Programs and texts

Abstract

them, the one that will probably come most readily to the minds of the readers of this journal: the use of computer programs to analyze the literary texts. Let me try by setting out four aspects of this analogy, to isolate the one aspect of their conjunction I wish to discuss. I think that whether I am right or wrong in the arguments I present concerning that aspect, there are general points at issue here that should interest anyone concerned with the general structure of natural languages. First, programs are, in some clear sense, texts. That is a simple truism, since they always consist (regardless of the programming language they are expressed in) of symbols that can be printed out on a page. But there is more to the truism than that: programs clearly have a grammar, as texts do, that must be known to the interpreter or compiler that reads them into the machine. Moreover, the clearest aspect of the analogy is shown by the editing programs that operate on programs-seen-as-texts, so as to alter them, in just the same way as editing programs work on ordinary texts, such as business letters. Texts, on the other hand, are in some sense programs. This assertion is less self-evident, and comes in two versions. In its strong, behavioral form, it asserts that texts used to communicate between people are in fact high-level assertions in a programming language. In this view, "Open the door," said to someone who obeys, is a high-level command which is seen as accessing some subprogram "in" the individual addressed, in a way similar to that in which a macro-function (say SQUAREROOT) accesses lower-order functions in order to carry out a detailed computation (opening a door in this case). This view owes something to the current fashion for "speech acts" (Searle, 1969) in which the behavioral aspect of language is emphasized, and attention is drawn to the way we can use it to create effects in others. Perhaps the clearest statement of the strong view is that "not only should we think of the production and comprehension of natural utterances as processes describable in algorithmic terms; but our utterances themselves should be thought of as pieces of program" (Longuet-Higgins, 1974). It should be emphasized that this view is a general claim about language and is therefore independent of the well-established modes of writing in which language is used to convey to a reader how to carry out a task: instruction manuals, the classic form, long predate programs for computers, so one hesitates to say that an instruction manual is "simulating a program." The weaker form of this view is to be found in an article entitled "Utterances as Programs" (Davies & Isard, 1972), in which an ingenious view is put forward that utterances resemble programs: the way in which certain pronouns are to be understood has strong analogies with the way in which variables are tied within program environments; that is to say, in copies of a database the variables may be tied in different ways in different copies, just as "I like you" has its variables tied differently according to who says it. It will be seen that this weaker view does not assert that utterances are programs, in the crude behavioral manner of the strong view, but rather that techniques for a language understanding program might have a strong similarity to certain wellunderstood techniques for manipulating program