2 Linguistic contributions to the study of mind: present
One difficulty in the psychological sciences lies in the familiarity of the phenomena with which they deal. A certain intellectual effort is required to see how such phenomena can pose serious problems or call for intricate explanatory theories. One is inclined to take them for granted as necessary or somehow “natural.”
The effects of this familiarity of phenomena have often been discussed. Wolfgang Köhler, for example, has suggested that psychologists do not open up “entirely new territories” in the manner of the natural sciences, “simply because man was acquainted with practically all territories of mental life a long time before the founding of scientific psychology . . . because at the very beginning of their work there were no entirely unknown mental facts left which they could have discovered.”1 The most elementary discoveries of classical physics have a certain shock value – man has no intuition about elliptical orbits or the gravitational constant. But “mental facts” of even a much deeper sort cannot be “discovered” by the psychologist, because they are a matter of intuitive acquaintance and, once pointed out, are obvious.
There is also a more subtle effect. Phenomena can be so familiar that we really do not see them at all, a matter that has been much discussed by literary theorists and philosophers. For example, Viktor Shklovskij in the early 1920s developed the idea that the function of poetic art is that of “making strange” the object depicted. “People living at the seashore grow so accustomed to the murmur of the waves that they never hear it. By the same token, we scarcely ever hear the words which we utter . . . We look at each other, but we do not see each other any more. Our perception of the world has withered away; what has remained is mere recognition.” Thus, the goal of the artist is to transfer what is depicted to the “sphere of new perception”; as an example, Shklovskij cites a story by Tolstoy in which social customs and institutions are “made strange” by the device of presenting them from the viewpoint of a narrator who happens to be a horse.2
The observation that “we look at each other, but we do not see each other any more” has perhaps itself achieved the status of “words which we utter but scarcely ever hear.” But familiarity, in this case as well, should not obscure the importance of the insight.
Wittgenstein makes a similar observation, pointing out that “the aspects of things that are most important for us are hidden because of their simplicity and familiarity (one is unable to notice something – because it is always before one’s eyes).”3 He sets himself to “supplying . . . remarks on the natural history of human beings: we are not contributing curiosities however, but observations which no one has doubted, but which have escaped remark only because they are always before our eyes.”4
Less noticed is the fact that we also lose sight of the need for explanation when phenomena are too familiar and “obvious.” We tend too easily to assume that explanations must be transparent and close to the surface. The greatest defect of classical philosophy of mind, both rationalist and empiricist, seems to me to be its unquestioned assumption that the properties and content of the mind are accessible to introspection; it is surprising to see how rarely this assumption has been challenged, insofar as the organization and function of the intellectual faculties are concerned, even with the Freudian revolution. Correspondingly, the far-reaching studies of language that were carried out under the influence of Cartesian rationalism suffered from a failure to appreciate either the abstractness of those structures that are “present to the mind” when an utterance is produced or understood, or the length and complexity of the chain of operations that relate the mental structures expressing the semantic content of the utterance to the physical realization.
A similar defect mars the study of language and mind in the modern period. It seems to me that the essential weakness in the structuralist and behaviorist approaches to these topics is the faith in the shallowness of explanations, the belief that the mind must be simpler in its structure than any known physical organ and that the most primitive of assumptions must be adequate to explain whatever phenomena can be observed. Thus, it is taken for granted without argument or evidence (or is presented as true by definition) that a language is a “habit structure” or a network of associative connections, or that knowledge of language is merely a matter of “knowing how,” a skill expressible as a system of dispositions to respond. Accordingly, knowledge of language must develop slowly through repetition and training, its apparent complexity resulting from the proliferation of very simple elements rather than from deeper principles of mental organization that may be as inaccessible to introspection as the mechanisms of digestion or coordinated movement. Although there is nothing inherently unreasonable in an attempt to account for knowledge and use of language in these terms, it also has no particular plausibility or a priori justification. There is no reason to react with uneasiness or disbelief if study of the knowledge of language and use of this knowledge should lead in an entirely different direction.
I think that in order to achieve progress in the study of language and human cognitive faculties in general it is necessary first to establish “psychic distance” from the “mental facts” to which Köhler referred, and then to explore the possibilities for developing explanatory theories, whatever they may suggest with regard to the complexity and abstractness of the underlying mechanisms. We must recognize that even the most familiar phenomena require explanation and that we have no privileged access to the underlying mechanisms, no more so than in physiology or physics. Only the most preliminary and tentative hypotheses can be offered concerning the nature of language, its use, and its acquisition. As native speakers, we have a vast amount of data available to us. For just this reason it is easy to fall into the trap of believing that there is nothing to be explained, that whatever organizing principles and underlying mechanisms may exist must be “given” as the data is given. Nothing could be further from the truth, and an attempt to characterize precisely the system of rules we have mastered that enables us to understand new sentences and produce a new sentence on an appropriate occasion will quickly dispel any dogmatism on this matter. The search for explanatory theories must begin with an attempt to determine these systems of rules and to reveal the principles that govern them.
The person who has acquired knowledge of a language has internalized a system of rules that relate sound and meaning in a particular way. The linguist constructing a grammar of a language is in effect proposing a hypothesis concerning this internalized system. The linguist’s hypothesis, if presented with sufficient explicitness and precision, will have certain empirical consequences with regard to the form of utterances and their interpretations by the native speaker. Evidently, knowledge of language – the internalized system of rules – is only one of the many factors that determine how an utterance will be used or understood in a particular situation. The linguist who is trying to determine what constitutes knowledge of a language – to construct a correct grammar – is studying one fundamental factor that is involved in performance, but not the only one. This idealization must be kept in mind when one is considering the problem of confirmation of grammars on the basis of empirical evidence. There is no reason why one should not also study the interaction of several factors involved in complex mental acts and underlying actual performance, but such a study is not likely to proceed very far unless the separate factors are themselves fairly well understood.
In a good sense, the grammar proposed by the linguist is an explanatory theory; it suggests an explanation for the fact that (under the idealization mentioned) a speaker of the language in question will perceive, interpret, form, or use an utterance in certain ways and not in other ways. One can also search for explanatory theories of a deeper sort. The native speaker has acquired a grammar on the basis of very restricted and degenerate evidence; the grammar has empirical consequences that extend far beyond the evidence. At one level, the phenomena with which the grammar deals are explained by the rules of the grammar itself and the interaction of these rules. At a deeper level, these same phenomena are explained by the principles that determine the selection of the grammar on the basis of the restricted and degenerate evidence available to the person who has acquired knowledge of the language, who has constructed for himself this particular grammar. The principles that determine the form of grammar and that select a grammar of the appropriate form on the basis of certain data constitute a subject that might, following a traditional usage, be termed “universal grammar.” The study of universal grammar, so understood, is a study of the nature of human intellectual capacities. It tries to formulate the necessary and sufficient conditions that a system must meet to qualify as a potential human language, conditions that are not accidentally true of the existing human languages, but that are rather rooted in the human “language capacity,” and thus constitute the innate organization that determines what counts as linguistic experience and what knowledge of language arises on the basis of this experience. Universal grammar, then, constitutes an explanatory theory of a much deeper sort than particular grammar, although the particular grammar of a language can also be regarded as an explanatory theory.5
In practice, the linguist is always involved in the study of both universal and particular grammar. When he constructs a descriptive, particular grammar in one way rather than another on the basis of what evidence he has available, he is guided, consciously or not, by certain assumptions as to the form of grammar, and these assumptions belong to the theory of universal grammar. Conversely, his formulation of principles of universal grammar must be justified by the study of their consequences when applied in particular grammars. Thus, at several levels the linguist is involved in the construction of explanatory theories, and at each level there is a clear psychological interpretation for his theoretical and descriptive work. At the level of particular grammar, he is attempting to characterize knowledge of a language, a certain cognitive system that has been developed – unconsciously, of course – by the normal speaker–hearer. At the level of universal grammar, he is trying to establish certain general properties of human intelligence. Linguistics, so characterized, is simply the subfield of psychology that deals with these aspects of mind.
I will try to give some indication of the kind of work now in progress that aims, on the one hand, to determine the systems of rules that constitute knowledge of a language, and on the other, to reveal the principles that govern these systems. Obviously, any conclusions that can be reached today regarding particular or universal grammar must be quite tentative and restricted in their coverage. And in a brief sketch such as this only the roughest outlines can be indicated. To try to give something of the flavor of what is being done today I will concentrate on problems that are current in that they can be formulated with some clarity and studied, though they still resist solution.
As I indicated in the first lecture, I believe that the most appropriate general framework for the study of problems of language and mind is the system of ideas developed as part of the rationalist psychology of the seventeenth and eighteenth centuries, elaborated in important respects by the romantics and then largely forgotten as attention shifted to other matters. According to this traditional conception, a system of propositions expressing the meaning of a sentence is produced in the mind as the sentence is realized as a physical signal, the two being related by certain formal operations that, in current terminology, we may call grammatical transformations. Continuing with current terminology, we can thus distinguish the surface structure of the sentence, the organization into categories and phrases that is directly associated with the physical signal, from the underlying deep structure, also a system of categories and phrases, but with a more abstract character. Thus, the surface structure of the sentence “A wise man is honest” might analyze it into the subject “a wise man” and the predicate “is honest.” The deep structure, however, will be rather different. It will, in particular, extract from the complex idea that constitutes the subject of the surface structure an underlying proposition with the subject “man” and the predicate “be wise.” In fact, the deep structure, in the traditional view, is a system of two propositions, neither of which is asserted, but which interrelate in such a way as to express the meaning of the sentence “A wise man is honest.” We might represent the deep structure in this sample case by formula 1, and the surface structure by formula 2, where paired brackets are labeled to show the category of phrase that they bound. (Many details are omitted.)
| 1 |  |
| 2 |  |
An alternative and equivalent notation, widely used, expresses the labeled bracketing of 1 and 2 in tree form, as 1′ and 2′ respectively:
| 1′ |  |
| 2′ |  |
If we understand the relation “subject-of” to hold between a phrase of the category noun phrase (NP) and the sentence (S) that directly dominates it, and the relation “predicate-of” to hold between a phrase of the category verb phrase (VP) and the sentence that directly dominates it, then structures 1 and 2 (equivalently, 1′ and 2′) specify the grammatical functions of subject and predicate in the intended way. The grammatical functions of the deep structure (1) play a central role in determining the meaning of the sentence. The phrase structure indicated in 2, on the other hand, is closely related to its phonetic shape – specifically, it determines the intonation contour of the utterance represented.
Knowledge of a language involves the ability to assign deep and surface structures to an infinite range of sentences, to relate these structures appropriately, and to assign a semantic interpretation and a phonetic interpretation to the paired deep and surface structures. This outline of the nature of grammar seems to be quite accurate as a first approximation to the characterization of “knowledge of a language.”
How are the deep and surface structures related? Clearly, in the simple example given, we can form the surface structure from the deep structure by performing such operations as the following:
| 3 |
|
Applying just operations a and b, we derive the structure underlying the sentence “a man who is wise is honest,” which is one possible realization of the underlying structure (1). If, furthermore, we apply the operation c (deriving “a man wise is honest”), we must, in English, also apply the subsidiary operation d, deriving the surface structure (2), which can then be phonetically interpreted.
If this approach is correct in general, then a person who knows a specific language has control of a grammar that generates (that is, characterizes) the infinite set of potential deep structures, maps them onto associated surface structures, and determines the semantic and phonetic interpretations of these abstract objects.6 From the information now available, it seems accurate to propose that the surface structure determines the phonetic interpretation completely and that the deep structure expresses those grammatical functions that play a role in determining the semantic interpretation, although certain aspects of the surface structure may also participate in determining the meaning of the sentence in ways that I will not discuss here. A grammar of this sort will therefore define a certain infinite correlation of sound and meaning. It constitutes a first step toward explaining how a person can understand an arbitrary sentence of his language.
Even this artificially simple example serves to illustrate some properties of grammars that appear to be general. An infinite class of deep structures much like 1 can be generated by very simple rules that express a few rudimentary grammatical functions, if we assign to these rules a recursive property – in particular, one that allows them to embed structures of the form [s . . .]s within other structures. Grammatical transformations will then iterate to form, ultimately, a surface structure that may be quite remote from the underlying deep structure. The deep structure may be highly abstract; it may have no close point-by-point correlation to the phonetic realization. Knowledge of a language – “linguistic competence,” in the technical sense of this term discussed briefly in the first lecture – involves a mastery of these grammatical processes.
With just this much of a framework, we can begin to formulate some of the problems that call for analysis and explanation. One major problem is posed by the fact that the surface structure generally gives very little indication in itself of the meaning of the sentence. There are, for example, numerous sentences that are ambiguous in some way that is not indicated by the surface structure. Consider sentence 4:
| 4 | I disapprove of John’s drinking. |
This sentence can refer either to the fact of John’s drinking or to its character. The ambiguity is resolved, in different ways, in sentences 5 and 6:
| 5 | I disapprove of John’s drinking the beer. |
| 6 | I disapprove of John’s excessive drinking. |
It is clear that grammatical processes are involved. Notice that we cannot simultaneously extend 4 in both of the ways illustrated in 5 and 6; that would give us 7:
| 7 | *I disapprove of John’s excessive drinking the beer.7 |
Our internalized grammar assigns two different abstract structures to 4, one of which is related to the structure that underlies 5, the other to the structure that underlies 6. Bu it is at the level of deep structure that the distinction is represented; it is obliterated by the transformations that map the deep structures onto the surface form associated with 4.
The processes that are involved in examples 4, 5, and 6 are quite common in English. Thus, the sentence “I disapprove of John’s cooking” may imply either that I think his wife should cook or that I think he uses too much garlic, for example. Again, the ambiguity is resolved if we extend the sentence in the manner indicated in 5 and 6.
The fact that 7 is deviant requires explanation. The explanation in this case would be provided, at the level of particular grammar, by formulation of the grammatical rules that assign alternative deep structures and that in each case permit one but not the other of the extensions to 5 or 6. We would then explain the deviance of 7 and the ambiguity of 4 by attributing this system of rules to the person who knows the language, as one aspect of his knowledge. We might, of course, try to move to a deeper level of explanation, asking how it is that the person has internalized these rules instead of others that would determine a different sound–meaning correlation and a different class of generated surface structures (including, perhaps, 7). This is a problem of universal grammar, in the sense described earlier. Using the terminology of note 5, the discussion at the level of particular grammar would be one of descriptive adequacy, and at the level of universal grammar it would be one of explanatory adequacy.
Notice that the internalized rules of English grammar have still further consequences in a case like the one just discussed. There are transformations of great generality that permit or require the deletion of repeated elements, in whole or in part, under well-defined conditions. Applied to structure 8, these rules derive 9.8
| 8 | I don’t like John’s cooking any more than Bill’s cooking. |
| 9 | I don’t like John’s cooking any more than Bill’s. |
Sentence 9 is ambiguous. It can mean either that I don’t like the fact that John cooks any more than I like the fact that Bill cooks, or that I don’t like the quality of John’s cooking any more than I like the quality of Bill’s cooking.9 However, it cannot mean that I don’t like the quality of John’s cooking any more than I like the fact that Bill cooks, or conversely, with “fact” and “quality” interchanged. That is, in the underlying structure (8) we must understand the ambiguous phrases “John’s cooking” and “Bill’s cooking” in the same way if we are to be able to delete “cooking.” It seems reasonable to assume that what is involved is some general condition on the applicability of deletion operations such as the one that gives 9 from 8, a rather abstract condition that takes into account not only the structure to which the operation applies but also the history of derivation of this structure.
Other examples can be found where a similar principle seems to be at work. Thus, consider sentence 10, which is presumably derived from either 11 or 12 and is therefore ambiguous:10
| 10 | I know a taller man than Bill. |
| 11 | I know a taller man than Bill does. |
| 12 | I know a taller man than Bill is. |
It seems clear that the ambiguity of 10 is not represented in the surface structure; the deletion of “does” in 11 leaves exactly the same structure as the deletion of “is” in 12. But now consider sentence 13.
| 13 | I know a taller man than Bill, and so does John. |
This sentence, like 9, is two-ways ambiguous rather than four-ways ambiguous. It can have the meaning of either 14 or 15, but not 16 or 17:11
| 14 | I know a taller man than Bill does and John knows a taller man than Bill does. |
| 15 | I know a taller man than Bill is and John knows a taller man than Bill is. |
| 16 | I know a taller man than Bill is and John knows a taller man than Bill does. |
| 17 | I know a taller man than Bill does and John knows a taller man than Bill is. |
But now a problem arises, as we can see by considering more carefully the derivation of 13. Let us refer to the deletion operation that gives 10 from 11 as T1, and to the deletion operation that gives 10 from 12 as T2. If we apply T1 to each of the conjuncts of 14, we derive 18:
| 18 | I know a taller man than Bill and John knows a taller man than Bill. |
Application of T2 to each of the conjuncts of 15 will also yield 18. But application of T1 to one conjunct and T2 to the other conjunct in 16 will also give 18, as will the same procedure (in the opposite order) when applied to the two conjuncts of 17. Thus, 18 can be derived by application of T1 and T2 to any of the four underlying forms, 14, 15, 16, or 17. The structure of 18 itself does not indicate which of these is the underlying form; the distinction has been eliminated by the deletion operations T1 and T2. But now consider the operation T3, which derives “I saw Bill and so did John” from “I saw Bill and John saw Bill.” Applying T3 to 18, we derive 13. However, we have noted that 13 can have the interpretation of 14 or 15, but not of 16 or 17. Thus we can see that T3 can apply to 18 only if either 14 or 15, but not 16 or 17, was the structure underlying 18 in the given derivations of 18. However, this information is not represented in 18 itself, as we have just observed. Therefore, to apply T3 to 18 we must know something about the history of derivation of 18 – we must have information that would not be contained in the labeled bracketing of 18 itself. What we must know, in fact, is that the two conjuncts of 18 derive from underlying structures in which the same element was deleted.12 It appears, once again, that some general condition on applicability of deletion transformations must be involved, a principle that somehow brings into consideration the history of derivation of deleted strings, perhaps certain properties of the deep structure from which they ultimately derive.
To see how complex the problem is, consider such sentences as “John’s intelligence, which is his most remarkable quality, exceeds his prudence” or “The book, which weighs five pounds, was written by John.” Presumably, the relative pronoun in the embedded appositive clause replaces a deleted noun phrase, and the condition on deletion that we are discussing implies that this noun phrase should be identical to the antecedent noun phrase “John’s intelligence” or “the book” in the underlying structure of the appositive clause. In each case, however, it can be argued that there is a difference between the antecedent and the noun phrase of the appositive clause. Thus, in the first case, we are referring to the degree of John’s intelligence in the main clause but to the quality of his intelligence in the embedded clause; and in the second case we are referring to the book as an abstract object in the main clause but as a concrete physical object in the embedded clause; one might expect these differences to be represented in deep structure, thus contradicting the principle to which we seem to be driven by the earlier examples. I will not go on with this discussion here, but the reader will discover, if he pursues the matter, that the problem is compounded when a richer class of cases is considered.
In fact, the correct principle is unknown in such cases as these, although some of the conditions it must meet are clear. The problem posed by these examples is a quite typical one. Attention to linguistic fact reveals certain properties of sentences, relating to their sound, their meaning, their deviance, and so on. Evidently no explanation for these facts will be forthcoming so long as we restrict ourselves to vague talk about “habits” and “skills” and “dispositions to respond,” or about the formation of sentences “by analogy.” We do not have the “habit” of understanding sentences 4, 9, and 13 in a certain way: it is unlikely that the reader has ever encountered sentences closely resembling these, but he understands them in a highly specific way nevertheless. To refer to the processes involved as “analogy” is simply to give a name to what remains a mystery. To explain such phenomena we must discover the rules that relate sound and meaning in the language in question – the grammar that has been internalized by the person who knows the language – and the general principles that determine the organization and function of these rules.
The misleading and inadequate character of surface structure becomes evident as soon as even the most simple patterns are studied. Consider, for example, sentence 19 – again, an artificially simple example:
| 19 | John was persuaded to leave. |
The deep structure underlying this sentence must indicate that the subject–predicate relation holds in an underlying proposition of the form of 20 (assuming grammatical functions to be represented in the same manner suggested earlier), and that the verb–object relation holds in an underlying proposition of the form of 21:
| 20 |  |
| 21 |  |
Thus, “John” is understood to be the subject of “leave” and the object of “persuade” in 19, and these facts are properly expressed in the deep structure underlying 19 if this deep structure embodies the propositions informally represented as 20 and 21. Although the deep structure must be constituted of such propositions, if the approach loosely outlined earlier is correct, there is no trace of them in the surface structure of the utterance. The various transformations that produce 19 have thoroughly obliterated the system of grammatical relations and functions that determine the meaning of the sentence.
The point becomes still more obvious if we take note of the variety of sentences that seem superficially to resemble 19, but that differ widely in the ways they are understood and the formal operations that apply to them. Suppose that “persuaded” in 19 is replaced by one of the following words:13
| 22 | expected, hired, tired, pleased, happy, lucky, eager, certain, easy |
With “expected” replacing “persuaded,” the sentence can mean roughly that the fact of John’s leaving was expected; but it is impossible to speak of the fact of John’s leaving being persuaded. With “hired,” the sentence has an entirely different meaning, roughly that the purpose of hiring John was so that he would leave – an interpretation that becomes more natural if we replace “leave” by a phrase like “fix the roof.” When “tired” is substituted, we derive a nonsentence; it becomes a sentence if “too tired” replaces “persuaded,” the sentence now implying that John didn’t leave. The word “pleased” is still different. In this case we can have “too pleased,” implying that John didn’t leave, but we can also extend the sentence to “John was too pleased to leave to suit me,” which is impossible in the earlier cases. “Happy” is rather like “pleased,” though one might argue that the verb–object relation holds between “please” and “John.” The sentence “John was lucky to leave” is interpreted in still another way. It means, roughly, that John was lucky in that he left, an interpretation that is impossible in the earlier cases; furthermore, we can construct such sentences as “John was a lucky fellow to leave (so early),” but none of the earlier examples can replace “lucky” in such sentences. “John was eager to leave” differs from the earlier cases in that it is formally associated with such expressions as “John was eager for Bill to leave” and “John’s eagerness (for Bill) to leave.” “John was certain to leave” can be paraphrased as “it was certain that John would leave”; of the other examples, only “expected” is subject to this interpretation, but “expected” obviously differs from “certain” in numerous other respects – for example, it appears in a sentence such as “They expected John to leave.” The word “easy” is of course entirely different; in this and only this case the verb–object relation holds between “leave” and “John.”
It is clear, in short, that the surface structure is often misleading and uninformative and that our knowledge of language involves properties of a much more abstract nature, not indicated directly in the surface structure. Furthermore, even such artificially simple examples as these show how hopeless it would be to try to account for linguistic competence in terms of “habits,” “dispositions,” “knowing how,” and other concepts associated with the study of behavior, as this study has been circumscribed, quite without warrant, in recent years.
Even at the level of sound structure, there is evidence that abstract representations are formed and manipulated in the mental operations involved in language use. We have a more detailed understanding of the nature of linguistic representation and the intricate conditions on rule application in this domain than in any other. The work of the past few years on sound structure seems to me to provide substantial evidence in support of the view that the form of particular grammars is determined, in highly significant ways, by a restrictive schematism that specifies the choice of relevant phonetic properties, the kinds of rules that can relate surface structure to phonetic representation, and the conditions on organization and application of these rules. It thus relates closely to the general topics discussed in the first lecture, topics that I will take up again below in considering the question of how this restrictive, universal schematism comes to be used in language acquisition. Furthermore, these investigations of sound structure, insofar as they support the conclusion that abstract phonological structures are manipulated by tightly organized and intricate systems of rules, are relevant to the very interesting problem of developing empirically adequate models of performance. They suggest that all current approaches to problems of perception and organization of behavior suffer from a failure to attribute sufficient depth and complexity to the mental processes that must be represented in any model that attempts to come to grips with the empirical phenomena. Space does not permit a detailed development of these topics, either with respect to the matter of phonological structure or with respect to its potential significance for cognitive psychology.14 However, one simple illustrative example, which is quite typical, may give some idea of the nature of the evidence that is available and the conclusions to which it points.
Recall that the syntactic rules of the language generate an infinite set of surface structures, each of which is a labeled bracketing of a string of minimal elements, such as 2, in which we may take the minimal elements to be the items a, wise, man, is, honest. Each of these items can itself be represented as a string of segments, for example man as the string of segments /m/, /æ/, /n/. Each of these segments may be regarded in turn as a set of specified features; thus, /m/ stands for the feature complex [+ consonantal], [– vocalic], [+ nasal], and so on. The segmental constitution of an item will be given by a lexical entry – a characterization of the inherent phonetic, semantic, and syntactic properties of the items in question. The lexicon of the language is the set of such lexical entries, with, perhaps, additional structure that need not concern us here. We are concerned now only with the phonetic properties of the lexical entry.
The lexical entry of an item must specify just those properties that are idiosyncratic, that are not determined by linguistic rule. For example, the lexical entry for man must indicate that its second segment is a low front vowel, but the degree of tenseness, diphthongization, nasalization, and so on, of this vowel need not be indicated in the lexical entry, since these are a matter of general rule, in part particular to various English dialects, in part common to all English dialects, in part a matter of universal phonology. Similarly, the lexical entry for man must indicate that it has an irregular plural, with the vowel shifting from low to mid. The segments of the lexical entry are abstract in the sense that the phonological rules of the language will frequently modify and elaborate them in a variety of ways; hence there need not be, in general, a simple point-by-point correspondence between the lexical entry and the actual phonetic representation. In discussing examples, I will use phonetic symbols in the usual way, each being regarded as a complex of a certain set of features. I will use the symbol / / to enclose lexical representations, and the symbol [ ] to enclose all representations derived from lexical representations by application of phonological rules, including, in particular, the final phonetic representation derived by application of the full set of phonological rules.
Consider first such words as sign-signify, paradigm-paradigmatic, and so on. For reasons that will become clearer as we proceed, it is the derived form, in this case, that is most closely related to the underlying abstract lexical representation. Suppose, then, that we tentatively assign to the stem in these forms the lexical representation /sign/ and /pærædigm/ where the symbols have their conventional phonetic interpretation. Thus, the underlying element /sign/ is realized as phonetic [sign] before -ify. However, it is realized as phonetic [sayn] in isolation. A similar observation holds of paradigm.
The forms of sign and paradigm in isolation are determined by certain phonological rules that, operating jointly, have the effect of converting the representation /ig/ to [ay] when followed by a word-final nasal. A careful analysis of English phonology shows that this process can be broken into a sequence of steps, including the following (the second and third of which, in fact, require further analysis).
| 23 |
|
Applying these rules to underlying /sign/ in isolation, we derive first [siγn] (where [γ] is the velar continuant) by 23a; then [s
n] by 23b; and finally [sayn] by 24c.
Rules 23a and 23b are of little interest, but 23c is a part of a very general system of rules of “vowel shift” that is quite central to English phonology. There are, for example, strong reasons for supposing that the stem underlying the forms divine-divinity is /divn/, where the segment // is weakened to [i] before -ity and becomes [ay] by rule 23c in isolation. Similarly, reptile derives from underlying /reptl/, which becomes [reptayl] by 23c in isolation and [reptil] before -ian, with the same shortening of vowel that takes place in divinity; and so on, in many other cases.
Consider next such words as ignite-ignition, expedite-expeditious, and contrite-contrition. Just as reptile and divine derive, by vowel shift, from /reptl/ and /divn/, so we can derive the first member of each of these pairs from /ignt/, /expedt/, and /contrt/, respectively. The rule that applies to give the phonetic realization is 23c, a special case of the general process of vowel shift. Evidently, the second member of each pair is derived by such processes as 24 and 25:
| 24 | Vowels become nontense before -ion, -ious, -ian, -ity, and so on. |
| 25 | The segment /t/ followed by a high front vowel is realized as [ |
The first of these rules is the one that gives [divin] from /divn/ in divinity and [reptil] from /reptl/ in reptilian. Similarly, it gives [ignit] from /ignt/ in ignition, [expedit] from /expedt/ in expeditious, and [contrit] from /contt/ in contrition. There is an obvious underlying generalization, namely that a vowel becomes nontense before an unstressed vowel that is not in a word-final syllable; when properly formulated, this rule, along with vowel shift and a few others, constitutes the central portion of the English phonological system.
The second rule, 25, applies to the element /ti/ in /ignition/, /expeditious/, and /contrition/, replacing it by [
] and giving, finally, the phonetic realizations [igni
n], [ekspdis], [kntrin], after the application of the rule that reduces unstressed vowels to []. In short, the segments realized as [ayt] in ignite, expedite, and contrite are realized as [i] in ignition, expeditious, and contrition.
But now consider the words right-righteous, phonetically [rayt]-[ray
s]. The latter form appears to deviate from the regular pattern in two respects, namely in vowel quality (we would expect [i] rather than [ay], by rule 24), and in the final consonant of the stem (we would expect [] rather than [], by rule 25). If right were subject to the same processes as expedite, we would have [ris] rather than [rays] as the phonetic realization, analogous to [ekspdis]. What is the explanation for this double deviation?
Notice first that rule 25 is not quite exact; there are, in fact, other cases in which /ti/ is realized as [] rather than as [], for example question [kwesn], contrasted with direction [drekn]. A more accurate formulation of 25 would be 26:
| 26 | /t/ followed by a high front vowel is realized as [ |
Returning to the form right, we see that the final consonant would be correctly determined as [] rather than [] if in the underlying representation there were a continuant preceding it – that is, if the underlying representation were /riϕt/, where ϕ is some continuant. The continuant ϕ must, furthermore, be distinct from any of the continuants that actually appear phonetically in this position, namely the dental, labial, or palatal continuants in the unitalicized portion of wrist, rift, or wished. We may assume, then, that ϕ is the velar continuant /x/, which does not, of course, appear phonetically in English. The underlying form, then, would be /rixt/.
Consider now the derivation of right. By rule 23b, the representation /rixt/ becomes [rt]. By rule 23c, the representation / rt / becomes [rayt], which is the phonetic realization of right.
Consider next the derivation of righteous. Assuming that it has the same affix as expeditious and repetitious, we can represent it lexically as /rixtious/ (I do not concern myself here with the proper representation for -ous). Let us suppose that the ordering of the rules so far discussed is the following: 23a, 24, 26, 23b, 23c, an ordering consistent with other relevant facts of English, given certain simplifications for convenience of exposition. Rule 23a is inapplicable and rule 24 is vacuous, when applied to the underlying form /rixtious/. Turning to rule 26, we see that it gives the form [rixous]. Rule 23b now applies, giving [rous], and rule 23c gives [rayous], which becomes [rays] by reduction of unstressed vowels. Thus by rules 26 and 23, which are independently motivated, the underlying representation /rixt/ will be realized phonetically as [rayt] in isolation and as [ray] in righteous, exactly as required.
These facts strongly suggest that the underlying phonological representation must be /rixt/ (in accord with the orthography and, of course, the history). A sequence of rules that must be in the grammar for other reasons gives the alternation right-righteous. Therefore, this alternation is not at all exceptional, but rather perfectly regular. Of course, the underlying representation is quite abstract; it is connected with the superficial phonetic shape of the signal only by a sequence of interpretive rules.
Putting the matter differently, suppose that a person knows English but does not happen to have the vocabulary item righteous. Hearing this form for the first time, he must assimilate it to the system he has learned. If he were presented with the derived form [ris], he would, of course, take the underlying representation to be exactly like that of expedite, contrite, and so on. But hearing [rays], he knows that this representation is impossible; although the consonantal distinction []-[] might easily be missed under ordinary conditions of language use, the vocalic distinction [i]-[ay] would surely be obvious. Knowing the rules of English and hearing the vocalic element [ay] instead of [i], he knows that either the form is a unique exception or it contains a sequence /i/ followed by velar and is subject to rule 26. The velar must be a continuant,15 that is, /x/. But given that the velar is a continuant, it follows, if the form is regular (the null hypothesis, always), that the consonant must be [], not [], by rule 26. Thus, the hearer should perceive [rays] rather than [rays], even if the information as to the medial consonant is lacking in the received signal. Furthermore, the pressure to preserve regularity of alternations should act to block the superficial analogy to expedite-expeditious and ignite-ignition, and to preserve [] as the phonetic realization of underlying /t/, as long as [ay] appears in place of expected [i], exactly as we observe to have occurred.
I do not mean this as a literal step-by-step account of how the form is learned, of course, but rather as a possible explanation of why the form resists a superficial (and in fact incorrect) analogy and preserves its status. We can explain the perception and preservation in the grammar of the []-[] contrast in righteous-expeditious on the basis of the perceived distinction between [ay] and [i] and the knowledge of a certain system of rules. The explanation rests on the assumption that the underlying representations are quite abstract, and the evidence cited suggests that this assumption is, in fact, correct.
A single example can hardly carry much conviction. A careful investigation of sound structure, however, shows that there are a number of examples of this sort, and that, in general, highly abstract underlying structures are related to phonetic representations by a long sequence of rules, just as on the syntactic level abstract deep structures are in general related to surface structures by a long sequence of grammatical transformations. Assuming the existence of abstract mental representations and interpretive operations of this sort, we can find a surprising degree of organization underlying what appears superficially to be a chaotic arrangement of data, and in certain cases we can also explain why linguistic expressions are heard, used, and understood in certain ways. One cannot hope to determine either the underlying abstract forms or the processes that relate them to signals by introspection; there is, furthermore, no reason why one should find this consequence in any way surprising.
The explanation sketched above is at the level of particular rather than universal grammar, as this distinction was formulated earlier. That is, we have accounted for a certain phenomenon on the basis of the assumption that certain rules appear in the internalized grammar, noting that these rules are, for the most part, independently motivated. Of course, considerations of universal grammar enter into this explanation insofar as they affect the choice of grammar on the basis of data. This interpenetration is unavoidable, as noted earlier. There are cases, however, where explicit principles of universal grammar enter more directly and clearly into a pattern of explanation. Thus, investigation of sound systems reveals certain very general principles of organization, some quite remarkable, governing phonological rules (see references in note 14). For example, it has been observed that certain phonological rules operate in a cycle, in a manner determined by the surface structure. Recall that the surface structure can be represented as a labeled bracketing of the utterance, such as 2. In English, the very intricate phonological rules that determine stress contours and vowel reduction apply to phrases bounded by paired brackets, in the surface structure, applying first to a minimal phrase of this sort, then to the next larger phrase, and so on, until the maximal domain of phonological processes is reached (in simple cases, the sentence itself). Thus, in the case of 2 the rules apply to the individual words (which, in a full description, would be assigned to categories and therefore bracketed), then to the phrases a wise man and is honest, and finally to the whole sentence. A few simple rules will give quite varied results, as the surface structures that determine their cyclic application vary.
Some simple effects of the principle of cyclic application are illustrated by such forms as those of 27:
| 27 |
|
The unitalicized vowels are reduced to [] in 27b, but they retain their original quality in 27a. In some cases, we can determine the original quality of the reduced vowels of 27b from other derived forms (for example, illustrative, demonstrative). The examples of 27a differ from those of 27b morphologically in that the former are derived from underlying forms (namely, relax, emend, elastic, connective) that contain primary stress on the unitalicized vowel when these underlying forms appear in isolation; those of 27b do not have this property. It is not difficult to show that vowel reduction in English, the replacement of a vowel by [], is contingent upon lack of stress. We can therefore account for the distinction between 27a and 27b by assuming the cyclic principle just formulated. In the case of 27a, on the first, innermost cycle, stress will be assigned by general rules to the unitalicized vowels. On the next cycle, stress is shifted,16 but the abstract stress assigned on the first cycle is sufficient to protect the vowel from reduction. In the examples of 27b, earlier cycles never assign an abstract stress to the unitalicized vowel, which thus reduces. Observe that it is an abstract stress that protects the vowel from reduction. The actual, phonetic stress on the unitalicized nonreduced vowels is very weak; it would be stress 4, in the usual convention. In general, vowels with this weak a phonetic stress reduce, but in this case the abstract stress assigned in the earlier cycle prevents reduction. Thus, it is the abstract underlying representation that determines the phonetic form, a primary role being played by the abstract stress that is virtually eliminated in the phonetic form.
In this case, we can provide an explanation for a certain aspect of perception and articulation in terms of a very general abstract principle, namely the principle of cyclic application of rules (see page 38). It is difficult to imagine how the language learner might derive this principle by “induction” from the data presented to him. In fact, many of the effects of this principle relate to perception and have little or no analogue in the physical signal itself, under normal conditions of language use, so that the phenomena on which the induction would have been based cannot be part of the experience of one who is not already making use of the principle. In fact, there is no procedure of induction or association that offers any hope of leading from such data as is available to a principle of this sort (unless, begging the question, we introduce the principle of cyclic application into the “inductive procedure” in some manner). Therefore, the conclusion seems warranted that the principle of cyclic application of phonological rules is an innate organizing principle of universal grammar that is used in determining the character of linguistic experience and in constructing a grammar that constitutes the acquired knowledge of language. At the same time, this principle of universal grammar offers an explanation for such phenomena as were noted in 27.
There is some evidence that a similar principle of cyclic application applies also on the syntactic level. John Ross has presented an ingenious analysis of some aspects of English pronominalization illustrating this.17 Let us assume that pronominalization involves a process of “deletion” analogous to those processes discussed earlier in connection with examples 8– 18. This process, to first approximation, replaces one of two identical noun phrases by an appropriate pronoun. Thus, the underlying structure 28 will be converted into 29, by pronominalization.
| 28 | John learned that John had won. |
| 29 | John learned that he had won. |
Abstracting away from properties of 28 that are not essential to this discussion, we can present it in the form 30, where x and y are the identical noun phrases and y is the one pronominalized, and where the brackets bound sentential expressions.
| 30 |  |
Notice that we cannot form 31 from 28 by pronominalization:18
| 31 | He learned that John had won. |
That is, we cannot have pronominalization in the case that would be represented as 32, using the conventions of 30:
| 32 |  |
Consider next the sentences of 33:
| 33 |  |
Continuing with the same conventions, the forms are represented underneath, in each case. Summarizing, we see that of the possible types 30, 32, 33a, b and 33c, d, all permit pronominalization except 32. These remarks belong to the particular grammar of English.
Notice that alongside 33d we also have sentence 34:
| 34 | Winning the race surprised John. |
Given the framework we have been assuming throughout, 34 must be derived from the structure “John’s winning the race surprised John.” Hence, in this case pronominalization can be a full deletion.
Consider now sentences 35 and 36:
| 35 | Our learning that John had won the race surprised him. |
| 36 | Learning that John had won the race surprised him. |
Sentence 35 can be understood with “him” referring to John, but 36 cannot. Thus, 35 can be derived by pronominalization from 37, but 36 is not derived from 28:
| 37 |  |
| 38 |  |
What might be the explanation for this phenomenon? As Ross observes, it can be explained in terms of the particular grammar of English if we assume, in addition, that certain transformations apply in a cycle, first to innermost phrases, then to larger phrases, and so on – that is, if we assume that these transformations apply to the deep structure by a process analogous to the process by which phonological rules apply to the surface structure.19 Making this assumption, let us consider the underlying structure 38. On the innermost cycle, pronominalization does not apply at all, since there is no second noun phrase identical to “John” in the most deeply embedded proposition. On the second cycle, we consider the phrase “[John’s learning [that John had won the race]].” This can be regarded as a structure of the form 30, giving 39 by pronominalization; it cannot be regarded as of the form 32, giving 40 by pronominalization, because the particular grammar of English does not permit pronominalization in the case of 32, as we have noted:
| 39 | John’s learning [that he had won the race] |
| 40 | his learning [that John had won the race] |
But 40 would have to be the form underlying 36. Hence, 36 cannot be derived by pronominalization from 38, although 35 can be derived from 37.
In this case, then, a principle of universal grammar combines with an independently established rule of particular English grammar to yield a certain rather surprising empirical consequence, namely that 35 and 36 must differ in the referential interpretation of the pronoun “him.” Once again, as in the formally somewhat analogous case of vowel reduction discussed earlier in connection with examples 27a and 27b, it is quite impossible to provide an explanation in terms of “habits” and “dispositions” and “analogy.” Rather, it seems that certain abstract and in part universal principles governing human mental faculties must be postulated to explain the phenomena in question. If the principle of cyclic application is indeed a regulative principle determining the form of knowledge of language for humans, a person who has learned the particular rules governing pronominalization in English would know, intuitively and without instruction or additional evidence, that 35 and 36 differ in the respect just noted.
The most challenging theoretical problem in linguistics is that of discovering the principles of universal grammar that interweave with the rules of particular grammars to provide explanations for phenomena that appear arbitrary and chaotic. Probably the most persuasive examples at this time (and also the most important ones, in that the principles involved are highly abstract and their operations quite intricate) are in the domain of phonology, but these are too complex to present within the scope of this lecture.20 Another syntactic example that illustrates the general problem in a fairly simple way is provided by the rules for formation of wh-questions in English.21
Consider such sentences as the following:
| 41 |
|
As examples a, b, and c show, a noun phrase in any of the three italicized positions in a sentence such as “John expected Bill to meet Tom” can be questioned. The process is essentially this:
| 42 |
|
All four of these processes apply nonvacuously in the case of 41b and 41c. Sentence 41b, for example, is formed by applying wh-placement to the noun phrase “someone” in “John expected someone to meet Tom.” Application of the process of wh-inversion (42b) gives “wh-someone John expected to meet Tom.” The process of auxiliary attraction (42c) gives “wh-someone did John expect to meet Tom.” Finally, the process of phonological interpretation (42d) gives 41b. Sentence 41d illustrates the fact that these processes can extract a noun phrase that is deeply embedded in a sentence – without limit, in fact.
Of the processes listed in 42, all but auxiliary attraction apply as well in the formation of relative clauses, giving such phrases as “the man who(m) John expected to meet Tom,” and so on.
Notice, however, that there are certain restrictions on the formation of questions and relatives in this manner. Consider, for example, the sentences of 43:
| 43 |
|
Suppose that we try applying the processes of interrogative and relative formation to the italicized noun phrases in 43. We should derive the following interrogatives and relatives from 43a–43f, respectively:
| 44 |
|
Of these, only bI and bR are fully acceptable, and cases a, c, d, and e are quite impossible, although it would be quite clear what they meant, were they grammatically permissible. It is not at all obvious how the speaker of English knows this to be so. Thus, sentences 43a and 43b are synonymous, yet only 43b is subject to the processes in question. And although these processes do not apply to 43d and 43f, they can be applied, with much more acceptable results, to the very similar sentences 45a and 45b:
| 45 |
|
In some unknown way, the speaker of English devises the principles of 42 on the basis of data available to him; still more mysterious, however, is the fact that he knows under what formal conditions these principles are applicable. It can hardly be seriously maintained that every normal speaker of English has had his behavior “shaped” in the indicated ways by appropriate reinforcement. The sentences of 43, 44, and 45 are as “unfamiliar” as the vast majority of those that we encounter in daily life, yet we know intuitively, without instruction or awareness, how they are to be treated by the system of grammatical rules that we have mastered.
It seems, once again, that there is a general principle that accounts for many such facts. Notice that in 43a the italicized noun phrase is contained within another noun phrase, namely “for him to understand this lecture,” which is the subject of the sentence. In 43b, however, a rule of extraposition has placed the phrase “for him to understand this lecture” outside of the subject noun phrase, and in the resulting structure this phrase is not a noun phrase at all, so that the italicized phrase in 43b is no longer contained within a noun phrase. Suppose we were to impose on grammatical transformations the condition that no noun phrase can be extracted from within another noun phrase – more generally, that if a transformation applies to a structure of the form
![[S...[A...]A...]S](../Images/85819eqnu1.png)
for any category A, then it must be so interpreted as to apply to the maximal phrase of the type A.23 Then the processes of 42 would be blocked, as required, in cases 43a, c, d, e, and f, but not in 43b. We will return shortly to 45.
There are other examples that support a principle of this sort, which we will refer to as the A-over-A principle. Consider the sentences of 46:
| 46 |
|
Each of these is ambiguous. Thus, 46a can mean that the car in the garage was kept by John, or that the car was kept in the garage by John. In the first case, the italicized phrase is part of a noun phrase, “the car in the garage”; in the latter case it is not. Similarly, 46b can mean that the man walking toward the railroad station was seen by Mary, or that the man was seen walking toward the railroad station by Mary (or, irrelevantly to this discussion, that Mary, while walking toward the railroad station, saw the man). Again, in the first case, the italicized phrase is part of a noun phrase, “the man walking toward the railroad station”; in the latter case, it is not. But now consider the two interrogatives of 47:
| 47 |
|
Each of these is unambiguous and can have only the interpretation of the underlying sentence in which the italicized phrase is not part of another noun phrase. The same is true of the relatives formed from 46, and these facts too would be explained by the A-over-A principle. There are many similar examples.
A slightly more subtle case that might, perhaps, be explained along the same lines is provided by such sentences as 48 and 49:
| 48 | John has the best proof of that theorem. |
| 49 | What theorem does John have the best proof of? |
In its most natural interpretation, sentence 48 describes a situation in which a number of people have proofs of that theorem, and John’s is the best. The sense thus suggests that “best” modifies the nominal phrase “proof of that theorem,” which contains another nominal phrase, “that theorem.”24 The A-over-A principle would therefore imply that the phrase “that theorem” not be subject to the processes of 42. Hence, 49 would not be derived by these processes from 48. And, in fact, sentence 49 has an interpretation rather different from that of 48. Sentence 49 is appropriate to a situation in which John has proofs of a number of theorems, and the questioner is asking which of these proofs is the best. The underlying structure, whatever it may be, would associate “best” with “proof,” not with “proof of that theorem,” so that “that theorem” is not embedded within a phrase of the same type and is therefore subject to questioning (and, similarly, to relativization).
The general principle just proposed has a certain explanatory force, as such examples illustrate. If postulated as a principle of universal grammar, it can explain why the particular rules of English operate to generate certain sentences while rejecting others, and to assign sound–meaning relations in ways that appear, superficially, to violate regular analogies. Putting the matter in different terms, if we assume that the A-over-A principle is a part of the innate schematism that determines the form of knowledge of language, we can account for certain aspects of the knowledge of English possessed by speakers who obviously have not been trained and who have not even been presented with data bearing on the phenomena in question in any relevant way, so far as can be ascertained.
Further analysis of data of English reveals, not unexpectedly, that this account is oversimplified and runs up against many difficulties. Consider, for example, sentences 50 and 51:
| 50 | John thought (that) Bill had read the book. |
| 51 | John wondered why Bill had read the book. |
In the case of 50, the italicized phrase is subject to interrogation and relativization, but not in the case of 51. It is unclear whether the phrases “that Bill had read the book” and “why Bill had read the book” are noun phrases. Suppose that they are not. Then sentence 50 is handled in accordance with the A-over-A principle, but not 51. To explain the blocking of the processes of 42 in the case of 51, we would have to assign the phrase “why Bill had read the book” to the same category as “the book.” In fact, there is a natural suggestion along these lines. Sentence 51 is typical in that the phrase from which the noun phrase is to be extracted is itself a wh- phrase, rather than a that- phrase. Suppose that the process of wh- placement (42a) assigns the element wh- not only to “the book” in 51 but also to the proposition containing it. Thus, both “wh- the book” and “why Bill had read the book” belong to the category wh-, which would now be regarded as a syntactic feature of a sort discussed in my Aspects of the Theory of Syntax, Chapter 2 (see note 6). Under these assumptions, the A-over-A principle will serve to explain the difference between 50 and 51.
Suppose that the phrases in question are noun phrases. Now it is 50, not 51, that poses the problem. Assuming that our analysis is correct so far, there must be some rule that assigns to the proposition “that Bill had read the book” a property of “transparency” that permits noun phrases to be extracted from it even though it is a noun phrase. There are, in fact, other examples that suggest the necessity for such a rule, presumably a rule of the particular grammar of English. Thus, consider sentences 52, 53, and 54:
| 52 | Who would you approve of my seeing? |
| 53 | What would you approve of John’s drinking? |
| 54 | *What would you approve of John’s excessive drinking of? |
Sentences 52 and 53 are formed by applying the processes of interrogation to a noun phrase contained in the larger phrases “my seeing –,” “John’s drinking –.” Hence, these larger noun phrases are transparent to the extraction operation. However, as 54 indicates, the italicized noun phrase in 55 is not transparent to this operation:
| 55 | You would approve of John’s excessive drinking of the beer. |
These examples are typical of many that suggest what the rule might be that assigns transparency. Earlier we discussed sentence 56 (sentence 4), pointing out that it is ambiguous:
| 56 | I disapprove of John’s drinking. |
Under one interpretation, the phrase “John’s drinking” has the internal structure of a noun phrase. Thus, the rule that inserts adjectives (3d) between a determiner and a noun applies, giving “John’s excessive drinking”; and, in fact, other determiners may replace “John’s” – “the,” “that,” “much of that,” and so on. Under this interpretation, the phrase “John’s drinking” behaves exactly like “John’s refusal to leave,” “John’s rejection of the offer,” and so on. Under the other interpretation, “John’s drinking (the beer)” does not have the internal structure of a noun phrase and is handled analogously to “John’s having read the book,” “John’s refusing to leave,” “John’s rejecting the offer,” and so on, none of which permits adjective insertion or replacement of “John’s” by other determiners. Suppose that we postulate a rule of English grammar that assigns transparency, in the sense just defined, to noun phrases that are also propositions lacking the internal structure of noun phrases. Thus, the phrases “that Bill had read the book” in 50, “my seeing –” in the structure underlying 52, and “John’s drinking –” in the structure underlying 53 would be assigned transparency; more precisely, the dominant noun phrase in these examples would not serve to block extraction by the A-over-A principle. In sentence 51, extraction would still be blocked by the category wh-, along the lines indicated earlier. And sentence 54 is ruled out because the relevant noun phrase of the underlying structure, “John’s excessive drinking of –,” does have the internal structure of a noun phrase, as just noted, and therefore is not subject to the special rule of English grammar that assigns transparency to the category NP when this category dominates a proposition that lacks the internal structure of an NP.
There are a few other cases that suggest the need for rules of particular grammar assigning transparency in this sense. Consider sentences 57 and 58:
| 57 |
|
| 58 |
|
The italicized noun phrases in 57 are not subject to the processes of interrogation and relativization, in accordance with the A-over-A principle, as we have already noted. In the case of 58, interrogation and relativization seem much more natural in these positions, at least in informal spoken English. Thus, the noun phrases containing the italicized phrases must be assigned transparency. It seems that what is involved is indefiniteness of the dominating noun phrase; if so, then for certain dialects there is a rule assigning transparency to a noun phrase of the form
| 59 |  |
There remain a number of very serious problems that seem to resist solution by such extensions and modifications of the A-over-A principle. Notice that this principle is formulated in a way that is not really well supported by the examples so far given. If the A-over-A principle were true in general, we would expect to find cases in which a phrase of category A cannot be extracted from a larger phrase of category A, for various choices of A. In fact, the examples given so far involve only A = noun phrase (or, perhaps, A = [+ wh-], as in the discussion of 51). Hence, an alternative formulation of the principle consistent with the facts just noted would assign nontransparency as an ad hoc property of certain types of noun phrases (and perhaps other constructions), rather than as a property of a category A dominating another category of the type A. Given just the facts so far presented, it would be proper to postulate the A-over-A principle instead of this alternative precisely because the A-over-A principle has a certain naturalness, whereas the alternative is entirely ad hoc, a listing of nontransparent structures. But there is crucial evidence, pointed out by John Ross (see reference in note 21), suggesting that the A-over-A principle is not correct. Ross points out that in the constructions from which noun phrases cannot be extracted, adjectives also cannot be extracted. Thus, consider the contexts “I believe that John saw –,” “I believe the claim that John saw –,” and “I wonder whether John saw –.” From the first of these, but not the second or third, we can extract a noun phrase in interrogation or relativization, a fact that we have been attempting to account for by modifications of the A-over-A principle. But the same is true of extraction of adjectives. Thus we can form “handsome though I believe that John is,” but not *“handsome though I believe the claim that John is,” *“handsome though I wonder whether John is,” etc. Whether one can extend the approach just discussed to account for this problem in some natural way, I do not know; at the moment, I see no approach that does not involve a perfectly ad hoc step. Perhaps this indicates that the approach through the A-over-A principle is incorrect, leaving us for the moment with only a collection of constructions in which extraction is, for some reason, impossible to accomplish.
Whatever the answer will prove to be, the complex of problems just discussed is a typical and important illustration of the kind of topic that is at the border of research today, in the sense mentioned at the outset of this lecture: that is, certain problems can be formulated clearly within a framework of ideas that is reasonably clear and well understood; certain partial solutions can be advanced; and a range of examples can be discovered where these solutions fail, leaving open for the time being the question whether what is needed is further elaboration and sharpening or a radically different approach.
I have so far discussed several kinds of conditions that transformations must meet: conditions of deletion, of the sort brought out by examples 8–18; the principle of cyclic application, illustrated by the discussion of examples 28–40 (with the phonological analogue discussed in connection with 27); and the A-over-A principle that was proposed as the basis for an explanation of such phenomena as are illustrated by examples 44–58. In each case, there is some reason to believe that the principle is appropriate, though there is no lack of evidence showing that the principle is inadequately formulated or, perhaps, misconceived. As a final illustration of this state of affairs, typical of the borderline of research that exists in linguistics as in any other field, consider a problem first discussed by Peter Rosenbaum (see reference in note 6). Consider the sentences of 60:
| 60 |
|
In interpreting these sentences, we supply a “missing subject” for the verbs “go,” “leave,” and “find,” respectively. In 60a, we understand the subject of “go” to be “John”; in 60b, we understand the subject of “leave” to be “Bill”; in 60c, we understand the subject of “find” and the subject of “wonder” to be “Bill.” In terms of the framework presupposed so far, it would be natural (though perhaps not necessary, as we will see below) to regard this missing subject as the actual subject in the deep structure, eliminated by a deletion operation. Thus, the underlying deep structures might be something like 61:
| 61 |
|
On the other hand, the facts indicate clearly that the sentences of 60 cannot derive from, say, 62:
| 62 |
|
It would be difficult to argue that in such cases there is an intrinsic semantic consideration ruling against such structures as 62. For example, one might interpret 62a as meaning that John agreed that someone should go; 62b as meaning that John persuaded Bill that he (John) would (should) leave; 62c as meaning that John’s finding Tom there caused Bill to wonder about John. There must be some general syntactic principle that rules against 62 as possible sources for 60 and that causes us to interpret 60 as based rather on 61. Rosenbaum suggests that what is involved is a certain condition on deletion operations, an “erasure principle” that prescribes roughly that the subject of an embedded proposition is deleted by the nearest noun phrase outside of this proposition, “nearness” being measured in terms of the number of branches in a representation such as 1′ or 2′.25 As he shows, a great many examples of varied sorts can be explained on this general assumption, which, like the others that I have been reviewing, involves a condition on transformations that would constitute part of universal grammar.
Here too, however, certain problems arise. Consider, for example, the following cases:26
| 63 | John promised Bill to leave. |
| 64 |
|
| 65 |
|
| 66 | John asked Bill for permission to leave. |
| 67 |
|
| 68 | John made an offer to Bill (received advice from Bill, received an invitation from Bill) to stay. |
| 69 | John helped Bill write the book. |
Sentence 63 violates the principle, since it is John, not Bill, who is to leave. In 64a, “John” is understood to be the subject of “work,” whereas in the apparently analogous sentence 64b the subject is understood to be “I.” In the case of 65a, it is “John” that is the understood subject of “wear”; in 65b, it is “I.” In the case of 66, “John” is the understood subject of “leave” and “Bill” of “permit,” underlying “permission,” presumably; in the case of 67a, “Bill” is the understood surface subject of the embedded proposition, but in 67b and 67c it is “John,” although “Bill” is the “nearest” noun phrase in all three cases, in Rosenbaum’s sense. In 68, it is “John” that is understood as the subject of “stay,” in apparent contradiction to the principle, though much depends on unresolved questions as to how these sentences are to be analyzed. The case of 69 is obscure in other ways. The erasure principle would suggest that “Bill” is the subject of “write,” although of course the sentence does not imply that Bill wrote the book – rather John and Bill did, together. But there is a difficulty in pursuing this interpretation. Thus, from 69 we can conclude that John helped write the book, but from the apparently analogous sentence “John helped the cat have kittens,” we cannot deduce that “John helped have kittens,” which is deviant, a fact that suggests that somehow there must be a grammatical relation between “John” and “write” in 69. To put it differently, the problem is how to account for “John helped write the book” as analogous to 60a, since obviously the analogue to 61a won’t do as a source.
Without pursuing the matter any further, we can see that although the erasure principle has much to recommend it and is probably somehow involved in the correct solution to this network of problems, there is much evidence still to be accounted for. As in the other cases mentioned, there are a variety of problems relating to the conditions that determine applicability of transformations, problems that still resist any near-definitive solution, though some interesting and illuminating proposals can be made that seem to go part of the way toward a general solution.
In discussing the nature of grammatical operations, I have restricted myself to syntactic and phonological examples, avoiding questions of semantic interpretation. If a grammar is to characterize the full linguistic competence of the speaker–hearer, it must comprise rules of semantic interpretation as well, but little is known of any depth regarding this aspect of grammar. In the references cited earlier (see note 6), it is proposed that a grammar consists of a syntactic component that specifies an infinite set of paired deep and surface structures and expresses the transformational relationship between these paired elements, a phonological component that assigns a phonetic representation to the surface structure, and a semantic component that assigns a semantic representation to the deep structure. As noted earlier (p. 27; see also pp. 94–97), I think there is strong evidence that aspects of the surface structure are also relevant to semantic interpretation.27 However this may be, there can be little doubt that a full grammar must contain fairly intricate rules of semantic interpretation, keyed, at least in part, to fairly specific properties of the lexical items and formal structures of the language in question. To mention just one example, consider sentence 70:
| 70 | John has lived in Princeton. |
From the assumption that this sentence has been properly used to make a statement, we can conclude that John is a person (one would not say that his dog has lived in Princeton); that Princeton is a place meeting certain physical and sociological conditions (given that “Princeton” is a proper noun); that John is now alive (I can say that I have lived in Princeton, but I cannot now say “Einstein has lived in Princeton” – rather, “Einstein lived in Princeton”); and so on. The semantic interpretation of 70 must be such as to account for these facts.
In part, such questions as these might be subsumed under a still-to-be-developed universal semantics, in which concepts and their relations are analyzed in a very general way; to take a classical example, it might be argued that the relation of meaning between “John is proud of what Bill did” and “John has some responsibility for Bill’s actions” should be explained in terms of the universal concepts of pride and responsibility, just as on the level of sound structure one might appeal to a principle of universal phonetics to account for the fact that when a velar consonant becomes palatal it ordinarily becomes strident (see references in note 14, for discussion). The proposal looks less attractive when applied to the case of 70, for example, with respect to the fact that proper use of 70 implies that John is now alive. When we try to pursue such questions, we soon become lost in a tangle of confused issues and murky problems, and it is difficult to propose answers that carry any conviction. For this reason, I am unable to discuss conditions on rules of semantic interpretation that might be analogous to the conditions on syntactic and phonological rules mentioned earlier.
Observe that I might well have been mistaken in the preceding remarks in assuming that the topics discussed belong to syntax rather than to the semantic component of a grammar, or to some domain in which semantic and syntactic rules interpenetrate. The issues are too clouded for us to be able to say that this is an empirical question, as matters now stand; but when they are sharpened, we may find that an empirical question can be posed. Consider, for example, the discussion of the erasure principle in syntax. Joseph Emonds has suggested (in unpublished work) that it is incorrect to assume, as I did, that the sentences of 60 are interpreted through reference to the underlying structures of 61. Rather, he argues that what I took to be the embedded proposition has no subject at all in the underlying form generated by the syntactic component, and a general rule of semantic interpretation takes the place of Rosenbaum’s erasure principle. Whether this is correct I do not know, but it is certainly a possibility. We can expect, as research continues into problems of grammar, that the boundaries that seem clear today may shift in unpredictable ways, or that some new basis for the organization of grammar may replace the framework that now seems appropriate.
The conditions on grammatical rules that I have been discussing are complex and only partially understood. It should be emphasized, however, that even some of the simplest and clearest conditions of the form of grammar are in no sense necessary properties of a system that fulfills the functions of human language. Correspondingly, the fact that they hold true of human languages in general and play a role in the acquired linguistic competence of the speaker–hearer cannot be lightly dismissed. Consider, for example, the simple fact that grammatical transformations are invariably structure-dependent in the sense that they apply to a string of words28 by virtue of the organization of these words into phrases. It is easy to imagine structure-independent operations that apply to a string of elements quite independently of its abstract structure as a system of phrases. For example, the rule that forms the interrogatives of 71 from the corresponding declaratives of 72 (see note 10) is a structure-dependent rule interchanging a noun phrase with the first element of the auxiliary.
| 71 |
|
| 72 |
|
In contrast, consider the operation that inverts the first and last words of a sentence, or that arranges the words of a sentence in increasing length in terms of phonetic segments (“alphabetizing” in some specified way for items of the same length), or that moves the left-most occurrence of the word “will” to the extreme left – call these O1, O2, and O3, respectively. Applying O1 to 72a, we derive 73a; applying O2 to 72b, we derive 73b; applying O3 to 72c, we derive 73c:
| 73 |
|
The operations O1, O2, and O3 are structure-independent. Innumerable other operations of this sort can be specified.
There is no a priori reason why human language should make use exclusively of structure-dependent operations, such as English interrogation, instead of structure-independent operations, such as O1, O2, and O3. One can hardly argue that the latter are more “complex” in some absolute sense; nor can they be shown to be more productive of ambiguity or more harmful to communicative efficiency. Yet no human language contains structure-independent operations among (or replacing) the structure-dependent grammatical transformations. The language-learner knows that the operation that gives 71 is a possible candidate for a grammar, whereas O1, O2, and O3, and any operations like them, need not be considered as tentative hypotheses.
If we establish the proper “psychic distance” from such elementary and commonplace phenomena as these, we will see that they really pose some nontrivial problems for human psychology. We can speculate about the reason for the reliance on structure-dependent operations,29 but we must recognize that any such speculation must involve assumptions regarding human cognitive capacities that are by no means obvious or necessary. And it is difficult to avoid the conclusion that whatever its function may be, the reliance on structure-dependent operations must be predetermined for the language-learner by a restrictive initial schematism of some sort that directs his attempts to acquire linguistic competence. Similar conclusions seem to me warranted, a fortiori, in the case of the deeper and more intricate principles discussed earlier, whatever their exact form may turn out to be.
To summarize: along the lines that have been outlined here, we might develop on the one hand a system of general principles of universal grammar,30 and on the other, particular grammars that are formed and interpreted in accordance with these principles. The interplay of universal principles and particular rules leads to empirical consequences such as those we have illustrated; at various levels of depth, these rules and principles provide explanations for facts about linguistic competence – the knowledge of language possessed by each normal speaker – and about some of the ways in which this knowledge is put to use in the performance of the speaker or hearer.
The principles of universal grammar provide a highly restrictive schema to which any human language must conform, as well as specific conditions determining how the grammar of any such language can be used. It is easy to imagine alternatives to the conditions that have been formulated (or those that are often tacitly assumed). These conditions have in the past generally escaped notice, and we know very little about them today. If we manage to establish the appropriate “psychic distance” from the relevant phenomena and succeed in “making them strange” to ourselves, we see at once that they pose very serious problems that cannot be talked or defined out of existence. Careful consideration of such problems as those sketched here indicates that to account for the normal use of language we must attribute to the speaker–hearer an intricate system of rules that involve mental operations of a very abstract nature, applying to representations that are quite remote from the physical signal. We observe, furthermore, that knowledge of language is acquired on the basis of degenerate and restricted data and that it is to a large extent independent of intelligence and of wide variations in individual experience.
If a scientist were faced with the problem of determining the nature of a device of unknown properties that operates on data of the sort available to a child and gives as “output” (that is, as a “final state of the device,” in this case) a particular grammar of the sort that it seems necessary to attribute to the person who knows the language, he would naturally search for inherent principles of organization that determine the form of the output on the basis of the limited data available. There is no reason to adopt a more prejudiced or dogmatic view when the device of unknown properties is the human mind; specifically, there is no reason to suppose, in advance of any argument, that the general empiricist assumptions that have dominated speculation about these matters have any particular privileged claim. No one has succeeded in showing why the highly specific empiricist assumptions about how knowledge is acquired should be taken seriously. They appear to offer no way to describe or account for the most characteristic and normal constructions of human intelligence, such as linguistic competence. On the other hand, certain highly specific assumptions about particular and universal grammar give some hope of accounting for the phenomena that we face when we consider knowledge and use of language. Speculating about the future, it seems not unlikely that continued research along the lines indicated here will bring to light a highly restrictive schematism that determines both the content of experience and the nature of the knowledge that arises from it, thus vindicating and elaborating some traditional thinking about problems of language and mind. It is to this matter, among others, that I shall turn in the final lecture.
1 W. Köhler, Dynamics in Psychology (New York: Liveright, 1940).
2 See V. Ehrlich, Russian Formalism, 2nd rev. edn. (New York: Humanities, 1965), pp. 176–77.
3 Ludwig Wittgenstein, Philosophical Investigations (New York: Oxford University Press, 1953), Section 129.
5 To bring out this difference in depth of explanation, I have suggested in my Current Issues in Linguistic Theory (New York: Humanities, 1965) that the term “level of descriptive adequacy” might be used for the study of the relation between grammars and data and the term “level of explanatory adequacy” for the relation between a theory of universal grammar and these data.
6 For a detailed development of this point of view, see J. Katz and P. Postal, An Integrated Theory of Linguistic Descriptions (Cambridge, Mass.: MIT Press, 1964) and my Aspects of the Theory of Syntax (Cambridge, Mass.: MIT Press, 1965). See also Peter S. Rosenbaum, The Grammar of English Predicate Complement Constructions (Cambridge, Mass.: MIT Press, 1967). These contain references to earlier work that they extend and modify. There has been a great deal of work in the past few years extending and modifying this general approach still further and exploring alternatives. At present the field is in considerable ferment, and it will probably be some time before the dust begins to settle and a number of outstanding issues are even tentatively resolved. Current work is too extensive for detailed reference to be called for in a sketch such as this. Some idea of its scope and general directions can be obtained from collections such as R. Jacobs and P. S. Rosenbaum, eds., Readings in English Transformational Grammar (Waltham, Mass.: Ginn and Company, 1970).
7 I use the asterisk in the conventional way, to indicate a sentence that deviates in some respect from grammatical rule.
8 Henceforth I shall generally delete brackets in giving a deep, surface, or intermediate structure, where this will not lead to confusion. One should think of 8 and 9 as each having a full labeled bracketing associated with it. Notice that 8 is not, of course, a deep structure, but rather the result of applying transformations to a more primitive abstract object.
9 There may also be other interpretations, based on other ambiguities in the structure “John’s cooking” – specifically, the cannibalistic interpretation and the interpretation of “cooking” as “that which is cooked.”
10 I should emphasize that when I speak of a sentence as derived by transformation from another sentence, I am speaking loosely and inaccurately. What I should say is that the structure associated with the first sentence is derived from the structure underlying the second. Thus, in the case now being discussed, it is the surface structure of 10 that is derived, on one analysis, from the abstract structure which, were it to undergo a different transformational development, would be converted into the surface structure of 11. That sentences are not derived from other sentences but rather from the structures underlying them has been explicitly assumed since the earliest work in transformational generative grammar about fifteen years ago, but informal statements such as those in the text here have misled many readers and have led to a good deal of confusion in the literature. Adding to the confusion, perhaps, is the fact that a very different theory of transformational relations developed by Zellig Harris, Henry Hiz, and others, does in fact regard the transformational operations as applied to sentences. See, for example, Z. S. Harris, “Co-occurrence and Transformation in Linguistic Structure,” in Language, Vol. 33, No. 3, 1957, pp. 283-340, and many later publications. For me, and most other speakers, sentence 12 is deviant. Nevertheless, the association structure that underlies 10 under one analysis must be postulated, perhaps deriving from the structure associated with “I know a man who is taller than Bill is.”
11 It also cannot have the meaning “I know a taller man than Bill and John likes ice cream.” Hence, if deep structure determines meaning (insofar as grammatical relations are involved), it must be that something like 14 or 15 is the immediately underlying structure for 13. It is a general property of deletion operations that some sort of recoverability is involved, a nontrivial matter with interesting empirical consequences. For some discussion, see my Current Issues, Section 2.2, and Aspects, Section 4.2.2. The problem posed by such examples as 9 and 13 was pointed out to me by John Ross. The first reference to the possibility that history of derivation may play a role in determining applicability of transformations appears in R. B. Lees, The Grammar of English Nominalizations (New York: Humanities, 1960), p. 76, in connection with his discussion – also the first – of the problem of identity of constituent structure as a factor in determining applicability of transformations.
12 If 18 itself is only two-ways ambiguous, a problem in fact arises at an even earlier point. The unnaturalness of 18 makes it difficult to determine this with any confidence.
13 See R. B. Lees, “A Multiply Ambiguous Adjectival Construction in English,” in Language, Vol. 36, No. 2, 1960, pp. 207-21, for a discussion of such structures.
14 For discussion of these topics, see my article “Some General Properties of Phonological Rules,” in Language, Vol. 47, No. 1, 1967. For a much fuller and more detailed discussion of phonological theory and its application to English, with examples drawn from many languages and some discussion of the history of the English sound system as well, see N. Chomsky and M. Halle, The Sound Pattern of English (New York: Harper & Row, 1968). The example in the text is discussed in detail, in the context of a more general framework of rules and principles, in Chapter 4, Section 4, of The Sound Pattern of English. See P. Postal, Aspects of Phonological Theory (New York: Harper & Row, 1968), for a general development of many related topics, along with a critical analysis of alternative approaches to the study of sound structure.
15 If it were a noncontinuant, it would have to be unvoiced, that is, /k/, since there are no voiced–voiceless consonant clusters in final position, by general rule. But it cannot be /k/, since /k/ remains in this position (for example, “direct,” “evict,” and so on).
16 In “connectivity,” it is on the third cycle that the stress is shifted. The second cycle merely reassigns stress to the same syllable that is stressed on the first cycle.
17 J. Ross, “On the Cyclic Nature of English Pronominalization,” in To Honor Roman Jakobson (New York: Humanities, 1967).
18 Of course, 31 is a sentence, but “he” in the sentence does not refer to John as it does in 29. Thus, 31 is not formed by pronominalization if the two occurrences of “John” are intended to be different in reference. We exclude this case from discussion here. For some remarks bearing on this problem, see my Aspects, pp. 144–47.
19 That transformational rules may be supposed to function in this way, itself a nontrivial fact if true, is suggested in my Aspects, Chapter 3. Ross’s observation suggests that this principle of application is not only possible but also necessary. Other interesting arguments to this effect are presented in R. Jacobs and P. S. Rosenbaum, eds., Readings in English Transformational Grammar, Chapter 28. The matter is far from settled. In general, understanding of syntactic structure is much more limited than that of phonological structure, descriptions are much more rudimentary, and, correspondingly, principles of universal syntax are much less firmly established than principles of universal phonology, though the latter, needless to say, must also be regarded as tentative. In part, this may be due to the inherent complexity of the subject matter. In part, it results from the fact that universal phonetics, which provides a kind of “empirical control” for phonological theory, is much more firmly grounded than universal semantics, which should, in principle, provide a partially analogous control for syntactic theory. In modern linguistics, phonetics (and, in part, phonology) has been studied in considerable depth and with much success, but the same cannot be said as yet for semantics, despite much interesting work.
20 See references in note 14. The issue is discussed in a general way in my “Explanatory Models in Linguistics,” in E. Nagel, P. Suppes, and A. Tarski, eds., Logic, Methodology, and Philosophy of Science (Stanford, Calif.: Stanford University Press, 1962); in my Current Issues, Section 2; in my Aspects, Chapter 1; and in other publications referred to in these references.
21 This matter is discussed in my Current Issues. There are several versions of this monograph. The first, presented at the International Congress of Linguistics, 1962, appears in the Proceedings of the Congress with the title of the session at which it was presented, “Logical Basis of Linguistic Theory,” ed. H. Lunt (New York: Humanities, 1964); a second appears in J. Fodor and J. J. Katz, eds., Structure of Language: Readings in the Philosophy of Language (Englewood Cliffs, N.J.: Prentice-Hall, 1964); the third, as a separate monograph (New York: Humanities, 1965). These versions differ in the treatment of the examples discussed here; none of the treatments is satisfactory, and the general problem remains open. New and interesting ideas on this matter are presented in J. Ross, “Constraints on Variables in Syntax,” MIT doctoral thesis (unpublished). I follow here the general lines of the earliest of the three versions of Current Issues, which, in retrospect, seems to me to offer the most promising approach of the three.
22 Actually, it seems that only indefinite singular noun phrases can be questioned (that is, “someone,” “something,” and so on), a fact that relates to the matter of recoverability of deletion mentioned in note 11. See my Current Issues for some discussion.
23 We might extend this principle to the effect that this transformation must also apply to the minimal phrase of the type S (sentence). Thus, the sentence 
can be transformed to “John was convinced that before dark Bill would leave” but not to “before dark John was convinced that Bill would leave,” which must have a different source. Like the original principle, this extension is not without its problems, but it has a certain amount of support nevertheless.
24 Space does not permit a discussion of the distinction implied here in the loose terminology, “noun phrase”–“nominal phrase,” but this is not crucial to the point at issue. See my “Remarks on Nominalization,” in R. Jacobs and P. S. Rosenbaum, eds., Readings in English Transformational Grammar. There are other interpretations of 49 (for example, with contrastive stress on “John”), and there are many open problems relating to such structures as these.
25 In yet-unpublished work, David Perlmutter has presented a strong argument that what is involved is not a condition on transformations but rather a condition on well-formed deep structures. The distinction is not crucial for what follows but would become important at a less superficial level of discussion.
26 Examples 63 and 67 are discussed by Rosenbaum; 64 was pointed out by Maurice Gross; 65 was pointed out in a different connection by Zeno Vendler, “Nominalizations,” in Transformations and Discourse Analysis Papers, No. 55 (Philadelphia: University of Pennsylvania, 1964), p. 67.
27 For some remarks concerning this problem, see my “Surface Structure and Semantic Interpretation,” in R. Jakobson, ed., Studies in General and Oriental Linguistics (Tokyo: TEC Corporation for Language and Educational Research, 1970). Literature on semantic interpretation of syntactic structures is expanding fairly rapidly. For recent discussion, see J. J. Katz, The Philosophy of Language (New York: Harper & Row, 1966); U. Weinreich, “Explorations in Semantic Theory,” in T. A. Sebeok, ed., Current Trends in Linguistics, Vol. III (New York: Humanities, 1966); J. J. Katz, “Recent Issues in Semantic Theory,” in Foundations of Language, Vol. 3, No. 2, May 1967, pp. 124–94; and many other papers.
28 More properly, to a string of minimal linguistic units that may or may not be words.
29 See G. A. Miller and N. Chomsky, “Finitary Models of Language Users, Part II,” in R. D. Luce, R. Bush, and E. Galanter, eds., Handbook of Mathematical Psychology, Vol. II (New York: Wiley, 1963), for some proposals regarding this matter.
30 Notice that we are interpreting “universal grammar” as a system of conditions on grammars. It may involve a skeletal substructure of rules that any human language must contain, but it also incorporates conditions that must be met by such grammars and principles that determine how they are interpreted. This formulation is something of a departure from a traditional view that took universal grammar to be simply a substructure of each particular grammar, a system of rules at the very core of each grammar. This traditional view has also received expression in recent work. It seems to me to have little merit. As far as information is available, there are heavy constraints on the form and interpretation of grammar at all levels, from the deep structures of syntax, through the transformational component, to the rules that interpret syntactic structures semantically and phonetically.