austere conception of a code. There is no reason why a code symbol might not have parts which, without becoming full-fledged symbols on their own, do play a role in the system. Thus they might play the role of mnemonic devices serving to put us in mind of features of the sentences represented by the symbols of which they are parts. For example, the code symbol for "Someone on board is sick" might contain the letter S to remind us of the word "sick," and, perhaps, the reversed letter E to remind those of us who have a background in logic of the word "someone." Thus, the flag for "Someone on board is sick" might be '[reversed E]S.' Now the suggestion at which I am obviously driving is that someone might introduce so-called sense-datum sentences as code symbols or "flags," and introduce the vocables and printables they contain to serve the role of reminding us of certain features of the sentences in ordinary perceptual discourse which the flags as wholes represent. In particular, the role of the vocable or printable "sense datum" would be that of indicating that the symbolized sentence contains the context ". . . looks . . . ," the vocable or printable "red" that the correlated sentence contains the context ". . . looks red . . ." and so on.

    9. Now to take this conception of sense-datum 'sentences' seriously is, of course, to take seriously the idea that there are no independent logical relations between sense-datum 'sentences.' It looks as though there were such independent logical relations, for these 'sentences' look like sentences, and they have as proper parts vocables or printables which function in ordinary usage as logical words. Certainly if sense-datum talk is a code, it is a code which is easily mistaken for a language proper. Let me illustrate. At first sight it certainly seems that

    A. The tomato presents S with a red sense datum

entails both

    B. There are red sense data


    C. The tomato presents S with a sense datum which has some specific shade of red.

This, however, on the kind of view I am considering, would be a mistake. (B) would follow -- even in the inverted commas sense of 'follows' appropriate to code symbols -- from (A) only because (B) is the flag for b, "Something looks red to somebody," which does follow from a, "The tomato looks red to Jones" which is represented in the code by (A). And (C) would 'follow' from (A), in spite of appearances, only if (C) were the flag for a sentence which follows from a.

    I shall have more to say about this example in a moment. The point to be stressed now is that to carry out this view consistently one must deny to such vocables and printables as "quality," "is," "red," "color," "crimson," "determinable," "determinate," "all," "some," "exists," etc., etc., as they occur in sense-datum talk, the full-blooded status of their counterparts in ordinary usage. They are rather clues which serve to remind us which sense-datum 'flag' it would be proper to fly along with which other sense-datum 'flags'. Thus, the vocables which make up the two 'flags'

    (D) All sense data are red


    (E) Some sense data are not red

remind us of the genuine logical incompatibility between, for example,

    (F) All elephants are grey


    (G) Some elephants are not grey,

and serve, therefore, as a clue to the impropriety of flying these two 'flags' together. For the sentences they symbolize are, presumably,

    (d) Everything looks red to everybody


    (e) There is a color other than red which something looks to somebody to have,

and these are incompatible.

    But one would have to be cautious in using these clues. Thus, from the fact that it is proper to infer

    (H) Some elephants have a determinate shade of pink


    (I) Some elephants are pink

it would clearly be a mistake to infer that the right to fly

    (K) Some sense data are pink

carries with it the right to fly

    (L) Some sense data have a determinate shade of pink.

    9. But if sense-datum sentences are really sense-datum 'sentences' -- i.e. code flags -- it follows, of course, that sense-datum talk neither clarifies nor explains facts of the form x looks f to S or x is f. That it would appear to do so would be because it would take an almost superhuman effort to keep from taking the vocables and printables which occur in the code (and let me now add to our earlier list the vocable "directly known") to be words which, if homonyms of words in ordinary usage, have their ordinary sense, and which, if invented, have a meaning specified by their relation to the others. One would be constantly tempted, that is, to treat sense-datum flags as though they were sentences in a theory, and sense-datum talk as a language which gets its use by coordinating sense-datum sentences with sentences in ordinary perception talk, as molecule talk gets its use by coordinating sentences about populations of molecules with talk about the pressure of gases on the walls of their containers. After all,

x looks red to S  .<-->. there is a class of red sense data which belong to x, and are sensed by S

has at least a superficial resemblance to

g exerts pressure on w  .<-->.  there is a class of molecules which make up g, and which are bouncing off w,

a resemblance which becomes even more striking once it is granted that the former is not an analysis of x looks red to S in terms of sense-data.

    There is, therefore, reason to believe that it is the fact that both codes and theories are contrived systems which are under the control of the language with which they are coordinated, which has given aid and comfort to the idea that sense-datum talk is "another language" for ordinary discourse about perception. Yet although the logical relations between sentences in a theoretical language are, in an important sense, under the control of logical relations between sentences in the observation language, nevertheless, within the framework of this control, the theoretical language has an autonomy which contradicts the very idea of a code. If this essential difference between theories and codes is overlooked, one may be tempted to try to eat his cake and have it. By thinking of sense-datum talk as merely another language, one draws on the fact that codes have no surplus value. By thinking of sense-datum talk as illuminating the "language of appearing" one draws on the fact that theoretic languages, though contrived, and depending for their meaningfulness on a coordination with the language of observation, have an explanatory function. Unfortunately, these two characteristics are incompatible; for it is just because theories have "surplus value" that they can provide explanations.

    No one, of course, who thinks -- as, for example, does Ayer -- of the existence of sense data as entailing the existence of "direct knowledge," would wish to say that sense data are theoretical entities. It could scarcely be a theoretical fact that I am directly knowing that a certain sense content is red. On the other hand, the idea that sense contents are theoretical entities is not obviously absurd -- so absurd as to preclude the above interpretation of the plausibility of the "another-language" approach. For even those who introduce the expression "sense content" by means of the context ". . . is directly known to be . . ." may fail to keep this fact in mind when putting this expression to use -- for example, by developing the idea that physical objects and persons alike are patterns of sense contents. In such a specific context, it is possible to forget that sense contents, thus introduced, are essentially sense data and not merely items which exemplify sense qualities. Indeed, one may even lapse into thinking of the sensing of sense contents, the givenness of sense data, as non-epistemic facts.

    I think it fair to say that those who offer the "another- language" interpretation of sense data find the illumination it provides to consist primarily in the fact that in the language of sense data, physical objects are patterns of sense contents, so that, viewed in this framework, there is no "iron curtain" between the knowing mind and the physical world. It is to elaborating plausible (if schematic) translations of physical- object statements into statements about sense contents, rather than to spelling out the force of such sentences as "Sense content s is directly known to be red," that the greater part of their philosophical ingenuity has been directed.

    However this may be, one thing can be said with confidence. If the language of sense data were merely a code, a notational device, then the cash value of any philosophical clarification it might provide must lie in its ability to illuminate logical relations within ordinary discourse about physical objects and our perception of them. Thus, the fact (if it were a fact) that a code can be constructed for ordinary perception talk which 'speaks' of a "relation of identity" between the components ("sense data") of "minds" and of "things," would presumably have as its cash value the insight that ordinary discourse about physical objects and perceivers could (in principle) be constructed from sentences of the form "There looks to be a physical object with a red and triangular facing surface over there" (the counterpart in ordinary language of the basic expressions of the code). In more traditional terms, the clarification would consist in making manifest the fact that persons and things are alike logical constructions out of lookings or appearings (not appearances!). But any claim to this effect soon runs into insuperable difficulties which become apparent once the role of "looks" or "appears" is understood. And it is to an examination of this role that I now turn.