The Problem of Universals in Language
by Charles F. Hockett
"The only useful generalizations about language are inductive generalizations" (Bloomfield, 1933, p.20). This admonition is clearly important, in the sense that we do not want to invert language universals, but to discover them, How to discover them is not so obvious. It would be fair to claim that the search is coterminous with the whole enterprise of linguistic in at least two ways. The first claim in which this claim is true is heuristic: we can never be sure, in any sort of linguistic study, that it will not reveal something of importance for the search. The second way in which the claim is plausible, if not automatically true, appears when we entertain one of the various possible definitions of linguistics as a branch of science: that branch devoted to the discovery of the place of human language in the universe. This definition leaves the field vague to the extent that the problem of linguistics remains unsolved. Only if, as is highly improbable, the problem were completely answered should we know exactly what linguistics is -- and at the same millennial moment there would cease to be any justification for the field. It is hard to discern any clear difference between "the search for language universals" and "the discovery of the place of the human language in the universe." They seem rather to be, respectively, a new-fangled and old-fashioned way of describing the same thing.
1.1 The assertion of a language universal must be founded on extrapolation as well as on empirical evidence.
Of course this is true in the trivial sense that we do not want to delay generalizing until we have full information on all the languages of the world. We should rather formulate generalizations as hypotheses, to be tested as new empirical information becomes available. But there is a deeper implication. If we had full information on all languages now spoken, there would remain languages recently extinct on which the information was inadequate. There is no point in imaging that we have adequate information also on these extinct languages, because that would be imagining the impossible. The universe seems to be so constructed that complete factual information is unattainable, at least in the sense that there are past events that have left only incomplete records. Surely we seek constantly to widen the empirical base for our generalizations; equally surely, we always want our generalizations to subsume some of the unobserved, and even some of the unobservable, along with all of the observed.
1.3 A feature can be widespread or even universal without being important
This is most easily shown by a trick. Suppose that all the languages of the world except English were to become extinct. Thereafter, any assertion true of English would also assent a (synchronic) language universal. Since languages no longer spoken may have lacked features we believe universal or widespread among those now spoken, mere frequency can hardly be a measure of importance.
1.4 The distinction between the universal and the merely widespread is not necessarily relevant
The reasoning is as for 1.3. Probably we all feel that the universality of certain features might be characterized as "accidental" -- they might just as well have turned out to be merely widespread. This does not tell us how to distinguish between the "accidentally" and the "essentially" universal. On the other hand, that which is empirically known to be merely widespread is thereby disqualified as an "essential" universal -- though careful study may show that it is symptomatic of one.
1.5 The search for universals cannot be usefully separated from the search for a meaningful taxonomy of languages.
(Here "taxonomy" refers to what might also be called "typology", not to genetic classification.) Suppose that some feature, believed to be important and universal, turns out to be lacking in a newly discovered language. The feature may still be important. To the extent that it is, its absence in the new language is a typological fact of importance about the language.
Conversely, if some feature is indeed universal, then it is taxonomically irrelevant.
Here is an example that illustrates both 1.4 and 1.5. It was at one time assumed that all languages distinguish between nouns and verbs -- by some suitable and sufficiently formal definition of those therms. One form of this assumption is that all languages have two distinct types of stems (in addition, possibly, to various other types), which by virtue of their behavior in inflection (if any) and in syntax can appropriately be labeled nouns and verbs. In this form, the generalization is rendered invalid by Nootka, where all inflectable stems have the same set of inflectional possibilities. The distinction between noun and verb at the level of stem is sufficiently widespread that its absence in Nootka is certainly worthy of typological note (1.5). But it turns out that even in Nootka something very much like the noun-verb contrast appears at the level of whole inflected words. Therefore, although Nootka forces the abandonment of the generalization in one form, it may still be that a modified form can be retained (1.4).
The Port Royal Grammar constituted both a putative description of language universals and the basis of a taxonomy. The underlying assumption was that every language must provide, by one means or another, for all points in the grammatico-logical scheme described in the Grammar. Latin, of course, stood as the origin in this particular coordinate system. Any other language could be characterized typologically by listing the ways in which its machinery for satisfying the universal scheme deviated from that of Latin. This classical view in general grammar and in taxonomy has been set aside not because it is false in some logical sense but because it has proved clumsy for many languages: it tends to conceal differences that we have come to believe are important, and to reveal some that we now think are trivial.
Given a taxonomy, if we find that languages of the most diverse types nonetheless manifest some feature in common, that feature may be important. It is not apt to be, however, if it is an easily diffusible item. Thus the fact that many languages all over the world have phonetically similar words for 'mama' is more significant than a similar widespread general phonetic shape for 'tea'.
In allowing for diffusion, we must also take into consideration that even features that do not diffuse readily may spread from one language to others when the speakers of the languages go through a long period of intimate contact. This fact, if no other, would seem to render suspect any generalizations based solely on the languages of Western Europe. And it is true that some such generalizations are refused by the merest glance at an appropriate non-European language. But contrastive study based exclusively on European languages also has a merit: our knowledge of those languages is currently deeper and more detailed than our knowledge of languages elsewhere, so that generalizing hypotheses can also be deeper. They may be due for a longer wait before an appropriately broad survey can confirm or confute them, but they are valuable nonetheless.
1.9 A universal feature is more apt to be important if there are communicative systems, especially nonhuman ones, that do not share it.
It may seem peculiar at first to propose that we can learn more about human language by studying the communicative systems of other animals; but a moment's reflection is enough to show that we can only know what a thing is by also knowing what it is not. As long as we confine out investigations to human language, we constantly run the risk of mistaking an "accidental" universal for an "essential" one -- and we bypass the task of clearly defining the universe within which our generalization are intended to apply. Suppose, on the other hand, that after discovering that a particular feature recurs in every language on which we have information, we find it lacking in some animal communicative system. In some cases, this might lead us to add the feature to our defining set of language. In any case, this seems to be one way of trying to avoid triviality in the assembling of our defining set.
1.10 The problem of language universals is not independent of our choice of assumptions and methodology in analyzing single languages.
5.2 Phonemes are not fruitful universals.
We can, indeed, speak quite validly of phonemes in the discussion of any language, but their status in the hierarchy of phonological units varies from one language to another, and also, to some extent, through varying preference or prejudice of analysts. The status of phonological components, on the other hand, is fixed once and for all by definition -- phonological components are the minimum (not further divisible) units of a phonological system. Given that all phonological patterning is hierarchical, the exact organization of the hierarchy, varying from one language to another, becomes a taxonomic consideration of importance, but not the basis of a generalization in the present context.
There are certain languages of the Caucasus (Kuipers, 1960) where one can, if one wishes, describe the phonological system in terms of perhaps a dozen phonological features organized into some seventy or eighty phonemes, which in turn occur in about twice that many syllables. Each syllable consists of one of the seventy-odd consonant phonemes, followed by one of the two vowel phonemes. It seems clear in such a case that the vowel "phonemes" are better regarded simply as two additional phonological features, so that a unit such as /ka/ is just a phoneme -- or, alternatively, that the term "phoneme" be discarded and one discuss the participation of features directly in syllables. Either way, one does not need both the term "phoneme" and the term "syllable". The case may be extreme, but it is real, and underscore the importance of the "anti-universal" given as 5.2.
5.5 Sound change is a universal. It is entailed by the basic design features of language, particularly by duality of patterning.