Is That a Fish in Your Ear?

One formerly fashionable way of avoiding the insoluble problem of fixing the meaning of a word was to imagine it as the compound product of sublinguistic mental units or “features” of meaning. Take the three words house, hut, and tent. They can all be used to refer to dwellings of some sort, but they refer to three different kinds of dwelling. The task of distinctive feature analysis was to find the minimal semantic constituents that would account for the meaning relations among these three semantically related terms. All three are “marked” with the feature [+dwelling], but only house also has the two features [+permanent] and [+brick]. Tent would be marked [–perma-nent] [–brick] and hut could be marked [+permanent] [–brick]. How wonderful it would be if all words in the language could be decomposed into atoms of meaning in this way. The meaning of a word would then be fully specified through the list of the distinctive features that mark it. If you could show that it was possible to account for the differences in the meanings of all the words in a language by the distribution of a finite set of semantic features, then you could go further still. You would be in a position to build a great Legoland of the mind, in which all possible meanings could be constructed out of irreducible, binary building blocks of sense.

To map some area of vocabulary (let alone a whole language) using only such elementary features of meaning is an enticing prospect, but it runs up against a fundamental problem: what criterion to use to establish the list of the elementary semantic features themselves. Common sense no doubt dictates that [±animate] and [±female] are among the distinctive features relevant to the meaning of the term woman and that [±chrome-plated] is not. But common sense appeals to our total experience of the nonlinguistic world as well as to our ability to find a way through the language maze: it is precisely the kind of fuzzy, vague, and informal knowledge that distinctive feature analysis seeks to overcome and replace. Despite the usefulness of binary decomposition for some kinds of linguistic description and (in far more complex form) in the “natural language processing” that computers can now perform, word meanings can never be fully specified by atomic distinctions alone. People are just too adept at using words to mean something else.

Such quasi-mathematical computation of “meaning” is equally unable to solve an even more basic problem, which is how to identify the very units whose meaning is to be specified. To ask what a word means (and translators often are asked to say what this or that word means) is to suppose that you know what word you are asking about, and that in turn requires you to know what a word is. The word word is certainly a familiar, convenient, and effective tool in the mental toolbox we use to talk about language. But it is uncommonly hard to say what it means.

Computers must know the answer, because they count words. That’s no consolation to us, however. What computers know about words is what they’ve been told, which comes down to this: a word is a string of alphabetic characters bounded on left and right by a space or one of these typographical symbols:—/ ? ! : ; , .[43] Computers don’t need to know what a word means to carry out the operations we ask of them. But we do! And if in some instance we really don’t, then we try to find out from a dictionary, from an acquaintance, or from listening to how other people speak. But all kinds of problems remain.

In languages such as English the identification of words is more art than science. Publishers have their own style sheets with rules for deciding whether couples have break-ups or break ups or breakups; but ordinary people also want to know if “to break up” should be counted as one, two, or three words. Yet nobody can really say.[44]

English prepositional verbs provide unending employment for language experts who want to determine what a word is. They come in three or four parts. Sometimes they stay together—“Did you remember to take out the trash?”—and sometimes they don’t: “I promised to take my daughter out to see a film.” Does that mean that “to take out” is a word (or three) or two different words—“to take out” and “to take … out”—(or six) that look the same? Compilers of alphabetical dictionaries adopt practical solutions, but not the same ones, leaving the underlying question—what word is this?—unresolved. Teachers of English as a second language know the best answer to the question of how many words there are in a prepositional verb. If you want to know how to use the language properly, don’t ask.