I. Some Questions and Answers
My function in this series of lectures, as I understand it, is to discuss new directions in phonology as I see them from the perspective of the phonological research going on at Stanford. I am happy to do so, but let me acknowledge first that I will not be able to reflect the full richness of the phonological research there under the auspices of Linguistics, Speech and Hearing Science, and other departments. To give just one example, I will not be referring directly to the exciting seminar conducted jointly by Will Leben and Orrin Robinson in the summer of 1975 on models of phonology proposed to remedy shortcomings in the model of Chomsky and Halle’s The Sound Pattern of English. The points I will make are my own, and my colleagues cannot be held responsible for them, but they will inevitably reflect the perspectives of the three research projects in which I am involved: Language Universals, Phonology Archive, and Child Phonology, which are supported at present principally by the National Science Foundation.
In recent years linguists interested in phonological theory have been asking new kinds of questions, questions that earlier theorists either had not thought of or had considered outside the realm of a general theory of phonology. The linguists who ask these questions feel strongly that phonological theory should provide answers, and they either attempt to find answers in current versions of phonological theory—and hence in their view help to confirm or give support to a particular theory—or suggest new models or modifications in existing theories that will support the answers. Accordingly, the kinds of questions being asked give good clues to what are likely to be new directions in phonological theory.
In order to make this point clear I would like to offer three examples of recent questions. The first was asked by the British linguist N. V. Smith. He wants to know why, in the development of a child’s phonological system, there are well-documented examples of “the loss of a contrast which has already been established” (Smith, 1973:4). In other words, he wants to know why the phonological development sometimes seems to go backward instead of proceeding in a step-by-step differentiation and growth in complexity. The example he has used most often comes from his own son’s development between the ages of two years and two months and two years and ten months. It is the boy’s pronunciation of the adult words side and light. At the beginning of the period he pronounces them alike, something like [dait]. Several months later he differentiates them into something like [dait] and [lait], respectively. Still later they merge again as [lait].
This kind of question is new in that it focuses on the explanation of child language development as one of the functions of phonological theory. Smith says explicitly “The child’s acquisition of phonology. . . is. . . of direct relevance for phonological theory” (1973:207), by which he means that the changes that take place in language acquisition offer evidence for the theoretical constructs set up to account for the adult language synchronically (1973:185). In fact, he even holds the view that the facts of “one nascent idiolect” are as relevant for phonological theory as “the facts of any other language, and to the extent that they have clear psychological validity, perhaps more relevant” (1973:185). Turned around the other way, he is saying that phonological theory is valid only if it accounts for the facts of child language development. This kind of demand may not sound strange to us now because of the increasingly sophisticated investigations of child phonology in recent years, but just a few years ago such a demand would have been rejected almost out of hand by most linguists.
Be that as it may, what kind of answer does Smith find to his question about the phenomenon of ‘recidivism’, as he calls it? His model of child phonology includes (a) a set of unique lexical representations in feature matrices that are identical with the adults’ surface phonetic system, and (b) an ordered set of ‘realization rules’ that are applied to classes of these representations to yield the child’s output. Smith does not attempt to explain how this model is ultimately replaced by the adult model with its more abstract representations, but he seems quite certain that the feature system of the child’s underlying representations does not change with development and that the set of realization rules at the age he studied is psychologically real or valid. Within this framework Smith finds his explanation for recidivism. The child, he claims, has all the distinctive features perceptually, but must gradually master them articulatorily, and the changes he makes in his realization rules reflect to a considerable extent his successive hypotheses on how to assign features in production.
In the case of his side:light example the child had distinct /s/ and /l/ in his lexical representations but his realization rules neutralized all alveolar consonants, including /t d s ∫ t∫ dʓ l r j/ to his |d|, with its voiced and voiceless positional variants. At this point the child pronounced both side and light as [dait]. He was producing the lateral  only in words in which all the consonants were liquids or semivowels, and these all appeared as , e.g., lorry → [lɔli], yellow → [lelo]. Next he began to pronounce the liquids and semivowels as lateral  also in the vicinity of coronal consonants, first varying freely with [d], then predominantly as [d]. At this point side and light were differentiated as [dait] and [lait]. Then the child, Smith thinks, hypothesized that the feature [continuant] rather than [fricative] was the crucial feature characteristic of /s/ and /∫/, and put into effect a realization rule that turned any /s/ or /∫/ in the vicinity of a coronal consonant into the lateral /l/, the only coronal continuant he could pronounce at that time, e.g., see → [li:], shade → [let], side → [lait]. At this point side was pronounced with  for the /s/, and side and light were once again homonyms, now as [lait], until at some still later time the boy could improve his realizations of /s/ and /∫/.
Smith’s central point in this example seems indisputable. In improving his pronunciation of the sibilants, the child tried out a lateral production and stuck with it for a while even at the cost of some homophony involving laterals of other origins. Also, Smith’s more general point in answering the question “Why recidivism?” seems well taken and instructive for phonological theory. He maintains that when a child seems to lose some articulatory ability in the course of his phonological development we should not assume that he has actually lost it but rather we should “invoke the psychological validity of the realization rules and the structural pressure of their longitudinal development” (1973:154). My reason for calling attention to the value of Smith’s question and answer is not that I think his answer always works or that I think his model of phonology is correct. In fact, there are many cases of recidivism—or ‘progressive phonological idioms’, as they have been called by Moskowitz—that his explanation does not fit, including the example most often cited, Hildegard Leopold’s pretty (Leopold, 1947; Moskowitz, 1973; Ferguson and Farwell, 1975), and the model as a whole is seriously inadequate in a number of respects. The question and answer are valuable for at least two reasons: the focus on discovering the nature and scope of phonological processes (such as those he is trying to represent in his realization rules) in human behavior, and the theoretical issues that his analysis has led him to reexamine.
Smith’s concern for justifying and explaining his realization rules represents one of the new directions in phonology in that it focuses on behavioral evidence along with linguistic formalisms. In fact in his chapter 4, “The Nature of the Acquisition of Phonology,” he spends twenty-two pages establishing the psychological validity of the realization rules and only six pages on support for them in terms of their formal linguistic properties.
Smith’s discussion of the implications of child phonology for phonological theory touches a number of different points, and one is worth emphasizing as indicative of another new direction in phonology. Smith concludes that distinctive features, in spite of all their explanatory power, are not the only basic phonological units, but rather that syllables, phonemes, and other units also have psychological reality and make possible the statement of valid generalizations not capturable with distinctive features alone. The new direction is toward recognition of multiple units and greater complexity in phonological behavior as opposed to currently misapplied notions of simplicity and linguistic evaluation measures.
The second question representing an example of new directions is very different. The Soviet linguist T. V. Gamkrelidze recently asked “Why are /p/ and /g/ so weak in the stop systems of natural languages?” (Gamkrelidze, 1974, 1975). The phenomenon he is referring to is the tendency for /p/ and /g/ to be marginal or lacking in stop systems with voice contrast. The full set /p t k b d g/ is very widespread among the world’s languages, but a number of languages have sets that are incomplete in one way or another, and the two commonest kinds of incomplete systems are /p t k b d -/ and /-t k b d g/. This kind of question is not new in a general sense, since one of the aims of phonological theory has always been to characterize the universal limits on sound systems of human languages and find explanations for the limits. The question does represent a new direction, however, in the sense that it is a very precise question, which requires reliable, comparable, accessible data from a large number of languages, preferably representing a systematic sampling of the world’s languages.
Gamkrelidze’s model of phonology is one in which phonemes of a relatively concrete, surface variety are basic units; distinctive features are also units with analytic autonomy; and markedness is a central concept. His model is very different from Smith’s, perhaps most strikingly in its use of notions of relative frequency and relative markedness; he does not deal with child phonology. Phonemes are identified by presence or absence of particular features, although some features may be interpreted as redundant and hence not involved in the basic identification of phonemes. Markedness is not tied only to features but to clustering of features, and may vary in a scale along a particular feature dimension, such as place of articulation. See Fig.2 for a typical Gamkrelidze diagram using arrows for markedness.
The relation between /ptk/ and /bdg/ in phonological analysis has usually been discussed in terms of correctly identifying the single feature that distinguishes them (e.g., voice, tenseness) or in describing the various articulatory, acoustic, and perceptual components or cues by which the feature is implemented (e.g., voice onset time, vowel duration, formant transitions). Linguists have shown very little interest in internal relations in the two series, and we are not prepared for a question about the differences between the p-b opposition and the k-g opposition and their linguistic relevance.
There can be no doubt that the phenomenon Gamkrelidze is asking about exists, and the demand that phonological theory provide an explanation seems reasonable. The data contained in many studies of child phonology development show an asymmetry between labials and velars in the acquisition of the voice distinction. In a number of ways, such as production errors and avoidance rules, children favor b over p but k over g (Ferguson, 1975b). Evidence for the disparity appears in the most unexpected places. The most likely explanation might be physiological—a consideration of relative ease of articulation—but a recent study by Saffran et al. (1975) found the same disparity in the perception skills of an adult male aphasic suffering from the syndrome pure word deafness’ or ‘auditory verbal agnosia’. The patient was given a variety of perception tasks to perform with both natural speech and synthetic speech stimuli. Although his spoken control of language was quite adequate, he had abnormal difficulties in identifying consonant differences in words and nonsense syllables. What is of interest to us in connection with Gamkrelidze’s question is that he was overwhelmingly better in identifying ka and ba than any others of the series (ga da ta pa) and his errors were generally in the direction of identifying pa as ba and all the others (ta da ga) as ka. In some way that phonological theory at present does not explain, b and k are polar extremes in the phonological space of the series /p t k b d g/.
But let us return to Gamkrelidze’s original question. His data on the tendency for /p/ and /g/ to be the gaps in phonological systems came from about thirty languages. Instead of looking for an explanation of the phenomenon, many linguists might doubt its reality, since the number of languages investigated was so small and the sample was quite possibly biased by the availability of grammars and informants, personal knowledge of the investigator, and consequent overrepresentation of particular language families or linguistic areas. Don Sherman of the Stanford Phonology Archive checked the phenomenon out on the 106 languages for which segment inventories were in our computer Archive at that time. Seventy-two had a voicing contrast in stops, of which sixty had the full /p t k b d g/ series. Of the twelve languages with incomplete series, seven lacked /g/ and four lacked /p/. Retrieval by phonetic segments as opposed to phonemically distinctive segments also agreed with the expectation: three languages lacked [g], one lacked [p], and all languages in the sample had the remaining segments at least as allophones or marginal sounds. (Sherman, 1975, reports on this as well as the data of a larger, noncomputerized archive, which gave comparable results.)
Gamkrelidze’s answer to his question is in terms of a scale of markedness. Certain feature combinations are more favored or optimal, i.e., unmarked’, than others. Thus in the series /p t k/ the velar end is the least marked, the labial end the most marked; while in the voiced series /b d g/, it is the labial end that is the least marked. Other things being equal, a voiceless stop is less marked or more favored than a voiced stop, but with the added scale of markedness that the combination of place of articulation and the voicing dimension shows, this basic marking is strengthened or weakened. Thus, /k/ is more favored over /g/ than /t/ is over /d/, and /p/ is correspondingly less favored over /b/. Gamkrelidze also applies this notion of relative markedness of feature combinations to fricatives and other consonants, and he sees the deeper explanation in terms of articulatory and perceptual constraints of human physiology.
Gamkrelidze’s answer may not be fully satisfying because it is not worked out in enough detail and needs evidence of other kinds to support it, but it is of interest in showing that the concept of markedness can be further explored with promising results. Smith, on the other hand, in his study of child phonology, was disappointed in his attempt to make use of markedness, concluding “it would seem. . . impossible to effect any interesting correlation between acquisitional phenomena and marking conventions” (1973:201). Markedness is, of course, not a single unified concept but a family of concepts intended to explain quite different phenomena of favoring one alternative over another under various conditions in phonology and in grammar in general. It would be interesting to have Smith check the /p g/ weakness phenomenon in his own child phonology data. From the data as given in this book there is no evidence for it at all, but the data were, of course, selected and analyzed with other aims in mind.
Even more interesting than the novel view of markedness is an aspect not realized by Gamkrelidze at the time he asked his question, namely, the way his /p g/ phenomenon shows up in child phonology and speech pathology. It is this kind of interconnectedness among phonological phenomena that gives us hope that some day we may have a comprehensive general theory of phonology. One of the new directions in phonological theory construction is certainly the incorporation of data from sources other than detailed phonological analyses of normal adult languages, and we should welcome it in spite of the additional knowledge and skills it requires of the phonologist.
The third question I want to cite is quite different from the other two in its focus on diachronic change. In 1971 the American linguist Alan Bell asked an old question in a new form when he wanted to know “Why are languages which lack CV syllables so rare?” Jakobson and Greenberg had both discussed the universality of CV syllable structure, and Greenberg had ventured a further related universal’ that if a language has clusters of n, consonants in a particular syllable type then it also has clusters of n-1 consonants in that position, except that CV→V does not hold (for citations of Jakobson and Greenberg, see Bell, 1971). The phonologicaThe patient was given aThe patient was given al theorist would be likely to ask the question “Why is CV a universal syllable type?” but Bell views the existence of particular types of syllable structure not as all-or-none, fully determined language states but rather as more or less likely outcomes of a variety of processes of language change. Accordingly he wanted the question really to be “Why are languages without CV syllables so rare that none are known?”
Bell’s model of phonology is essentially that of generative phonology with two differences. First, he works only with the systematic phonetic level at which he assumes each segment is specified for the feature ‘syllabic’ and for its syllable adherence, i.e., what ‘nucleus’ it belongs to (Bell, 1971:114). Second, he adds the notions of language state and diachronic process in a Markov chain model in which it is possible to assign probabilities to the various processes that lead to a change of state. See Fig.3 for an example of one of Bell’s diagrams.
Finding the answer to the question was time-consuming and full of frustrations, but the final result was a much better understanding of processes of change in syllable structure and possible synchronic states as well as the development of a model that could be applied to other phonological questions. Bell used data from 144 languages plus another 25 for illustrating particular phenomena. He devised a new typology of syllable types that was more delicate and more revealing than existing ones: a parametric typology in terms of maximum and minimum numbers of segments in initial, medial, and final clusters. He proceeded to identify three main categories of processes that lead to change in syllable structure: cluster formation processes, cluster simplification processes, and word combination processes. Within each category he assigned relative probabilities to the processes included. This was done by estimating, calculating against the actual language data, and then correcting. Finally the relative weighting of the three main categories of processes was expressed by constants that likewise were estimated and adjusted against the language data.
The answer that emerged from Bell’s computations was that the CV syllable is not universal or near-universal by its inherent nature, but because of the high probability of various processes that lead to it, or, more exactly, by the high probability of the processes that lead away from certain other heavily disfavored syllable structures. CV syllables as such may disappear by vowel loss (and are more likely to do so than CVC syllables, for example), the positions of occurrence of CV syllables in a given language may be highly constrained, and even when there are not distributional constraints the frequency of occurrence of CV syllables may be low. The explanation in terms of the summation of probabilities of alternative processes of change seems much more convincing.
This third question and answer certainly represent some new directions in phonological theory. In particular I would call attention to the principled relationship between diachrony and synchrony and the use of probabilistic models. In recent years we have seen renewed and increased interest in the theoretical foundations of sound change, and it is becoming apparent how useful it is to regard every phonological state as the outcome of processes of change or indeed as the manifestation of changes in progress. Greenberg’s state-and-process model, which was the basis for both Bell’s paper and Larry Hyman’s very different recent paper on nasals (Hyman, 1975), provides an excellent starting point for a new direction in phonological theory. The use of probabilistic models as opposed to the traditional categorical models of phonology is also slowly invading the field. Several decades ago Martin Joos explained to us that linguistic structure and even linguistic change have no place for probabilities and frequency; what really matters is all-or-none (Joos, 1950). A generation of structuralists followed by a generation of generativists have by and large agreed with this position. Now, however, we have Labov and his colleagues demonstrating the linguistic relevance of variable rules, and Bell’s question and answer suggesting the use of a Markov model for certain aspects of language change. It certainly looks as though at least some parts of phonology are following the path of many scientific disciplines toward quantification and calculation of probabilities for predicted outcomes.
In this lecture I have used three questions, one asked by a British, one by a Soviet, and one by an American linguist, in order to point to new directions in phonological theory, and although the topics were very different the three questions all point toward similar new directions: First, the broadening of the concerns of phonology to include in a central place such formerly marginal fields as child language development and speech pathology. Second, the abandoning of our high evaluation of certain narrow forms of simplicity in favor of much more complexly ordered models, which will in part require reliable data from large numbers of languages and the use of probabilistic models. Third, the focusing on behavioral processes in both synchronic and diachronic phonology as opposed to inventories of elements or the formal characteristics of rule systems. Finally, the cheerful adoption of a pluralistic phonology, i.e., willingness to make use of apparently conflicting or overlapping models, in the hope that the transitional probabilities of our science favor the reaching of a more stable state of phonological theory some time in the future.
II. Lessons from Fricatives
In the discussion period after yesterday’s lecture, Sandy Schane asked a question, which was in effect “Why worry about child phonology?” Let us start with his question today. Schane acknowledged that the study of child phonology was of some intrinsic interest, but he was really asking “What does that have to do with the goals of phonological theory as many of us understand them?” or “What does that have to do with writing phonological components in the grammars of languages?” That made me realize that probably many of you had not experienced the impact of the book that Jakobson wrote back in 1941. I would like to start today by trying to relive for you the effect it had on some of us at that time. Although the book was largely wrong in detail, its impact was important and the kind of inspiration it gave can be just as great today, even though our goals in phonological theorizing may be somewhat different.
In 1941 Roman Jakobson wrote the little monograph entitled Kindersprache Aphasie und allgemeine Lautgesetze and published it in Sweden, in an out-of-the-way series, and in German (Jakobson, 1941/1968). Not many American linguists read it, and those who did read it had varying reactions to it. Some said it was nonsense. I can remember several distinguished American linguists saying that there are no such things as allgemeine Lautgesetze (‘general sound laws’). Another reaction was that it was simply the truth: many people accepted what Jakobson said in that book as fact about how language is acquired as though it had been based on a solid body of data. Needless to say some of us did not sympathize very much with either of those two reactions to Kindersprache, but felt “This is great, this is fascinating, let us begin studying the phonological principles he is talking about and see to what extent they are true. Perhaps we will be led in new and important directions.” And that is what we have been doing at Stanford, first with Ruth Weir’s work (Weir, 1962) and then after my arrival in 1967, in the Child Phonology Project.
Now what were the claims Jakobson made in his little book? One of the first was that babbling is more or less random, or unstructured from a phonological point of view, that babies are likely to make any sounds at all when they babble, including those that occur in all sorts of languages, not just the sounds of the language they were exposed to, and that this is very different from the onset of true speech. In fact, he pointed out that sometimes there is even a period of silence, when the child stops babbling for a while before he starts true speech. In short, he made strong statements to the effect that in babbling almost any sound may be uttered, whereas true speech, when it comes, is very structured.
This leads to a second claim Jakobson seemed to make, that from the very first real utterances of the child the phonology is structured in exactly the same sense that adult phonology is structured. Not that the facts of the child’s phonology will be the same in detail as the facts of the adult language, but that the nature of the structure is the same. Even if the child has only two words as a total active lexicon, those two words will have a phonological structure that in essence is the same as adult language. He meant by that, of course, that there will be distinctiveness—or ‘contrast’, as American phonologists often call it—i.e., there will be phonological oppositions used to distinguish lexical items. Also, there will be variations conditioned by position, or ‘environment’, i.e., there will be allophonic variation in both child and adult phonology. There will be limitations on distribution; phonotactic constraints such as neutralization, typical word shapes, and the like in both; and finally there will be subsystems and marginal phenomena in both. Here it is of interest that Jakobson points out that in child phonology marginal phenomena have an extra role in that the marginal material is sometimes the source of core material; that is, may move from the margin into the main body of the phonological organization.
At the time Jakobson wrote, the dominant position in the study of child phonology was that the child’s mistakes in the acquisition of phonology were random, that there was no structure. Jakobson took the extreme opposite view, that true speech on the part of the child is organized in exactly the same sense that adult phonology is organized, and also—a further claim—far from being random or unpredictable, the ‘mistakes’ that the child makes approaching the adult phonology are systematic, lawful, rule-governed—whatever terminology you like to use. Patterns can be found and established for individual children, for whole languages, and even for all languages.
And that brings us to another important claim that is very explicit on Jakobson’s part. Jakobson said that if you have the correct analytical framework of distinctive features and oppositions you will find that there is a universal order of acquisition. Children acquiring a language, no matter what the language is, no matter where it is or when it is, will follow the same general order of the acquisition of features and oppositions in relationship to the phonological organization of the adult language. That is an astonishing claim. Of course, when you look even at any two children acquiring English you find it does not seem to work very well, but put in Jakobson’s terms, it was a claim that made sense, and it is this particular claim of Jakobson’s claims that has perhaps been most widely either denounced or accepted as correct.
The universal order of acquisition was tied to another of Jakobson’s claims. He said that the time when a particular sound or opposition is acquired by the child will be related to the distribution of that sound or opposition among the languages of the world. Let me use some examples of his to make his point clear. Since a sound like t, an apical stop of the t d sort, is practically universal in human language—almost every language has an apical stop—then that should be one of the first sounds acquired in the universal order of acquisition, no matter what the language. Or the opposite extreme, if a sound like the fricative Czech ř is very rare in the world’s languages—there are very few languages that have a segment of that kind in their inventory—Jakobson would claim this is one of the latest sounds to be acquired by a child learning a language. If a child is learning Czech, the ř should be among the latest acquired of the sounds.
He made still another claim related to the universal order of acquisition. Sounds are not only always acquired in a given order but also are lost in the reverse of that order; in other words, the last one acquired is the first one lost, in the case of aphasia or other circumstances in which phonology is disappearing. So we can make an exact mirror kind of statement to our earlier examples: we can say the Czech ř will be among the first sounds lost with a Czech aphasic and a t or d would be among the last. Jakobson claims that there is one set of underlying principles that account for the universal order of acquisition, the universal order of dissolution, the relation between these and the distribution of sounds in the languages of the world, and so on. There is a set of underlying principles to which he gave the quaint name “Laws of Irreversible Solidarity.”
Now I would like to talk about the kind of research we do at Stanford in relation to Jakobson’s theories. Maybe at this point we can say what he said that was wrong and what was right, and give you some sort of feeling for the kind of research we might want to do under his inspiration. First, what was wrong? Babbling is not random, and it tends to improve in intonation; it develops toward clearer syllabic sequences in later stages, and among children who continue to babble along with true speech, there are cases where a certain sound appears more frequently in babbling just before it appears in true speech (cf. Kaplan and Kaplan, 1971). That was just one instance where Jakobson was wrong. In fact, since we are going to talk about fricatives today, let us note that although Jakobson said children babble almost any sound, they rarely babble voiceless fricatives, particularly those formed in the forward part of the mouth, f s ∫ and so forth (Cruttenden, 1970; Ferguson, 1975a). There is almost none of that in babbling. I suspect that if Jakobson realized how wrong he was on babbling, he would now extend his notion of structuredness to earlier years and claim that babbling is also structured and regular and systematic, although not quite in the same sense as ‘true speech’. Or perhaps he would say it has incipient structure or is preparing the way for full structure.
He was wrong about aphasia also, by and large. I know very few students of aphasia today who would accept his formulations as far as phonology is concerned. Aphasia is too complex a thing. Some of you here are taking courses in aphasia and would know this better than I. Aphasics can have many strange things go wrong in phonology or syntax, but they rarely seem to follow the nice orderly pattern that Jakobson would like.
How about cases in which Jakobson simply did not talk about things that were important, where you might say he was inadequate? He never talks about which positions certain sounds come in, beginning or end of a word, between vowels, and so forth. That turns out to be important, since there are different patterns of acquisition in different positions, sometimes quite unlike what Jakobson suggests, so that some of the rules are broken, so to speak (cf. Ferguson, 1975a; Macken, 1975). He does not even talk about such an obvious thing as clusters. Individual segments are not acquired the same way in clusters as they are as single segments. Jakobson also says almost nothing about perception vs. production. The examples he gives are almost all in terms of production: the child learns to make certain sounds or to make use of a certain opposition at a particular point in his development. Jakobson does not talk about whether the child has already mastered the sounds or oppositions in perception, or whether perception and production are learned the same way, or what the relation between the two might be at all. In fact, Jakobson pays almost no attention to what is meant by the notion of acquired: when a given sound, a given opposition, or a given feature is acquired, does that mean the child pronounces it in all the words where he should pronounce it, and does so with a pronunciation that is acceptable in the adult community, or just that he sometimes gets it right in some words or even makes a phonological opposition in his own speech, not commensurate with the adult phonetics? When Jakobson says stops are acquired before fricatives, does he mean all stops before any fricatives, or at least some stops before the first fricative? The lack of precision on this point is quite disappointing, because when it comes to trying to check Jakobson’s predictions, which seem so remarkably clear and explicit, you find you really cannot. Almost always the actual data are too complicated to check out his predictions.
So then what is right about Jakobson’s claims? Why am I still so inspired? Why do I still think Jakobson is so great that I want to go on doing research on child phonology inspired by him? Well, for one thing there is a surprising amount of order. In spite of the fact that individual children do differ in order of acquisition, in spite of the complexities I just alluded to, it is astonishing that overall there tends to be a kind of universal order of acquisition, and to a considerable extent it seems to reflect what Jakobson said. That is a fantastic achievement. I do not know of any other sphere of human sociocultural behavior where you can make that kind of nice precise prediction and find it confirmed; even though to some extent it is wrong and does not take care of many of the facts, it is successful enough to be inspiring.
Secondly, he is right in that there is a relationship between the order of acquisition and notions of markedness or frequency of occurrence in the languages of the world. What is predicted about t d and Czech ř is found to be true. When you study the acquisition of Czech by Czech children, the ř sound is among the last to be acquired. On the other hand, a t- or d- like sound is among the first to be acquired by any child.
This viewpoint of Jakobson’s is so clear and insightful that it is possible to interpret apparently quite different new phonological theories and bring them into line with what Jakobson was saying. For instance, if you want to throw out phonemes and oppositions as the things learned, and say it is only features that are learned, as some theories of phonology might suggest, you can still reread Jakobson, and it will make sense that the distribution of the languages of the world and phenomena such as pathology are all related by some common principles. Or if you take a position like Stampe’s, it can be done. His view of phonology is radically different from Jakobson’s: he does not basically predict a universal order of acquisition—although recently he is moving a little closer to that. He talks about the child’s being born with a finite number of unordered natural processes, and as the child grows up he must progressively suppress them or limit their operation or order and reorder them until he eventually achieves the adult phonology, but Stampe is still basically following Jakobson’s line that there are underlying principles—natural processes, if you will—that account for the interrelationships of acquisition and other kinds of phonological behavior (Stampe, 1969).
I think it must be clear to you what I mean by what I call the Jakobson vision. His vision was that somehow at the bottom of all kinds of phonological phenomena, no matter where we look, whether it is speech pathology or language change through time or the way children acquire language, at the bottom of all phonological phenomena there are some general principles that hold; it is the job of the phonologist to discover them, and you can get at them no matter where you start. You can study speech pathology or language acquisition, or write descriptive grammars, or study the phonological history of families of languages, or whatever, and somehow you will be getting at the same basic, underlying principles. That is what Jakobson was saying, that is the vision that captured my imagination, and even if Jakobson was wrong in many details, and even if phonological theory itself keeps changing, I think that his vision is still as compelling now as it was in 1941.
Now what do we do in the way of research on child phonology at Stanford? Other places do important work but Stanford’s is what I want to talk about. We do lots of kinds of things. Let me hasten to explain that we do not spend all our time studying children’s fricatives, although that is what I am talking about today. One feature of our approach is not just to do our own isolated little experiments and then try to generalize on the basis of those experiments. We prefer to make use of work from very different lines of research. For example, we go through in detail the diary studies done by others in the past, as psychologist or linguist parents of particular children followed their phonological development through time, and we do some longitudinal studies ourselves. We look at large-scale studies where investigators have checked thousands of schoolchildren to see what sounds they are pronouncing satisfactorily in English, and we have done some cross-sectional studies. We want to try to incorporate findings from that kind of research in our thinking. There are also some carefully worked out psycholinguistic experiments where the research design is intended to be foolproof, where you have control groups and all that machinery, and you have randomized everything. We want to find out what those studies have to tell us, and we do a few of them ourselves.
If you are going to investigate fricatives, we first need longitudinal data. We have followed seven English-learning children through a period of about a year, beginning at an age when they had at least one fricative, on that fairly lengthy process of fricative acquisition. We now have tape recordings and transcriptions of these seven children through that period of time, and we are writing up that body of data from a number of points of view. We also carry out some psycholinguistic experiments of our own kind: we have about 28 children, whom we put through various kinds of perception and production experiments on particular hypotheses. In some instances the number of subjects or the number of trials is too small for fully convincing statistical results, but we still find these experiments one of the most fruitful sources of information about how children acquire phonology. Suppose, for example, we think, because Jakobson or some current phonological theory suggests it, that such and such must occur in such and such a way. We devise a test to see if in fact that is what happens, and so we have that kind of specialized information about particular aspects of fricative acquisition from 28 kids—unfortunately not including the seven we followed in our longitudinal studies. Finally, I tend to be suspicious, no matter what kind of general linguistics research is involved, if we look at only one language. I have always been suspicious of efforts of finding universals by looking deeply at one language. I do not deny that it can be done, but I myself am nervous about that kind of approach and so we always like to have at least two languages, preferably a much larger number. So, in addition to studying English-speaking children and their acquiring of fricatives and other sounds, we also study Spanish-speaking children and hope to add some Cantonese-learning children before long. There are enough Spanish-speaking children in the neighborhood of Palo Alto so that we can have a number of children to follow longitudinally. By this time I think the attrition rate has gotten the number down to only four children from whom we have data for a number of months, fewer children than we would like to have, but nevertheless a good bit of data and again with specific psycholinguistic testing of particular hypotheses.1
I think that is enough to give you a feeling for the kind of data that we look for and how we go about obtaining it. I might add that we ourselves do not do any testing of pathological cases or deviant phonologies, but it so happens that Dave Ingram, Mary Louise Edwards, and others who have been associated with our project have in fact done some of the same things with children whose phonologies are deviant, and they relate those findings to what we are doing, so that there is a connecting link between our own Stanford-sponsored work and work with speech pathology.
Now why fricatives? One reason is that they are late in acquisition; as I said a few minutes ago Jakobson said that children will in general acquire stops before fricatives. (Have you ever tried working with a child a year and a half old, trying to get him to do psycholinguistic tests of one sort or another? You can realize why we long for the mature two-and-a-half-year-old so we can work with fewer handicaps.) Also, fricatives present a number of special problems. They do not seem to be acquired quite so neatly and directly as stops are, and so in that sense the study of them should be more revealing. Finally, Jakobson made a number of predictions about them and we can check them out, or at least use them as a starting point. I assume that you all know what the fricatives in English are, but let me set down the symbols ƒ υ θ ð s z ∫ ʓ. Unfortunately when we started the study we did not include the affricates t∫ and dʓ along with the fricatives, so for some tests we have data on the affricates, and for others we do not.
Now what are the predictions that Jakobson made? First, as we have said several times, he predicted that stops would be acquired before fricatives. He also predicted that this would be true in each place of articulation, i.e., if there is both a stop and a fricative in a given place of articulation in any language the child would acquire the stop first. Also, he implicitly makes the claim that voiceless fricatives would be acquired in principle before voiced ones, because voiceless fricatives are somehow less marked or more favored among the world’s languages. A claim that he made quite explicitly was that s would be the first fricative to be acquired. He gave a number of reasons why that would be so, and you can imagine what those reasons would be in terms of distribution in the world’s languages, and so on. There are some complicating factors, since he would acknowledge that a labial stop would often be acquired before an apical stop, but he maintained that the s would be acquired first among the fricatives.
You remember that Jakobson also insisted on the systematic nature of the relationship between adult phonology and child phonology. Accordingly, he made predictions about the kinds of substitutions children will make for adult sounds. In the case of fricatives he predicted that the commonest substitutions would be stops, as would be expected from the order of acquisition. Finally, since Jakobson viewed acquisition in terms of adding oppositions, he claimed that as a particular opposition—or the distinctive feature that implements it—is acquired, the feature will tend to spread rapidly throughout the system. In other words, when [± continuant] or [± voice] is acquired it will tend to spread across the board to phonemes already acquired, insofar as the new opposition is relevant.
Those were some of his predictions. Now, what are the facts when you look at our acquisition? In part, Jakobson is right. It is true in general that stops are acquired before fricatives, and it is true in general that voiceless fricatives are acquired before voiced, but it is not true that s is always acquired first. More children learning English acquire ƒ first. Some children do acquire s first, and that brings up a point that I will come back to, that there is more individual variation than linguists would like to consider.
Let us look at the order in which fricatives are acquired. In general ƒ s ∫ are acquired first, then υ z, and finally θ ð ʓ. This is a rather strange order in some respects, and raises a number of interesting questions. Let us just note that it agrees only in part with Jakobsons predictions.
Next, what are the most common substitutions for the fricatives? You remember Jakobson would predict stops. The most common substitutions are these:
|Adult sound||Child’s substitution|
These are not the only substitutions by any means; many others occur, but these are the commonest ones. A lot of questions are raised for the Jakobsonian viewpoint: For example, why do we have a b as substitute for a v but s as substitute for f? While this pattern holds for English-learning children it is not true for our Spanish-learning children. The Spanish-learning children have a great tendency to produce p as a substitute for f, which is relatively rare with children learning English. And for those of you who do not see any connection between child phonology development and phonological theory in general I would call your attention to this question: Why is it that when speakers of some languages try to pronounce an f in a foreign language that they do not have in their own language, they produce a p as a substitute for the while speakers of other languages do not do so? For example, speakers of Philippine languages typically substitute p for f in English. In the same way here, both Spanish and English have a p and an f, and yet English-learning children by and large do not substitute a p for an while Spanish-learning children, at least in our experience, very often do so. It seems to me that there is some kind of fundamental phonological question here; I would assume the most likely explanation is that there is something about the phonological systems of Spanish and English that is a determining factor in having one pattern in Spanish acquisition and another in English.
We could talk about these substitution rules at some length. It turns out there is a series of possible substitution patterns: avoidance, deletion, stop, continuant, correct fricative. Let me talk about the first one. Children avoid pronouncing certain sounds that for one reason or another they think they cannot pronounce properly. They are perfectly happy to imitate wrongly or make substitutions for some sounds, and other sounds they just will not choose to speak. We found this out in connection with children pronouncing word-initial stops. A given child would fluctuate in pronunciation between p and b, but saying predominantly p, and had no productive contrast between p and b. Yet, if you looked at the words that particular child was using, almost all the ones he had in the p-b category were words that began with b in the adult language, and he was avoiding words that begin with p in the adult language. At some level he was saying, “There’s a difference between these two things and I am going to choose this one.” So, in a primitive way the child begins to acquire voicing opposition. If we had time, I would like to talk a little more about what I call avoidance rules, or saliency rules, because I think these are a part of adult phonology as well. If I asked all of you to write down particular sounds that you avoid, either in English or in some other language you have learned, you would probably be able to mention a few, particularly certain clusters or certain combinations with stress. Also, there are particular words you avoid, not only particular sounds, but particular lexical items. This avoidance is a part of all phonological systems in my view. It is a phenomenon particularly evident with fricatives: many children in the early stage of their acquisition of fricatives characteristically either avoid producing all words with fricatives or avoid one class, for example, all f’s.
Then some fricatives are deleted. Instead of having a substitute, the word is said with no sound in the place where the fricative should be. This is done more often with voiced fricatives than with voiceless ones, and there are other kinds of limitations to it; nevertheless, it is a very common phenomenon.
Then we have the stop and continuant substitutions. ‘Continuant’ is a loose term here by which I mean to indicate less closure and less friction noise than a fricative, for example, a w for an f. The Smith child referred to yesterday used a w substitution for an f in the early part of his acquisition of fricatives.
Now the order I wrote is not an arbitrary order; Ingram (1975) claims that as any child goes through the process of acquiring fricatives he follows this order of substitution patterns, although no one child ever goes through the whole sweep of possibilities. Thus, in acquiring a fricative a child always avoids before he deletes, deletes before he has a stop, and has a stop before a continuant. For example, a given child first avoids words with f, then may substitute w and finally go to f, or he may first delete s, then have a d, and finally go to an s. Our data do not bear this out completely, but to a surprising extent. This is of particular interest in speech pathology, since in deviant phonologies some of these early patterns may persist very late.
I said earlier that Jakobson did not talk about position or ‘environment’. In the case of fricatives this is important because it turns out that fricatives are more easily acquired between vowels or after vowels than stops are, so that in word-final position a child may acquire fricatives before he acquires stops. This tells us something about the nature of fricatives (Ferguson, 1975a).
Also, Jakobson did not discuss clusters, but the patterns of acquisition of fricative clusters are likewise instructive. Let me take one example, clusters with s. Three common kinds of initial s-clusters occur in English: s plus stop (sp- st- sk-), s plus nasal (sm- sn-), and s plus ‘glide’ (sl- sw-). The first question is whether the child treats each cluster as a sequence of two segments—as the phonologists assume the adult does—or as a singleton of special characteristics. The best evidence for the singleton hypothesis comes from children’s pronunciation of sw-. It is amazing how often the sw- cluster is handled as a single unit by children acquiring the language, and is represented by an f. The frication of the s and the labiality of the w combine in the f, however you want to phrase that in your brand of phonology. I think we have a lot to learn about the acquisition of clusters. If I understand Sarah Hawkins’s research correctly, even schoolchildren as old as twelve years of age who seem to have the whole cluster system of English do not have the same temporal relationships between parts of the clusters that adults do (Hawkins, 1973). Apparently in the long and gradual acquisition of the temporal characteristic of clusters, they seem to follow quite a different path from combinations of single segments.
Let us take sp- and sn- clusters as examples of two major types. With sp- clusters, the commonest pattern is the deletion of the s and pronunciation of simple p, and as the p gradually acquires an s, the p is typically lengthened and often less aspirated than the p found elsewhere, so that there is an opposition between p- and the reflex of sp-, but not in terms of successive segments for /sp/. It is as though the child is doing the best he can to pick out this and that characteristic of the sp- cluster until he finally gets everything into it. With initial sn- what characteristically happens is that the child first says just n, then n with a voiceless onset, that is, the child gets the voicelessness of the s, and then puts it into the n first before succeeding with the full s onset. These developmental sequences, like Jakobson examples, help us to predict; they show the incredible degree of pattern in the acquisition of phonology and suggest facts of underlying structure in adult phonology.
I would like to make a point about individual patterns of acquisition, individual strategies or paths toward acquisition. I said that most children acquire f first, before s, but not all children do. A very good example of another way of acquiring fricatives is Hildegard Leopold’s, where her first fricative was ∫, and it came up in a funny way. Her family played a kind of game with a toy train: as they pulled the train along, I think the parents said “choo-choo” or “sh-sh” to imitate the sound of the steam. In any case, she began referring to the train game as ∫ and then extended the meaning to movements of similar kinds, to many things that had nothing to do with trains. Similarly they said “sh!” to her, meaning ‘be quiet’, just before she went to sleep, so that her ∫ came to mean ‘Go to sleep’ and probably was related also to English sleep and German schlafen in that bilingual family’s use. In any case, she had a pair of ∫ homonyms. These items are so marginal that people might argue that they have nothing to do with real fricative acquisition, which blossoms much later, but that is not true. The ∫ stayed along as her first continuant, and in final position after a vowel it comes to be the representative of any final fricative and some nonfricatives; for example, adult final f, s, and ∫ are represented by her final ∫. It is the only fricative she had in her system for a long time, and the final ∫ is in complementary distribution with her initial [j] which represents another set of sounds altogether, these two functioning in some sense as two parts of the same unit (Ferguson, 1968). It is this kind of phenomenon, incidentally, that makes me unhappy with Smith’s notion that the child has no phonological system of his own, but only a system of realization rules to get from the adult input to the child’s output. In many senses, the child does have his own phonological organization.
Now let us talk about the perception and production of fricatives. It is generally assumed that children learn how to perceive an opposition before they know how to produce it afterward. That sounds like a very reasonable assumption, that before a child learns how to make an opposition he learns how to recognize it and identify it. Now, of course, some of you who take a position along the lines of the so-called motor theory will question that assumption, saying that in order to perceive a sound difference you have to be able to make the relevant articulatory movements yourself. But in general most linguists probably make the assumption of perception before production. First, I would like to make the point that infants apparently can discriminate some speech-sound differences. The article most people refer to that began a whole sweep of neonate experimentation, Eimas and associates (1971), showed that children as young as one month of age (when you cannot ask for their intuitions, you have to measure their nonnutritive sucking responses and the like) could apparently react differently to ba and pa. Even if there are some problems with that particular experiment and its claims, I think the literature in general suggests that at a very early age children are able to make certain very precise discriminations among special sounds. The research literature suggests not only a different kind of processing for speech and nonspeech material but even the possibility of some very specific built-in feature detectors, of the kind we have been so surprised in the past few years to find in visual perception. The point I want to make here is that this one aspect of perception comes very early in the development of child phonology, but it is not the same as what Shvachkin has called ‘phonemic perception’, that is, the ability not only to discriminate between two different sounds but also to sort lexical items on the basis of containing them, store the lexical items appropriately, and use the sounds consistently for recognition of the lexical items. This kind of linguistic perception can be tested before the child is able to produce the sounds at all, as well as at any later period in the development. Testing of this kind is relevant for evaluating Smith’s claim that at a very early age the child has the perceptual system complete and that phonological development is essentially perfecting production. At Stanford we have used the Shvachkin technique of testing perception, in the linguistic sense of perception, in the early stages of language acquisition. Let me summarize our work in one or two sentences for our purposes here.
It is generally agreed that children at a relatively early age, although not necessarily prespeech, are able to differentiate among l, r, w, and y, but are not able to reproduce those oppositions consistently until much later. The situation is much more complicated than that—there are all sorts of individual variations—but let me just say that it is pretty clear that liquids are perceived as different relatively early, but produced correctly relatively late. If all sounds went like that we could be much more sympathetic with Smith’s position. However, fricatives do not work that way. We did some Shvachkin tests of fricatives with our children. These tests are a real nuisance. (You make little toys and name them mak, bak, or the like, and you get the child convinced that one thing is really called a mak and another a bak, and then get him to give you evidence that he correctly identifies and remembers the two names. Unfortunately sometimes children do not pay attention, or deliberately do what you do not want, because that is what they think you’re after, or adopt some irrelevant strategy such as always choosing the first one mentioned.) Nevertheless, we are quite convinced that the linguistic perception of fricatives is not early, as is the perception of liquids. In some of our tests it comes at about the same time, sometimes even the production for a given child is a little ahead of the full mastery of perception for a particular opposition. Now presumably this means that the relationship is complex and difficult to tap, but it certainly goes against the notion that perception always precedes production by some great interval in the system. Fricatives are a very good place to test that, because fricatives happen to be a place where perception and production are developed relatively late, and so we have this clear demonstration that it is not true that the perceptual always precedes the productional by great intervals.
I would like to make a final observation: much child phonology research is relevant, among other things, to speech pathology. One little example is that individual differences in early phonological development are so great that we may be able to use some of them as diagnostic devices to predict whether a given child is going to have something wrong with his later phonological development. I will mention two illustrations: If an English-learning child (as opposed to a Spanish-learning or Tagalog-learning child) substitutes p for f the chances are that he or she is going to be a slow phonology learner. If an English-learning child has a lot of reduplication in the earliest stages as opposed to a moderate amount, then again the indications are he is likely to be a slow phonology learner. In this respect our findings may not have anything to do with phonological theory in the narrow sense but with the usefulness of linguistics to society: a careful phonological description of a child’s language behavior at an early time may turn out to be very useful in predicting pathologies or individual lines of phonological development.
III. Universal Nasal Processes
Today I would like to talk, not along the Child Phonology line of research at Stanford, but along the line of Language Universals research and our Phonology Archive. First, I should explain that this lecture will be dedicated to two members of the faculty here this summer, professors Robins and Schane, because the former wrote a paper some years ago on nasalization in Sundanese, which has been reprinted and commented on in many ways (Robins, 1957), and the latter has written at considerable length on the status of nasalized vowels in French and more general theoretical problems related to that (e.g., Schane, 1968).
I will talk about three universal nasal processes. Before we get to the first one, let me say a word about nasal consonants in general. Languages have nasal consonants. That is the kind of statement one would like to make in universal, all-or-none terms. It would be very satisfying to say that all languages have nasal consonants, but if I try to get away with a statement like that there will be a specialist in American Indian languages who will speak up and point out that there are a half dozen languages in the Northwest Coast area, of several unrelated families, that in fact lack nasals. It is no help to learn that some of them are known to have had nasals not so long ago and the nasals changed into voiced stops: the fact is that some languages in the world just do not have nasals. Nevertheless, the statement is a good way to start to talk about nasal universals, somewhat in the same philosophy as some of the comments made in connection with Alan Bell’s question in the first lecture. It is still a universal-type statement if you can say ‘in almost all situations’, as long as it then becomes part of some more general framework in which you can specify the conditions under which it will not happen or some several dimensions along which you can predict varying probabilities. In any case, languages in general do have nasal consonants, and I would like to make a few observations about them. The number of nasal consonants is typically no greater than the number of positions in the basic obstruent series in the language, that is if you have labial, alveolar, palatal, and velar obstruents there will be no more than four different kinds of nasals as far as place of articulation is concerned in the language. Once again there are a handful of exceptions, although actually not one that clearly breaks that rule without a questionable line of analysis (Ferguson, 1966).
Another observation may be along the lines of which nasal is preferred to another, that is, which is more favored or less ‘marked’. If we were going to use the Gamkrelidze arrows to show which is preferred we could write
n is the most favored, or at least marked, and then m a very close second to that, but still second for a number of reasons, and the ŋ a very poor third, and after that the list really tapers off very rapidly so that palatals, retroflex, and other kinds of nasal consonants are much less preferred. We could say many more interesting things about nasal consonants, but that is enough at this point (Ferguson, 1975c).
Now we will get to the first of the nasal processes; I call it Nasal Spread. What I would like to assert is that in every language there is some evidence for the fact that the nasality of nasal consonants spreads to the surrounding vowels. This is a universal process in the sense that there is always a tendency for nasality to spread in this way although the extent and the details vary from language to language. Those of you who care about experimental phonetics and the physiology of speech production will be able to hazard some reasons for that: it is a question of the timing of the velic opening vs. other articulatory movements that have to be made, and this kind of process belongs with other kinds of familiar phonological processes of assimilation where a characteristic of one stretch of speech spreads over into surrounding sounds. Let us pause a while and look at this phenomenon.
It is the kind of phenomenon that phonologists do not pay much attention to by and large, but I would like to discuss it a bit. Suppose we imagine two languages with identical phonologies in terms of inventories of phonemes, and rules, and so forth, except that one has a lot more Nasal Spread than the other; that is, in one there is just a little bit of nasalization of vowels next to nasal consonants and the other has a lot of it, although it still does not affect the phonemic status or the nature of the rules or inventories involved. I think we would feel that there is a significant difference between these two phonologies, that we have put our finger on something that characterizes two languages as different from one another (one thing that linguistics is trying to do is to achieve better means of characterizing the differences between languages of the world). Here, it turns out, if we examine the degree of Nasal Spread in different languages the degree may be very different, and unexpectedly so. For example, look at English, American English in particular. There is a considerable amount of Nasal Spread, generally from the nasal consonant to the preceding vowel and more with some vowels than with others, and more depending upon which consonant follows the nasal and so on, but there is so much Nasal Spread that we must put English toward the high end of the scale. On the other hand, look at Swedish: it has as many nasal consonants as English has and is not so drastically different phonologically from English, but there is almost no Nasal Spread (some, to be sure; you remember we agreed on a conspiratorial working assumption that all languages have Nasal Spread). Swedish has very little, and I suggest that one of the significant differences between the Swedish and English phonological systems is degree of Nasal Spread. Some people have hazarded the guess—a quick and easy guess—that facts like this have to do with whether there are distinctive nasal vowels in the language. There is something to that as a guess, I am sure, but let us think of some languages that have nasal vowels and see if there is the same kind of variation in Nasal Spread among them. In French there is very little Nasal Spread, that is, when you have a nasal consonant next to an oral vowel that oral vowel does not get much nasality attached to it, some but not very much. It has very little nasalization compared to the vowel in a corresponding position in English, next to a nasal consonant; but if we look at a language like Hindi, which also has distinctive nasal vowels, there is quite a bit of Nasal Spread involved under these conditions, so that languages apparently differ systematically or in patterned ways in the degree of Nasal Spread.
Another thing I would like to observe: several degrees of vowel nasality are recognizable, at least in phonetic terms. Bengali is a favorite language of mine for examples. It has seven oral vowels and seven nasal vowels, and the contrast between oral and nasal vowels is neutralized next to nasal consonants under a variety of conditions. The interesting observation I want to make here is that the vowel next to a nasal consonant—where the opposition is neutralized—is often more heavily nasalized than a distinctively nasal vowel elsewhere. People who like neutralizations to come out with the unmarked member are troubled by that kind of phonology, but in any case that is the way Bengali works. The point to be made here is not the phonological status of nasalization in Bengali, but simply that it is fairly easy to discern three levels of nasalization: zero nasality, phonemic nasality, and nonphonemic nasality, of which the last is the greatest in this language, as it so happens. In general, it is fairly easy to recognize two levels of nasality in addition to nonnasal, and those who have listened to both French nasal vowels and Portuguese nasal vowels can notice that the French sound more strongly nasal than the Portuguese. I do not know exactly what the difference is. Decades ago Henry Sweet said it had to do with pressure in the pharynx (Sweet, 1877:8, 211; 1913:468). Finally, people whom I respect tell me that in Chinantec there are three levels of nasality where there is really a distinctive function involved—oral vowels, nasalized vowels, and supernasalized vowels—no matter what kind of analysis in terms of underlying nasal consonants, etc., may be adopted (Merrifield, 1963). In short, there is a tendency for nasality to spread to neighboring vowels, and this important phenomenon comes in measurable doses in various ways in different languages (Clumeck, 1975; Ferguson, 1975c).
The second process I call Nasal Loss. By this I mean the process by which consonants drop out of languages. Once again we can consider this a universal process. Where you have clusters involving nasals there is some tendency in every language for the nasal consonants to drop out of them, and where the nasal comes at the end of the word there is some tendency for the nasal to drop. Just as we saw that Nasal Spread was part of a family of assimilation processes, we can regard Nasal Loss as part of a whole family of weakening processes, including cluster simplification in general and erosion of consonants in final position.
You might find that a little misleading, because you might think I mean that nasals are particularly weak and likely to be lost, but nothing could be further from the truth. In fact, homorganic nasal obstruent clusters are among the world’s most frequently occurring clusters. For example, if a language has any medial clusters at all, it is likely to have one consisting of nasal followed by oral stop in the same place of articulation. Similarly, single nasals are not at all rare at the end of a word, in fact, the other way around. If a language has any final consonants at all, the chances are it has a final nasal. If we watch languages losing final consonants, the chances are they get rid of obstruents first, leaving only s, a couple of liquids, and n, and then it is just the n, and then finally the n goes, so that when I am talking about nasal loss here I am not saying that nasals are weak and very likely to be lost. No, very often they get lost after other things, but Nasal Loss is nevertheless a universal process in that it tends to happen, and when the conditions are favorable enough, it gets carried through.
Let us talk first about the characteristic way in which nasals get lost at the end of a word. Suppose a language has m’s, n’s, and ŋ’s, all as possible final consonants. One way you could imagine final nasal loss taking place would be for each of them to get a little bit weaker, but keeping a three-way contrast, until finally all the final nasals just fade away. In fact, however, languages never seem to work that way. What happens instead is that there are mergers. Instead of the nasals getting weaker, the number of contrasts keeps decreasing and what tends to happen is that m becomes n and n becomes ŋ, and finally ŋ becomes a nasal vowel. This is pretty certainly what has happened in the history of French and more spectacularly in the history of Chinese dialects, where, although there are exceptions here and there, generally the route has been for a gradual merger of nasal endings with the final stages being velar nasal and nasal vowel before complete disappearance of nasality (Chen, 1973; Ruhlen, 1973; Ferguson, 1975c).
It is fun to look around for a language in which final vowel loss is happening, where the process is working itself out before your eyes. Spanish is the language I have in mind. There is only one possible nasal in final position in Spanish. It is difficult to get speakers of Spanish to produce the contrast between final m and final n. In teaching English to Spanish speakers you get the students to bring their lips together and say a nice clear final m and the next minute they are again saying n for m. This is a very deep-seated neutralization in Spanish. Millions of Spanish speakers have only an n in final position, but other millions of Spanish speakers have gone a step further and have ŋ in final position, and still other Spanish speakers have a fluctuation between n and ŋ or fluctuation between a final ŋ and a final nasal vowel (Cedergren and Sankoff, 1975). We see a very good example of this regular process of Nasal Loss in final position.
It would be very nice if once again we could say “That’s the way final nasals always get lost.” Unfortunately, that is not always the way final nasals get lost, although it may represent one of those universal statements that are valid if you put them in a larger framework. Let me offer one counterexample to show that this is not the way it always works. At almost any period, starting some centuries B.C. and coming down to the present time, Greek has been losing its final n; in fact Greek has been trying to lose its final n for over two thousand years: it almost makes it every once in awhile, but the speakers pull the n back into the language because they know how to read or write or because on occasion they speak a more formal variety of Greek, which has it, or whatever, so that the final n is still there in present-day spoken Greek, at least some of the time. Now we would expect the loss of final n in Greek to happen the way we said it generally tends to happen—that is, the n becomes a velar nasal, then a nasal vowel, and disappears. I think you can look in vain among present-day Greek dialects to find someplace where utterance-final n is pronounced like a velar nasal, or where the final n has become a nasalized vowel. Instead what happens is that people just drop the n; they just drop it altogether, it’s gone. It is either all the way there or all the way not there, and many words vary, having two forms, under some circumstances with n and under others without.
So here we find there are at least two ways you can have a final Nasal Loss, but one way is clearly dominant, and occurs over and over again, and the other way occurs in a much more limited number of languages under different sets of circumstances. Seeking to understand why Nasal Loss occurs one way in one language and another way in another is an example of what phonology is all about.
What happens to Nasal Loss in clusters? If you have a cluster consisting of a nasal consonant followed by a nonnasal consonant, there is a definite sequence to the loss of the nasal and a definite preference as to where it disappears first. It depends on whether the following non-nasal consonant is voiceless or voiced. (It also depends on whether it is a fricative or a stop, and other things.) In general the nasal consonant disappears (and affects the preceding vowel) first before voiceless consonants and only later, if at all, before voiced consonants. In fact, there is very different treatment between clusters with a voiceless consonant and clusters with a voiced consonant. Frequently, for example, instead of Nasal Loss in the voiced cluster, you have some kind of assimilation resulting in a geminate nasal consonant, or something else takes place. Whatever the details are, and there are many, depending on the language, what is really striking is the difference between what happens before a voiceless consonant and what happens before a voiced one. The evidence comes from changes through time not only in languages but also in child phonology. The way children acquire nasal clusters, in English and other languages (English is best documented) is very different in a cluster that has a voiceless consonant from one that has a voiced consonant. To call on Smith’s data again, his child dropped the nasal before the voiceless stop particularly in the early stages, but before the voiced stop the nasal stayed and the voice stop dropped, and that is a very common thing in child language development.
Nasal vowels as separate ‘phonemes’ or lexically distinctive segments come into languages almost always as the result of the interaction of Nasal Spread and Nasal Loss. That is, if in a given language nasality is spreading far to the vowel system and the vowels are becoming fairly heavily nasalized, and also the other kind of weakening process is coming in so that final nasals are lost or nasals in clusters are lost—if those two processes match in just the right way, you will have one or more nasalized vowels, without a surrounding environment conditioning the nasality. So one useful way of looking at the coming into existence of distinctive nasal vowels in a language is to see it as the interaction of the universal processes of Nasal Spread and Nasal Loss.
I am not concerned here with what phonological level this is happening at—whether it is happening at the most surfacy kind of level or several notches deeper. What I am concerned with is the nature of the processes in general, whether in some particular language it may have gone only so far that there are still morphophonemic alternations around that give you a nasal consonant in some sort of abstract notation, or whether it has gone still further and there are no such alternations around so you are forced to have a nasal vowel at an underlying level, or whether you do not use that kind of terminology or analysis at all. What I am looking at is simply the process of coming into existence of nasal vowels in languages. They seem to be the result of Nasal Spread and Nasal Loss in various kinds of overlapping interactions (Greenberg, 1966; Ferguson, 1966, 1975c).
I made some comments in the beginning about nasal consonants; now let me make some about nasal vowels. Some languages have distinctive nasal vowels. If there are nasal vowels in a language, we can be sure there will be no more nasal vowels than the number of oral vowels in the language. Bengali, for example, has seven oral vowels and seven nasal vowels, which is a maximum system in terms of the number of nasal vowels in relation to the number of oral vowels. Very often you find systems with fewer nasal vowels than oral vowels. If there are fewer nasal vowels, you can be sure one will be some kind of [a]. You might find a really rare exception, but if the nasal vowels come into the language by the route we are talking about, which is almost always the way they come in, then the language will have at least one [a] vowel. The frequency of occurrence of nasal vowels relative to the frequency of oral vowels is very low, incredibly low. In Bengali, which has an equal number of oral vowel and nasal vowel phonemes, if you count occurrences in running texts you will find at most 1/20 as many nasal vowels as oral, and with some nasal vowels as low as 1/50 the number of corresponding oral vowel occurrences. Finally, we could observe that in most languages with nasal vowels, there are morpho-phonemic alternations with nasal consonants.
Now we have arrived at what I regard as an interesting question, and one of the reasons I have been doing all this talk about processes. Suppose someone asked you as a linguist or phonological theoretician or whatever, why, when we have many languages in the world that have, say, five-vowel systems, say /a e i o u/ (there are hundreds of such languages), do we not have any languages in the world that have instead as a vowel system /a ə ã ǝ̃/ and /a͌/ with second-degree nasalization. We just said we could have two degrees of nasalization and we know about distinctive use of vowel length. Why couldn’t we have a system of this kind? In fact, why couldn’t we have a nine-vowel system with length and double length, and nasality and double nasality? That would work perfectly well as far as bearing information loads and identifying items in the lexicon are concerned; everything would work fine. There is no obvious reason you could not have a phonology of that sort to do what a language has to do, and yet we all know that is a very unlikely phonology. We can answer the question along different lines. Someone asked after the first lecture: “What did Bell mean by looking at inherent nature vs. processes?” I would like to look at the question both ways.
First, you could give an answer based on ‘inherent nature’. I do not want to use that phrase if it turns people off, but you could give an answer that says the system would not work for good articulatory and perceptual reasons. People who have done experiments with the ability to perceive nasal vowels have found unambiguously that it is harder to distinguish vowel-quality differences among nasal vowels than among oral vowels. Similarly, while nasality by itself is quite easy to perceive, to recognize two degrees of it would put a strain on our perception apparatus. Thus perceptual problems would account for the tendency to merge and neutralize nasal vowels more than oral vowels, and for some of the other characteristics of nasal vowels we talked about, such as their lower frequency with respect to oral vowels. Or you could use the same approach from an articulatory point of view and you could probably show that it is harder to produce nasal vowels: not only do you have to put the tongue in the right position but you have to do something with the velic opening. I don’t know, but I am suggesting that you could find answers along the line of the inherent nature of nasal vowels to explain why we have vowel systems of the kind we have instead of other easily imaginable vowel systems based on nasality.
You could also give an answer based on the process point of view. Where do nasal vowels come from? In almost all cases they come from Nasal Spread and Nasal Loss in clusters and in final positions. That means there are never going to be more nasal vowels than oral vowels to begin with because they come from oral vowels by nasalization. What is more, they are going to be less frequent in occurrence because their origin is oral vowels followed by a nasal, and there will be fewer instances of that than oral vowels not followed by a nasal in a given language. So the total result is going to be, once again, fewer occurrences of nasal vowels. I would like to agree, I think, with whoever asked that question the first day, and say that both are legitimate ways of looking at the question “How do you characterize language states in general?” or “What is universally true about how a language must work?” Both approaches are important, and in some important fundamental sense they must be the same thing. They are not unrelated; they must have the same underlying principles, Jakobson’s Laws of Irreversible Solidarity, if we only knew what they were.
The third process is Nasal Syllabicization, a clumsy name for the tendency for nasal consonants to become syllabic. We all know that some consonants can become syllabic although it is usually vowels that are syllabic, and we can even make some principled statements about which consonants are more likely to do this than other consonants. One might be inclined to say offhand that liquids, l’s and r’s, are more likely to do this than n’s and m’s. This sounds like a reasonable kind of statement because it would agree with dominance orders or sonority scales or whatever, in which liquids are somehow more sonorous than nasals are more sonorous than fricatives are more sonorous than stops. Scales of that sort account for certain facts of distribution, certain changes, and so forth. It would be pleasant if it turned out that you would be more likely to get liquids than nasals becoming syllabic consonants. Interestingly enough, if you check it out with languages it turns out not to be so. In over sixty languages where we could get good evidence of the occurrence of syllabic consonants, only three languages had liquids that were syllabic consonants but not nasals. Many had both, and many had nasal syllabic consonants and not liquids. Obviously, whatever order or preference list is involved here, it is not the one we have for sonority. That makes us a little unhappy because I at least was raised to believe that Sanskrit was the language you should pay attention to for things like that; Sanskrit does have syllabic l’s and r’s and does not have syllabic nasals although some people reconstruct earlier stages of Indo-European where there were syllabic nasals as well. I am not talking about phonemically distinctive syllabic consonants, whatever that may mean. I am just talking about the existence of syllabic consonants at a systematic phonetic level, or whatever you want to call it, in a given language, so that English, for example, has syllabic nasal consonants, regardless of what kind of phonological interpretation you want to give them—as in my pronunciation of the on in button. And it is quite clear that languages are more likely to have syllabic nasals than to have syllabic liquids.
Which is the most likely nasal consonant to become syllabic? I think given what I have said so far and our personal experience with languages, we probably are not able to make a reasonable guess. If you guess n you are wrong. Maybe it is a good time to point out that English syllabic nasals are not very typical of syllabic nasals around the world. Just as we tend to think of all nasal vowels as working the way French ones do, since French is the language with nasal vowels that is most at hand, we tend to think of English syllabic nasals as the norm. However, by and large French nasal vowels are just on the edge of normality’, they are not typical of nasal vowel systems around the world, and the same thing is true of English syllabic consonants. For example, English syllabic consonants generally assimilate to the preceding consonant, not the following consonant, but most languages are the opposite. In English, it is more likely that n will become syllabic rather than m or η. To put it impressionistically in terms of my own behavior, I am more likely to have a syllabic n in button than a syllabic m in open or a syllabic η in broken. If you look at enough languages to keep a proper score, n is definitely the least likely of the three nasal consonants to become a syllabic nasal; m is by far most likely to become syllabic, and η and n are relatively close to one another, but n is clearly in third place. This conclusion is the result of looking with great care at some 65 languages that have syllabic nasals, and it constitutes a universal statement in the sense we have been using, that is, not exceptionless, but illustrating the way languages fundamentally work (Bell, 1970).
It should now be clear that if you want to look for universal processes of this kind, or even if you want to look for generalizations about possible language states, the only way to do it is to have reliable, comparable, accessible data around that you can turn to and get answers. Strangely enough, linguists traditionally do not like that. If, for example, we want to know when it is most likely for an n to become syllabic what we apparently prefer to do is think about languages that we know, go to our own file cabinets, 3x5 cards, or whatever, call up a friend on the phone and say “Hey, Joe” or “Hey, Sue, do you know any languages where n becomes syllabic under such and such circumstances?” Or if we get really serious we go to the library and look up some grammars that we think might have syllabic nasals in them. I know of no other scientific discipline that operates in that strange way. Somewhere we should have information of this sort in books and articles, put in a comparable format where you can get at it, so you can ask the librarian or the archivist or the computer and the reply will be “Oh, yes: there are 66 languages that have syllabic n under these circumstances and they are A, B, C,. . . and four of them have interesting exceptions.” It would seem obvious that we could work much faster at getting to an understanding of phonological systems if we had that kind of thing available. That is why we set up the Phonology Archive at Stanford.
That was the thinking in back of the project, but actually getting it started was not easy. It is not just that linguists do not like the idea, for whatever historical reasons in the development of our discipline, but there are some real practical difficulties. Some we deliberately did not pay much attention to. For example, if you ask most linguists about the feasibility of a phonology archive, they say you cannot do it because the published grammars and articles are not reliable, they are just full of mistakes. Our response to that objection has always been something like this: “Then why do we go on publishing them? Presumably, we have a body of published descriptions of languages; if they are no good, then why are we in the business at all? There must be something there—some parts of the information must be reliable. We don’t have to go and check every single fact every single time we want to find one.” On that basis we decided to use only published sources. Sometimes we have them checked by linguists who know the language well and can add comments, but basically it is published documentation, not individual interpretations of a particular native speaker or particular linguist. Thus, the sources are publicly available for any user of the Archive to consult and evaluate.
Also there is the question of what theoretical interpretation you are going to have. Different linguists want to ask very different questions about phonology. I am sure that the way I put some of the material in my lectures here has been very different from the way you have heard some other phonologists put questions, and what is more, ten years from now, or twenty-five years from now, they are going to ask very different theoretical questions from the ones any of us are asking now. So the question is how to put things into the computer or the archive, so that people of different theoretical persuasions can get at them. That is not an easy question to answer. We did not want to have an archive that would be good just for our theory, if we had one; or just for a current, widely accepted theory; or even for what the theory is going to be ten years from now. We wanted to put in observable data that we all agreed on, which would be grist for anyone’s theoretical mill, so to speak. That is asking a lot, but we worked for two or three years to come up with a format that we think would do it, and the Archive is now operational.
I would like to make a final point or two, about the kind of questions you want to ask. There is a great tendency for all of us to ask pseudo-questions or red herrings. I will now reveal my own biases since some of you here can accept the red herrings about which I get upset and some of my questions seem pointless to you. Let me give some examples of what I mean by pseudo-questions. “Is nasality a feature of segments or is it a prosodic entity of some kind?” That is the kind of question people ask seriously. I do not deny their right to ask it, but it is the kind of question I would regard as a pseudo-question or a red herring as far as our Archive would be concerned. It is my own bias that any feature in phonology might function either as a segmental feature or as a prosodic entity of some kind. Now to be sure, stopness is much more likely than stress to be a segmental feature, and stress is much more likely to be some kind of prosodic feature. But there is no knowing. For example, we tend to think about [± voice] as a segmental feature; in fact, one day at a staff meeting of the Universals Project at Stanford I made the mistake of saying “I don’t know of any languages where a given morpheme or word is plus voice or minus voice throughout—whose voicing really works as a kind of accent,” and Joe Greenberg, as he always does, came up with an example: “How about Yambe in New Guinea?” It turns out that Yambe really does work that way, and I was happy to have an example for adult language, since the phenomenon is well attested in child phonology.
In the very early stages of acquisition when they are trying to master the notion of voicing, there are a number of children who will extend it over the whole word. For example, Hildegard Leopold’s pretty, tick-tock, and some other words she said voiceless or whispered throughout, while most of her words were not devoiced that way. She had either plus voice or minus voice for whole words until she mastered the feature and got it localized down into segments. So even an apparently segmental feature such as voicing can, in some instances, function prosodically. Nasality is a good example—it is right in between. When I think about Bengali, one of the really fine languages on the face of the earth, with its seven oral vowels and seven nasal vowels, it is really quite debatable whether nasality is a prosody or not, because there are severe prosodiclike constraints on its occurrence. For example, you can have sequences of vowels and semivowels in the language, up to seven in a row, but they generally have to be all oral or all nasal, and we have already mentioned the neutralization of nasality in vowels next to nasal consonants. I suspect that in every Indo-Aryan language there are some reasons for wanting to regard nasality as prosodic and some reasons for wanting to regard it as segmental. My concern would be not to decide once and for all, as some linguists would like to do, whether nasality is really a prosodic feature or really a segmental feature, but rather to find out the nature and extent of the variation in segmentality or prosodicity of nasality in different languages, the conditions under which it may be one or the other, and so on. And that is the kind of information that somehow has to be available in the archive, not obscured by premature decisions of coding.
Another red herring kind of question would be “Is this a major class feature?” Linguists tend to get preoccupied with what the features are that really divide all segments into major classes such as consonants and vowels, true major features. Nasals are particularly interesting in this connection. What major class do they belong to? I would say they may belong with both liquids and stops. They belong with liquids in the sense that they may become syllabic, as we said a while ago, and they do a lot of other things that liquids do; they belong with stops in that they tend to share a place of articulation features and other kinds of phenomena. Nasals just do not fit that neatly into one major class. Arabic is a good illustration of that. Most varieties have m and n as nasal consonants, and are these a separate class of nasals or do they belong with the liquids l and r or with obstruents b f t d, etc.? There is one principled way of answering that question that is of considerable interest. In Semitic phonology there are constraints on the co-occurrence of phonemes within a given root, depending on how close they are phonetically. If you have a b as the first consonant of the root, you cannot have a b as the second, and you are not even likely to have an f, whereas you can have a t or d very well as the second consonant. The roots are typically three consonants long and there are severe co-occurrence restrictions on root consonants in Semitic in general and in Arabic in particular (Greenberg, 1950). Now if you calculate the phonologically similar classes as found by looking at what can occur with what in a given root, you will find that the m clearly belongs with the b and f: the frequency of co-occurrence is very low for this compared to the co-occurrence of one of these with consonants outside that class. But if you look at n, t, and d there are no restrictions whatsoever, they occur quite freely, but n does not co-occur with l and r. This is why Cantineau, the Arabic phonologist, said n does not belong with t and d in Arabic, it belongs with the l and r, whereas m belongs with the b and f; and once having looked at it this way, he found a lot of other phonetic evidence that made sense (Cantineau, 1946). It is that kind of classification that one might miss if one were setting up things like major class features in constructing an archive. Major class features are for the user of the archive to construct if he wants to, on the basis of the material he can get from the archive.
IV. What Is Phonology?
Today I would like to sketch a beautiful, detailed, explicit model of what phonology is and provide a satisfying set of methods for finding out more about it, and the lecture would be called “Models and Methods.” Unfortunately such a session would take more time than we have and require a more intensive intellectual effort than I can provide.
In lieu of that I will just ramble on in a fairly unstructured way about models and methods, and call the lecture “What Is Phonology?” The word phonology is ambiguous in that it means either the study or the thing studied. I mean both, and it will generally be clear from what I say whether I am referring to the study or the object of study.
What is phonology? I do not want to give a formal definition, but we at least have to agree informally in order to know what we are talking about. I would say that the study of phonology is the study of sound systems in human language, or, better, the systematic study of sound systems in human language. If you like the word scientific with a capital S you can say the scientific study of sound systems in human language; I do not care one way or the other on that point. The object of the study is just that—sound systems in human language. Now how old is the systematic study of phonology? When did it start? We could pick 1957 or 1921 or any other landmark. My view is, however, that it started at least as far back as the second millennium B.C.
I am thinking of the study that must have gone into creating the first more-or-less alphabetic systems of writing, at the eastern end of the Mediterranean. I cannot imagine how people could have thought up writing systems of that kind without their having done some systematic study of sound systems of human language. There is no way you can accidentally stumble on the alphabetic principle. You have to do a lot of thinking and introspecting, listening and experimenting, before you come up with it, and I think that the invention of alphabetic writing reflected both of the principal goals of the study of phonology, as I see them. Those two principal goals are: (1) to achieve a better way to characterize or represent particular sound systems, showing how any one particular sound system differs from another or all other sound systems; and (2) to characterize sound systems in general, that is, the nature of sound systems in human language, the universal constraints on them, their universal properties, and so forth. Those early alphabet inventors were doing both those things, that is, they were figuring out a way to characterize their own particular language (or languages, because actually the alphabet got used right away for different varieties of the same language and even different languages), and they were also trying to figure out how language works in general. Now, of course, this is only speculation because they did not write it down and say so; I am just asserting that they could not have done what they did without having those goals. If we want explicit written evidence and not just circumstantial evidence that people were doing phonology, then we probably do not have to go back quite so far, maybe the fifth century B.C. or something of that sort. For instance, we could take the universal favorite Panini; everyone says he was a great phonologist, and I think he was. He really cared about how a sound system works. It is a little hard to be absolutely sure that he was interested in how sound systems work in all human language, and not just in how the Sanskrit system works, but I am willing to give him the benefit of the doubt. I think he was doing both things and doing them very explicitly. Of course he had predecessors and colleagues; whomever you want to credit with the first explicit study of phonology, it is way back in time, and we are still now slowly building on what was found in the past about how sound systems work.
Now we have to have some justification for talking about phonology all by itself. Is phonology autonomous in some sense? As you know, people sometimes quarrel over whether phonology is autonomous, or for that matter whether syntax is autonomous or other things are and I want to reply both yes and no, in a fairly commonsense way. Phonology is not autonomous in the sense that people do not make sounds in isolation from the rest of language; particular people make sounds in particular words on particular occasions in particular constructions that mean particular things; apart from that you do not have sound systems. Sound systems do not make sense outside of language use and in that sense the study of phonology can never be completely autonomous. On the other hand, it does turn out in a very commonsensical way that we can give some analytical autonomy to phonology. Wherever we may want to draw the line in our theory, we have to give some autonomy to phonology. We all know the person who has learned to speak a foreign language so fluently and so well that he is more at home in that language than he is in the one he started with, and yet he still speaks the second language with a terrible accent. We have all met the linguist or phonetician who learns to pronounce the language so well that you cannot distinguish his pronunciation from that of a native speaker, yet he cannot construct two sentences in the language properly. It is possible to see a kind of commonsense autonomy in which phonology is somehow separate from the rest of language even though it always appears in language use.
Now another issue. Phonology is more complicated than we think. That sentence is really enough for the whole lecture today. Phonology is more complicated than we think, and the more we find out about it the more complicated it turns out to be. I do not think that is a bad thing at all. In fact, I tend to make good value judgments about it. The more you study phonology the more beautiful and the more awe-inspiring it turns out to be. I do not mind at all using a religious word when I say the attitude of a phonologist to phonology is reverence. The more you study human phonologies the more fascinatingly complex and orderly and impressive they turn out to be. I would like to make that point very explicitly at the beginning, just because, time after time, we find ourselves slipping back into wanting to make phonology simpler. There is a great tendency to want to make everything as simple as possible, and although I am not in favor of multiplying entities unnecessarily, I am in favor of recognizing not only that phonology is very complex but that the deeper we look into it the more complexity we will find.
How do people acquire phonology? Children do not learn phonologies as such, that is they do not directly learn oppositions or phonemes or features or rules or whatever it is that phonology consists of. Children learn how to say things. They learn words and how to put them in constructions that mean things on appropriate occasions. But in some marvelous mysterious way, while they are doing that, they sort out this other machinery that is there and gradually begin to operate with features and rules and so on. I think it is very important to remember this point. Jakobson talks about acquiring oppositions; that is in some ways a bad metaphor. It is not as though the oppositions are there, and the child learns first this opposition and then that one. The words are there and the child learns the words, and in some wonderful way the child hypothesizes about the phonology; he organizes phonological regularities from the words he is learning. I would rather change the metaphor a little bit and say that the child learns words and things like words, but he constructs a phonological system.
Now why do I insist on that backwards way of saying it? After all, I am interested in phonology, and presumably the child does in some sense acquire it. I would say we were forced to put the metaphor in that backwards way at Stanford when we kept studying how children acquire language. We looked for such things as oppositions and phonemes and familiar entities of that sort, but what we found was strangely different. Suppose you want to find out whether a child has a p-b opposition, a very reasonable thing to look for. One thing you could do would be to look around and see if he has a minimal pair. In very young children with fewer than fifty words total vocabulary that is not so easy to do, and you must often settle for other evidence. What we found out was disconcerting. Usually you cannot answer the question in a simple, direct way. For example, if you look at how particular words beginning with labial stops are pronounced, one word may begin pretty consistently with a p—or very close to p. Another word, a different lexical item, may vary all over the place, sometimes pronounced with a p, sometimes with a ph, sometimes with a b, and so forth; another word is sometimes pronounced with a b, sometimes with a β and perhaps still another is pretty consistent with a b, so it turns out that almost every word has its own phonological characterization and its own phonological history.
Facts like that finally led us to realize, against all our intentions and all our training and background, that something similar holds true even in adult phonology, that at the basis of an individual adult’s phonology is really something like the phonologically unanalyzed phonetic shapes of whole words, no matter how much phonological order the individual may put into it. All of us make some use of phonetic shapes of whole words unphonologized. We can all repeat a new word we have never heard before, maybe one in another language, blurt it out, not following some of the phonological rules we have in our language, and finally say it again with awareness and note, “Oh yeah, that ends in such-and- such a way.” Let us say you come from a part of Texas where you pronounce thing [θæŋ] although you pronounce sing [SIŋ] (people tell me there are parts of Texas like that). It is perfectly possible that you could discover, when you are age 45 or so that thing does not rhyme with sing; suddenly you notice it. You have been saying it all your life, but you have not become phonologically aware of it; it has not been explicit. Maybe that is not a very convincing example.
Let us consider TOT, the tip-of-the-tongue phenomenon. When you try to remember a word, you know what it is, but you cannot get it out—somebody’s name, for example, or a word you do not use very often. There are some ways in which you try to get access to it in a phonologically oriented way. It is true, that certain segments may come to you, but it is also possible that the strangest, most extraneous kind of phonetic information comes to your mind. To take a very obvious example, the position of stress in the word may come to mind. You may not be able to recall any of the segmental material upon which you would presumably base your phonological rule of stress placement, but you still remember the stress on the word. It is obvious that at the early stages children have whole phonetic word shapes and only gradually do the word shapes get phonologically analyzed by the child. As we grow older we keep adding phonological organization until the phonetic shapes of the words become relatively unimportant, but we still have those, too, somewhere underneath.
As the child continues to grow his phonology develops and he makes more complex organizational hypotheses about the sounds he uses, largely out of awareness. The child shows by his behavior that he has some sort of productive rules. He pronounces something aspirated or stressed or nasalized or long depending on the conditioning environment. As the child is doing all this he is also getting into a network of social variation in the use of the language. The people talking to the child are speaking different regional and social dialects and situational and functional registers depending on where they come from and what the occasion is, e.g., talking to a child or to an adult. The child is taking in all that kind of data at the same time and hypothesizing about it. It is a different kind of hypothesis making; he has to say, “my [æ] is his [εə]” when it is Uncle Joe who is talking to him; he has to make all kinds of generalizations of this sort about the language he hears and speaks. Again, some of this variation is just learned as whole word shapes, some is phonologically organized, and so on, but there is a whole network of social variability in pronunciation that the child is acquiring, and that is an important part of acquiring the phonology of any language.
In our own society we eventually learn how to read and write, too; that is still probably not true of most people in the world, but it is increasingly true, and this is so important we just cannot brush it aside. It is part of the social variability business, too, but it has such a tremendous impact that as phonologists we have to recognize it. As soon as a speech community begins to write and read its own language, the spelling representation begins to affect the speakers’ phonology. Examples are legion, but linguists prefer to ignore them, pretending there is only an occasional spelling pronunciation. That is not the way it is. We begin to add to our phonetic/phonological representation—if you want to use that word of words—a written dimension in addition to all the other stuff we are already loading it down with, and that enriches or complicates the kind of phonological knowledge or competence we have.
All of this can eventually get awareness added to it. All these wonderful phonological hypothesizings are done at first basically out of awareness. The child cannot talk about it. But as the child gets older, he gets to the point where he can talk about it. I am sure that some of you, especially if you have tried to teach reading or the first grade, have had the experience of watching the child discover what you mean by saying two things end the same way or begin the same way. Maybe you have had the same frustrating experiences I have had, talking to a reading teacher or a reading specialist who says, “This child can’t make [or can’t hear] the difference between b and p.” I want to say, “What do you mean? This child is in the first grade; when he says ball or Paul he knows what is meant, he says it right and reacts consistently when others say it. What do you mean by sáying the child can’t do that?” The reading teacher, however, does mean something. It turns out that if you try to get the child to tell where the difference is between ball and Paul, the child has no idea—he may not even know whether the difference is at the beginning of the word or the end of the word. The child has not yet acquired the awareness of segmentation, which is a prerequisite for learning how to read (Mattingly, 1972). We are so used to written representation that we think the segmentation is natural and consciously accessible. Awareness of the sort of phonological organization we have in our heads is something that grows gradually and has its own patterns of growth, differing from the growth of phonological organization as a rule. It is also worth noting that individuals differ a lot in the degree of phonological awareness they have. It is rather pleasant to think that it is phonologists and linguists who have the greatest awareness. Maybe, in fact, that is all the study of phonology is: making more precise or explicit the phonological awareness we can have inside our heads.
No two of us have exactly the same phonology. We each learn a different set of vocabulary items from different people around us under different circumstances; we do our internal phonologizing in different ways; and we reach different levels of phonological awareness. As a result, every one of us has a unique phonology. That is something that is very hard to persuade linguists of; I know, since I have tried on a number of occasions, and the resistance is very strong. I say something like, “We all have a different phonology and it doesn’t make sense to say that a child acquires English phonology by age X”—and a little later (once after almost two hours of talking about it) one of the linguists present says, “Yes, but there must be a time when the child really has acquired the phonology of the language”—so I really have not made the point. It is strange that we do not mind this when we are talking about lexicon; everyone admits that no two of us have exactly the same vocabulary, and that we do not use the words we have in common in exactly the same way. Phonology is like that too. There is, of course, a great deal of overlap among individual phonologies in a speech community, and I would suspect that the amount of overlap is roughly comparable to that in lexicon. The overlap makes it possible to construct an impressive set of common phonological rules, but we must not kid ourselves, for example, that we all speak the phonology of English. We all speak our own phonologies, and the way we organize them internally happens to overlap a lot. Studying the nature of these overlappings is interesting and important in itself, and it is a necessary part of learning how phonological systems work.
Let us move to the question of phonological change. When a change in the phonological behavior of a speech community takes place—and this is happening all the time—it must start with particular sounds in particular words said by particular speakers on particular occasions, and then it may spread from some sounds to other sounds, or from the same sounds in some words to the same sounds in other words, or from some groups of speakers saying those sounds in those words to other groups of speakers saying the same sounds in those words or different words; it may spread from one occasion of use to another. It just does not happen that a sound change takes place in all words for all speakers for all occasions. So any model we have of phonology has to account for this complicated way that phonological change takes place in a speech community. There is no model of phonology currently available that really can do justice to these facts about phonological change.
How do we explain the changes that take place anyway? Which kinds of change are possible and which are not possible? What constraints are there on sound changes and what explanation can we offer for them? I would hazard a guess that there are at least three quite different kinds of constraints or explanations. One I call phonetic, and by this I mean things that have to do with the physiology of speech production and speech perception. A lot of the changes are explainable in these terms or along this dimension. I do not know where to draw the boundary line in this fairly peripheral kind of processing—but presumably we want to draw it somewhere before we get to the cortex. In any case, it has to do with the way nerves and muscles and all the rest operate in both perception and production. I think phonetic is as good a word as any for that.
For another kind of constraint or explanation I use the word cognitive—a poor choice, perhaps, but cognitive already has so many meanings I guess it will not hurt to use another one. By cognitive constraints or explanations I refer to the kind of processing that goes on, let us say, in the central nervous system. But I am not concerned with the physiology at the moment; I am concerned with identifying the kind of constraints that are true not only for sound systems but also for other kinds of human cognitive behavior, including processes of classification, hierarchization, and so on, which are not limited to phonology but appear elsewhere in language and for that matter in nonlanguage kinds of behavior.
Finally, the third kind I call social. I take it as a universal assumption that in all societies social groupings tend to be reflected by language use. Also in all speech communities some people tend to change their speech to talk more like other people or less like other people, depending on social factors. These universal kinds of social constraints are an important factor in phonological change.
The characterizations we make, the definitions for phonology we use, must include processes of change and must include ultimately these three kinds of dimensions, what I am calling crudely phonetic, cognitive, and social.
Now I would like to run through some characteristics of the study of phonology in the future. I think this is what I am supposed to talk about—New Directions in Phonology. As it so happens some of the ‘new’ directions are very old, but some may be really new. Let me run through a list of characteristics of phonology and methods of research that typify the new directions.
First, phonology is polysystemic—a good working term, I think. I say that with considerable reluctance because I was raised in a tradition that said—what arrant nonsense!—phonology is a unified system even if you have some fairly distinct little subsystems. While that is a good and insightful notion, I think it is basically wrong. Years ago when I was studying Bengali I noticed, as any phonologist studying Bengali would notice, that the stem vowels of the verbs alternate in highly regular ways, between i~e, e~æ, u~o, o~ɔ, and a~e, depending on the suffix. Part of the conditioning seems to be phonological and part morphological, but if you are a little imaginative in constructing the underlying representations (or their equivalents in other theoretical frameworks) it can be completely phonological. So that within the verb system all verb stems have an automatic vowel alternation such that you only need to posit five underlying vowels. There are seven vowel qualities, but you do not need to talk about seven vowels in analyzing Bengali verb stems. But those seven surface vowel qualities are identified by any Bengali with the seven distinctively different vowels in the nonverbs of the language, which are all over the place and do not alternate so neatly. I remember a Firthian analysis of Bengali verbs that really irritated me because it insisted that there were only five vowels in the verb even though there may be seven vowels elsewhere in the language. I said, “That’s crazy; anybody knows there are seven vowels in the verb, along with some regular alternations.” Somehow we have to get both those facts strongly and importantly in any phonological description of Bengali. It is not accidental or superficial that there are only five stem-differentiating vowels in the verb system and seven vowels elsewhere, and that the seven surface vowels of verbs are phonetically identical with the seven vowels that contrast elsewhere. The interlocking systems must somehow be captured in a way that I cannot feel any current phonological theory does. That is one example of polysystemic analysis.
It would take too long to give an autobiographical list of sins where I feel I have forced something into a single system when it was many systems, but I think we increasingly have to reckon with interacting systems within our phonologies. In child phonology we often have to recognize different systems in initial position and final position, or different systems of chronological layering. In whole languages there can be vocabularies of different origins that differ in phonology or in marginal phenomena of baby talk, or animal calls, and some that in some respects are similar to core phenomena in a system, but in other respects work quite differently.
I am not offering just a recapitulation of Firthian phonology by any means, but Firth and his followers at least had their eyes on some important characteristics about phonology when they made fun of American linguists’ squeezing everything into a segmental, monosystemic framework.
Second, phonology is variable. The fact of variation has to be included systematically in any type of phonological theory. And if we need a nice example we can take a Labovian variable, so that we all know what we are talking about. It is a part of the competence of any speaker of New York English that the degree of post-vocalic r in his speech will correlate with the degree of carefulness of speech, social class, and some other social factors, and it is interesting to study the patterns of such variation. The pattern of this variation in New York is different from the pattern elsewhere, so to that extent it represents regional dialect variation. Also, it differs by social class, so that it is social dialect variation. Finally, the variation depends on degree of formality and who the addressee is, so it is registral variation. Some way of handling geographical and social dialect variation as well as register variation has to be built into our understanding of phonology and the kind of model we have of phonology.
An interesting kind of variation is the modification of pronunciation adults make when talking to children. A simple example again out of my own experience: I spoke with a very marked Philadelphia accent when I was growing up, and as I grew older and went to the university some of the features of the Philadelphia accent were cancelled out or modified, toned down a little bit. When I was quite young you could tell after a sentence or so that I came from Philadelphia; when I was older you might have had to listen to quite a number of sentences and focus very carefully on my pronunciation of ‘short a or some other special feature before you could figure out that I did indeed come from Philadelphia. It would seem then that I had gotten rid of those early Philadelphianisms, or toned them down. But when I was talking to my own children when they were very young, I found myself reinstating many of those Philadelphianisms that I thought had vanished. Not that I intended to; in fact I was dismayed to find I was doing that. Then I observed that a lot of other people do the same kind of thing, and it seemed clear that stored away somewhere in my competence was a set of rules that said, as it were, “Suppress certain phenomena in ordinary conversation but throw a switch and let them up and out again in certain situations.” And this is not a strange peculiar property of Philadelphians, or of linguists, or of Ferguson. It is a phenomenon that is found in all speech communities on the part of all speakers. We all do it, and it belongs somehow in our theoretical framework of phonology.
An example that I have used over and over again (e.g., Ferguson, 1972) is the difference between [r] and [γ] in Christian and Muslim dialects of Arabic in the city of Baghdad. The Muslim dialect has a contrasting r and gamma; the Christian dialect also has a contrasting r and gamma, so they have the same inventory of phonemes, so to speak, but there are many words in Muslim Arabic with an r that have a gamma in the Christian Arabic. By no means all; there are some that have r also in Christian Arabic. It is also true that many Christian speakers can switch to the Muslim pronunciation, so that if a group of Christians are speaking to each other freely, and a Muslim walks in, they may all switch over automatically to speaking Muslim Arabic, which is the dominant variety. What does this mean? It means that the Christian Arab in Baghdad has a set of lexical items in his repertoire that are labelled “pronounce gamma if there are Christians there, but pronounce r if there are Muslims there.” He also has a set of lexical items that are always pronouncd with r, and another set that are always pronounced with gamma. Now that is an incredibly complicated thing to do, yet people do it largely out of awareness—although every Christian will recognize right away as soon as he hears the gamma in a certain word that he is speaking with a coreligionist. The signal is there. That kind of variation is not strange and peculiar—only the example is strange because it is not familiar to us—the phenomenon is a part of all languages, all speech communities, all phonological systems; it has to do with the way phonological systems function.
The third thing is that phonology has multiple basic units. It is fun to argue about which is the basic unit: is it the distinctive feature or the phoneme, the opposition or the segment. Those are all good candidates, and one can think of many others; but I think the more one seriously tries to represent what is going on in phonological systems the more one is driven to recognize that one cannot get along with just one or two basic units. There is a whole list of basic units that we have to find. It is quite clear in some languages that a unit like the word plays an important role in the phonology, and you cannot do without it; that is you cannot derive it from other kinds of units in the same system. Linguists tried to get along without syllables for a long time; I was raised in the tradition that said syllables were not really of interest to linguists; you could always characterize them by sets of abbreviations of segmental units. Then it turns out that in many languages you have to have syllables as the unit in which phonological rules work, and that the syllable is basic in child language processing (e.g., Fudge, 1969). I am sure you are familiar with the studies (e.g., Gleitman and Rozin, 1973) that show that in teaching children to read it works better, by and large, to teach them in terms of syllables rather than phoneme-length segments. All I am saying is that we have to have many more levels, units, and components in our phonology than we ever dreamed. We have to have units larger as well as smaller than the distinctive features of phonemes.
Now a fourth characteristic, and this is not so much a characteristic of phonology as it is a current bias of mine. Given our present primitive understanding of how phonology works, I think we always have to look at phonological phenomena in at least two ways: as the organization of elements and relations, and as processes. Sometimes those two ways of looking at things are easily translatable, one into the other; and sometimes it is quite clear that some things are better, more helpfully, more conveniently stated in organizational terms or in processual terms, but sometimes there is an important insight in doing it both ways, and sometimes it is not clear which way is better. So for the time being I have adopted a policy of trying to say everything both ways. In a way it is reminiscent of Hockett’s talking about item-and-arrangement, process, and so on (Hockett, 1955). One of the great things generative phonology has done for us is to give us the notation X→Y/Z for the notion that X becomes Y under condition Z; it is a wonderful kind of notation and turns out to be useful and suggestive in many ways.
We should not lose sight of the fact that we can use the same notation and conceptualization in talking about synchronic processes in which there is no real time involved, and diachronic processes in which there is real time involved. We can talk about processes with this notation and conceptualization if they are within the same system and if they are between systems. Let me give just one example of this last point. When we talked about substitution rules in child phonology development, that was going outside the system. We were not talking about processes in the same sense linguists usually talk about phonological rules. We were talking about the processes relating to the adult model and the child’s output. So we have a kind of four-way classification that is between synchrony and diachrony, between within-system and across-system. What is fascinating about making lists of phonological processes from different languages and in those four categories is that hundreds, perhaps thousands turn out to be the same. They do not all turn out to be the same. Some of them will happen in only one or several of the four categories but not in the others, but that is what is fundamental about understanding phonological processes. We have to find out which of these processes are common across the board and which are not, and try to find explanations of why that should be.
Just one more characteristic of phonology—it is likely to put less emphasis on distinctive vs. nondistinctive, or phonemic vs. phonetic. I did not say we were abandoning the notion that features of distinctiveness have some particular role in phonology, but we have in general tended to emphasize that idea so much that we have lost important information about how phonological systems work. Again, we were forced to recognize this in doing child phonology studies. We found out we could understand what we were doing only if we looked at extreme phonetic detail before we could phonologize. In opposition to some people who are saying emphatically that we must draw the line ever more carefully between phonetic and phonemic interpretations of child language development, I would say no, we must practice more erasing of that line here and there to understand how some of the phenomena work, although we must somehow do it without losing the important content of distinctive function.
Now just a few words about method. How do you find the data? How do you get the facts in phonology? I am a proponent now, as I think I have been as far back as I can remember, for all four ways that people use, and I am particularly opposed to people who say only one of those four ways is the right way to do it. The first way is introspection. “Say it to yourself and try to figure out what is going on.” It is easy to downgrade that and to note that you can always kid yourself. Introspection has lots of problems. Psychologists have discovered and rediscovered that time and time again. It is a tricky approach to use to find out about human behavior, but still there are many things to be found out by introspection that cannot be found out in other ways, and introspection has provided some of the most insightful and instructive work in the whole field of phonology. Another way is the old-fashioned structured elicitation of the anthropologist. “Tell me a folktale would you, and let me get it on my tape recorder.” Of course, it is easy to put that down too. The situation is unnatural, and what you get is going to be distorted. The critics ask why not at least get into a conversation with the informant. I will just repeat the argument I used for introspection. In spite of all the difficulties of doing it that way, some of the most important work in phonology has been done that way, and it is still a very intensive way to get the kind of data we want and need in phonology.
The third way is naturalistic observation. We are finally getting better at that. We are finally getting to the point where we observe real language in actual use and have a record of it in reliable form. I think that is great. I do not know that I have to defend that method here. There are some people who would say, maybe you do not even want to do that. I remember an argument in which someone said that ordinary, sloppy, casual, quick conversation is the worst place to start; it is the farthest removed from the norm. First you have got to get super-careful pronunciation in just the right way, and then you can write the rules when you get back to the other one, maybe. This represents two points of view, essentially the same ones that Hockett characterized a long time ago as clarity norm vs. frequency norm. Finally, there is another method, which an increasing number of people are using these days and which we have to learn to do better: carefully designed experimentation to get the data we need in such a way that even nonlinguists will be convinced that the material is valid and reliable.
In addition to predicting the use of all four methods, I think that phonologists will tend to be more sensitive in certain ways than many have been in the past. First they will be more sensitive to physiological phenomena. It is all very well to hypothesize and speculate about the nature of phonological systems, but we can no longer afford to do so in as much ignorance as we have often shown about human anatomy and physiology. Also we will have to be more sociolinguistically sensitive. Phonologists have made incredible blunders by failing to cope with sociolinguistic factors, and we are still wasting a lot of time arguing about noncomparable data because we are sociolinguistically insensitive to the limitations involved. I like all four methods of getting data, and I would like all four to be done with physiological and sociolinguistic sensitivity; I do not see how any one can really quarrel with that. We are in favor of virtue.
I would like to say at least a word about whether I think the prognosis for phonology is good, bad, or indifferent. We should stop and take a look at the field and see whether we are really getting anywhere. Is it going to turn out that this field does not really exist or that phonology has to mark time while some other branch of linguistics moves ahead? No, I have high hopes for phonology. I think the prognosis for phonological theory is excellent; I feel more optimism about phonology at the present time than I do about syntax and semantics. Needless to say, in every branch of linguistics the theoretical positions are in disarray; that is, there are alternatives that seem to be mutually exclusive. You try one alternative for a while and it turns out to be very disappointing; you try another and it does not seem to work any better. But let us realize that in the field of phonology we have an enormous amount of data from all kinds of languages, and we have a degree of sophistication in phonological analysis that we have never had before. We have ways of asking questions about phonology that have become a matter of course, although a few years ago they were not known. I would say that the future of phonology is very bright indeed. The patient is not in serious condition, but on the way to good health, and maybe we should not talk about its being a patient at all.
WPLU = Working Papers on Language Universals
PRCLD = Papers and Reports on Child Language Development
Bell, Alan. 1970. Development of syllabic nasals: the case of the Bantu noun class prefixes mu-, mi-, and ma-. WPLU 2:B1-B18.
———. 1971. Some patterns of occurrence and formation of syllable structures. WPLU 6:23-137.
Cantineau, Jean. 1946. Esquisse d’une phonologie de l’arabe classique. Bulletin de la Société de Linguistique de Paris 43:93-140.
Cedergren, Henrietta J., and Sankoff, David. 1975. Nasals: a sociolinguistic study of change in progress. In Ferguson et al., eds. (1975), Nasálfest, pp. 67-80.
Chen, Matthew. 1973. Cross-dialectal comparison: a case study and some theoretical considerations. J. Chinese Ling. 1:38-63.
Chomsky, Noam, and Halle, Morris. 1968. The Sound Pattern of English. New York: Harper & Row.
Clumeck, Harold. 1975. A cross-linguistic investigation of vowel nasalization: an instrumental study. In Ferguson et al., eds. (1975), Nasálfest, pp. 133-53.
Cruttenden, Alan. 1970. A phonetic study of babbling. British Journal of Disorders of Communication 5(2): 110-17.
Eimas, Peter D.; Siqueland, Einar R.; Jusczyk, Peter; and Vigorito, James. 1971. Speech perception in infants. Science 171:303-306.
Ferguson, Charles A. 1966. Some assumptions about nasals: a sample study in phonological universals. In J. H. Greenberg, ed., Universals of Language, pp. 53-60. Cambridge: MIT Press.
———. 1968. Contrastive analysis and language development. Monog. Series on Lang, and Ling. 21:101-12.
———. 1972. Some requirements of a theory of language behavior. Presidential address, L.S.A.
———. 1975a. Fricatives in child language acquisition. In Luigi Heilmann, ed., Proceedings of the 8th Int’l. Congress of Linguistics (Bologna-Florence, 1972), pp. 647-64. Bologna: Società editrice il Mulino.
———. 1975b. Sound patterns in language acquisition. In D. P. Dato, ed., Developmental Psycholinguistics: Theory and Applications, pp. 1-16. Georgetown University Round Table 1975.
———. 1975c. Universal tendencies and ‘normal’ nasality. In C. A. Ferguson et al., eds., Nasálfest, pp. 175-96.
Ferguson, Charles A., and Farwell, Carol B. 1975. Words and sounds in early language acquisition. Language 5(2):419-39.
Ferguson, Charles A.; Hyman, Larry M.; and Ohala, John J., eds. 1975. Nasálfest: Papers from a Symposium on Nasals and Nasalization. Stanford, California: Language Universals Project, Department of Linguistics, Stanford University.
Fudge, E. 1969. Syllables. J. Linguistics 5:253-86.
Gamkrelidze, T. V. 1974. Sootnošenie smycnyx frikativnyx v fonologičeskoj sisteme. Moscow: Institut Russkogo Jazyki, Akademija Nauk SSSR.
———. 1975. On the correlation of stops and fricatives in a phonological system. Lingua 35:231-62.
Gleitman, Lila R., and Rozin, Paul. 1973. Teaching reading by use of a syllabary. Reading Research Quarterly 8:447-83.
Greenberg, Joseph H. 1950. The patterning of root morphemes in Semitic. Word 6:162-81.
———. 1966. Synchronic and diachronic universals in phonology. Language 42(2):508-18.
Hawkins, Sarah. 1973. Temporal coordination of consonants in the speech of children: preliminary data. J. of Phonetics 1 (3): 181-217.
Hockett, Charles F. 1955. A Manual of Phonology. Baltimore: Waverly Press.
Hyman, Larry. 1975. Nasal states and nasal processes. In Ferguson et al., eds. (1975), Nasálfest, pp. 249-64.
Ingram, David. 1975. The acquisition of fricatives and affricates by normal and linguistically deviant children. In A. Caramazza, and E. Zuriff, eds., The Acquisition and Breakdown of Language. Baltimore: The Johns Hopkins University Press.
Jakobson, Roman. 1941. Kindersprache, Aphasie und allgemeine Lautgesetze. Uppsala. Trans, by Allan R. Keiler: Child Language Aphasia and Phonological Universals. The Hague: Mouton, 1968.
Joos, Martin. 1950. Description of language design. Journal of Acoustical Society of America 22:701-708.
Kaplan, Eleanor L., and Kaplan, George A. 1971. Is there any such thing as a pre-linguistic child? In John Eliot, ed., Human Development and Cognitive Processes. New York: Holt, Rinehart and Winston.
Leopold, Werner F. 1947. Speech Development of a Bilingual Child: A Linguist’s Record. Vol. II: Sound-Learning of the Two-Year Old Child. Northwestern University Humanities Series, Vol. 11. New York: AMS Press.
Macken, Marlys A. 1975. The acquisition of intervocalic consonants in Mexican Spanish: a cross-sectional study based on imitation data. PRCLD 9:29-42.
Mattingly, Ignatius G. 1972. Reading, the linguistic process and linguistic awareness. In J. F. Kavanagh and I. G. Mattingly, eds., Language by Ear and Eye: The Relationships between Speech and Reading. Cambridge: MIT Press.
Merrifield, William R. 1963. Palantla Chinantec syllable types. Anthropological Linguistics 5.5:1-16.
Moskowitz, A. I. 1973. Acquisition of phonology and syntax: a preliminary study. In Hintikka et al., eds., Approaches to Natural Language, pp. 48-84. Holland: Reidel Publishing Co.
Robins, R. H. 1957. Vowel nasality in Sundanese: a grammatical study. In Studies in Linguistic Analysis, pp. 87-103. Oxford: Blackwell.
Ruhlen, Merritt. 1973. Nasal vowels. WPLU 12:1-36.
Saffran, Eleanor M.; Marin, Oscar S. M.; and Yeni-Komshian, Grace H. 1975. An analysis of speech perception in word deafness. Brain and Language, in press.
Schane, Sanford. 1968. French Phonology and Morphology. Cambridge: MIT Press.
Sherman, Donald. 1975. Stop and fricative systems: a discussion of paradigmatic gaps and the question of language sampling. WPLU 17:1-31.
Smith, N. V. 1973. The Acquisition of Phonology. Cambridge: Cambridge University Press.
———. 1973. The Acquisition of Phonology: A Case Study. Cambridge: Cambridge University Press.
Stampe, David. 1969. The acquisition of phonetic representation. Papers from the Fifth Regional Meeting of the Chicago Linguistic Society 5:443-54.
Sweet, Henry. 1877. A Handbook of Phonetics. Oxford: Clarendon Press.
———. 1913. Collected Papers. Oxford: Clarendon Press.
Weir, Ruth H. 1962. language in the Crib. The Hague: Mouton.
1. At the time this paper was given, only four children were participating in the study; later one child returned and another was added. By the completion of the study, complete data had been gathered from six chi ldren over a nine-month period.