WHY I DO NOT BELIEVE IN PHONEMES:
ON THE COGNITIVE VALIDITY OF LINGUISTIC THEORIES OF PHONOLOGY[1]
Ethnolinguistic Notes, Series 3, Number 17, 1983

George W. Grace

For some time I have been concerned that the theories of language which have been developed by linguistics give a misleading view of the nature of the abilities which language presupposes and hence, ultimately, of the nature of the mind. In various previous writings I have tried to show that the assumptions underlying some of these theories are not literally true. However, so far I have said very little about what have perhaps been the most influential of all of these theories the theories of phonology. Therefore, it is my purpose here to turn to phonology and to propose a different kind of hypothesis to account for the abilities which it requires. If it should prove to be valid, this hypothesis will throw a quite different light on the question of the nature of mind.

Human language is a phenomenon of great importance. It is also one which is very inaccessible to direct observation. It is manifested only in its results, and it is only through those results that we are able to study it at all. There is not even unanimous agreement among linguists as to where it is located. Some hold that it is a phenomenon of the human mind that its locus is the minds of its speakers (although the mind is, in its turn, also observable only through its own results). Others would seemingly opt for the idea that language or at least individual languages are superorganic phenomena. It is not clear to me whether or not superorganic phenomena require any locus at all.

In any case, it seems clear that language, whatever and wherever it is, is not directly accessible to observation. Under these circumstances, the only strategy available is to observe the phenomena which result from it and then to construct models to account for them. That is, we must contrive some way of talking to use for talking about language (for ways of talking about things, cf.Grace 1982b,c).

The crucial point is that the observable results of language those phenomena which we are actually able to observe- are at a very great remove from human language, itself. Under such circumstances, the kinds of models which one actually constructs will depend relatively much on the state of mind in which one approaches the task, and relatively little upon the nature of the object in this case, language itself.

There is no serious harm in that as long as the models serve the purposes for which they were designed. Where the harm comes is when they are converted to quite different, and unanticipated, purposes. I think that that is happening in the case of linguistic models of language.

The fact is, linguistic theories as presently constituted are not an accurate guide to the nature of human language and, in fact, were never really designed to be one. The primary concern of linguistics throughout most of its history has been with the differential characteristics of individual languages, and recent attempts to construct a theory of the nature of language in general are still heavily influenced by that earlier concern.

But there are questions for which the nature of language becomes a serious concern. Consider, for example, the problem of determining in what ways particular abilities which animals of other species have shown themselves capable of acquiring differ from the abilities required by human language. Or, for that matter, consider the whole problem of the evolution of language, that is, the (presumably gradual) accumulation of those abilities which language requires. Or, more generally, consider the recently growing interest in the part language has played in the development of those abilities, especially the intellectual abilities, which characterize the human species. These new concerns have aroused a more general interest in the abilities which underlie language i.e., the abilities which are presupposed by language.

At first glance, linguistics would seem to be in a good position to contribute to such inquiries. As it happens, contemporary linguistic theory has accorded particular status to what has been called "linguistic competence", which is conceived of as "knowledge" which underlies the ability to speak a particular language. This language-specific competence, in turn, is supposed to be underlain and shaped by certain native abilities. It is upon these native abilities that the capacity for language rests. It is quite natural then that these native abilities, and linguistic competence in general, should be considered important evidence in the identification of the human syndrome. And yet, I want to claim that this competence is representative of only a hypothetical idealized speaker, and that the abilities required of this idealized speaker are quite unrepresentative of the abilities of real speakers. Consequently, I believe, current linguistic conceptions of the abilities involved in the capacity to use language are likely to constitute as much a distraction as an aid to attempts to discover the nature of human abilities.

"Pictures" of How Things Work

My whole point in this paper turns on the claim that the conclusions which we reach are significantly shaped by the assumptions which we bring to the inquiry in the first place. Therefore, the kind of argument that I will be presenting will several times require me to attempt to describe sets of underlying assumptions, or as I prefer to put it, "pictures" of how the things in question work or of the basic principles according to which those things are organized.

However, I need first of all to say something about the problem inherent in attempting to describe such "pictures". The main problem is evidenced by the fact that the very people who are supposed to operate with any particular such picture usually reject whatever description is proposed to represent it. A familiar example would be Noam Chomsky's descriptions of the assumptions of American structuralist linguistics. I do not find it surprising that such descriptions are usually rejected. Presented in such skeletal form, they seem extremely simplistic- so simplistic that it would be embarrassing to have anyone imagine that they accurately described what any experienced professional actually believed. And certainly no experienced professional does believe that they represent the whole story; if they did, there would be nothing further left to do.

However, I believe that what we as scientists do constantly do is to accept some simple model such as those which I will sketch in this paper as representing the essential nature of the relationship the principle upon which the system is organized. Once such an assumption has been made, we are apt to regard all instances which the principle does not seem to fit as elaborations upon what continues to be regarded as the general rule. One can even imagine two individuals or two schools of scholars who agree in detail about the interpretation of all individual cases, but who are in fundamental disagreement as to what the organizing principle of the system is. Therefore, I feel that if I am creating straw men in the sense that no sophisticated person would believe that the pictures I will sketch accurately represent the whole truth, such pictures can nevertheless exercise a very serious influence upon the conduct of inquiry.

Thus, then, I think that the following picture, even though it is doubtless an oversimplification of what any individual linguist actually believes, represents an approximate consensus about the general principles according to which language is organized. I think it is a fair representation of the picture which we would present, if not to fellow intellectuals, at least to the person in the street or the beginning student.

In this paper I will be dealing only with the third of the three points below, that dealing with the nature of phonology. The first two points are given in order to establish the context within which our theories of phonology were designed to perform. My limiting the discussion to phonology should not be taken to suggest that I believe that the problem of misrepresenting the nature of language is limited to phonology, or even that it concerns phonology in a more serious way than the other aspects of language. In fact, a large part of what I have written over the past few years has dealt with this same misrepresentation by linguistic models for the other half of the linguistic sign.

However, phonological theories, and particularly the concept of the phoneme, have provided a cornerstone for linguistic theory generally. If for no other reason, then, they seem to merit more careful scrutiny.

The main points of what I propose as linguistics' picture of how language is organized are as follows:

(1) Language exists in the form of a (largish) number of units called "languages". The individual language is what we might call the domain of systematization i.e., the principal systems of phonology, syntax, lexicon, and whatever else one recognizes morphology, semantics, pragmatics are attributable to particular languages. Differently put, they are organized by individual language. [2]

(2) The main task of linguistics is to describe these units, or at least to determine how such units should be described- specifically, what sort of formal properties are called for in the language of description.

(3) One part of a language is a lexicon, which consists of entities (call them "words") composed of a form and a meaning. The form is not a written form, but a spoken one. I call it a "lexification". Specifically, each lexification consists of a particular sequence of segments, and this is particularly clear in structuralist theory all of the segments in the lexifications of all of the words of a given language are drawn from a small fixed inventory of such segments, called "phonemes". Phonemes are thus analogous to letters, the phonemic inventory of the language is analogous to an alphabet, and the phonemic representation of a particular word is analogous to its spelling. A major part of what I want to question is that such phonemic alphabets are actually part of languages.

Plan of this Paper

What I will attempt to do here is as follows:

1. I will attempt to explain the principal assumptions of phonological theory, at least in part, by showing that they derive from deeper-lying assumptions which are part of our particular cultural tradition.

2. I will present a much simpler hypothesis of the nature of phonological competence a hypothesis which I believe still accounts for all of the relevant facts.

3. I will discuss certain other facts which seem to be better explained by my hypothesis.

Assumptions Underlying the Principal Assumptions of Phonological Theory

I will attempt here to explain some of the assumptions of phonological theory as due in part to the influence of further, underlying, assumptions. That is, the package of assumptions involved in the concept of the phoneme become more understandable if we assume that those who first made them were influenced by certain implicitly held pictures of how things work. I will point out three key assumptions of phonological theory, and attempt to identify the sources of each in our cultural tradition.

1. An assumption that phonological segmentation is a given, that is, that there is a correct segmentation which is obvious to all. (For this assumption and all of the others here, there is of course the qualification that the assumption as stated is not regarded as precisely true in all cases. The actual assumption is that it is near enough to being true to be considered an accurate characterization of the general structural principle involved.)

I attribute this assumption primarily to the influence of the alphabetic writing tradition. The phonological segmentation which we attribute to linguistic expressions ordinarily closely approximates what would be expected in an alphabetic system of writing for the language.

2. An assumption that the phonological segments in the expressions of a particular language are tokens of types, and that these types are categories which stand to one another in a relation of categoric exclusion. That is, they do not overlap. (In generative phonology there can be overlap in surface representations, but underlying representations are characterized by distinctive features specifications).

I attribute this assumption also in part to the influence of the alphabetic writing tradition.

However, I believe that there is a much more general and more fundamental picture that is at work here. The organizing principle depicted in this picture applies only incidentally to phonology, but I believe that it has some effect here as well. Although I have alluded to it previously e.g., in Grace 1982a, 1983a, 1983b I have never given a very clear description of it. Therefore, I will try to describe it more fully here even though its applications to phonology are only peripheral. It is, in fact, a picture of the epistemological functioning of language.

In this picture the universe is a closed system, or near enough to being one for us to act as if it were. That is, there is a single finite universe which is common to all of mankind and, therefore, to the speakers of all languages. Any given language is analogous to a map for that universe. That is, the language stands in a relation to that universe which is like the relation between a map and the territory which it is a map of. This is true because the language (by means of its vocabulary) divides up the real universe in much the way that a political map divides up the earth's surface it represents an exhaustive partition of reality.

Each political unit which appears on a map corresponds to a defined territory. Analogously, a lexical item in a particular language corresponds to a certain territory within the real universe. Therefore, it follows that a word can be defined by giving an exact specification of the boundaries of the real world territory which it represents. Translation from one language to another is essentially a matter of expressing the same points in reality in terms of the categories of the second map rather than those of the first [3]

I think it is obvious that this picture is reflected in our current phonological theories. We have seen that phonological theory assumes that words are naturally divided into segments and that these segments group into phonemes. Beyond that, we may say that, at least in structuralist phonology, the set of phonemes constitute a mapping of phonetic space (with, perhaps, some margins of security, and some unmapped territory such as clicks on the maps for most languages). It seems also generally to be assumed, although that is not required by the theory, that the way in which a foreign word will be perceived (as when it is to be borrowed) can be determined just as translation equivalents are to be determined i.e, by "translating" each segment from one phonemic mapping into the other.

I think that the influence of this mapping picture of the world can also be discerned in the concept of distinctive features. One characteristic of this picture is that it focuses what I think is clearly an excessive attention on the drawing of boundaries on fixing the line of demarcation between one category and another. No doubt it is semantics which has suffered most seriously from this misguided concern, but, if my interpretation is right, it has also affected phonology.

3. An assumption that the phonological competence required of speakers of a language involves both an analytic knowledge as well as a holistic knowledge of the lexifications of its words. I refer here to the two modes of knowing for linguistic forms which I distinguished in Grace 1981. There is holistic knowing, i. e., knowing whatever it is as a unit that is by memorization and analytic knowing, which is to know it in terms of its constituent parts. It is obviously the case that all lexifications must be known holistically. We must know, for example, that "dog" is the lexification for the category Canis familiaris. However, it is not obvious that any kind of analytic knowledge is required. All that is necessary is the ability to recognize the word (in context) and the ability to pronounce it.

The explanation which I propose for our attributing analytic knowledge of lexifications to language speakers is that we are accustomed to making a particular assumption about how the mind works. The assumption which I have in mind is an assumption that skills what Gilbert Ryle has referred to as "knowledge-how" can be analyzed in terms of underlying knowledge-that knowledge of entities and rules of operation. The exercise of the skill is then conceived of as the appropriate application of the rules.

In his classical paper, "Knowing how and knowing that" (Ryle 1946 [page citations from the 1971 reprinting]), Ryle argued that "Philosophers have not done justice to the distinction... between knowing that something is the case and knowing how to do things. In their theories of knowledge they concentrate on the discovery of truths or facts, and they ignore the discovery of ways and methods of doing things or else they try to reduce it to the discovery of facts." (1971:215) That is, they have tended to make the unwarranted assumption that some kind of knowledge of principles, rules, or the like, must underlie and account for skilled performance. However, to make such an assumption, he argues, is to violate the canon of scientific parsimony. That is, it adds entities beyond those which are necessary to account for the facts.

The reason why this is non-parsimonious is that we must in any case postulate the existence of knowledge-how. If we assume the existence of principles, rules, or whatever, to govern the performance in question, it will still be necessary to postulate a skill in translating the principles into performance, since it is obviously possible to know the rules for doing something without being able to put them into execution.

Turning to the question of the nature of phonology, it is obvious, I believe, that it is possible for someone to have memorized the entire phonological description of a language- underlying representations with full feature specification, rules, etc. and still to be unable to pronounce the language skillfully and fluently. The latter skill would still need to be acquired. My argument, therefore, is that there is no need to postulate knowledge of this phonological information at all that to do so is a violation of the canon of parsimony. Therefore it seems to me that the only phonological equipment which it is necessary to attribute to the fluent speaker of a language is a knowledge of the lexifications of the language and the physical skills necessary to pronounce them.

An Alternative Hypothesis of the Nature of Phonological Competence

From what has just been said it seems reasonable to hypothesize that it is not necessary for speakers of a language to have analytic knowledge of the lexifications of the words of their language. On this basis I can propose a simpler hypothesis of how humans organize sounds in their minds that is adequate to explain all of the facts I am aware of which are not attributable to literacy. My hypothesis is as follows:

First of all, pronouncing a language is a kind of skilled behavior. In order to pronounce a word one must coordinate a number of different muscular movements, which must be correctly executed and correctly sequenced. The amount of acquired skill involved is apparent from the existence of foreign accents in the speech of persons who have not adequately acquired the skills peculiar to a particular language (or dialect, etc.).

This skill, I propose, involves "chunking" of coordinated sequences of movements of the speech organs. The pronunciation of each chunk can be governed by a single conscious instruction. This instruction calls for the execution of the coordinated set of muscle movements required for the pronunciation of that particular chunk of sound. If I were to suspend my disbelief and imagine for a moment that the phoneme is the unit which corresponds to a chunk, then I might say that we have a subroutine for the pronunciation of each phoneme. Thus, if I wanted to pronounce "pot", I would have my brain send out instructions something like this: "Execute /p/, /a/, /t/" (meaning, execute the subroutine for pronouncing /p/, then that for /a/, then that for /t/).

However, that is not exactly the way I see it as working. I do think that the same chunks must recur in more than one word. I imagine that that has been the case since long before the beginning of anything deserving of the name language. I cannot imagine any vocal-auditory system in which there was more than one possible utterance where the skills required for the pronunciation of each were totally independent. And I feel sure that even today there are pronunciation skills which apply to a sound system as a whole. (That is why it is hard to pronounce a French word in an English sentence without either pausing to get one's speech organs set for the task or pronouncing it with an English accent).

But in addition to the global package of skills, I imagine that as the vocabulary (inventory of utterances) of our pre language grew, it quickly became impractical to maintain entirely distinct subroutines for each vocabulary item, and soon we had the situation where vocabulary item A began with the same subroutine as item B, although not the same as item C, etc. That is, as the vocabulary grew, the number of distinct subroutines required was maintained at a manageable level. So far, that seems quite in accord with the phonemic hypothesis. However, I do not think any more of that hypothesis is required, and my own hypothesis would make the following changes in its assumptions:

First, it would not assume that the chunks for which subroutines exist correspond to what the phoneme hypothesis recognizes as segments.

Second, it would not assume that instances of what would be analyzed as different allophones of the same phoneme would necessarily belong to the same subroutines.

Third, it would not assume that what we would consider different speakers of the same dialect would necessarily have the same chunks in their systems.

Fourth, although I assume that there would be miminal pairs distinguished by the difference in one pair of chunks, I do not assume that it is the business of chunks to be different, or that speakers know them in terms of their contrasting features. I imagine the distinctions to be an incidental result of the way the chunking came about.

Facts Seemingly Better Explained by my Hypothesis

I now turn to observations about phenomena which surely belong to the realm of phonology, but which phonological theory seems not to explain or not to explain as satisfactorily as my own hypothesis.

1. Informants who simply cannot seem to make sense of questions which assume that phonology is discretely organized. I am not sure how even one such case can be compatible with traditional phonological theory. I think of one informant in particular who had trouble deciding whether or not any two words shared a common sound, and he showed no evidence of expecting that they ever would. The particular case I have in mind was in New Caledonia in 1955. His language was not written. It, like other New Caledonia languages, has a difficult phonology. The man in question was what I regarded as a poor informant, but he was the only informant available for his language that I had access to so I stuck with him. Some of my New Caledonian informants (for other languages) sometimes made some attempt to write in their languages, basing their spellings on French. I do not think this one ever tried that. Moreover he seemed (all of this is thinking back as best I can) the type of person who would not have done much reading, and presumably writing, in French either.

The point is, he seemed to me always to be thinking in terms of continuities rather than the quantum mechanics (Joos's term) that linguistics assumes we have in our heads. I now wonder whether he is the exception, or whether it is the "good informants" who are. In any case, if I have interpreted the case accurately, it is not clear how theories which assume that phonology is organized in terms of discrete entities can tolerate any such exceptions at all.

The hypothesis which I have proposed, of course, does not assume the existence of discrete entities, or well-definedness of boundaries.

2. Different speech organ sets in different languages. A phenomenon which I have always thought to be both striking and surely important is this: In order to pronounce correctly a word in one language in the middle of a sentence in another, it is necessary to pause and do something which feels like re-setting the state of the speech organs. I think an adequate theory of phonology should account for this. In fact, I am convinced that we will never have more than a very limited understanding of diachronic phonology until this phenomenon is adequately understood.

I believe that the hypothesis which I proposed does accurately represent this aspect of the nature of language. However, it is likely to strike linguists as deficient because it does not provide a way of talking about the specifics of the differences in the sounds used in speaking different languages- the main thing which current phonological theories do provide a way of talking about. As I pointed out above, these phonological theories (like linguistic theories generally) have their roots precisely in a concern with the differential characteristics of different language systems.

However, my contention is that their way of talking about the differences misrepresents the essential nature of those differences. Most specifically, I do not know how these theories could characterize the speech organ sets which we are discussing here. Current phonological theory is simply designed for a different purpose from that of my hypothesis.

3. Difficulties in recognizing the "natural" segmentation.This difficulty shows up in various guises. Consider, for example, the difficulty we have had in determining whether certain English vowel nuclei were single segments or diphthongs. My suspicion has been that the best answer for one dialect may not be the best for others. There is also obviously much uncertainty about the status of such phonetic concepts as "affricate" or "aspirate".

For a more personal example, I have run into some problems in trying to interpret my own dialect in terms of mainstream analyses of the phonology of the English language. One example is this: Some linguists analyze words such as "boy" as having an open o preceding the glide: others analyze it as having a closed o. I had a strong feeling that the closed o interpretation was wrong for me, but for a long time could not find clear evidence as to why. (I have since found such evidence, but that is irrelevant to the present point).

One pair I considered in trying to resolve this problem was boy/bowie (before I found that bowie is "correctly" pronounced like buoy [as I (incorrectly, I believe) pronounce the latter]). One could substitute (Myrna) Loy and (Robert) Lowie. But people told me the second member of each pair was a disyllable, while the first was a monosyllable. I could never figure out whether that was the case or not. I think that in my dialect many, if not most, monosyllables can be pronounced alternatively as disyllables. Incidentally, in pursuit of this matter, I found several cases where speakers of other dialects proposed phonemic analyses for particular words that were quite different from what I was proposing even though our pronunciations had seemed quite similar to me.

There is one kind of evidence which I would like to make more of, but unfortunately what I have been able to find is very skimpy. I said above that, as I understand it, the established view in linguistics is that when speakers of one language attempt to interpret the pronunciation of a lexification of another language, the principle which they follow is to translate segment by segment into their own phonemic mapping of phonetic space.

If an actual study were made with naive speakers that is, speakers who did not know how the word was spelled or how it was analyzed phonemically in the source language I do not believe that it would bear that view out with anything approaching consistency. What I think it would show is that naive perceptions of phonetic similarity between pronunciations in different languages, would accord very poorly with the officially determined numbers of segments in the pronunciations being compared. One experience which I have in mind consists of the pronunciations that World War II American soldiers made of words from European languages when they did not know the official analysis of that word in the original language particularly, when they did not know what the analysis into segments was supposed to be.

I can not cite any specific examples now. I only remember noticing that it was the case. However, one more recent experience bears on the same point. It relates to something said to me by a man from Biak Island in Cenderawasih Bay in Irian Jaya, Indonesia. This man had been adopted by an American army unit in World War II and had traveled with them to Japan where he had remained during a part of the occupation. During this time he had acquired a fairly fluent command of colloquial American English. I do not know to what extent he could write the language. Anyway, he once remarked to me that the Biak language and English shared one word the word "water". Now, the Biak word is analyzed as consisting of just three segments /war/. The English word is analyzed as consisting of five perhaps a surface representation of four. Moreover, the English word presumably is a disyllable, while the Biak word is a monosyllable. Still, I had to admit that the two did sound a good bit alike.

I would argue that an adequate theory of phonology should have made it possible to predict that the two words would sound similar. Now, it is true that I can give a sort of ex post facto phonetic explanation which I think makes the fact that they did sound similar more understandable. In that explanation I would ask you to bear in mind that we were, of course, speaking of a typically American pronunciation of "water", and I would point out that the Biak /r/ is quite noticeably trilled.

The hypothesis which I have proposed, of course, does not assume that there is any correct segmentation.

4. "Defective" phonemic distribution. Considered from the perspective of the traditional theory, it is difficult to understand the failure of certain phonemes in certain languages to occur in certain phonological environments. For example, English (ng) {no phonetic symbols available here} does not occur pre-vocalically; h does not occur post-vocalically; s does not occur initially before r, while (sh) does not occur initially before most other consonants; (ch) occurs initially, while ts does not; b, d, and g do not occur after initial s, etc., etc.

These distributional gaps seem to be largely unsystematic and, if one adheres to the usual conception of phonological knowledge, to constitute an uneconomical utilization of the phonological resources of the language. However, the conception which I propose, since it allows for units (articulatory chunks) which are generally longer than the traditional segment, and since it does not require collation of phonetically similar segments occurring in complementary environments into single entities, does not make the kinds of distribution I have described appear to be in any way "defective". It is not clear what a defective distribution might be because it is not clear what a non-defective distribution might be.

Conclusions

I have proposed here that our phonological theories give a misleading view of the nature of language. More precisely, what is misrepresented are the abilities which underlie language- abilities which are presumably an essential part of the definition of humanness.

I have proposed that the lexifications of linguistic signs are not naturally segmented, and that the idea that they are was inspired by our familiarity with alphabetic writing. I have proposed further that our knowledge of the lexifications of our languages does not emphasize oppositions or distinctive features as phonological theories assume that they do, and that the idea that they do has also been inspired by alphabetic writing and by a conception of categorization as being a particular partition of a common universe. Finally, I have proposed that speakers of a language do not necessarily have any sort of analytical knowledge of the lexifications of that language. That is, their only knowledge (if skill in pronouncing and recognizing knowledge how is not to be counted as knowledge) of the lexifications may be a holistic one.

Of course, none of this is intended to suggest that it is not possible to teach people to talk about the lexifications of their own languages using the way of talking which linguistics has devised. Certainly it is. In fact, the way of talking about orthography which is generally associated with instruction in writing in an alphabetic system is very similar mutatis mutandis to the way of talking of phonemic theory.

Furthermore, I imagine that sometimes, even without formal instruction, individuals who have had sufficient informal exposure to alphabetic writing have worked out similar analyses for themselves. Indeed, if this sort of analysis is as natural as it is sometimes represented as being, perhaps speakers perhaps even great numbers of speakers throughout human history have hit upon it without even having had the stimulus of exposure to alphabetic writing.

However, any abilities which are not required of all language-bearers cannot be part of the abilities which were prerequisite to the existence of language. They must be left out of account in any determination of the nature of language. And it would be doubly true that any abilities which are to be explained only as consequent upon exposure to alphabetic writing have nothing to contribute to the definition of language.

I am likewise not suggesting that our way of talking and the concepts that it provides are not useful for analytic purposes. Surely they are. They provide a useful way to characterize particular phonological systems (even though I am proposing that these systems are really only systems of articulatory gestures) so as to be able to compare and contrast them. And as Ryle pointed out in the work referred to, analyzing knowledge-how as knowledge-that is a well-established and useful pedagogical device. It has a quite natural role in what Stephen Krashen calls language "learning" as opposed to "acquisition".

But to say that certain concepts are useful analytic and pedagogical devices is not the same thing as saying that they represent something which really exists. It is perfectly proper to talk as if these terms refer to things that really exist as long as we are playing the same game as long, that is, as we are using them for the purposes for which they were intended.

But sometimes it matters whether they really exist or not. One such time is when we are interested in determining what can be learned about the nature of the human species from the abilities which language presupposes. At such a time we must shift our attention to the conditions of the real world as nearly as we can determine them. I believe that when we examine them in the right light, we will find that our phonological theories as well as most of the rest of our theories of language structure give us a seriously over-intellectualized picture of the abilities involved in the knowledge and use of language. And this over-intellectualization of language has had a part in leading scholars in a variety of fields to an over-intellectualized picture of the workings of the human mind. That, I believe, places a serious responsibility upon the discipline of linguistics.

More On The Reality Of Phonemes

There seem to be three [4] main points which come up in the comments which I have received on my ELN17 (Grace 1983c). They can be stated approximately as follows:

1. Does not the very fact that alphabetic writing was invented prove that the languages in question already had units for the letters to represent? Isn't it necessary to infer that they had phonological units very like our phonemes?

2. Are my articulatory "subroutines" actually anything more than phonemes in a slightly different guise?

3. Why is the validity of the phoneme dependent upon its being psychologically real? Cannot languages be thought of as having formal structures which have nothing to do with psychological reality and which can be perfectly proper objects of study in their own right?

I will try to answer each of these in turn:

1. Of course, we in linguistics have long been accustomed to the assumption that the invention of alphabetic writing was a simple response to the phonological structure of languages. However, I do not think that there is any obvious reason to believe that that was necessarily the case. As far as I know alphabetic writing has not been invented independently more than once; moreover, forms of writing based on other principles have also been invented. It is my impression that the mode in the case of truly independently invented systems would be either a syllabic base or one in which only consonants are represented as segmental units with vowels not represented at all, or represented as something which we might equate with suprasegmentals. But writing systems do not have to be based at all exactly on already-existing linguistic units. For example, it seems that Chinese characters do not correspond very exactly to any kind of linguistic structural units.

I suggest that cartography would provide a better metaphor for the problem of representing the utterances of a language in writing. I suggest cartography because it also presents a problem for which there is no perfect solution in this case the problem of representing the curved surface of a sphere on a flat surface. The response to the problem has been a number of different projections, each being more appropriate than the others for some particular uses. I suggest that the problem of representing a language in writing is similar there are several different projections that have been tried, and none is equally satisfactory for all languages. In fact, none is completely satisfactory for any language.

2. I think my subroutines differ from phonemes in a number of important ways. (1) They are not part of a generalized theory of the language which is supposed to be invoked in both the production and the interpretation of speech. Rather, they are part of the motor skill involved in fluent pronunciation. Therefore, they probably are to be thought of as having their neurological connections with the motor functions. Moreover, according to my hypothesis the signantia need be known only as gestalts; i.e., although individuals are free to analyze them in any way they desire, analysis of any sort is not a necessary part of linguistic competence. (2) The subroutines are not conventionalized that is, nothing requires that they be at all alike from one individual to another except that they must generate the same output within tolerable limits (whatever these are). (3) They are not typically one segment (in the sense of articulatory phonetics) in length. I am not very clear on this, but I imagine them typically to be considerably longer some comprising whole words and quite variable in length some probably containing one or more others within them. I imagine their numbers not to be at all constrained by any principle of parsimony. (4) These subroutines are supposed to account just for our highly skilled pronunciations. I imagine that many of us have signantia which we produce in normal speech which involve coordinated movements of the speech organs which have not been fully "grooved" that is, which we pronounce less fluently. There would be a strict conception of a person's phonological system in which it would be possible for that system not to provide for some word or words which that person habitually uses. E.g., That might be the case with some Americans who pronounce "Bach" with an [x]. Some of them may not pronounce it with quite the practiced ease with which they pronounce the bulk of their vocabularies i.e., which is characteristic of words which fall fully within their systems. I have discussed some such phenomena in my own speech in Grace 1981 (cf. pp. 79-81).

I should repeat that what I intend to be proposing is a more parsimonious hypothesis, one which is intended to account with the simplest assumptions possible for those facts which must be accounted for. I put it forward most tentatively with the hope that it will help to show that the phoneme may not be a necessary assumption and to stimulate the re-thinking that I think is required.

3. Does language have a formal structure which is independent of the structure which is psychologically real for its speakers? Frank Lichtenberk has suggested a framework in which to put this question. He invokes Karl Popper's concepts of "World 2" ("the psychological world"-Popper 1982:114) and "World 3" (the world of the products of the human mind" ibid. ["World 1" is the physical world the world of matter and energy]). Lichtenberk suggests that language can be studied both as a World 3 phenomenon, as has usually been the practice of linguistics, and also as a World 2 phenomenon, perhaps most appropriately under the rubric of psycholinguistics.

I do not find Popper's conceptualization as I understand it entirely adequate for the problem under discussion, but it provides a useful beginning. He makes it clear (at least in Popper and Eccles 1977: 16) that he regards human language as a World 3 phenomenon. However, it is clearly among the very earliest of the World 3 phenomena to emerge, and in fact is surely prerequisite to most of the rest. It seems clear to me that for Popper the quintessential World 3 phenomena are "linguistically formulated thoughts". (For example, he says [1982: 116] "I will take the world of linguistically formulated human knowledge as being most characteristic of World 3."). He speaks frequently of the contents of libraries as examples of World 3 phenomena. I think that what Popper is most concerned with under the heading "World 3" are realities (scientific theories and the like) created by humans through the medium of language. Therefore, I will attempt to illuminate the question before us by introducing a further distinction.

I will propose that language is not one, but two different, products of the human mind i.e., that we habitually use the term "language" to designate the products of two different creative processes. I will call these, respectively, Language 1 and Language 2. To put the matter simply, I will say that humankind created language by using it, thereby giving us Language 1, and subsequently created language in a new guise by talking about it. Thus was born Language 2. The creation of Language 1 was what we ordinarily think of as the evolution of language. I describe it as "creation through use" on the assumption that speaking evolved at least in part out of things which our prehuman ancestors used to do which performed some of the functions which language now performs. However, this assumption is not essential to anything else in my argument.

As I understand Popper, Language 2 is more like the other things he usually has in mind when he talks about World 3 phenomena than is Language 1. World 3, he tells us (1982: 116) "is the world of problems, theories and arguments". As I conceive of it, linguists did not invent Language 2, but they have elaborated it and, to a considerable extent, taken possession of it and come to assert proprietary rights over it. I will try to give an account of how phenomena like Language 2 come into being.

To the extent that a community discusses a given topic with regularity, their manner of speaking of it is likely to become conventionalized. With sufficient use, as in the case of a formal scientific discipline, this conventionalization will lead to what Richard Rorty calls "normal discourse". Rorty writes (Rorty 1979: 320), "...normal discourse is that which is conducted within an agreed-upon set of conventions about what counts as a relevant contribution, what counts as answering a question, what counts as having a good argument for that answer or a good criticism of it. Abnormal discourse is what happens when someone joins in the discourse who is ignorant of these conventions or who sets them aside."

I take discourse about language within the framework of linguistics to be an example of normal discourse. [5] It is a set of conventions for talking about language. Such a set of conventions by implication embodies certain assumptions about the nature of the object of discussion. That is, the agreed-upon conventions about what counts as a contribution, what counts as an answer, a good argument, etc., imply agreed-upon assumptions about the nature of the object. Another way to put this would be to say that we talk about language as if it had the characteristics that our agreed-upon conventions imply.

I have suggested that Language 2 is a way of talking about language rather than suggesting that it is a way of thinking about it. An important point lurks therein, I believe, but one which I have been very slow to grasp. I can identify three main sources which led me to an appreciation of this point.

The first is the following quote from Benjamin Lee Whorf: "The revolutionary changes that have occurred since 1890 in the world of science especially in physics but also in chemistry, biology, and the sciences of man have been due not so much to new facts as to new ways of thinking about facts....

"I say new ways of thinking about facts, but a more nearly accurate statement would say new ways of talking about facts. It is this use of language upon data that is central to scientific progress." (Carroll 1956: 220)

The second source is Rorty, with what is implied in his substitution of the concept of "normal discourse" for Kuhn's "normal science".

The third is the following statement by Berger and Luckmann (1967 : 152-53): "The most important vehicle of reality- maintenance is conversation....It is important to stress, however, that the greater part of reality-maintenance in conversation is implicit, not explicit. Most conversation does not in so many words define the nature of the world. Rather, it takes place against the background of a world that is silently taken for granted. Thus an exchange such as, 'Well, it's time for me to get to the station,' and 'Fine, darling, have a good day at the office' implies an entire world within which these apparently simple propositions make sense. By virtue of this implication the exchange confirms the subjective reality of this world."

What I take to be the significant message in what Whorf and Rorty say is that what a science most essentially is, is a way of talking about something it is a set of conventions governing a ("normal") discourse. What it is most essential to learn in order to become a qualified professional in a scientific discipline (and I certainly intend to include linguistics here) is the required manner of discourse. The manner of talking of a discipline is much more accessible to observation than its manner of thinking (if, indeed, there is any common manner of thinking). For that reason it, rather than a manner of thinking, is what apprentices may observe and imitate, and it, rather than a manner of thinking, is what the apprentices can be required to display and be judged upon. In principle, we always retain our freedom to think as we see fit, no matter how we must talk. [6]

What Berger and Luckmann bring home to me is that the explicit statements that we make about the nature of reality (or of our particular subdomain of it) are rather insignificant in comparison with what we convey about it by implication. Everything we say implies that we believe in the existence of a kind of world in which what we are saying makes sense. A whole way of talking, such as the conventions governing the normal discourse of a scientific discipline, says by implication a lot about the nature of the world or at least of that part of the world which the discipline involves itself in. We talk as if the world were this way. (I am sure, by the way, that this is the primary means by which children acquire the world view of their society. They project from what they hear said the kind of world in which all of that makes sense to put it in other terms, they construct an appropriate reality.)

Now we can get back to the main point. It is (and that is what ELN17 was mainly about) that when we talk about what we call "language", we're actually talking about Language 2, although we may be inclined to assume that we are talking about Language 1, or that what we are doing is equivalent to talking about Language 1. Of course, Language 2 is a constructed reality which attempts faithfully to reflect the characteristics of Language 1, and it does so to a considerable extent. Unfortunately, however, we do not know exactly how much.

What I am suggesting is that phonemes are phenomena of Language 2, but not of Language 1. Language 2 is what linguistics has created to serve as its object (in the sense that all sciences create their objects). It stands in the relation of a model to Language 1, but it is a model designed, consciously or unconsciously, to highlight what was taken to be of particular interest. It seems to me that what Language 2 is designed to bring out and develop fully about language (i.e., Language 1) are those properties in the design of human languages which give them the potential to serve as repositories of decontextualized information. Even the languages of pre-literate peoples are looked at in this perspective in terms of how they structure information. Questions which are not relevant to this concern are much more likely not to get asked. (I have said something about this perspective in my remarks on "the intertranslatability postulate and its consequences" [in Grace 1983a] and "the Carnapian model of language" [in Grace 1983b]).

My answer to the third question is, therefore, that it is indeed possible to think (and more especially, talk) of languages as having formal structures which have nothing to do with psychological reality and to study these structures. However, what is required in order to do so is to create a model of language structure to create what I call "Language 2", and to study that. To subject Language 2 phenomena to scientific linguistic analysis, of course, requires that we observe complex rules about the kinds of observations to make and which of the possible observations authorizes which of the possible descriptive statements. In other words, this analysis is subject to the normal discourse conventions of linguistics.

That does not mean that any structural features which are posited in this framework are necessarily false. In fact, if they are posited validly while playing the game according to the conventions, they are true within the framework of the conventions (wherever those as if conditions hold). But there is no general answer to the question of their status outside the framework of the conventions where the basic assumptions of the conventions do not hold the question may be uninterpretable or the unanswerable except on a case by case basis.

It seems that we are left in this situation; there is no problem about the reality of the phoneme as an element of Language 2. It was built into Language 2 when the latter was created. On the other hand, it has seemed to me over the past several years, when I have been wanting to raise the question of the language-1-reality of the phoneme, that there was no clear way within the normal discourse of linguistics to raise and discuss it. If it is assumed that appeals to psychological reality are not relevant, but that the question is one of formal structure, then it is still not clear to me what conventions could apply what could count as a relevant contribution or what would be accepted as a good argument (a dilemma of somewhat Goedelian cast).

WHAT ARE LANGUAGES?

The question of what languages are has implications for all of linguistics, but I've been particularly interested in its implications for historical linguistics. In historical linguistics languages are assumed to be entities that maintain their identity over time (even while undergoing changes); they are said to split but not to merge, to come into "contact" with other languages, to "borrow" lexical items and even grammatical features from other languages. The question that interests me here is what lies behind these metaphors what, really, are these languages and what really is going on when they are said to "split", be in "contact", "borrow", etc.? This Note is concerned with the first of those questions.

Linguistics, synchronic as well as diachronic, has consistently assumed that human language is divided up (without remainder) into individual units that we call "languages". But where are these languages to be found; where are they located? Nowadays, at least, there is widespread agreement that each individual language is located, not in any one place, but scattered among the separate minds/brains of many individual humans. It is supposed to take the form of "knowledge", although mainly not conscious knowledge. The humans in question are those who are said to "speak", or simply to "know", the particular language.

Thus, we assume that a language exists as knowledge in the minds/brains of individuals, each of whom has acquired this knowledge by processing his/her own personal linguistic experiences. However, since the experiences of different individuals can't have been identical, it seems inevitable that each language will exist in many different forms each built up from the experiences of a particular individual. Still, linguistics has always assumed that the language in the mind/brain of one individual is in some significant sense the same language as that in the minds/brains of all of the other individuals whom linguists would count as speakers of that language. But what might be the sense in which this could be true? In what sense and to what extent can a language that is found scattered, and in non-identical forms, among the minds/brains of many different individuals be a uniform system?

There is also another question to be asked: To what extent must it be uniform in order to be able to perform its functions? The answers to these questions seem to depend in large part on the form taken by the speakers' knowledge.

Therefore, we must ask: What form does this knowledge take? What I will do here is propose and discuss three hypothetical solutions which, I hope, can serve as reference points for future discussion.

First, in Grace 1992 (pp. 3-6) I spoke of the conception of languages as what I called "governing systems", that is, systems that govern the construction of a particular set (infinitely large, of course) of sentences, each with its own meaning. That is, the governing system is supposed to authorize the sentences and warrant the meaning(s) of each sentence so authorized. The governing system is traditionally supposed to consist of a syntactic system and lexicon, which constitute the means by which it is supposed to be able to accomplish this function. The governing-system conception of the way languages work, then, presupposes the traditional view discussed in Grace 1993 of the knowledge possessed by its speakers: that is, the view that their knowledge consists of a syntax and a lexicon. Let us consider that as one hypothesis of the nature of linguistic knowledge (and therefore of the individual language). Thus:

HYPOTHESIS 1 . Individuals' knowledge of a language takes the form of a syntactic system and a lexicon.

DISCUSSION: I tried in Grace 1993 to show that that conception of speakers' linguistic knowledge was untenable. The syntax-lexicon model has been developed by linguists for the structural analysis of artifacts of speech. Or more exactly, the analysis is of selected aspects of the artifacts. What are usually analyzed are written texts or what has been spoken and preserved as an artifact by transcribing or audio recording. An analysis (language description) produced by applying this model effectively defines and characterizes the language for the linguistic profession.

For convenience, I will refer to structural-analysis-of-artifacts-of-speech descriptions [7] as SAAS descriptions and descriptions of the knowledge-of-language in the minds/brains of individual humans as KOL descriptions [8]. What Hypothesis 1 amounts to is the claim that a person's KOL should be expected to conform to the same model as that used for SAAS descriptions. However, the point that I (and many others) have tried to make is that there seems to be no good reason to take seriously the claim that KOL in any way resembles any SAAS description. That is, it seems very reasonable to expect that the knowledge-of-language in the minds/brains of individual humans will prove to take a form radically different from that represented in SAAS language descriptions.

In Grace 1993 I argued that the ability to use language requires the individual to call up as wholes utterances or utterance partials, contemplating each as a unitary phenomenon so as to evaluate the likely effect of each on the contemplated audience. KOL hypotheses should, I think, be required to satisfy the conditions that emerge from that fact (as I believe it to be).

A second hypothesis would be that these utterances and utterance partials are taken from a repertoire of what I will call "discourse components" possessed by the speaker. Thus:

HYPOTHESIS 2 . Individuals' knowledge of a language takes the form of a repertoire of discourse components and their uses.

DISCUSSION: What I have in mind in speaking of "discourse components" includes such things as have been called "speech formulas" (as described, for example, in Pawley 1985, 1992), "grammatical constructions" (as described in Lakoff 1987), "schemata" (as described in Tyler 1987), "semantic formulas" (Pawley 1992), "catch phrases", "idioms", "cliches", "proverbs", etc.

I assume that there are some discourse components that are sufficient alone to constitute an utterance (more accurately, to serve as the vehicle of an utterance). Others have to enter into combinations in order to constitute a suitable vehicle for utterance. It appears that in these combinations the components may be simultaneous, or one can precede the other, or various kinds of overlapping and embedding can occur.

In asserting that the individual's knowledge-of-language takes the form of a repertoire of discourse components, Hypothesis 2 implies that KOL descriptions are possible and that they would consist of nothing more than (a list of?) appropriately characterized discourse components.

Hypothesis 2 suggests that the individual constructs a repertoire of discourse components by analyzing the utterances that s/he hears and extracting recurrent parts. Possibly there are clues of some sort as to what parts to extract, clues that would lead all speakers of the language to extract largely the same parts and therefore to have very comparable repertoires. Again possibly there are not, and possibly different speakers have very different discourse components in their repertoires.

Although it's impossible now to say much about what possible different forms the individual discourse components might take (that would be a problem for future research), I will offer a few comments.

First, some discourse components seemingly can have other discourse components as parts (i.e., some discourse components can have variable (non-lexified) sub-components).

Second, knowing a discourse component presumably includes knowing what it can combine with.

Third, it seems likely that everything that would count as a lexical item in the Hypothesis-1 approach would be eligible for inclusion as a discourse component under Hypothesis 2.

Fourth, an interesting question is whether the repertoire of discourse components could include grammatical constructions with no lexification at all (e.g., "noun + intransitive verb"). I don't know for sure. If so, could it even go so far as to incorporate in some form or other everything postulated by Hypothesis 1 (since I've suggested already that it could include all of the lexical items)? I don't know. Would it include abstract schemata such as the three-part (beginning-middle-end) structure of narrative and many written forms (and even conversations, I think) in our culture? Again, I don't know.

Finally, have some KOLs in some periods in history had significantly larger numbers of more abstract discourse components than others? I have suspected that discourse components may have tended to become more abstract in the history of our own culture as people have been called upon more frequently to talk to people they haven't previously known and about an increasingly wide range of subjects. Today, we often have to talk about subjects in which we aren't expert (i.e., which involve us in discourses e.g., medical discourse while consulting a physician for which our repertoires of discourse components are not well developed). What I'm suggesting is that language use in our culture has become exceptionally ad hoc and that that has probably required our repertoires of possible utterances to provide much more flexibility (to move somewhat more in the direction of Hypothesis-1 type knowledge?).

The third hypothesis is:

HYPOTHESIS 3 : Individuals' knowledge of language is nothing more than a large memory store of experiences with the emphasis on experiences in which language was used. For linguistic purposes, we may think of this as essentially a store of utterances. We interpret what's said by recalling other cases where the same thing or something similar was said. We decide what to say by recalling similar situations and what was said in them (and what the consequences were).

DISCUSSION: Clearly individuals don't remember all of the details of any utterance they have heard, nor do they remember every utterance individually. The point of the hypothesis is that our KOL at the given point in time doesn't (as Hypothesis 2 would have it) consist of a fixed inventory of precisely-specified components which require only to be assembled according to the applicable instructions in order to produce (the linguistic expressions that serve as vehicles of) utterances. Rather, Hypothesis 3 maintains, what we have is more like a store of utterance examples .

If we don't remember all details of every utterance we have heard, what if this hypothesis is valid do we have in our memory stores? Again, anything approaching a full answer to that question would have to be a matter for future research. However, as far as I can judge it seems likely that other things being equal, we remember particular utterances (or parts of utterances) better in proportion to: (1) how often we've heard them; (2) how recently we've heard them; and (3) the amount of emotional impact hearing them had on us. By "remember better" I mean either that the details are preserved more accurately, or that the entire utterance (or utterance part) is more accessible to ready recall, or (probably most often) both.

It does seem clear that we can often remember some parts of an utterance quite clearly, but not remember other parts.

Even though I've suggested that the memory store contains partial as well as complete utterances, which may make Hypothesis 3 seem hard to distinguish from Hypothesis 2, there is a fundamental difference between the two. According to Hypothesis 2 what we know are discourse components , which are potential utterance parts which we have only to assemble in the right ways to produce utterances. According to Hypothesis 3, what we know are examples , and our task is to use these examples as a basis for designing our own utterances. I don't know very much about how this is (to the extent that the hypothesis is valid) accomplished, but I would suppose that the principal basis is analogy. For a simple example of what I have in mind, given utterances A, B, C in the memory store, we decide that we want to produce an utterance that will be similar to A, but differ from it in just the way that B differs from C.

EVALUATION: I believe that I have made it clear that I don't find Hypothesis 1 at all plausible, and why. What about 2 and 3? My feeling is that neither is entirely right. Hypothesis 2 presents all discourse components as being defined with the same preciseness and all as equally accessible. I don't think that this can be right. Actually, I think that some discourse components are much more familiar than others, and that we are much more certain of the actual composition of some than of others. Furthermore, I think that we clearly don't just pull ready-made components out of storage and put them together, but that we sometimes have to do some adapting of the sort which is emphasized in Hypothesis 3.

On the other hand, Hypothesis 3 leaves too many questions unanswered for one to feel very satisfied with it. Just how are the utterances that we remember, and remember better, selected? Where our memory has selected out parts of utterances, how are these parts selected? Just how does the process by which we use our available examples to design new utterances work? Of course, it would be unreasonable to ask for precise answers now since no work has been done on this, but the kind of suggestions that I've been able to give here aren't precise enough for us to feel very sure just what the hypothesis is.

If I were to give my best judgement at this time, it would be that the KOL is some kind of compromise between what is proposed in the latter two hypotheses that it has aspects of both.

IMPLICATIONS: What are the implications of these hypotheses for the question with which we began? We have been talking about the knowledge-of-language (KOL) of the individual, but it has been assumed that the "language" of which the individual has knowledge is something which extends far beyond any individual. What are these supra-individual entities that we know as "English", "Japanese", etc.? What kind of answer would each of the hypotheses suggest for that question?

First of all, I think that Hypothesis 1 most strongly requires the internal unity of such languages. The "governing system" role that each is called upon (by that hypothesis) to play seems to presuppose a tightly-structured system. It would seem that the possibility for variation within such a system would be strictly limited that adding more varieties to a governing system would soon loosen it up to the point that it would lose the ability to govern anything (more and more rules becoming optional). The presupposition that a language must be a governing system implies that all of language is divided up into quite distinct languages, and that therefore, it must be possible to sort each individual speaker into one or another pigeon hole (or perhaps two or more in the case of multilinguals, but with each multilingual's KOL neatly partitioned among the pigeon holes). [9]

Hypothesis-1 literature assumes that what the language learner will have to learn is specifiable beforehand in a fairly precise way, because s/he has to learn a particular language whose syntax and lexicon (which are what s/he will have to learn) are already established and non-negotiable.

Hypothesis 2 doesn't require such tightly structured systems or so much internal unity. What one must learn what the KOL consists of is not highly structured. It is a repertoire, much more like a dictionary than a grammar. In such a case we can imagine different individuals who have grown up in fairly close proximity having substantially different repertoires of discourse components, according to the accidents of their experience. But we can imagine them communicating with each other fairly successfully by relying on the discourse components that they do share (presumably still a large number). Hypothesis 2 doesn't seem to require that all KOLs in the world must belong to one of a fixed number (usually estimated as between five and ten thousand) of entities of the kind we've been talking about: i.e., "languages". (I imagine that advocates of Hypothesis 2 if there were any would say that KOLs will be found to cluster into different languages just to the extent that the speakers whom they represent belong to distinguishable socially -based clusters).

Hypothesis 3 doesn't seem to require any external system at all. Presumably, the ability of some people to (more or less) understand one another results from the fact that some people have heard the same utterances or (much more often) have heard similar ones.

CONCLUDING REMARKS: These hypotheses are intended as ideal types each taking a single principle and building on it rather than as truth claims. I don't believe that any one of them is entirely correct. However, I think there is some approximation to the truth in all three. I think that, as is implied by Hypotheses 1 and 2, there is more structure in our KOL than Hypothesis 3 admits. However, as implied by Hypothesis 3, there is not a sharp line between linguistically relevant and linguistically irrelevant knowledge as Hypotheses 1 and 2 imply. I think that, as 3 suggests, we have a great variety of information much seeming quite unrelated to language that we can, and at times do, draw upon in speaking and listening.

I would say that the KOL in reality appears to combine characteristics of Hypotheses 2 and 3. However, it was pointed out above that the definition of 2 is loose enough to permit most or all of 1 to be accommodated within it. And it appears to me that individuals who show greater virtuosity in their use of language tend to have a more analytic knowledge of the language. That is, I suspect that their KOLs may include discourse components with more of the characteristics proposed by Hypothesis 1 than does the KOL of the average speaker and/or may include greater numbers of such components. In fact, I think that what is regarded in our culture as "good" command of a language is likely to be closely related to the presence of Hypothesis-1 characteristics. However, I don't believe that the Hypothesis-1 characteristics predominate in any KOL.

Although none of the hypotheses is presumably accurate in itself, I hope that the three of them do succeed in bracketing the truth that the truth lies somewhere in between. In particular, I hope that it may be useful to examine separately what each of them implies about the processes of linguistic change and differentiation. Since the Hypothesis-1 picture of language has dominated discourse up until now, I think examining the implications of all three may be all the more interesting if, as I believe, both Hypotheses 2 and 3 are closer to the truth than is Hypothesis 1.

AFTERWORD

I shouldn't end without emphasizing that admitting that SAAS descriptions aren't accurate descriptions of the kind of knowledge that speakers have of their languages won't deprive them of all value. Such descriptive accounts came about in the first place because they are useful. In particular, they're an important practical help for anyone attempting to use (or to learn) a strange language. In the early stages particularly, understanding what we hear or read is a matter of word-by-word decoding (i.e., translation). And to speak, except for the simplest utterances, we have no option in the earliest stages but to conceive our utterances first in our own language and then translate essentially word-for-word. Dictionaries and grammars are in principle the best kinds of aid I can imagine for translation tasks of this particular kind. (In fact as I've often mentioned, I think that it is precisely from the experience of using dictionaries and grammars in such language-learning-via-translation tasks that the idea developed that all language use is a kind of translation, and that we have internalized grammars and dictionaries that we use for that purpose).

But SAAS descriptions have uses beyond such practical applications. They provide the best way we have of talking about language differences. They are our way of talking about the range of differences among languages of the world, specifying characteristics that are universal and, for those that are variable, specifying just what variants exist in the world. They also permit us to talk about the differences between related languages both between contemporaneous ones and between ancestor and descendant and thus provide our means of talking about linguistic change. We would be left in a most awkward situation if we were no longer permitted to use SAAS descriptive statements.

But I do wish we knew better what lies behind them. I continue to think the KOLs must be the ultimate objective, and although I suppose the SAAS descriptions are somehow resultants of the structure and content of the KOLs, the relation seems to be far from direct and straightforward.

 


Notes:

[1.] This is a revision of a talk which I gave on October 4, 1983, to the University of Hawaii Linguistics Department Tuesday Seminar entitled, "Why I do not believe that languages have phonemes".

[2.] It seems worthwhile to note the questionableness of the assumption that all major language systems have the same domain that which we refer to as the "language". I want to argue, on the contrary, that if systems of phonemes did exist, the language is not the level of organization to which they belong.

As one kind of example, it is apparent that in the English language, some speakers make distinctions which others do not. Each of the following pairs of lexifications constitutes a minimal pair for some speakers and a homophonous pair for others: pin/pen, cot/caught, watt/what, horse/hoarse, merry/Mary, merry/marry. However, these pairs are not isolated lexical items; each exemplifies a systematic difference in the pronunciations of different speakers. In each of these cases, there are some speakers of English (those who do not make the distinction) for whom the second member of the pair as pronounced by those who do have the distinction is systematically unpronounceable.

The obvious solution to this particular problem, of course, is to say that the lexification does have different phonemic spellings in different dialects, but that there is a regularity to it all. The different phonemic systems are related by regular sound correspondences, and therefore the phonemic spelling in one system is largely predictable from that in another. We might, therefore, amend the picture described above to say that where there are different phonemic spellings of what would otherwise qualify as a single linguistic sign, and the different phonemic spellings reflect different phonemic systems, and the correspondences among the spellings are explainable in terms of regular sound correspondences, then all of the systems belong to one superordinate system, and it is to that superordinate system that the words belong. That is, the subordinate systems correspond to a level lower than the language, but they are systematically related in such a way as to constitute a superordinate system.

However, to take this course has an unexpected consequence. If we decide the question of same or different lexification (i.e., what would popularly be called same or different "word") on the basis, not of a single phonemic spelling, but any set of phonemic spellings which are related by regular correspondences, we find that there is no reason to stop at the boundaries of a single language. The spellings, both orthographic and phonemic, of (to take one example among many) Eng. sociology, Fr. sociologie, Germ. Soziologie, etc. are related by quite regular correspondences. Lexifications such as this spread very readily from language to language, and it is usually impossible to guess in which language any of them first appeared.

Thus, if phonemes and lexifications both are to be thought of as existing as part of some kind of system, the language (langue) does not seem to be the right system for the specification of either. The phoneme seems to belong to a smaller system, so that a language might consist of a family of related phoneme systems, but the lexification seems readily to assume a distribution which far overreaches the bounds of a particular language. (This seems to be further evidence of the exaggerated role that linguistics has ascribed to the units which we call languages in the functioning of language.)

[3 .] The implications of this picture may be more clearly seen if we compare it with another, contrasting, picture, which I think has an at least equal claim to validity. According to the second picture, the universe is for all practical purposes an open system. That is, we can only know the characteristics of the universe through our experiences, but these experiences are, and presumably are destined always to be, only a very limited and largely accidental sampling of the set of experiences which the universe is capable of providing.

Given these circumstances, the strategy by which we cope with the problem of getting around and functioning in this vastness is to produce a theory of reality, and to act according to its dictates. This is, I believe, the same strategy that our closest non-human relatives, on the one hand, and science, on the other, employ. It seems, therefore, perhaps to be an inevitable strategy.

I said that other animals must employ this strategy of constructing for themselves, on the basis of their own experience, the realities in relation to which they live their own lives. However, once language has become available, individuals become able to pool their experiences and to coordinate them into a single common reality which is represented in their language. The basis of this is the vocabulary of the language. The meanings (what I call the "conventional senses") of the lexical items are the elements which make up this reality. Particular experiences, then, are interpreted in terms of these elements.

[4.] Another point which has been made several times is that my characterizations of the assumptions of linguistics repeatedly fail to represent the most sophisticated views to be found in linguistics. They do not present the best information available or the views of those linguists who have most carefully studied the specific matter in question. That is no doubt true, but my intention in the passages to which such criticism applies was to describe the "picture of the world" that generally informs linguistic discourse which is projected by normal linguistic discourse rather than the most expert view to be found in the profession. These more sophisticated views, if they figure sufficiently prominently in future discussions, will surely eventually work their way into the picture.

[5.] Perhaps a brief explanatory note is in order here. Rorty's "normal discourse" is modeled on Thomas Kuhn's "normal science" the science practiced within the framework of a "paradigm". When attempts have been made to interpret events in the history of linguistics in terms of Kuhn's model, the objection has sometimes been raised that linguistics has never had a paradigm in the full Kuhnian sense. That is probably a fair point, but it does not affect what has seemed to me the principle lesson for us in Kuhn. That lesson was that we can never have direct general access to actual reality, and that our only recourse, therefore, was to construct models which purport to represent actual reality, and then to check them as best we can. Moreover, so little of these models has been explicitly checked, or is likely ever to be, that we in effect live and work within realities of our own invention. (Or as Kuhn said, serious research cannot even begin until we assume that we have firm answers to all of the important questions).

[6.] But how independent, in fact, are ways of speaking and ways of thinking? How accurate an index of the thinking of the members of a profession is their way of speaking? To answer that question is not important to the main issues here, but it is of course of general interest.

It seems to me that we can at least say that it is not completely easy to keep one's beliefs separate from one's statements. I would also like to quote a statement by Samuel E. Martin, on a different but I think related, matter. Martin writes (1964: 407), "In a number of situations the Japanese has an explicit (and often very effective) way to soothe people's feelings; in many of these same situations, the Korean, like the American, says nothing at all. Both are apt to suspect the sincerity of the Japanese, and this is unjust, since it is virtually impossible to say 'thank you' all day long and not end up with a vague feeling of gratitude, or to excuse yourself time after time without a certain humility setting in of itself."

[7.] I should point out that most SAAS descriptions intend to include potential artifacts of speech as well as those that happen to occur in their corpora.

[8.] There is some temptation to suggest an analogy between this contrast between SAAS and KOL descriptions and Noam Chomsky's (1986) distinction between externalized language (E-language) and internalized language (I -language). However, Chomsky's assumptions about the nature of KOL are very different from those made here.

Moreover, although SAAS descriptions are usually made from the utterances of a number of speakers and purport to represent a (supra-individual) language, it would be possible to make an SAAS description of the speech of a single person. I'm not sure that the output of an individual could count as an E-language for Chomsky.

[9.] If a language is the sum of KOLs of its speakers, consider the problem presented to Hypothesis 1 by dialect chains. We need to divide a chain up into languages (both for description and for historical/comparative treatment). Suppose we reach a point where we have decided on the number of languages and selected (geographical, presumably, in most cases) centers for each. Now, consider the case of two centers such that there is a dialect that is linguistically intermediate between the two so that it could equally well be assigned to either. Which one we assign it to makes a difference in the two languages since each is a sum of the KOLs attributed to it. Now this seems not to be a problem at all if the KOLs are of the Hypothesis-3 type, and not much if they are of Hypothesis-2 type. However, for Hypothesis 1 and the associated governing system assumption, a more serious problem seems to be raised. Just what are the rules of the governing system?

Again, suppose we've made a decision to assign the dialect to language A, but after some years, conditions have changed so that it seems clear that it's now closer to the other dialects of B. If we reassign the dialect from A to B, both languages are automatically changed (since each language is the sum of the KOLs which make it up). Therefore, their descriptions should be revised. (Or, if one thinks of the descriptions as existing prior to the intervention of the linguist, the revision occurs automatically. In fact, if instead of "descriptions" we refer to their "grammars", everyone would presumably agree that the revision was automatic). This seems awkward for Hypothesis 1. For example, the languages will have changed, but the individual speakers' KOLs won't have. Nobody speaks any differently, and yet the governing systems of both must be assumed to have changed.

 


REFERENCES :

Berger, Peter L. and Thomas Luckmann. 1967. The social construction of reality: A treatise on the sociology of knowledge. Anchor Books. Garden City NY: Doubleday and Co.

Carroll, John B. (ed.). 1956. Language, thought, and reality: Selected writings of Benjamin Lee Whorf. New York: Wiley.

Grace, George W. 1981. Ordinary language. Printout.

Grace, George W. 1983a. More on the nature of language: The intertranslatability postulate and its consequences. Ethnolinguistic Notes, Series 3, Number 9. Printout.

Grace, George W. 1983b. The linguistic construction of reality. Ethnolinguistic Notes, Series 3, Number 11. Printout.

Grace, George W. 1983c. Why I do not believe in phonemes: On the cognitive validity of linguistic theories of phonology. Ethnolinguistic Notes, Series 3, Number 17. Printout.

Martin, Samuel E. 1964. Speech levels in Japan and Korea. In Dell Hymes (ed.). Language in culture and society: A reader in linguistics and anthropology. New York, etc.: Harper and Row, pp. 407-415.

Popper, Karl R. 1982. The open universe: An argument for indeterminism. Totowa NJ: Rowman and Littlefield.

Popper, Karl R. and John C. Eccles. 1977. The self and its brain. Berlin, etc.: Springer International.

Rorty, Richard. 1979. Philosophy and the mirror of nature. Princeton NJ: Princeton University Press.

Grace, George W. 1981. An essay on language. Columbia SC: Hornbeam Press.

Grace, George W. 1982a. The question of the nature of language. Ethnolinguistic Notes, Series 3, Number 4. Printout.

Grace, George W. 1982b. Ways of talking about things. Ethnolinguistic Notes, Series 3, Number 5. Printout.

Grace, George W. 1982c. More on ways of talking about things: Contexts. Ethnolinguistic Notes, Series 3, Number 7. Printout.

Grace, George W. 1983a. More on the nature of language: The intertranslatability postulate and its consequences. Ethnolinguistic Notes, Series 3, Number 9. Printout.

Grace, George W. 1983b. The linguistic construction of reality. Ethnolinguistic Notes, Series 3, Number 11. Printout.

Ryle, Gilbert. 1971. Knowing how and knowing that. In Collected Papers, Volume II. New York: Barnes and Noble Books.

Chomsky, Noam. 1986. Knowledge of language: Its nature, origin, and use. Convergence: A series founded, planned, and edited by Ruth Nanda Anshen. New York, etc.: Praeger.

Grace, George W. 1992. Another attempt to explain why I have misgivings about what we tell people about language. Ethnolinguistic Notes, Series 3, Number 43. Printout. 21 pp. Also internet World Wide Web page (Click Here).

Grace, George W. 1993. What is the language faculty? Ethnolinguistic Notes, Series 3, Number 44. Printout. 8 pp. Also internet World Wide Web page (Click Here).

Lakoff, George. 1987. Women, fire and dangerous things: What categories reveal about the mind. Chicago: University of Chicago Press.

Pawley, Andrew. 1985. On speech formulas and linguistic competence. Lenguas Modernas 12: 84-104.

Pawley, Andrew. 1992. Formulaic speech. In William Bright (Editor in Chief), International Encyclopedia of Linguistics, Vol. 2, pp. 22-25. Oxford/New York: Oxford University Press.

Tyler, Stephen A. 1987. The unspeakable: Discourse, dialogue, and rhetoric in the postmodern world. Madison WI: The University of Wisconsin Press.

 

see also Grace, George W. 1987 The Linguistic Construction of Reality (pdf)

 

HOME