When we study human language, we are approaching what some might call the “human essence,” the distinctive qualities of mind that are, so far as we know, unique to man and that are inseparable from any critical phase of human existence, personal or social. Hence the fascination of this study, and, no less, its frustration. The frustration arises from the fact that despite much progress, we remain as incapable as ever before of coming to grips with the core problem of human language, which I take to be this: having mastered a language, one is able to understand an indefinite number of expressions that are new to one’s experience, that bear no simple physical resemblance and are in no simple way analogous to the expressions that constitute one’s linguistic experience; and one is able, with greater or less facility, to produce such expressions on an appropriate occasion, despite their novelty and independently of detectable stimulus configurations, and to be understood by others who share this still mysterious ability. The normal use of language is, in this sense, a creative activity. This creative aspect of normal language use is one fundamental factor that distinguishes human language from any known system of animal communication.
It is important to bear in mind that the creation of linguistic expressions that are novel but appropriate is the normal mode of language use. If some individual were to restrict himself largely to a definite set of linguistic patterns, to a set of habitual responses to stimulus configurations, or to “analogies” in the sense of modern linguistics, we would regard him as mentally defective, as being less human than animal. He would immediately be set apart from normal humans by his inability to understand normal discourse, or to take part in it in the normal way – the normal way being innovative, free from control by external stimuli, and appropriate to new and ever changing situations.
It is not a novel insight that human speech is distinguished by these qualities, though it is an insight that must be recaptured time and time again. With each advance in our understanding of the mechanisms of language, thought, and behavior, comes a tendency to believe that we have found the key to understanding man’s apparently unique qualities of mind. These advances are real, but an honest appraisal will show, I think, that they are far from providing such a key. We do not understand, and, for all we know, we may never come to understand what makes it possible for a normal human intelligence to use language as an instrument for the free expression of thought and feeling; or, for that matter, what qualities of mind are involved in the creative acts of intelligence that are characteristic, not unique and exceptional, in a truly human existence.
I think that this is an important fact to stress, not only for linguists and psychologists whose research centers on these issues, but, even more, for those who hope to learn something useful in their own work and thinking from research into language and thought. It is particularly important that the limitations of understanding be clear to those involved in teaching, in the universities, and even more important, in the schools. There are strong pressures to make use of new educational technology and to design curriculum and teaching methods in the light of the latest scientific advances. In itself, this is not objectionable. It is important, nevertheless, to remain alert to a very real danger: that new knowledge and technique will define the nature of what is taught and how it is taught, rather than contribute to the realization of educational goals that are set on other grounds and in other terms. Let me be concrete. Technique and even technology is available for rapid and efficient inculcation of skilled behavior, in language teaching, teaching of arithmetic, and other domains. There is, consequently, a real temptation to reconstruct curriculum in the terms defined by the new technology. And it is not too difficult to invent a rationale, making use of the concepts of “controlling behavior,” enhancing skills, and so on. Nor is it difficult to construct objective tests that are sure to demonstrate the effectiveness of such methods in reaching certain goals that are incorporated in these tests. But successes of this sort will not demonstrate that an important educational goal has been achieved. They will not demonstrate that it is important to concentrate on developing skilled behavior in the student. What little we know about human intelligence would at least suggest something quite different: that by diminishing the range and complexity of materials presented to the inquiring mind, by setting behavior in fixed patterns, these methods may harm and distort the normal development of creative abilities. I do not want to dwell on the matter. I am sure that any of you will be able to find examples from your own experience. It is perfectly proper to try to exploit genuine advances in knowledge, and within some given field of study, it is inevitable, and quite proper, that research should be directed by considerations of feasibility as well as considerations of ultimate significance. It is also highly likely, if not inevitable, that considerations of feasibility and significance will lead in divergent paths. For those who wish to apply the achievements of one discipline to the problems of another, it is important to make very clear the exact nature not only of what has been achieved, but equally important, the limitations of what has been achieved.
I mentioned a moment ago that the creative aspect of normal use of language is not a new discovery. It provides one important pillar for Descartes’ theory of mind, for his study of the limits of mechanical explanation. The latter, in turn, provides one crucial element in the construction of the anti-authoritarian social and political philosophy of the Enlightenment. And, in fact, there were even some efforts to found a theory of artistic creativity on the creative aspect of normal language use. Schlegel, for example, argues that poetry has a unique position among the arts, a fact illustrated, he claims, by the use of the term “poetical” to refer to the element of creative imagination in any artistic effort, as distinct, say, from the term “musical,” which would be used metaphorically to refer to a sensual element. To explain this asymmetry, he observes that every mode of artistic expression makes use of a certain medium and that the medium of poetry – language – is unique in that language, as an expression of the human mind rather than a product of nature, is boundless in scope and is constructed on the basis of a recursive principle that permits each creation to serve as the basis for a new creative act. Hence the central position among the arts of the art forms whose medium is language.
The belief that language, with its inherent creative aspect, is a unique human possession did not go unchallenged, of course. One expositor of Cartesian philosophy, Antoine Le Grand, refers to the opinion “of some people of the East Indies, who think that Apes and Baboons, which are with them in great numbers, are imbued with understanding, and that they can speak but will not for fear they should be employed, and set to work.” If there is a more serious argument in support of the claim that human language capacity is shared with other primates, then I am unaware of it. In fact, whatever evidence we do have seems to me to support the view that the ability to acquire and use language is a species-specific human capacity, that there are very deep and restrictive principles that determine the nature of human language and are rooted in the specific character of the human mind. Obviously arguments bearing on this hypothesis cannot be definitive or conclusive, but it appears to me, nevertheless, that even in the present stage of our knowledge, the evidence is not inconsiderable.
There are any number of questions that might lead one to undertake a study of language. Personally, I am primarily intrigued by the possibility of learning something, from the study of language, that will bring to light inherent properties of the human mind. We cannot now say anything particularly informative about the normal creative use of language in itself. But I think that we are slowly coming to understand the mechanisms that make possible this creative use of language, the use of language as an instrument of free thought and expression. Speaking again from a personal point of view, to me the most interesting aspects of contemporary work in grammar are the attempts to formulate principles of organization of language which, it is proposed, are universal reflections of properties of mind; and the attempt to show that on this assumption, certain facts about particular languages can be explained. Viewed in this way, linguistics is simply a part of human psychology: the field that seeks to determine the nature of human mental capacities and to study how these capacities are put to work. Many psychologists would reject a characterization of their discipline in these terms, but this reaction seems to me to indicate a serious inadequacy in their conception of psychology, rather than a defect in the formulation itself. In any event, it seems to me that these are proper terms in which to set the goals of contemporary linguistics, and to discuss its achievements and its failings.
I think it is now possible to make some fairly definite proposals about the organization of human language and to put them to empirical test. The theory of transformational-generative grammar, as it is evolving along diverse and sometimes conflicting paths, has put forth such proposals; and there has been, in the past few years, some very productive and suggestive work that attempts to refine and reconstruct these formulations of the processes and structures that underlie human language.
The theory of grammar is concerned with the question, What is the nature of a person’s knowledge of his language, the knowledge that enables him to make use of language in the normal, creative fashion? A person who knows a language has mastered a system of rules that assigns sound and meaning in a definite way for an infinite class of possible sentences. Each language thus consists (in part) of a certain pairing of sound and meaning over an infinite domain. Of course, the person who knows the language has no consciousness of having mastered these rules or of putting them to use, nor is there any reason to suppose that this knowledge of the rules of language can be brought to consciousness. Through introspection, a person may accumulate various kinds of evidence about the sound–meaning relation determined by the rules of the language that he has mastered; there is no reason to suppose that he can go much beyond this surface level of data so as to discover, through introspection, the underlying rules and principles that determine the relation of sound and meaning. Rather, to discover these rules and principles is a typical problem of science. We have a collection of data regarding sound–meaning correspondence, the form and interpretation of linguistic expressions, in various languages. We try to determine, for each language, a system of rules that will account for such data. More deeply, we try to establish the principles that govern the formation of such systems of rules for any human language.
The system of rules that specifies the sound–meaning relation for a given language can be called the “grammar” – or, to use a more technical term, the “generative grammar” – of this language. To say that a grammar “generates” a certain set of structures is simply to say that it specifies this set in a precise way. In this sense, we may say that the grammar of a language generates an infinite set of “structural descriptions,” each structural description being an abstract object of some sort that determines a particular sound, a particular meaning, and whatever formal properties and configurations serve to mediate the relation between sound and meaning. For example, the grammar of English generates structural descriptions for the sentences I am now speaking; or, to take a simpler case for purposes of illustration, the grammar of English would generate a structural description for each of these sentences:
| 1 | John is certain that Bill will leave. |
| 2 | John is certain to leave. |
Each of us has mastered and internally represented a system of grammar that assigns structural descriptions to these sentences; we use this knowledge, totally without awareness or even the possibility of awareness, in producing these sentences or understanding them when they are produced by others. The structural descriptions include a phonetic representation of the sentences and a specification of their meaning. In the case of the cited examples 1 and 2, the structural descriptions must convey roughly the following information: they must indicate that in the case of 1, a given psychological state (namely, being certain that Bill will leave) is attributed to John; whereas in the case of 2, a given logical property (namely, the property of being certain) is attributed to the proposition that John will leave. Despite the superficial similarity of form of these two sentences, the structural descriptions generated by the grammar must indicate that their meanings are very different: one attributes a psychological state to John, the other attributes a logical property to an abstract proposition. The second sentence might be paraphrased in a very different form:
| 3 | That John will leave is certain. |
For the first there is no such paraphrase. In the paraphrase 3 the “logical form” of 2 is expressed more directly, one might say. The grammatical relations in 2 and 3, are very similar, despite the difference of surface form; the grammatical relations in 1 and 2 are very different, despite the similarity of surface form. Such facts as these provide the starting point for an investigation of the grammatical structure of English – and more generally, for the investigation of the general properties of human language.
To carry the discussion of properties of language further, let me introduce the term “surface structure” to refer to a representation of the phrases that constitute a linguistic expression and the categories to which these phrases belong. In sentence 1, the phrases of the surface structure include: “that Bill will leave,” which is a full proposition; the noun phrases “Bill” and “John”; the verb phrases “will leave” and “is certain that Bill will leave,” and so on. In sentence 2, the surface structure includes the verb phrases “to leave” and “is certain to leave”; but the surface structure of 2 includes no proposition of the form “John will leave,” even though this proposition expresses part of the meaning of “John is certain to leave,” and appears as a phrase in the surface structure of its paraphrase, “that John will leave is certain.” In this sense, surface structure does not necessarily provide an accurate indication of the structures and relations that determine the meaning of a sentence; in the case of sentence 2, “John is certain to leave,” the surface structure fails to indicate that the proposition “John will leave” expresses a part of the meaning of the sentence – although in the other two examples that I gave the surface structure comes rather close to indicating the semantically significant relations.
Continuing, let me introduce the further technical term “deep structure” to refer to a representation of the phrases that play a more central role in the semantic interpretation of a sentence. In the case of 1 and 3, the deep structure might not be very different from the surface structure. In the case of 2, the deep structure will be very different from the surface structure, in that it will include some such proposition as “John will leave” and the predicate “is certain” applied to this proposition, though nothing of the sort appears in the surface structure. In general, apart from the simplest examples, the surface structures of sentences are very different from their deep structures.
The grammar of English will generate, for each sentence, a deep structure, and will contain rules showing how this deep structure is related to a surface structure. The rules expressing the relation of deep and surface structure are called “grammatical transformations.” Hence the term “transformational-generative grammar.” In addition to rules defining deep structures, surface structures, and the relation between them, the grammar of English contains further rules that relate these “syntactic objects” (namely, paired deep and surface structures) to phonetic representations on the one hand, and to representations of meaning on the other. A person who has acquired knowledge of English has internalized these rules and makes use of them when he understands or produces the sentences just given as examples, and an indefinite range of others.
Evidence in support of this approach is provided by the observation that interesting properties of English sentences can be explained directly in terms of the deep structures assigned to them. Thus consider once again the two sentences 1 (“John is certain that Bill will leave”) and 2 (“John is certain to leave”). Recall that in the case of the first, the deep structure and surface structure are virtually identical, whereas in the case of the second, they are very different. Observe also that in the case of the first, there is a corresponding nominal phrase, namely, “John’s certainty that Bill will leave (surprised me)”; but in the case of the second, there is no corresponding nominal phrase. We cannot say “John’s certainty to leave surprised me.” The latter nominal phrase is intelligible, I suppose, but it is not well formed in English. The speaker of English can easily make himself aware of this fact, though the reason for it will very likely escape him. This fact is a special case of a very general property of English: namely, nominal phrases exist corresponding to sentences that are very close in surface form to deep structure, but not corresponding to such sentences that are remote in surface form from deep structure. Thus “John is certain that Bill will leave,” being close in surface form to its deep structure, corresponds to the nominal phrase “John’s certainty that Bill will leave”; but there is no such phrase as “John’s certainty to leave” corresponding to “John is certain to leave,” which is remote from its deep structure.
The notions of “closeness” and “remoteness” can be made quite precise. When we have made them precise, we have an explanation for the fact that nominalizations exist in certain cases but not in others – though were they to exist in these other cases, they would often be perfectly intelligible. The explanation turns on the notion of deep structure: in effect, it states that nominalizations must reflect the properties of deep structure. There are many examples that illustrate this phenomenon. What is important is the evidence it provides in support of the view that deep structures which are often quite abstract exist and play a central role in the grammatical processes that we use in producing and interpreting sentences. Such facts, then, support the hypothesis that deep structures of the sort postulated in transformational-generative grammar are real mental structures. These deep structures, along with the transformation rules that relate them to surface structure and the rules relating deep and surface structures to representations of sound and meaning, are the rules that have been mastered by the person who has learned a language. They constitute his knowledge of the language; they are put to use when he speaks and understands.
The examples I have given so far illustrate the role of deep structure in determining meaning, and show that even in very simple cases, the deep structure may be remote from the surface form. There is a great deal of evidence indicating that the phonetic form of a sentence is determined by its surface structure, by principles of an extremely interesting and intricate sort that I will not try to discuss here. From such evidence it is fair to conclude that surface structure determines phonetic form, and that the grammatical relations represented in deep structure are those that determine meaning. Furthermore, as already noted, there are certain grammatical processes, such as the process of nominalization, that can be stated only in terms of abstract deep structures.
The situation is complicated, however, by the fact that surface structure also plays a role in determining semantic interpretation.1 The study of this question is one of the most controversial aspects of current work, and, in my opinion, likely to be one of the most fruitful. As an illustration, consider some of the properties of the present perfect aspect in English – for example, such sentences as “John has lived in Princeton.” An interesting and rarely noted feature of this aspect is that in such cases it carries the presupposition that the subject is alive. Thus it is proper for me to say “I have lived in Princeton” but, knowing that Einstein is dead, I would not say “Einstein has lived in Princeton.” Rather, I would say “Einstein lived in Princeton.” (As always, there are complications, but this is accurate as a first approximation.) But now consider active and passive forms with present perfect aspect. Knowing that John is dead and Bill alive, I can say “Bill has often been visited by John, but not “John has often visited Bill”; rather, “John often visited Bill.” I can say “I have been taught physics by Einstein” but not “Einstein has taught me physics”; rather, “Einstein taught me physics.” In general, active and passive are synonymous and have essentially the same deep structures. But in these cases, active and passive forms differ in the presuppositions they express; put simply, the presupposition is that the person denoted by the surface subject is alive. In this respect, the surface structure contributes to the meaning of the sentence in that it is relevant to determining what is presupposed in the use of a sentence.
Carrying the matter further, observe that the situation is different when we have a conjoined subject. Thus given that Hilary is alive and Marco Polo dead, it is proper to say “Hilary has climbed Mt. Everest” but not “Marco Polo has climbed Mt. Everest”; rather, again, “Marco Polo climbed Mt. Everest.” (Again, I overlook certain subtleties and complications.) But now consider the sentence “Marco Polo and Hilary (among others) have climbed Mt. Everest.” In this case, there is no expressed presupposition that Marco Polo is alive, as there is none in the passive “Mt. Everest has been climbed by Marco Polo (among others).”
Notice further that the situation changes considerably when we shift from the normal intonation, as in the cases I have just given, to an intonation contour that contains a contrastive or expressive stress. The effect of such intonation on presupposition is fairly complex. Let me illustrate with a simple case. Consider the sentence “The Yankees played the Red Sox in Boston.” With normal intonation, the point of main stress and highest pitch is the word “Boston” and the sentence might be an answer to such questions as “where did the Yankees play the Red Sox?” (“in Boston”); “what did the Yankees do?” (“they played the Red Sox in Boston”); “what happened?” (“the Yankees played the Red Sox in Boston”). But suppose that contrastive stress is placed on “Red Sox,” so that we have “The Yankees played the RED SOX in Boston.” Now, the sentence can be the answer only to “Who did the Yankees play in Boston?” Note that the sentence presupposes that the Yankees played someone in Boston; if there was no game at all, it is improper, not just false, to say “The Yankees played the RED SOX in Boston.” In contrast, if there was no game at all, it is false, but not improper, to say “The Yankees played the Red Sox in Boston,” with normal intonation. Thus contrastive stress carries a presupposition in a sense in which normal intonation does not, though normal intonation also carries a presupposition in another sense; thus it would be improper to answer the question “Who played the Red Sox in Boston?” with “The Yankees played the Red Sox in Boston” (normal intonation). The same property of contrastive stress is shown by the so-called cleft sentence construction. Thus the sentence “It was the YANKEES who played the Red Sox in Boston” has primary stress on “Yankees,” and presupposes that someone played the Red Sox in Boston. The sentence is improper, not just false, if there was no game at all. These phenomena have generally been overlooked when the semantic role of contrastive stress has been noted.
To further illustrate the role of surface structure in determining meaning, consider such sentences as this: “John is tall for a pygmy.” This sentence presupposes that John is a pygmy, and that pygmies tend to be short; hence given our knowledge of the Watusi, it would be anomalous to say “John is tall for a Watusi.” On the other hand, consider what happens when we insert the word “even” in the sentence. Inserting it before “John” we derive: “Even John is tall for a pygmy.” Again, the presupposition is that John is a pygmy and that pygmies are short. But consider: “John is tall even for a pygmy.” This presupposes that pygmies are tall; it is therefore a strange sentence, given our knowledge of the facts, as compared, say, to “John is tall even for a Watusi,” which is quite all right. The point is that the position of “even” in the sentence “John is tall for a pygmy” determines the presupposition with respect to the average height of pygmies.
But the placement of the word “even” is a matter of surface structure. We can see this from the fact that the word “even” can appear in association with phrases that do not have any representation at the level of deep structure: Consider, for example, the sentence “John isn’t certain to leave at 10; in fact, he isn’t even certain to leave at all.” Here, the word “even” is associated with “certain to leave,” a phrase which, as noted earlier, does not appear at the level of deep structure. Hence in this case as well properties of surface structure play a role in determining what is presupposed by a certain sentence.
The role of surface structure in determining meaning is illustrated once again by the phenomenon of pronominalization.2 Thus if I say “Each of the men hates his brothers,” the word “his” may refer to one of the men; but if I say “The men each hate his brothers,” the word “his” must refer to some other person, not otherwise referred to in the sentence. However, the evidence is strong that “each of the men” and “the men each” derive from the same deep structure. Similarly, it has been noted that placement of stress plays an important role in determining pronominal reference. Consider the following discourse: “John washed the car; I was afraid someone ELSE would do it.” The sentence implies that I hoped that John would wash the car, and I’m happy that he did. But now consider the following: “John washed the car; I was AFRAID someone else would do it.” With stress on “afraid,” the sentence implies that I hoped that John would not wash the car. The reference of “someone else” is different in the two cases. There are many other examples that illustrate the role of surface structure in determining pronominal reference.
To complicate matters still further, deep structure too plays a role in determining pronominal reference. Thus consider the sentence “John appeared to Bill to like him.” Here, the pronoun “him” may refer to Bill but not John. Compare “John appealed to Bill to like him.” Here, the pronoun may refer to John but not Bill. Thus we can say “John appealed to Mary to like him,” but not “John appeared to Mary to like him,” where “him” refers to “John”; on the other hand, we can say “John appeared to Mary to like her,” but not “John appealed to Mary to like her,” where “her” refers to Mary. Similarly, in “John appealed to Bill to like himself,” the reflexive refers to Bill; but in “John appeared to Bill to like himself,” it refers to John. These sentences are approximately the same in surface structure; it is the differences in deep structure that determine the pronominal reference.
Hence pronominal reference depends on both deep and surface structure. A person who knows English has mastered a system of rules which make use of properties of deep and surface structure in determining pronominal reference. Again, he cannot discover these rules by introspection. In fact, these rules are still unknown, though some of their properties are clear.
To summarize: the generative grammar of a language specifies an infinite set of structural descriptions, each of which contains a deep structure, a surface structure, a phonetic representation, a semantic representation, and other formal structures. The rules relating deep and surface structure – the so-called “grammatical transformations” – have been investigated in some detail, and are fairly well understood. The rules that relate surface structure and phonetic representation are also reasonably well understood (though I do not want to imply that the matter is beyond dispute: far from it). It seems that both deep and surface structure enter into the determination of meaning. Deep structure provides the grammatical relations of predication, modification, and so on, that enter into the determination of meaning. On the other hand, it appears that matters of focus and presupposition, topic and comment, the scope of logical elements, and pronominal reference are determined, in part at least, by surface structure. The rules that relate syntactic structures to representations of meaning are not at all well understood. In fact, the notion “representation of meaning” or “semantic representation” is itself highly controversial. It is not clear at all that it is possible to distinguish sharply between the contribution of grammar to the determination of meaning, and the contribution of so-called “pragmatic considerations,” questions of fact and belief and context of utterance. It is perhaps worth mentioning that rather similar questions can be raised about the notion “phonetic representation.” Although the latter is one of the best-established and least controversial notions of linguistic theory, we can, nevertheless, raise the question whether or not it is a legitimate abstraction, whether a deeper understanding of the use of language might not show that factors that go beyond grammatical structure enter into the determination of perceptual representations and physical form in an inextricable fashion, and cannot be separated, without distortion, from the formal rules that interpret surface structure as phonetic form.
So far, the study of language has progressed on the basis of a certain abstraction: namely, we abstract away from conditions of use of language and consider formal structures and the formal operations that relate them. Among these formal structures are those of syntax, namely, deep and surface structures; and also the phonetic and semantic representations, which we take to be certain formal objects related to syntactic structures by certain well-defined operations. This process of abstraction is in no way illegitimate, but one must understand that it expresses a point of view, a hypothesis about the nature of mind, that is not a priori obvious. It expresses the working hypothesis that we can proceed with the study of “knowledge of language” – what is often called “linguistic competence” – in abstraction from the problems of how language is used. The working hypothesis is justified by the success that is achieved when it is adopted. A great deal has been learned about the mechanisms of language, and, I would say, about the nature of mind, on the basis of this hypothesis. But we must be aware that in part, at least, this approach to language is forced upon us by the fact that our concepts fail us when we try to study the use of language. We are reduced to platitudes, or to observations which, though perhaps quite interesting, do not lend themselves to systematic study by means of the intellectual tools presently available to us. On the other hand, we can bring to the study of formal structures and their relations a wealth of experience and understanding. It may be that at this point we are facing a problem of conflict between significance and feasibility, a conflict of the sort that I mentioned earlier in this paper. I do not believe that this is the case, but it is possible. I feel fairly confident that the abstraction to the study of formal mechanisms of language is appropriate; my confidence arises from the fact that many quite elegant results have been achieved on the basis of this abstraction. Still, caution is in order. It may be that the next great advance in the study of language will require the forging of new intellectual tools that permit us to bring into consideration a variety of questions that have been cast into the waste-bin of “pragmatics,” so that we could proceed to study questions that we know how to formulate in an intelligible fashion.
As noted, I think that the abstraction to linguistic competence is legitimate. To go further, I believe that the inability of modern psychology to come to grips with the problems of human intelligence is in part, at least, a result of its unwillingness to undertake the study of abstract structures and mechanisms of mind. Notice that the approach to linguistic structure that I have been outlining has a highly traditional flavor to it. I think it is no distortion to say that this approach makes precise a point of view that was inherent in the very important work of the seventeenth- and eighteenth-century universal grammarians, and that was developed, in various ways, in rationalist and romantic philosophy of language and mind. The approach deviates in many ways from a more modern, and in my opinion quite erroneous, conception that knowledge of language can be accounted for as a system of habits, or in terms of stimulus-response connections, principles of “analogy” and “generalization,” and other notions that have been explored in twentieth-century linguistics and psychology, and that develop from traditional empiricist speculation. The fatal inadequacy of all such approaches, I believe, results from their unwillingness to undertake the abstract study of linguistic competence. Had the physical sciences limited themselves by similar methodological strictures, we would still be in the era of Babylonian astronomy.
One traditional concept that has reemerged in current work is that of “universal grammar,” and I want to conclude by saying just a word about this topic. There are two kinds of evidence suggesting that deep-seated formal conditions are satisfied by the grammars of all languages. The first kind of evidence is provided by the study of a wide range of languages. In attempting to construct generative grammars for languages of widely varied kinds, investigators have repeatedly been led to rather similar assumptions as to the form and organization of such generative systems. But a more persuasive kind of evidence bearing on universal grammar is provided by the study of a single language. It may at first seem paradoxical that the intensive study of a single language should provide evidence regarding universal grammar, but a little thought about the matter shows that this is a very natural consequence.
To see this, consider the problem of determining the mental capacities that make language acquisition possible. If the study of grammar – of linguistic competence – involves an abstraction from language use, then the study of the mental capacities that make acquisition of grammar possible involves a further, second-order abstraction. I see no fault in this. We may formulate the problem of determining the intrinsic characteristics of a device of unknown properties that accepts as “input” the kind of data available to the child learning his first language, and produces as “output” the generative grammar of that language. The “output,” in this case, is the internally represented grammar, mastery of which constitutes knowledge of the language. If we undertake to study the intrinsic structure of a language-acquisition device without dogma or prejudice, we arrive at conclusions which, though of course only tentative, still seem to me both significant and reasonably well-founded. We must attribute to this device enough structure so that the grammar can be constructed within the empirically given constraints of time and available data, and we must meet the empirical condition that different speakers of the same language, with somewhat different experience and training, nevertheless acquire grammars that are remarkably similar, as we can determine from the ease with which they communicate and the correspondences among them in the interpretation of new sentences. It is immediately obvious that the data available to the child is quite limited – the number of seconds in his lifetime is trivially small as compared with the range of sentences that he can immediately understand and can produce in the appropriate manner. Having some knowledge of the characteristics of the acquired grammars and the limitations on the available data, we can formulate quite reasonable and fairly strong empirical hypotheses regarding the internal structure of the language-acquisition device that constructs the postulated grammars from the given data. When we study this question in detail, we are, I believe, led to attribute to the device a very rich system of constraints on the form of a possible grammar; otherwise, it is impossible to explain how children come to construct grammars of the kind that seem empirically adequate under the given conditions of time and access to data. But if we assume, furthermore, that children are not genetically predisposed to learn one rather than another language, then the conclusions we reach regarding the language-acquisition device are conclusions regarding universal grammar. These conclusions can be falsified by showing that they fail to account for the construction of grammars of other languages, for example. And these conclusions are further verified if they serve to explain facts about other languages. This line of argument seems to me very reasonable in a general way, and when pursued in detail it leads us to strong empirical hypotheses concerning universal grammar, even from the study of a particular language.
I have discussed an approach to the study of language that takes this study to be a branch of theoretical human psychology. Its goal is to exhibit and clarify the mental capacities that make it possible for a human to learn and use a language. As far as we know, these capacities are unique to man, and have no significant analogue in any other organism. If the conclusions of this research are anywhere near correct, then humans must be endowed with a very rich and explicit set of mental attributes that determine a specific form of language on the basis of very slight and rather degenerate data. Furthermore, they make use of the mentally represented language in a highly creative way, constrained by its rules but free to express new thoughts that relate to past experience or present sensations only in a remote and abstract fashion. If this is correct, there is no hope in the study of the “control” of human behavior by stimulus conditions, schedules of reinforcement, establishment of habit structures, patterns of behavior, and so on. Of course, one can design a restricted environment in which such control and such patterns can be demonstrated, but there is no reason to suppose that any more is learned about the range of human potentialities by such methods than would be learned by observing humans in a prison or an army – or in many a schoolroom. The essential properties of the human mind will always escape such investigation. And if I can be pardoned a final “nonprofessional” comment, I am very happy with this outcome.
1 I discuss this matter in some detail in “Deep Structure and Semantic Interpretation,” in R. Jakobson and S. Kawamoto, eds., Studies in General and Oriental Linguistics, commemorative volume for Shiro Hattori, TEC Corporation for Language and Educational Research, Tokyo, 1970.
2 The examples that follow are due to Ray Dougherty, Adrian Akmajian, and Ray Jackendoff. See my article in Jakobson and Kawamoto, eds., Studies in General and Oriental Linguistics, for references.