The advent of transformational generative grammar
Chomsky’s formal introduction to linguistics began in 1947, when he read the proofs of Harris’s Methods of Structural Linguistics (1951; henceforth Methods). As Harris described the book in the introduction:
This volume presents methods of research used in descriptive, or more exactly, structural, linguistics. It is thus a discussion of the operations which the linguist may carry out in the course of his investigations, rather than a theory of the structural analyses which result from these investigations. The research methods are arranged here in the form of the successive procedures of analysis imposed by the working linguist upon his data.
The operations that Harris details in Methods are essentially taxonomic procedures of segmentation and classification.
The basic operations are those of segmentation and classification. Segmentation is carried out at limits determined by the independence of the resulting segments in terms of some particular criterion. If X has a limited distribution in respect to Y, or if the occurrence of X depends upon (correlates completely with) the occurrence of a particular environment Z, we may therefore not have to recognize X as an independent segment at the level under discussion [footnote omitted]. Classification is used to group together elements which substitute for or are complementary to one another [footnote omitted]. (p. 367)
Harris goes on to say in a footnote that ‘the class of elements then becomes a new element of our description on the next higher level of inclusive representation’. These operations yield a grammar of lists: ‘In one of its simplest forms of presentation, a synchronic description of a language can consist essentially of a number of lists’ (p. 376). These include a segmentphoneme list, a phoneme distribution list, several morphophonemic lists, lists dealing with type and sequences of morphemes, a component and construction list and a sentence list—the list of utterance structures. The purpose of this grammar of lists is to state all the regularities in a language, derived from an exhaustive analysis of a presumably representative corpus of utterances.
In striking contrast, Chomsky’s first major work in linguistics, his master’s thesis ‘The morphophonemics of Modern Hebrew’ (henceforth MMH), virtually bypasses these taxonomic procedures and focuses instead on the construction of a grammar that describes grammatical sentences in a language, of which the linguist’s analysed corpus is merely a special subclass. In this work Chomsky equates the linguistic analysis of a language with ‘the process of converting an open set of sentences—the linguist’s incomplete and in general expandable corpus—into a closed set—the set of grammatical sentences—and of characterizing this latter set in some interesting way’. He goes on to distinguish linguistic method (e.g., as discussed in Harris’s book) from linguistic description.
Accordingly we might distinguish and consider separately two aspects of the linguistic analysis of a language, a process of ‘discovery’ consisting of the application of the mixture of formal and experimental procedures constituting linguistic method, and a process of ‘description’ consisting of the construction of a grammar describing the sentences which we know from step one to be grammatical, and framed in accordance with the criteria related to its special purposes. (p. 2)
In MMH Chomsky assumes that the two processes are sequential steps in linguistic analysis. Yet his work on Hebrew concerns only the second step. ‘The outline of Modern Hebrew grammar given below is an example of the second step in linguistic analysis, artificially isolated’ (p. 3).
In attempting to motivate the distinction between discovery and description, Chomsky notes that for a linguistic grammar ‘considerations of elegance’ are operative in both processes.
However it will still be useful to consider the processes of discovery and description separately For the most reasonable way to approach to the investigation and analysis of the notions of simplicity in terms of which ‘grammatical in L’ is defined (i.e., those notions of elegance that are relevant to the very formulation of the procedures of linguistics) seems to be to assume, for some language, that the grammatical sentences are fixed (i.e., that the process of discovery has been completed) and to determine the effect on grammar-formulation of explicit considerations of simplicity imposed on the grammatical statement [footnote omitted]. (p. 3)
In effect, these remarks indicate how Chomsky’s initial work in linguistics is already moving in the direction of a theory of structural analyses (presumably, but not necessarily, derived from the application of linguistic method)—in contrast to Harris’s Methods, where the distinction between method and theory is made (see the first quote from Methods on page 653) but the issue of theory is not pursued.
Another point where Chomsky’s early practice diverges from Harris’s method concerns a distinction which Harris sets up in section 20.21 of Methods. Harris notes that an investigator in linguistics faces a choice of purposes: to state ‘all the regularities which can be found in any stretch of speech, so as to show their interdependences (e.g., in order to predict successfully features of the language as a whole)’, or to synthesize ‘utterances in the language such as those constructed by native speakers’ on the basis of some minimal information (p. 365). The procedures presented in Methods are geared most naturally to the former purpose. Moreover, as Chomsky notes in the introduction to The Logical Structure of Linguistic Theory (1955a, published 1975; henceforth LSLT), ‘Harris did not elaborate on the suggestion that a grammar can be regarded as a device for “synthesizing utterances,” an idea that does not, strictly speaking, seem compatible with the general approach of Methods’ (p. 50: n. 45).
In MMH Chomsky gives an explicit interpretation to Harris’s notion of ‘synthesizing utterances’ in terms of a generative grammar. Chomsky describes the origin of this work as follows:
Harris suggested that I undertake a systematic structural grammar of some language. I chose Hebrew, which I knew fairly well. For a time I worked with an informant and applied methods of structural linguistics as I was then coming to understand them. The results, however, seemed to me rather dull and unsatisfying. Having no very clear idea as to how to proceed further, I abandoned these efforts and did what seemed natural; namely, I tried to construct a system of rules for generating the phonetic forms of sentences, that is, what is now called a generative grammar. I thought it might be possible to devise a system of recursive rules to describe the form and structure of sentences, recasting the devices in Harris’s methods for this purpose, [footnote omitted] and thus perhaps to achieve the kind of explanatory force that I recalled from historical grammar [footnote omitted]. (LSLT, p. 25)
Although the introductory section of MMH suggests that ‘the application of the mixture of formal and experimental procedures constituting linguistic method’ (e.g., Harris’s) constitute step one in linguistic analysis, Chomsky’s actual practice demonstrated that step one could be bypassed. It is the implementation of ‘step two’ in linguistic analysis that leads Chomsky to the formulation of a theory of structural analyses in LSLT.
Note that Chomsky’s notion of ‘synthesizing utterances’ differs significantly from Harris’s. In section 20.3 (‘Description of the language structure’), Harris characterizes linguistic analysis as follows:
The work of analysis leads right up to the statements which enable anyone to synthesize or predict utterances in a language. These statements form a deductive system with axiomatically defined initial elements and with theorems concerning the relations among them. The final theorems would indicate the structure of the utterances of the language in terms of the preceding parts of the system. (pp. 372–3)
Although the mention of ‘deductive system’ has a distinctly contemporary ring—recall Chomsky’s many discussions of the deductive structure of the current principles and parameters theory of UG,5—Harris’s notion of synthesizing utterances involves an actual synthetic process, starting with phonetic units and proceeding from them to the construction of phonemes, morphemes and so on to the construction of utterances. For Chomsky, however, the notion ‘synthesizing utterances’ is interpreted analytically. Chomsky is quite explicit about this in the introductory section of MMH, in the version submitted to the University of Pennsylvania:
the theoretical linguistic system constructed is available for descriptions of small parts of the realm of experience constituted by the totality of its individuals. In one of the most interesting cases, the part to be described is an individual language. As above, the system can be applied in two ways. One can begin with the elementary phonetic units, construct phonemes, morphemes, syntactic classes, etc., proceeding synthetically. Or one can state the most general unit (i.e., the sentence) in terms of its constituents (e.g., the particular phrases of the language), further analyze these into their constituents, etc., until finally every possible sentence is represented in terms of phonetic units, thus proceeding analytically. Again, the choice will depend on considerations of elegance and adequacy (p. 3)
For Chomsky MMH is ‘an attempt to carry out an analytic statement of Modern Hebrew grammar’. In context, it constitutes a radical departure from the kind of linguistics Chomsky had been learning as a graduate student. We have some indication of how radical a departure from the fact that Chomsky himself considered the work he was doing in MMH as ‘more or less as a private hobby, having no relationship to “real linguistics”’ (LSLT, p. 29).
Adopting the analytic approach to grammar construction led Chomsky to postulate abstract underlying morphological forms, which were simply outside the range of structures adducible via Harris’s methods of segmentation and classification. Note that many of the intermediate derived forms are also beyond the range of Harris’s methods. Consider for example the sample derivation given in section 6, A.2 (where line 0 gives the underlying morphophonemic form):
The derivation of kamá under this analysis involves the sequential application of a set of morphophonemic rules where the output of one rule serves as the input of another. The derivations proceed analytically from a morphophonemic representation to a phonemic representation—in contrast to Harris’s synthetic procedure which begins with phonetic analysis and progresses to phonemic and then morphophonemic level of representation, presumably without intermediate representations between the phonetic/phonemic or phonemic/morphophonemic levels. Recall that Chomsky’s morphophonemic analysis is itself embedded in a broader syntactic analysis of Hebrew sentences, sketched in the introductory sections of MMH though not elaborated there.
Chomsky’s use of abstract underlying representations (in contrast to Harris) is also a distinguishing feature of his syntax.
By 1952, I was working on generative grammar of English, and shortly obtained results that I found quite exciting, though they were entirely divorced from the systems of procedures of analysis on which I was working at the same time; in particular, results on the system of auxiliary verbs in simple declaratives, interrogatives, negatives, passives, and on complex verb constructions such as ‘want (NP) to (VP)’, ‘consider NP (to be) Predicate’ [footnote omitted]. As in the case of my earlier work on morphophonemics of Hebrew, it was possible, so it became clear, to discover systems of rules that made sense of the distribution of forms, principles that served to explain the collection of superficially chaotic and anomalous facts. In this case, too, investigation led to more abstract underlying structures that were far removed from anything that might be obtained by systematic application of procedures of analysis of the sort that I was investigating. (LSLT, pp. 30–1)
The systems of rules Chomsky is referring to here are of course those of phrase structure and transformational structure as he was then developing them.
It is well known that Chomsky’s theory of transformations is very different from Harris’s. One of the most salient differences between the Harris and Chomsky theories is that grammatical transformations for Harris were essentially equivalence relations among sentences, whereas for Chomsky they were rules of grammar. Thus Harris’s first published discussion of grammatical transformations in section 2.33 of Harris 1952 uses the term in the title of the section but not in the text of that section. Instead, he discusses the procedures of establishing equivalences and providing several examples:
But what is ‘equivalence’? Two ELEMENTS are equivalent if they occur in the same environment within the sentence. Two SENTENCES in a text are equivalent simply if they both occur in the text (unless we discover structural details fine enough to show that two sentences are equivalent only if they occur in similar structural positions in the text). Similarly, two sentences in a language are equivalent if they both occur in the language. In particular, we will say that sentences of the form A are equivalent to sentences of the form B, if for each sentence A we can find a sentence B containing the same morphemes except for differences due to the difference in form between A and B. For example, N1 VN2 is equivalent to N2 is V-en by N1 because for any sentence like Casals plays the cello we can find a sentence The cello is played by Casals.
In Harris 1957 he gives a more formal definition of what constitutes a transformational relationship:
If two or more constructions (or sequences of constructions) which contain the same n classes (whatever else they may contain) occur with the same n-tuples of members of these classes in the same sentence environment (see below), we say that the constructions are transforms of each other, and that each may be derived from any other of them by a particular transformation. (p. 147)
Moreover, this relationship is predicated in terms of strings rather than constituent structure, as in Harris 1965, where he explicitly rejects constituent structure analysis in favour of string analysis. Thus for Harris, a transformation is a relation that holds between pairs of sentences.
In striking contrast, Chomsky defines a transformation as a mapping of a string with its phrase structure interpretation (i.e., a phrase-marker) onto another phrase-marker. Such mappings may iterate and thus cannot be simply relations between sentences. Furthermore, they crucially rely on constituent structure analysis, unlike Harrisian transformations. Whereas Harris uses transformations as a means for describing the syntax of texts, Chomsky employs transformational analysis to get at a more explanatory analysis of the structure of language. As he notes in Syntactic Structures, for example, ‘transformational analysis brings out the fact that negatives and interrogatives have fundamentally the same “structure,” and it can make use of this fact to simplify the description of English syntax’ (pp. 64–5). This kind of insight, which involves positing abstract underlying structures and intermediate stages of derivation is simply beyond the power of Harris’s transformational analysis, which is restricted to relations between pairs of surface strings.
Perhaps the greatest difference between Harris and Chomsky concerns their interpretation of linguistic analysis. From the very outset, Chomsky adopts a realist interpretation of grammar. Language is somehow a real phenomenon of the physical world and as such has a particular structure—i.e., it is structured one way and not just any way. Thus there is a fact of the matter for all grammatical descriptions because they make empirical claims. Harris, in contrast, does not adopt the realist position.
The methods described here do not eliminate non-uniqueness in linguistic descriptions. It is possible for different linguists, working on the same material, to set up different phonemic and morphemic elements, to break phonemes into simultaneous components or not to do so, to equate two sequences of morphemes as being mutually substitutable or not to do so. The only result of such differences will be a correlative difference in the final statement as to what the utterances consist of. The use of these procedures is merely to make explicit what choices each linguist makes, so that if two analysts come out with different phoneme lists for a given language we should have exact statements of what positional variants were assigned by each to what phonemes and wherein lay their differences of assignment.
The methods presented here are consistent, but not the only possible ones of arranging linguistic description. (p. 2)
It is obvious from this passage that for Harris there is no fact-of-the-matter in linguistic descriptions. He goes on to say that ‘the particular way of arranging the facts about a language which is offered here will undoubtedly prove more convenient for some languages than for others’. Thus they avoid
the undesirable effect of forcing all languages to fit a single Procrustean bed, and of hiding their differences by imposing on all of them alike a single set of logical categories. If such categories were implied, especially to the meanings of forms in various languages, it would be easy to extract parallel results from no matter how divergent forms of speech.
Because Harris’s procedures ‘are merely ways of arranging the original data; and since they go only by formal distinctions there is no opportunity for uncontrolled interpreting of the data or for forcing of the meaning’. As a result, ‘the data, when arranged according to these procedures, will show different structures for different languages’. Thus Harris seems to deny the generality of his methods of analysis as part of a theory of linguistic structure. His focus here appears to be on the apparent diversity of linguistic forms rather than on some underlying unity.
In rather striking contrast, consider the opening paragraph of Chapter 2, ‘The nature of linguistic theory’, of LSLT:
Descriptive linguistics is concerned with three fundamental problems. On the one hand, the descriptive linguist is interested in constructing grammars for particular languages. At the same time, he is interested in giving a general theory of linguistic structure of which each of these grammars is an exemplification. Finally, he must be concerned with the problems of justifying and validating the results of his inquiries, and demonstrating that the grammars that he constructs are in some sense the correct ones. All three of these problems will occupy us in this investigation of linguistic structure. (p. 77)
The agenda that Chomsky maps out here is clearly more ambitious than Harris’s. In addition to providing linguistic descriptions of particular languages, linguistic theory under Chomsky’s conception must provide a general theory of linguistic structure— something which Harris says his procedures will not provide. And furthermore linguistic theory must deal with the problems of justification and validation under the realist interpretation (which follows from the assumption that a grammar can be correct or incorrect).
Chomsky’s realist interpretation of linguistic theory is grounded in the psychological interpretation of grammar: a grammar exists as part of the real world in the mind/brain of a speaker and constitutes that speaker’s knowledge of his/her language. Although this psychological interpretation is not discussed in Chomsky’s published writings until 1965 (in the first chapter of Aspects, written in 1958–912) Chomsky has noted that during the period in which he wrote LSLT (that is, circa 1955), ‘it lay in the immediate background of my own thinking’. However, he refrained from raising the issue at that time because it seemed to him ‘too audacious’. The earliest reference to the psychological interpretation of generative grammar occurs in the last section of Robert Lees’s review of Syntactic Structures in Language (1957). The final section (titled ‘Learning theory’) begins: ‘Perhaps the most baffling and certainly in the long run by far the most interesting implications of Chomsky’s theories will be found in their cohesions with the field of human psychology.’ Lees goes on to discuss Chomsky’s model of grammar as a scientific theory of human linguistic behaviour. He concludes the section and the review by raising the fundamental issue of language acquisition.
We come now to the point of this lengthy discussion of induction versus theory construction. Though it is possible, it is certainly not an easy task for the psychologist to explain the mechanism by means of which a child, confronted with a vast and perplexing array of different stimuli, manages to learn certain things which can be generalized by induction from repeated occurrences. We would not ordinarily suppose that young children are capable of constructing scientific theories. Yet in the case of this typically human and culturally universal phenomenon of speech, the simplest model that we can construct to account for it reveals that a grammar is of the same order as a predictive theory. If we are to account adequately for the indubitable fact that a child by the age of five or six has somehow reconstructed for himself the theory of his language, it would seem that our notions of human learning are due for some considerable sophistication.
To Chomsky it seemed obvious
that language acquisition is based on the child’s discovery of what from a formal point of view is a deep and abstract theory—a generative grammar of his language—many of the concepts and principles of which are only remotely related to experience by long and intricate chains of unconscious quasi-inferential steps.
Furthermore, Chomsky assumed that much of this theory must be innate in the mind/brain of the language learner.
A consideration of the character of the grammar that is acquired, the degenerate quality and narrowly limited extent of the available data, the striking uniformity of the resulting grammars, and their independence of intelligence, motivation, and emotional state, over wide ranges of variation, leave little hope that much of the structure of language can be learned by an organism initially uninformed as to its general character.
The psychological interpretation of grammar leads to a clarification of the notion ‘language’—though it should be noted that this clarification does not appear in print until Chomsky 1986a.16 The psychological interpretation concerns the internalized grammar of a speaker, a structure of the mind/brain—what Chomsky calls an I-language, where ‘I’ is intended to suggest internal, individual and intensional. In contrast, structural and descriptive linguistics is concerned with a rather different notion of language, ‘as a collection of actions, or utterances, or linguistic forms (words, sentences) paired with meanings, or as a system of linguistic forms or events’—what Chomsky calls an E language, where ‘E’ is intended to suggest external and extensional. The latter notion has no status under the psychological interpretation.
In Chomsky’s earliest work on generative grammar, before he explicitly adopted the psychological interpretation, the discussion of the concept language’ tends in the direction of E-language—not surprising for someone trained in the tradition of structural linguistics. Consider, for example, the way Chomsky defines language and grammar in LSLT.
We define a language to be a set (in general, infinite) of strings in a finite alphabet, each string being of finite length. We define a grammar of the language L as a finite device which generates all and only the strings of L in a determinate way. (p. 71)
Given such definitions, a primary goal of linguistic analysis is to construct a system of rules from which the language can be derived. This assumes that a language is in some sense a well-defined object—a false assumption simply because there is no way independent of the grammar postulated by the linguist to determine exactly all and only the grammatical sentences of a language.
Under the psychological interpretation, grammar construction is subject to a further constraint. In addition to accounting for linguistic phenomena, a grammar must meet a criterion of learnability. Thus the system of rules in a grammar must plausibly be learnable given the primary language data available to a child learning a first language. Given that the theory of grammar under the psychological interpretation constitutes a theory of the language faculty, the innate mental structure of a child employs in grammar construction, the more linguistic phenomena that can be derived from the theory of grammar, the less that needs to be accounted for by constructing language-specific rules. Therefore, the psychological interpretation leads to a shift in focus from rule systems for particular languages to the more general properties of the language faculty. What happened in actual practice in the development of generative grammar was that investigations of general constraints on rule systems lead to a significant simplification of the rule systems themselves, as will be discussed in the following section.
