George W. Grace
University of Hawaii
QUESTION: Except for a brief discussion in Grace 1997c we've said nothing about sound change in all of this talk about linguistic change. However, we could hardly think of any discussion of linguistic change as complete without some mention of the question of the regularity of sound change.
ANSWER: I agree.
QUESTION: I think we can begin by stipulating that there's a good bit of empirical evidence that sound changes are often regular--enough evidence that some linguists have even claimed that sound laws know no exceptions. It seems that that claim was an overstatement, but we have no clear understanding of just how regular such change is or under what conditions. However, maybe the really key question here is why it should be regular at all.
ANSWER: I'm going to undertake a first step toward answering that question, but we need to cover some preliminaries first so that we understand just what it is that we're talking about. You might want to begin by asking me what linguists mean when they speak of sound changes' being "regular".
QUESTION: All right, what do they mean? What does the regularity hypothesis in fact hypothesize?
ANSWER: The hypothesis is that (except for isolated cases where extraneous factors intervene) changes in the sounds of a language conform precisely to rules specifying what's happened to particular (pre-change) phonemes in particular (pre-change) environments. Much of historical linguistics has operated from the assumption that there are rules (the problem being to find them) specifying what's happened to every phoneme of a language in every environment.
QUESTION: Why do you word your answer in terms of phonemes when you've said (e.g., in Grace 1983 and 1997c) that you don't believe in them?
ANSWER: It's true that I don't believe that phonemes and other concepts of traditional phonological description have much or anything to do with what people must know--and know how to do--in order to speak a language. However, phonemic theory is essentially a refinement of the point of view of the scholars who developed comparative linguistics and who conceived the regularity hypothesis in the first place. To talk in phonemic terms is to present the hypothesis pretty much as it was conceived, and that makes it easier to talk about. I'll wait until later to attempt to take a new look at the phenomena in the perspective we've been using in this series of Ethnolinguistic Notes.
QUESTION: All right. To get back to the subject, what was the basis for the regularity hypothesis?
ANSWER: Well, we might say that the basis was two-fold--that there were both what we might call empirical and theoretical arguments supporting it. You briefly mentioned the existence of empirical evidence above. It may be said to begin with the observation that, given a linguistic expression in one language, we often can successfully predict what form a cognate in another, related, language will take. (For example, if we know that the Tongan words for "six" and "seven" are, respectively, ono and fitu, our prediction would be that if Hawaiian had cognates, their forms would be ono and hiku, and that in fact turns out to be the case).
QUESTION: Certainly there are many examples of such successful predictions that could be cited from all over the world, but there are also many cases where some of the segments in what appears to be a cognate expression don't conform to the predictions. Don't these cases appear to refute any claim that regularity in sound change is more than sporadic?
ANSWER: Yes, they do, and for that very reason demonstrations in several such cases that what had appeared to be irregular changes were in fact regular made a strong impression on the linguists of the time. Scholars such as Grassmann and Verner showed that once the right environments were identified, certain changes that had appeared to be irregular actually conformed to quite regular rules. These demonstrations appeared to round out the empirical case; their success suggested that all sound changes are in fact regular--that all that's required is for linguists to devote enough time and ingenuity to discover the rule.
QUESTION: To summarize, you're saying that the following two facts provided the empirical basis that the regularity hypothesis rests on: (1) the fact that often the shape of cognates was straightforwardly predictable and (2) the fact that in a number of cases where it had at first seemed not to be predictable, it had subsequently turned out to be so when these cases had been analyzed with sufficient care. Is that a fair summary?
QUESTION: You said that there were both empirical and theoretical bases for the hypothesis. We've covered the empirical one, now what's the theoretical one?
ANSWER: The theoretical basis was a supposed explanation for the supposed regularity. This explanation was the existence of a "mechanism" (to use Hockett's term) by which sound change supposedly proceeded. Underlying it is (once again) the assumption that speakers function in terms of phonemes. The hypothesized "mechanism" involved speakers having an articulatory target for each allophone (i.e., for each phoneme in each of the environments in which it occurs), and these targets drifting over time.
QUESTION: What do you mean by "each of the environments in which it occurs"?
ANSWER: That's a good question, of course, and the answer was never made very clear. But for the scheme to work at all the relevant environments have to be fairly restricted, most often to the one or two phonemes immediately preceding or following.
QUESTION: Why do you say that this restriction would be necessary "for the scheme to work at all"?
ANSWER: Because if the environment were extended to include, say, the entire word, then the regularity hypothesis would be indistinguishable from the supposedly opposite hypothesis that "every word has its own history".
QUESTION: You probably should explain why this hypothesized mechanism would provide a theoretical basis for the regularity hypothesis.
ANSWER: It would do so because if the change proceeded in just the postulated way, it would necessarily be regular (compare the definition of regularity above).
One consequence of this assumption, by the way, was that some linguists considered any phonological change that was not regular to be by definition not the result of "sound change" but of some other process.
QUESTION: All right, but we've agreed that sound changes are often regular. If the mechanism you've just described isn't available as an explanation, what explanation would you propose?
ANSWER: Once again we're at a point where I don't know how to go on without the ground being prepared. In this case what is necessary is to get our thinking free from the phonemic hypothesis and the assumptions that accompany it.
QUESTION: In order to accomplish that, you'll presumably have to tell us more about your alternative to the phonemic perspective. You've said elsewhere that the individual's sound system is a kind of memory store, and that the core of it is stored motor skills. You've said that it's basically an inventory of what you were calling "articulatory gestures"--i.e., articulatory movements that the individual has become practiced at. However, this is still pretty sketchy.
ANSWER: I realize that it is. However, before we go on, I want to point out that the term "articulatory gestures" is an unfortunate one. I began using it a long time ago, but the term is now understood to mean something quite different from what I have in mind. What I have in mind is more like an executable routine; maybe a better term would be "articulatory routines" or "chunks".
QUESTION: How does one go about acquiring these routines? This sound system?
ANSWER: Well, I picture it somewhat like this. From early in the process of building up his/her KOL (i.e., knowledge of language = the store of remembered utterances [see Grace 1997a]), the child practices imitating these utterances. Gradually, s/he becomes more skilled in producing utterances that sound right to other members of the community. That means in particular that the linguistic expressions used can be promptly recognized and that the pronunciation is characteristic of the members of that community.
QUESTION: When you say that "the pronunciation is characteristic of the members of that community", isn't that just another way of saying that the child has acquired the phonological system of the community?
ANSWER: That depends on what you mean by "phonological system of the community". As you know, I don't really believe in phonological systems in the usual sense because I haven't been convinced that a considerably simpler hypothesis wouldn't adequately account for the facts.
QUESTION: What kind of "simpler hypothesis" do you have in mind?
ANSWER: I believe that part of knowing a linguistic expression is knowing a pronunciation for it, and I believe that the pronunciations known by one individual correspond to those used by the other members of the community.
QUESTION: What do you mean by saying they "correspond to" the pronunciations of others? Why not say that they're the same pronunciations as those used by others?
ANSWER: Because they aren't the same. To begin with, people have different voices, and different voices can't produce acoustically identical pronunciations. For that matter, the same person can't even produce two pronunciations that are acoustically identical.
QUESTION: Of course they aren't identical in the strictest phonetic sense, but aren't they equivalent in some sense? Isn't there some system more general than the individual by which they're just different realizations of the same pronunciation?
ANSWER: According to the received view there is, of course; in fact, more than one. I'll comment on the kinds of systems that have been proposed, but in order to do so, I need to begin by describing an assumption made by the received view.
QUESTION: Go ahead.
ANSWER: This assumption is that the stream of speech is segmented. The assumed segments are about letter size. By that I mean that the number of segments in an expression (e.g., a word or a sentence) approximates the number of letters that would be used to write it in an alphabetic orthography.
Each segment is supposed to consist of an articulatory target where each of a number of different articulators must simultaneously assume its own prescribed position. Voiced bilabial stops or high front unrounded vowels would represent examples of such articulatory targets (of course, the prescribed articulation for each target can be specified in greater detail). It is this moment of (theoretical) simultaneity that is supposed to count. Each segment (except the final one) is then followed in real time by a period of transition during which the articulators move to assume their positions for the next target (or the next one in which they have a crucial role).
QUESTION: And you question the assumption that the stream of speech is so segmented?
ANSWER: Yes. I question in particular that it's so segmented in the speakers' knowledge of language.
QUESTION: What's the evidence? For example, can this segmentation be observed when the movements of the speech organs or the acoustic stream are monitored?
ANSWER: Well, I'm not a phonetician, but as far as I'm aware there's nothing to distinguish the targets from the transitions; the speech organs seem to be in continuous movement as long as the person is speaking (acoustic analysis also seems to confirm this). However, that in itself doesn't necessarily invalidate the hypothesis of segmentation of course.
QUESTION: All right, now I'll let you continue with your answer to the question. You'd said that the language learner couldn't pronounce anything in the same way as anyone else because no two pronunciations could be identical, and I'd asked if there wasn't some more general system in which they could count as the same. You were going to describe the kinds of general systems that have been proposed.
ANSWER: Yes. These systems are designed for segments (i.e., the supposed articulatory targets). Articulatory phonetics is one such system. It provides a classificatory scheme for segments. For example, corresponding segments abstracted from two different utterances might be said to be the same in the sense of being classified alike in terms of articulation--say, as voiceless interdental fricatives.
But what you probably have in mind is the notion that there's a set of phonemes and phonological rules that's shared by the speakers of the same language (or, if [in light of previous Notes in this series] we want to avoid the concept "language", say, by the members of the community from whom the child is learning to speak), and that each of the supposed segments in a given utterance is a realization of one of these phonemes. That's what I'm not prepared to assume.
QUESTION: Why? I don't want to get us off the subject, but maybe a very brief reminder of why you're not prepared to make that assumption would be helpful here.
ANSWER: Very briefly, because it's a violation of scientific parsimony to attribute more structure to people's knowledge of language than is required to account for the kinds of competence that they display.(1)
QUESTION: And what kinds of competence do people display?
ANSWER: That question leads to another: what people are we talking about? Does the phonemic hypothesis that we're considering hypothesize that a phonemic system is a necessary part of the linguistic competence of a speaker of a language (i.e., that having such a system is a necessary condition for qualifying as a normal speaker of any language past or present)? That's the hypothesis (or version of the hypothesis, if you prefer) that I'm not prepared to accept. If you can formulate some weaker version, I might be willing to accept that.
For example, I would hold that our assumptions about phonological competence shouldn't attempt to account for all of the performance of the most skilled users of language. No doubt the KOL (knowledge of language) of the most gifted creative artists often contains all kinds of analytical and other knowledge. I emphatically am not claiming that any particular kinds of knowledge might be impossible for exceptional individuals.
Another point: what would qualify as normal competence in one culture may differ from that in another. For example, there are forms of disguised speech in the USA that depend on the letters used in spelling the words (i.e., the letters as distinct from the sounds they represent in the words in question). If the use of such disguised speech is seen as part of normal competence, then an ability to spell must be considered to be a part of normal competence in this culture. However, we can be sure that it was not a part of normal competence in preliterate cultures.(2)
In short, I assume that the speakers that we must try to account for are those without any of these kinds of special competence. We should be thinking in terms of the minimal knowledge required to qualify (at any time and place since full-fledged language has existed) as normal speakers.
QUESTION: All right, what do your imagine this minimal knowledge to consist of?
ANSWER: Let me begin by saying that I think of it as being more a matter of knowing how to do something than factual knowledge ("knowledge-how" rather than "knowledge-that"). I think that what one has to learn how to do is (1) recognize the linguistic expressions in one's memory store--and innovative expressions based on them--when they're pronounced by (at least the most significant) others, and (2) pronounce those expressions--or innovations based on them--in such a way that they are recognizable by others.
QUESTION: Well, let's set the ability to recognize all of the different pronunciations of different speakers aside for a moment and just consider the knowledge needed to pronounce all of the possible expressions oneself. There are an infinite number of possible expressions. It's obvious that one can't learn to pronounce each of them individually. It seems that it has to be possible to break these down into phonemes or some other kind of phonetic signals that are finite in number and are used again and again in the pronunciations of the entire repertoire of expressions. Then the pronunciations of these phonemes--or whatever--would be all that the learner has to learn.
ANSWER: I'm not exactly arguing against that. What I believe is that the learner has to develop skill in producing certain coordinated movements of speech organs, and I've found it easier to talk about it as if these could be classified into a finite number of what we're calling "articulatory routines".(3) Usually each of the articulatory routines required to pronounce a particular linguistic expression will be one that's also used in the pronunciation of many other linguistic expressions.
QUESTION: How do your articulatory routines differ from phonemes?
ANSWER: In several ways. First of all, they're physical skills and maybe nothing more. I think that for many linguists, the phoneme is a more intellectual unit of knowledge--more like "knowledge-that" than "knowledge-how".
Second, I don't think of the articulatory routines as being a single segment in length. In fact, I imagine them to be of varying lengths, and generally not at all the same length for different speakers of the same language. I don't think of the kind of economy and elegance that are key considerations in classical phonemic analysis as playing any role here.
Thirdly, some of the classical questions of phonemic analysis consequently don't play any role here: for example, whether or not somewhat similar portions of different articulatory routines are to be counted as being in some kind of allo- relation to each other (analogous to that of allophones of the same phoneme). Again there's nothing analogous to the question of whether a particular portion of various routines is to be counted as a single (e.g., affricate) segment or as a sequence (e.g., a stop plus a fricative) of two.
QUESTION: In your scheme would it even be permissible to say that the difference between English mate and met is the same one as that between bait and bet?
ANSWER: Let me preface my answer to that question with a disclaimer. I can't really make any judgments about the facts of "English" (at any rate, without further qualification) because I think that "English" (i.e., "the English language") is ultimately a fiction.
However, there's a considerable literature in which quite a store of what have been generally accepted as facts about English has been built up. It'll be a lot easier to be more concrete in this discussion (and to answer your specific question) if I assume some of these "canonical" facts to be valid. That is, I'll assume that there is an English language in which the words under discussion occur and have the phonological properties usually attributed to them.
Now to answer your question: as far as I can see now, we'd be justified in saying that in most KOLs of English-speaking individuals bait contrasts with bet and that mate contrasts with met, and even that the contrast is the same one in both cases.
QUESTION: All right, what about the other skill you say the learner must master: that of recognizing linguistic expressions in all of the varied pronunciations of the other speakers of the language?
ANSWER: That's really a matter of learning to operate with sound correspondences, some of which are more subtle and therefore less likely to come to the speaker's conscious attention than others.
QUESTION: Can you give examples of the kind of sound correspondences that you're talking about?
ANSWER: We might start with some less subtle ones. Again, I'll resort to what I assume to be generally-accepted "facts" about "English". Consider some of the (largely regional dialectal) differences in sound systems of English speakers. Some speakers of English don't distinguish between words such as pen and pin. Others don't distinguish among Mary, merry, and marry; still others don't distinguish hoarse and horse, whine and wine, etc. In each of these cases, it's not a question of certain individual words, but of more generalized articulatory routines.
Dialect differences such as these represent regular sound correspondences, and speakers must learn in effect to apply the correspondence in understanding and repeating what others have said. (Of course, not all differences of pronunciation can be accounted for by regular sound correspondences, there are also isolated cases: for example, the two pronunciations of either and neither).
However, it's important to recognize that we constantly operate with much subtler correspondences, some of them isolated cases but many quite regular, because no two pronunciations of the same linguistic expression are ever identical.
QUESTION: Even though you don't believe in phonemes, I'd think that the simplest way to describe each of the correspondences that you cite would be in terms of phonemes: a particular phonemic contrast is neutralized before nasals in some dialects and not in others, etc. How can you conceive of regular sound correspondences other than in phonemic terms?
ANSWER: I don't disagree with your statement that the easiest way to describe the correspondences is in terms of phonemes. There can be no doubt that the concepts of linguistic descriptions are convenient for talking about quite a few things. This is an example of the effectiveness of what Andrew Pawley (1991) has called "subject matter codes". A highly-elaborated subject matter code for talking about human language has been developed in Western culture--in considerable part by linguists--and it's very difficult to say much about language without drawing on this code (and thus tacitly accepting its assumptions). However, this all-too-convenient code is based on the analysis of the artifacts of speech ("texts"), and not on the knowledge that underlies the speakers' ability to use language.
What I am maintaining is that the effective correspondences aren't between phonemes, but between the pronunciations of linguistic expressions (or at least of the meaningful units that compose them).(4) If it's possible on the basis of A's pronunciation of an expression to make valid predictions about how B will pronounce the same expression, then there are regular correspondences between A's and B's phonologies.(5)
QUESTION: All right, I'll let that stand. But we seem to be engaged in endless digressions. What else stands in the way of our getting back to the question of why sound change should be regular?
ANSWER: Actually, we're back to it right now. My explanation is that the regularity of sound change is explained by the very regularity of sound correspondences that we've been discussing.
QUESTION: Wait. That's the exact opposite of the traditional explanation. The traditional explanation has been that regular sound correspondences exist because of regular sound change; you're saying that regular sound change occurs because of regular sound correspondences. You'd better explain.
ANSWER: The explanation isn't very complicated. All I'm saying is that the people who talk to one another in the same "language" (to use the standard subject-matter code)--let's call them collectively "the community"--have to be connected by regular sound correspondences. The community at any one time is made up of people of different ages, each linked to each other one (in our hypothetical situation, at least) by a pattern of regular sound correspondences.
However beyond that, there have also existed in the recent past--and still exist in the memory of the older members of the community--regular sound correspondences linking these older members to (still older) people now dead; and there will be similar links between the younger members of the community and people yet to be born. As we said above, each child learning the language learns to produce utterances that "sound right" to members of the community--that is, which show regular sound correspondences to the speech of other members.
In short, people whose generations haven't overlapped at all are still connected by a chain of links of regular sound correspondences.
QUESTION: I understand you to be saying that regularity of sound correspondences is transitive; that is, if there are regular correspondences between speakers A and B, and again between B and C, then there will necessarily be regular correspondences between A and C (and so on for any number of additional speakers D, E, F, etc.). And since A, B, C, etc. may belong to any number of different generations, sound change is necessarily regular?
ANSWER: That, with certain qualifications, is what I'm proposing. One qualification is that, as we mentioned above, there will surely be some isolated cases of pronunciation differences that don't conform to any regular correspondences (our example was the two pronunciations of either and neither), and these will tend to accumulate as the number of links needed to connect the speakers increases. Likewise there will be less and less overlap in the utterances that speakers produce as the time gap between them increases, and therefore the evidence of the regularity of their sound correspondences will diminish accordingly.
The other qualification is that our discussion so far has tacitly assumed ideal conditions of a certain kind--conditions that have rarely if ever obtained anywhere for any length of time.
QUESTION: What conditions do you mean?
ANSWER: The picture that's been presented so far assumes, at least tacitly, a monolingual community that has no effective contact with any other community, that has no significant internal variation, and where each succeeding generation of speakers apparently consists of the offspring of the preceding generation. Any deviation from those ideal conditions could be expected to complicate the pattern of observed linguistic change.
QUESTION: What do you mean by "observed" change?
ANSWER: I mean that what gets counted as linguistic change is derived from the comparison of language states (états de langue) or more precisely, of linguistic descriptions which are supposed to characterize language states. (Here I use the term "linguistic description" to refer to any descriptive material that may exist to represent a language--this can be as little as a short word list).
QUESTION: You seem to be implying that the "observed" change is different from the actual change; you probably should explain farther.
ANSWER: My only point is that if, when you think about the problem of regularity in change, what you're imagining is something like our ideal situation of a homogeneous, monolingual community with no contact with other languages or dialects, then what you're imagining probably bears little resemblance to the actual situations represented by many of the available linguistic descriptions. For further discussion of this problem see Grace 1997b.
QUESTION: Then, what is your final word about the regularity of sound change?
ANSWER: That the necessity for the speakers to operate with regular sound correspondences does, when taken by itself, constitute a strong force leading toward such regularity, but that the results of this force are likely to become considerably less evident in real-life situations because of the presence of other factors.
1. The pronunciation of meaningful units has to be memorized--that is to say, learned holistically--in any case. (For present purposes, there'd be no harm in just assuming that all of the "meaningful units" so memorized are "morphemes", although in fact holistic learning no doubt frequently applies to much longer units as well). The point about parsimony is that the phonemic hypothesis would require that the speaker acquire both a holistic and an analytic knowledge of the pronunciation of the meaningful unit.
I don't think of my articulatory routines as implying analytic knowledge--at least, in the same sense. What I imagine to happen the first time one experiences the need to pronounce a particular unit is that one starts with some kind of image of the target pronunciation in mind and then attempts to execute that pronunciation, and that in that attempt one draws upon whatever articulatory skills one possesses (many of these skills having been developed in previous pronunciations of expressions in the language). I conceive of these skills as something that an observer could (theoretically) analyze into what we're calling "articulatory routines".(Back up)
2. I particularly think that one of our concerns should be with the biological basis of language. I don't think we should assume that our species has undergone any significant evolutionary changes in the few years since writing was invented. Consequently, if we're concerned with identifying the innate language competences of humans, we need to be fairly meticulous in ruling out anything in the behavior (and inferred knowledge) of the humans whom we study that could result from anything at all that they might know about writing (even that it's possible). (Back up)
3. I should point out that although I speak of articulatory routines as if each were a well-defined unit, I'm really not sure how easily these skills in manipulating the speech organs that are a necessary basis for fluent pronunciation can be sorted neatly into a fixed number of "routines". Nevertheless, I find it convenient to talk in terms of such entities as long as this way of talking isn't otherwise misleading.
Note that I've discussed my conception of articulatory routines (under the name "articulatory gestures") in slightly more detail in Grace 1983 (esp. pp. 289-91) and 1984 (esp. 317-19). (Back up)
4. There's another point that needs to be made here: one shouldn't think of these correspondences as they are known to the speaker as having the precision of a mathematical formula. Rather, I imagine them to involve our poorly understood capacity for pattern recognition. Our ability to recognize the same tune as played by different instruments (and also to recognize which instrument is playing it on any given occasion) and the ability to recognize variations on a musical theme both seem to me to offer suggestive analogies. I can't begin to give a formal account of how this capacity works, but I have discussed it briefly in Grace 1981, esp. pp. 67-71, where I proposed that it's "a result of the natural functioning of the perceptual apparatus" (1981: 71). (Back up)
5. I assume that we could even go so far as to say that there are regular correspondences between a certain person's pronunciations under some extraordinary condition--such as when he or she had a bad cold--and his/her "normal" pronunciations. (Back up)
Grace, George W. 1981. An essay on language. Columbia SC: Hornbeam Press. (Back up)
Grace, George W. 1983. Why I do not believe in phonemes: On the cognitive validity of linguistic theories of phonology. Ethnolinguistic Notes, Series 3, Number 17. Printout. Also (1996) Internet WWW page at http://www2.hawaii.edu/~grace/eln17.html. (Back up)
Grace, George W. 1984. More on the reality of phonemes. Ethnolinguistic Notes, Series 3, Number 19. Printout. Also (1996) Internet WWW page at http://www2.hawaii.edu/~grace/eln19.html. (Back up)
Grace, George W. 1997a. Linguistic change: 5. The individual's knowledge of language. Ethnolinguistic Notes, Series 4, Number 6. Internet WWW page at http://www2.hawaii.edu/~grace/elniv6.html. (Back up)
Grace, George W. 1997b. Linguistic change: 6. The individual's knowledge and the traditional notion of languages. Ethnolinguistic Notes, Series 4, Number 7. Internet WWW page at http://www2.hawaii.edu/~grace/elniv7.html. (Back up)
Grace, George W. 1997c. Linguistic change: 7. How does "the language" change? Ethnolinguistic Notes, Series 4, Number 8. Internet WWW page at http://www2.hawaii.edu/~grace/elniv8.html. (Back up)
Pawley, Andrew. 1991. "How to talk cricket: On linguistic competence in a subject matter. In Robert Blust (ed.). Currents in Pacific linguistics: papers on Austronesian languages and ethnolinguistics in honour of George W. Grace. Pacific Linguistics, series C-117, pp. 339-68. (Back up)
|Home Page||The Ethnolinguistic Notes||The Ethnolinguistic Notes, Series 1 and 2||Ethnolinguistic Notes, Series 3||The Ethnolinguistic Notes, Series 4||Reflections: Language Evolution|
|Reflections: Knowledge of Language||Personal Page||The Human Predicament||Why Write Unpublishable Things?||Modest Proposals||Odds and Ends||Pictures|
First put on the Web on 3 December 1997
Revised 16 December 1997
Spelling corrections made 6 January 1998