The Application of Transformational Grammar

to Literary Analysis

John T. Grinder, Suzette Haden Elgin,

《Guide to Transformational Grammar》,1973

In this chapter we discuss a special area of language, which we will refer to as literary language, an area often discussed in the literature under the terms poetics and stylistics. As we use the term here, it should be understood to refer to the body of linguistic material left over when one has excluded gibberish on the one hand and ordinary discourse to he other. This description hardly constitutes a definition and would, of course, be of little value in assisting someone to determine whether a particular piece of linguistic material is or is not literary language.

   We will not attempt to describe any decision procedure that might allow the automation of the process of deciding when a given linguistic object is literary language; rather, we will make an overt appeal to the reader’s intuitions regarding the identification of such material. The aim of this chapter is thus somewhat more modest than many of those that have preceded it. We wish to argue for an extension of the principles of analysis which have been used thus far in the book to this body of language known as literary language. We wish to claim that literary language, like all other language, is within the proper domain of linguistics, and that the grammar which generates literary language and that which generates ordinary discourse are one and the same in the sense that one includes the other.

    While it may at first seem obvious that one would use the same system of analysis for both literary and nonliterary language, there has been much resistance to the incorporation of literary language in the set of material of be subjected to explicit analysis. It seems clear that the resistance of opponents to such incorporation is based on two facts: They do not believe that literary language constitutes a systematic body of linguistic material, and they do not believe that literary and nonliterary language are systematically related.

     Since it appears that it is the strikingly different surface structure characteristics displayed by literary language versus nonliterary language that people feel most acutely represent the difference between the two types of language, and since it is in poetry that one finds the most extreme surface differences (with respect to normal discourse), we will confine the discussion in this chapter to poetry. It should be understood, however, that much of what we have to say with regard to poetry is intended to apply to the entire range of literary language.

    If you are a native speaker of English, you will probably say that you know what a poem is; that is, you know one when you see one. But, if someone asks you to tell him, specifically or even generally, what one is, you are likely to find yourself in difficulty. That difficulty, the inability to specifically define a poem as such, has never yet been resolved, and transformational grammar is no closer to such a definition than any other sort of grammar is.

    If, then, we are forced to rely upon the intuition that we recognize a poem when we are confronted with it, can we make some sort of specific statement about the reasons for that recognition?

    First of all, we recognize a poem as such because of the recognition conventions that we have been taught. Some of these conventions will be culture-specific: in English literature they include things like the following:

    1. A small block of language is arranged in isolation on a page, surrounded by a broad margin of white.

    2. Instead of the type running from side to side of the page in uniform strips that fill it margin to margin, the right-hand side of the text is chopped off, either in varying lengths or some patterned length that does not extend all the way across the page.

    3. At the top of the block of type there is a title,, a few words in length and centered a space or two above the balance of the text.

    4. The language used shows patternings of sound and rhythm, such as rhyme, assonance, alliteration, and meter.

    Also in English literature, poetry may exhibit the phenomenon of special line-patterning to produce specific forms, such as the sonnet.

    Having identified some linguistic object as a poem, we have learned to expect certain things of it. We expect to find a surface difference from ordinary discourse. We expect to find surface strings that we would not accept as well-formed in ordinary discourse, but that we will accept within the poem. This characteristic, this difference, is referred to in transformational grammar as deviance.

    The deviance may be very slight indeed. In much traditional poetry it is confined entirely to the presence of rhyme and metric pattern. If, on the other hand, you examine some of the more experimental poets, you may be tempted to conclude that a line of poetry is a completely unrestricted thing. Poetic license appears to some to be a license to throw down any random selection of words, slap t title at the top, and call the result a poem. This impression is misleading, but it is exceedingly common and vastly seductive.

     Another characteristic we can expect to find is that many things will matter to us about this chunk of language that do not matter at all in ordinary conversation. Assume, for example, that we are at the dinner table and someone says, “Pass the peas, please.” The result should be only that we pass the requested vegetable. On the other hand, if we find that same surface string in a poem, we are going to notice that the line is made up of four monosyllables, that three of them begin with the same sound, that the last two rhyme, and so on. The number of such things we notice will vary with the amount of interest and training that we have in the interpretation of poetry, but at least some of them will be noticed by any native speaker. Moreover, those characteristics that are overlooked at first will be readily understood by any native speaker once they have been pointed out to him. Thus, it would seem that the intuitional experience of an individual carefully examining a section of poetry is not unlike the experience of an individual discovering a previously overlooked ambiguity in ordinary discourse. If we consider the “pass the peas, please.” Example, it is clear that the fact that such sound patterning is overlooked in ordinary discourse but is important in literary language suggests one major difference between the two types of language. Namely, when dealing with literary language, every feature of the linguistic object─phonological, semantic. Orthographic, or any other(s)─must be carefully considered and assumed to be relevant.

    This in itself would be enough to distinguish poetic language from ordinary discourse, since it would be highly inefficient to have to consider all features of stretches of language in ordinary communication. Buildings would burn down while firemen pondered the significance of the phonological sequences and imagery of the call for help. Conversation would be impossible: business transactions would come to a halt. In other words, while the function of the phonological sequences “selected” by the speaker in ordinary discourse is simply to allow the listener to recover the Deep Structure (meaning) of that sequence, the sequence of phonological units occurring in line of poetry serves a double function. It allows the listener or reader to recover the Deep Structure, and at the same time it affords him a kind of sensual ante esthetic pleasure.

    It is clear that stylistic recognitions are going to vary greatly from person to person, just as is the case with judgments of grammaticality. Intensive training in the interpretation of literary language will cause an individual to see a great deal more of the total content of a piece of literary language as relevant than he would have done otherwise. Classes in the appreciation and analysis of literature are based upon this fact. But even the infant human, who presumably has had no such training, will react instinctively to such stylistic factors as rhythm and repetition.

    Another characteristic of poetic language marks it off irrevocably from ordinary discourse: It cannot be paraphrased or tampered with without being destroyed in the process. This is not true of ordinary speech. Consider the following sentences:

E1. (a)  I’d like you to build me a house.

(b)        I want you to build a house for me.

(c)         What I want you to do is build me some place to live in.

(d)        I need a house built and I’d like you to do it for me─ I plan to live in it myself.

You could expand this list for a long time, and still there would be no change in the items of information involved. You want a house, you want the person you are speaking to build it, and you intend the house as a dwelling for yourself. No one of the sentences is particularly better than another. It is difficult to imagine anyone fighting about whether he had actually said “I’d like you to build me a house,” or “I want you to build a house for me,” because the differences are of little importance. But consider the following sequence:   

   E2. Build thee more stately mansions. O my soul.

   This is, of course, a line from a poem by Oliver Wendell homes. It is addressed to the soul of the poet (wherever and whatever that may be) and it conveys the message that the poet wishes the soul to construct for itself dwellings that are more stately. Now consider the following:

   E3. (a)  Build thee more stately houses, O my soul.

(b)        Construct thee statelier mansions, O my soul.

(c)         Build bigger and better houses for yourself, O my soul.

The information content of the sentences in E3 is the same as that of E2, but

not one of the three sequences of E3 is substitutable for the original line. This characteristic, the fact that paraphrase is not possible, is one of the most important ways we have of distinguishing poetic language from ordinary language. Once a poem has been written, a sort of web of relevance relations is established for it. The set of phonological items becomes entangled with the semantic ones, the syntactic arrangements are not separable from the typographic ones, and the entire balance of the web is destroyed if any part of it is tampered with.

    Now we have established the fact that poetry appears to be different from ordinary language and that, for various reasons, we are able to recognize it as poetry and deal with it on that basis. At this point we come to the question that is an issue of dispute in transformational grammar today, namely, just how this “deviant” body of language is to be dealt with.

    Notice that the word deviant itself is a relational word. That is, something must be deviant with respect to some standard, not in isolation. In the case of poetry, it is probably the nonliterary language that serves as that standard. However, nonliterary language is not characterized by transformational grammarians in some statistical manner, but rather by s set of explicit rules─a grammar. Thus, we arrive at a more interesting characterization of the deviancy of poetic language: namely, poetry is seen as a result of the extension, omission, or alteration of members of the set of transformations applied to the set of Deep Structures (meanings). The interesting question is, what sort of alterations in the set of transformations are possible? We will argue that the alterations are systematic and can be systematically described.

    There have been a number of traditional ways of dealing with the problem. There is the approach that appeals to “divine inspiration” (states flatly that since we cannot understand, we can only admire) and treats poetry as completely outside the rest of language and therefore no part of linguistics in any way. If you happen to belong to this group, there is no point in your reading any farther. There is the idea of fiddling around with the surface differences and doing research based upon them, which leads to things like counting the number of times Tennyson used the letter “l” in any given body of text. There is the painful explication de texte that is the systematic destruction of a poem or literary work. All these things have their place, but they are the periphery, the symptoms of literature. They are a way of going around and around the issue without touching it.

    The transformational linguist hopes to go considerably beyond this respectful skirting of the issues. He hopes to take up the question of literary language, not in isolation from the rest of the language but as a living and vital part of it. This involves his assuming that poetry is generated by a grammar, just as ordinary discourse is, and this involves a problem. Consider the following:

E4. (a)  He danced his dance.

(b)        He danced his did.

(c)         He danced his paper bag.

E4(a) is ordinary language, E4(b) is a line from a poem by e. e. cummings(Poems

1923-1954, Harcourt Brace Jovanovich, Inc.), and E4(c), unless some evidence can be found to the contrary, is nonsense. (It should be noted that there is nothing about E4(b), in isolation, to indicate that it is a line of poetry or to distinguish it from E4(c): this is a case of special knowledge.)

   But, if you are going to allow a grammar to generate a line of poetry like He danced his did, and you are going to treat that as acceptable, how are you going to prevent a sequence like He danced his paper bag? On the surface it appears that he danced is paper bag should be a lot less deviant than he danced his did, since it is at least structurally identifiable as a proper expansion of an English sentence. No Phrase Structure Grammar of English, however, offers as a possible expansion the following sequence of rules:

R1. (a)  S→NP VP

(b)        VP→ Verb Determiner Verb  

It is precisely these two rules that would be required to generate He danced his did.

   One solution to this problem that has been proposed is that each poem should be set up as a dialect of the language being used, with its own grammar. Thus, just as the sequence He ain’t is marked as deviant in our grammar, but is certainly acceptable in several American dialects, He danced his did is assumed to be an acceptable sequence in the dialect of the poem, which presumably contains rule R1(b).

    This solution effectively isolates the dangerous linguistic object and makes it impossible for it to contaminate the grammar of ordinary discourse. It also allows a grammarian to take s straightforward, old-fashioned English sentence like e. e. cummings in this line has used a verb where in ordinary discourse a noun would be used and write it out in symbols that certainly look more impressive than the sentence did. Moreover, it handily makes it unnecessary for the grammarian to do anything more, since every poem becomes a wild card in its private, observationally adequate box.

    But if we take this proposal, that each poem is to be treated as a separate dialect, seriously, let’s stop and consider what this is going to force us to do.

    We must remember that no one has as yet found any really satisfactory way of defining, in any formal manner, just what constitutes literary language, or poetic language, or a poem. Our judgments in this regard will not be uniform from speaker to speaker. It should therefore be clear that there is no way to draw a line what will restrain the scope of the proposed “separate dialect” system. Without a formal definition of the poem, which would allow us to make principled decisions as to what is to be excluded or included, we cannot proceed in any useful way. For example, if a chunk of language set off in the middle of a page is to be called a poem, what do you do about such things as advertisements that may have similar characteristic?  (And in fact much advertising copy not only cannot be excluded from literary language but constitutes one of the most effective bodies of non-ordinary language in contemporary society.)

    Consider also what sort of language mechanism would be required in order to make the dialect system operate. We would have to assume that each time a poetic stretch of language turned up in a body of otherwise nonliterary language, the individual who had been previously using the ordinary grammar of English would suddenly switch to a new dialect, only to switch back to ordinary English at the end. There would have to be a continual flip from normal grammar to these multitudinous dialects and vice versa.

    These criticisms, however justified, are not in themselves sufficient reason to support the claim that the grammar which generates literary language is the same that generates ordinary language. They constitute negative reasons; what is needed is a set of positive ones. We will next consider some of these.

    First of all there is the obvious fact that both types of language, literary and ordinary, use the same phonology, the phonology of the given native language. The first letter of the word peas is a voiceless bilabial stop in English poetry, just as it is in ordinary English discourse. And just as in ordinary discourse the sequence ‘mbiq’ is not interpretable as an English word, so in literary language the same holds true.[1]

    A second fact is that both ordinary and poetic language use the same lexicon. Consider the following lines from Shakespeare(Sonnet 44):

E5. When I have seen the hungry ocean gain

    Advantage on the kingdom of the shore….

The word ocean , which occurs in the first line of this pair, has exactly the same lexical reading as it does in the following sentence:

E6. I can see the ocean from here.

Certainly it is not likely that in ordinary discourse one would say, “I can see the hungry ocean from here.” Nonetheless, this use of the word hungry is understood as a metaphorical extension of the same word in ordinary discourse and not as some new and unique word having some other lexical reading.

    It is also the case that poetry follows the same syntactic rules, in exactly the same way, as does ordinary discourse. If a poet wishes to form within the boundaries of a poem a command, a question, a relative clause, a comparative, or any other structure of English, he has open to him only those syntactic processes that he would have if he were using ordinary discourse. It is not possible even to imagine some new way of asking a question within the boundaries of a poem that would not be one of the mechanisms for forming a question in ordinary discourse, for example.[2]

  We propose, then, that a poem (or any other piece of literary language) begins, just like any other piece of language, as a tree structure, This tree structure contains all semantic elements of the total content of the poem. This Deep Structure is then related to its Surface Structure by a derivational process─the series of transformations of the English language─just as is any other Deep Structure. The resulting Surface Structure, however, is then subject to a series of optional transformations that are forbidden for ordinary language. We will return to this point and discuss these supplementary transformations later in this chapter.

   If we establish that the Deep Structure of a poem undergoes the same basic transformational operations as does any ordinary Deep Structure, and if we establish that the most immediately obvious “deviancies” are due to a very late set of optional processes, the problem that remains is that of deciding just where the essential difference between poetry and ordinary discourse lies.[3] The word essential must be carefully stressed, since the majority of the differences noticed are not essential. An example of a purely stylistic and therefore superficial, difference would be the following:

E7. (a)  Three little girls came through the woods.

   (b)  Through the woods came three little girls.

This is certainly not an essential difference, Even if it is pushed further, as in the following example, it remains stylistic:

E8. Little girls three through came the woods.

  There are reasons for such stylistic changes. They are not random, but have significance for the poet and are intended to be noticed, but they are surface phenomena. We must look elsewhere for the basic difference we are seeking.

  The question to ask is this: Ignoring the surface manifestations that make poetry different, and keeping in mind the characteristics (such as resistance to paraphrase and relevancy of all factors present) that we have already established as differences, can we select some single most important factor that sets off poetry from ordinary discourse? The most satisfactory answer seems to be: the characteristic of saying a great deal very well in a very small space. Let us call this characteristic compression[4].

All the things that happen to language in a poem and that do not happen to language in ordinary discourse are motivated by the attempt to establish and maintain this elusive compression without losing the Deep Structure (meaning) in the process. Since we know that the total semantic content of the poem is present in the Deep Structure, we must assume that those elements that have been affected transformationally by systematic processes such as deletion, permutation, insertion, and substitution.

   We will now return to the discussion of these transformations that apply only to poetic language. It is important at this point to state that although the transformations we are about to discuss are extremely rare outside of poetry, it is not possible to exclude them entirely from the language of prose. For example, we would not be surprised to find them operating in the novels of James Joyce or Gertrude Stein. The question then becomes a venerable one; that is, are we to say that such works are prose, making extensive use of poetic techniques, or that they are really poetry, making use of prose techniques? We will not attempt to settle this question here.

   The following schema represents the grammar as it deals with poetic language, whether in poetry or in other literary contexts:

F1.                Deep Structure (Semantic Representation)

   Transformational component 1    Successive stages in the syntactic derivation linked

                                    ↓                by transformations t1…tn

                             Surface structure

   Phonological Component             Successive stages in the phonological derivation linked by phonological rules

                                   ↓

 Phonologically interpreted surface structures

   Transformational Component 2    Successive stages in the literary derivation linked by transformations t’1…

                                                        t’n, all of which are optional

                                   ↓

                             Literary Language

   The transformations listed in schema F1 as  t1…tn  include those that have already been discussed in this text. Now we want to consider those transformations that are shown above as t’1…t’n.

     We will confine our discussion to two transformations (OVERLAP and TECHNIQUE DELETION), which appear to be well motivated for the grammar of English and which can be demonstrated to be extensions of transformational processes already needed independently by the grammar.

Overlap Deletion 

OVERLAP DELETION is a transformation that applies within a body of poetic language when two phonologically identical sequences occur immediately contiguous to one another under specific conditions, and which operates to delete one of the pair of identical items. Consider the following pair of sentences from ordinary discourse:

E9. (a)  I will jump in the lake, and then Bill will jump in the lake.

(b)        I will jump in the lake, and then bill will.

E9(b) is derived transformationally from E9(a) by deletion of the identical sequence jump in the lake. This transformation is called VERP PHRASE DELETION.

     The constraint on this sort of deletion is very rigid, as is necessary for ordinary communication. Ordinary discourse does not present itself for review and interpretation as literary language does; it is transient and can rely upon no special consideration. This is not true for poetry, since it is expected that the reader of listener will have the time and make the effort to take into account all factors that enter into the interpretation of the poem. It is therefore reasonable to expect that constraints upon deletion in poetry will be less rigid than those upon ordinary discourse.

     The following lines (by e. e. cummings, “There is a here and,” Poems 1923-1954, Harcourt Brace Jovanovich, Inc.) demonstrate the extension of this process of deletion under identity to poetic language:

E10.  …the ocean

      wanders the streets are so

      ancient…

   In E10 the two Deep Structures are the following:

E11. (a)  The ocean wanders the streets.

    (b)  The streets are so ancient.

The transformation OVERLAP DELETION has applied to delete one instance of the streets under identity with the other.[5]

The condition of phonological identity of the immediately contiguous items is very rigid, as shown by E12.

E12. (a)  I will touch you/you will touch me→ 

               I will touch you will touch me.

          (b)  I will touch him/he will touch me→

              *I will touch him will touch me.

         It is clear that identity of syntactic function is not required, since in the examples we find one of the twin sequences functioning as the object of a preceding verb or preposition, and the second functioning as the subject of the upcoming verb, and the transformation still applies. For English, however, it appears to be necessary that the two items have the same lexical reading.[6] To show that this is true, consider the following examples:

E13. (a)  Never tell a lie.

(b)  Lie in my bed→

(*c)  Never tell a lie in my bed.

Although application of OVERLAP DELETION to E13(a) and (b) certainly yields a comprehensive and grammatical English sentence and a perfectly possible line of English poetry, it is at once obvious that the resulting sequence E13(c) cannot be understood as containing the original pair of Deep Structures. Since the whole point of OVERLAP DELETION is to achieve compression by allowing the possibility of the single combined sequence yielding the meanings of both original sequences, the transformation cannot apply; this result is perfectly consistent with the basic concept that transformations may not change meaning.

   There is still an additional constraint upon this transformation in English. Consider the following line by T. S Eliot (“Burnt Norton,” Four Quartets, Harcourt Brace Jovanovich, Inc.):

E14.  Only through time time is conquered.

   The item time in this line would appear to satisfy all the conditions previously specified for the transformation; nonetheless, it cannot be reduced to Only through time is conquered without destroying the deep structure. If we examine the Deep Structure of the Eliot line, we will be able to see the essential difference here. The Eliot line has a Deep Structure something like the following:

E15.  Someone conquers time only through time.

      [+PRO]

This sequence would be dominated by a single S-node at the time application of OVERLAP DELETION took place, which is not true of any of the preceding examples. We can therefore add the additional constraint for English that the transformation can apply only if the two identical items originate in separate underlying strings.

   OVERLAP DELETION may the be formulated for English as follows:

R2. Within the context of poetic language, if a given sequence s1 is followed immediately by another sequence s’1, delete s’1, where

(i)                 s1 and s’1 are phonologically identical.

(ii)               S1 and s’1 share the same lexical reading.

(iii)             S1 and s’1 are dominated by separate S-nodes at the time OVERLAP DELETION applies.

You will notice at once that this transformation is formally rather different from the transformations we discussed before. Fore example: the effect of this transformation would be to reduce a structure like E16:

E16.      S             S

            ∕   \          /    \

  ∆     ∆        ∆     ∆

  to a structure[7] like E17:

E17.

           S     S

          /   \  /  \

         ∆    ∆    ∆

    Nonetheless, although this transformation is formally unusual─to such an extent that a new formalism would be required to allow us to write the Stxuctural Index and description of the rule─there is nothing whatever about it that is not already a part of the grammar of ordinary English discourse.[8]

Technique Deletion

  We have seen that overlap deletion is the logical extension of the transformations that effect deletion under identity in ordinary discourse, such as VERB PHRASE DELETION. The second transformation we will consider, TECHNIQUE DELETION, is the logical extension of the transformation in ordinary discourse known as UNSPECIFIED AGENT DELETION.

 Consider the following:

E18. (a) John wants to leave.

    (b) The door opened.

   The deleted item in E18(a) is uniquely recoverable; it can be only John. This is an example of EQUI-UP DELETION. But the sentence of E18(b) is a different matter. Here there is no single lexical item that can be specified as having served as the Deep Subject of the verb open. It is, instead, the case that there is a small set of such items and, as native speakers of English, we know what items may or may not be members of this set. We know that the item must be someone or something, and we know roughly what characteristics the someone or something must have. We know it opened the door or the newborn baby opened the door.

   The same process is operative in poetic language, but operative in a less rigid manner. Consider the following:[9]

E19. this small horse newly

    he is fresh from his mother’s flanks

   If the poet intended this sequence to mean this small horse newly purchased, he is in much the same situation is the speaker who intends the door opened to mean the diagonalized matrix opened the door, or who intends John wants to leave to mean John wants Mary to leave. Instead, we can safely assume that there is a small set of possible items that could have occurred after newly and that, because of the overt items, we can specify that set fairly closely. The set of eligible items is probably the following:

E20. born, dropped, foaled, nursed

We can determine this because of the semantics of the material present and because we are provided by the poet with the following phonological sets:

F2. (a)  /n/…newly

(b)        /r/…horse, fresh, mother

(c)         /f/…fresh, from, flanks

(d)        /l/…small, newly, flanks

Given these phonological sets, we can safely eliminate dropped from the set of candidates E20, leaving us with the other three. This is a degree of ambiguity that is tolerable in poetic language, and is in fact desirable, since the several possibilities for the deleted item allow a high degree of compression.

To apply this transformation, the poet relies upon the stylistic devices at his command, such as rhyme, meter, and parallelism. If the transformation is misapplied, he will have changed the meaning of the underlying string or produced a degree of ambiguity that is not tolerable, in the sense that he has made even approximate recovery of the deleted item(s) impossible. Since this process is less constrained than deletion in ordinary discourse, it is likely that it will be more often misapplied than is UNSPECIFIED AGENT DELETION. This does not mean that it is unsystematic, but rather that the performance factor figures very highly in the application of this transformation.

   Now consider the following example:

E21. snowflakes round and round through air

The missing item here is the verb. The degree of ambiguity is relatively high, but given the phrase round and round we know that it is restricted to a verb of motion and that the motion must fall within the category circular. If the poet meant the verb to be explode, then he should have told us so; he has misapplied the transformation.

  It is more likely that we will be given additional information, as in E22.

E22. snowflakes round and round through air like ballerinas

  The set of possible lexical items is now specified more completely[10]. To the original circular-motion idea of round and round, we have now added the idea of “dance,” and we can safely claim that the set of items contains at least the following:

E23. whirl, twirl, spin, dance

   Let us further restrict this set of items, as follows:

E24. two snowflakes round and round through air like twin ballerinas

We now have the lexical item twirl reinforced by its phonological similarity to the word twirl and by its orthographic similarity to the word two. We also have the item spin reinforced by the fact that it rhymes with the word twin. We can now say that the set of possible deleted items contains the two members twirl, spin and is no more ambiguous than a set containing the two category items someone and something.

   A tentative formulation of the transformation TECHNIQUE DELETION for English might be the following:

R3. Where a lexical item(i) is one of a small finite set of possible items. And (ii) can be reinforced in the Surface Structure by such technical devices as rhyme, assonance and alliteration, and meter, that lexical item may be deleted from the surface string of poetic language.

   It should at once be clear that it would be even more difficult to write the formal Structural Index and description of structural change for this transformation that it was to do so for OVERLAP DELETION. This is a difficulty caused by the limitations of our formalism; it should not be allowed to obscure the fact that this is a transformational process of English, that it is systematic rather than random, and that it is an extension of similar processes already present in the grammar of the ordinary discourse of English.

   It is obvious that the process of deletion is far more complex in poetry than in ordinary language and that it is based upon more intricate conditions. As we have said before, this is possible because the language of poetry is available for extended consideration and examination.

   Notice that in giving a formulation to the transformations involved in the enumeration of possible strings of literary language, we have employed the same vocabulary and the same notation that has been used throughout this text. That is, while the set of transformations we are discussing in this chapter differs from those considered in previous chapters in that they are defined on the output of the phonological component and are always optional, they appear to be otherwise indistinguishable.

Other Techniques

It must be pointed out that it appears that some techniques are employed in literary language which cannot be readily accounted for by means of transformations, either of the type proposed in this chapter or of the type associated with the enumeration of strings of nonliterary language. We will illustrate by example.

    Ross(1967b) studied extensively the set of restrictions that are usually referred to as Universal Constraints on transformations. It is characteristic of these constraints that they are of the form shown in R4:

R4.  Given some string s1 and some transformation t1 such that s1 is properly analyzable with respect to the Structural Index of t1, t1 may not be applied to s1 if such and such a condition obtains.

     Consider a specific example of such a universal constraint on transformations, that constraint known as the Conjoined Structure Constraint, or CSC. We will use the strings of E25 as an example, along with the transformation PASSIVE.

E25. (a) Sam kissed Sue.

    (b) Sam kissed Sue and her sister.

The CSC can be stated now in the format suggested by R5:

R5.  Given some arbitrary string s1 and some arbitrary movement transformation t1, t1 may not apply to s1 if the application of t1 to s1 would split a conjunction.

The transformation PASSIVE is an example of a movement rule, and is thus relevant. The arbitrary string E25(a) contains no conjunction; however, the string E25(b) does, namely, the sequence Sue and her sister. Now consider the effect of the application of PASSIVE to the two strings.

E26. (a) Sue was kissed by Sam.

   (*b) Sue was kissed and her sister by Sam.

It is clear from the linear sequence of elements E25(b) that PASSIVE has applied to a conjunction in such a way as to split it. The effect of the transformation is the production of a deviant string, since the transformation has applied in spite of a constraint to the contrary. The CSC is not limited in its application to the movement transformations contained within the grammar of English, but is proposed as a constraint on movement transformations in all human languages. This is a claim of importance, and one that we will discuss in detail in Chapter 13.

   Consider, however, the following lines from Milton:[11]

E27. …Is not the earth

     With various living creatures and the air

     replenished…

Clearly, the lines from Milton violate the CSC. It is equally clear that we can be sure that random violations of the CSC do not occur in Milton or in any other literary work. As we mentioned earlier in this chapter. The traditional characterization of literary language as being deviant seems to us to be interpretable only in the sense of deviancy with respect to the set of well-formed strings enumerated by the nonliterary grammar. We have been attempting to show that the transformations that enumerate the set of well-formed literary strings are defined over the output of the phonologically interpreted strings of the nonliterary portion of the grammar. In the case of the example from Milton, we can give an explicit characterization of what has happened.

   Without becoming involved in disputes regarding the consciousness with which the poet dose what he does, we can say that Milton’s example makes it clear that high-level constraints defined over the set of nonliterary transformations may be relaxed in certain instances to achieve a particular literary effect. The determination of the set of circumstances where such constraints may be relaxed would contribute much to the understanding of the limits of literary language.[12]

    Thus we return to the point of this chapter. We have explained a number of literary language phenomena. We have proposed two transformations that operate only in poetic language, but which are extensions of transformational processes within ordinary discourse. We have done this in an attempt to support the claim that the conceptual schema and mechanisms developed by transformational syntax provide an insightful approach to the explicit analysis of literary language. We assume that by     an explicit analysis of literary language we are giving serious content to the traditional claim that it is the poet who is master of his language.

[1] This is not to imply that a poet might not use a device such as a sequence of sounds that violates the phonological rules of English; he certainly might. But the fact that such a sequence was a violation would remain true for the poem, just as it would for an ordinary sentence. The rule that everything must be considered relevant will cause the reader to note these deviant sound sequences and assume that there is a reason for their being present. It will not cause him to accept them as members of Set A for English..

[2] To claim that the structures of literary language are not the result of operation of rule s of ordinary syntax would cause us to have within the grammar duplicates of every rule. That is, we would have to postulate a literary and a nonliterary PASSIVE rule, a literary and nonliterary EQUI-UP DELETION rule, and so on.

[3] At this point we are referring specifically to poetry and not to all literary language.

[4]  Notice that, having isolated this characteristic, we are no nearer a definition of poetry, since compression might also be the defining characteristic of telegrams and headlines.

[5] Since it is not possible to tell from the resulting surface string which of the twin items has been deleted, we will arbitrarily decide that it is the second.

[6] This constraint does not necessarily hold for other languages. For example, in Japanese it is possible for OVERLAP DELETION to apply not only to individual items with different lexical readings but also to portions of those items. The analogous procedure for English would allow us to apply OVERLAP DELETION to the sequence friendship/shipmates and delete only the identical segment ship. The result for English would be the clearly unsuccessful *friendshipmates.

[7] This structure was suggested to us by Professor Yuki Kuroda.

[8] The reader may be familiar with the device known as double syntax, which was used extensively by Shakespeare and has a long and respectable history in English literature. Double syntax appears to be equivalent to OVERLAP DELETION plus one additional constraint. Consider the following example:

     For the victims of war we offer nothing

     but the best.

   In this example the reader or listener has the possibility of stopping either at the end of the first line given or at the end of the second, with resulting complete strings in both cases. The difference between double syntax and OVERLAP DELETION is, however, that for double syntax the resulting combined sequence must be grammatical in the traditional sense of the word. Thus, for the victims of war we offer nothing but the best is just as grammatical as for the victims of war we offer nothing. However. Although the ocean wanders the streets is grammatical, the ocean wanders the streets are so ancient is not. It would appear, then, that for English, the constraint that the combined sequences be grammatical has been dropped from the grammar, in just the same way that a phonological or syntactic rule might be so dropped.

[9] A very similar sequence appears in the work of e. e. cummings.

[10] There are a number of formal mechanisms for describing this process. For example, we could borrow from the work of Weinreich (1966) the mechanism called a Transfer Feature, which would operate to extend feature markings from one morpheme to another under specific conditions.

[11] This example, for which we are indebted to Ann Banfield, is From Paradise Lost, VIII:369-371

[12] It seems reasonably certain that even literary language will have to exhibit some form of the constraint first formulated by Perlmutter (1968a) to the effect that an otherwise permissible transformation will be blocked if it results in a surface string that is identical to one resulting from a totally different derivation. We might then propose that the Milton example is possible, precisely because there can be no derivation resulting in a sequence that would have the earth replenished with the conjoined NP living creatures and the air. However, much more research will be necessary before any such claim can be made precise.