4.3.1　Transformational-Generative grammar and FLT

Transformational grammar was an influential theory of grammar formulated by Chomsky in the late 1950s（Chomsky，1957，1966）.The theory inspired a considerable amount of psycholingulstic work in the 1960s and early 1970s.

As a language teacher，one has behind us two powerful traditions in language teaching.The first is called rotememorization，either the nineteenth-century and earlier variety which entailed memorizing lists of words and statements of rules，or the later behaviorist-psychology and structuralist linguistics pattern-practice.Variety that is common still today.The second is the intuitive-generalizing style of teaching.This tradition has been less popular：first，because it is harder to understand and needs a good teacher and good presentation to work，and second，because until recently it was considered heretical to suggest that people were in any interesting way different from rats.It was assumed that people learned languages，both native and second，as they and rats learned everything else by repetition，by exercise，and by fitting new things into an old pattern already learned.The more one repeated something，the better it was learned.It was also assumed to be dangerous to let one think about sentences one was learning because one would not form a pattern correctly，since one would not establish a direct stimulus（heard sentence of situation）-response relationship.It was assumed that language was just another kind of stimulus-response；the speaker heard a sentence or felt some sensation，and this triggered，without the intervention of any kind of reflection，a response，also verbal.He was just like the rat pushing a switch for food.However，if the speaker thought about a sentence，or wondered why it was grammatical before he said it，or were concerned about its relationship to other sentences，he would break this stimulusresponse link and would not be using language as a native speaker does.The same learning model was assumed for first-language learning too：the child learned the pattern by repeating the sentences he heard from his parents.When he had repeated all the sentences that had been said in the first five years，and had memorized those，he had acquired the grammar of his parents and was a native speaker.Second-language learning was viewed as the same sort of processas first-language learning.It was important to present to the second-language learner only correct sentences.Otherwise，like a rat presented with contradictory stimuli，he might not know which response was correct.

In the last ten years，a new linguistic theory，transformational grammar，has arisen，in direct opposition to the behavioralstructuralist theories within which so many language texts have been written.As far as language and language learning are concerned，it has much in common with the beliefs of the rational grammarians.One very significant added element in the modern theory，of course，is the existence of transformational rules，the form of which and the sphere of applicability of which are strictly controlled by the theory itself，avoiding the fuzziness，arbitrariness and the ad hoc treatments of data that characterize the work of the“intuitive”rationalist school.In non-transformational intuitive grammar，sentences are related to one another on no grounds except the intuition of the writer.Sometimes the relationship postulated would be considered correct by modern analyses；sometimes they were absurd.The theory provided no automatic way of distinguishing truth from fiction：there was no formal definition of a possible underlying structure or a possible rule.Moreover，though the theory assumed an underlying universal logical structure，there were no constraints on the form of this structure，or on the permissible types of rules changing it to the superficial structure.Virtually anything could be treated as an underlying structure，and—in theory at least—virtually any operation could be performed on it.One sentence might have many possible analyses，with no means of discriminating among them as to which was right，and with no assumption made that there must be only one underlying structure for a non-ambiguous superficial sentence.Transformational theory，as exemplified most fully perhaps in Chomsky's Aspects of the Theory of Syntax，formalized these intuitive concepts so that they could be checked，constrained，and tested.

In this work，Chomsky assumed two levels of grammar and a set of transformational rules mediating between them.At the level of deep structure，everything is present that enables one to know what the sentence means：identical words，later deleted，abstract elements of various sorts that leave syntactic markings elsewhere in the sentence，and so forth.Transformational rules delete these under various conditions，producing surface structures.

①I saw a boy that hates potato chips.

Therefore，the transformational analysis of a sentence like①assumes a deep structure in which the noun phrase a boy occurs twice：once where it will occur in the surface structure，as the object of the main clause，and once where that occurs in the superficial structure.These two nouns are co-referential；both refer to the same boy necessarily or else the sentence would be meaningless，or rather，nonsensical，because infinitely ambiguous：without a principle of recoverability of deletion—allowing only identical elements to be deleted—one would never know what that referred to，and the sentence could be interpreted in infinite ways.But since all languages contain this principle，it is inconceivable that that could have replaced any noun but a boy，identical in reference to the subject，which is still present superficially.It is evident that that refers to boy，an animate human masculine singular noun，because of constraints on what can occur in the relative clause.If that were really just an infinitely ambiguous form not going back to a deep-structure noun boy，the ungrammaticality of ②-④would be unrelated to that of sentences like ⑤-⑦below—a patently absurd situation，since the speaker of English knows the same thing is wrong with both sets of sentences：

②*I saw a boy that elapsed yesterday.

③*1 saw a boy that was pregnant.

④*1 saw a boy that laughed at themselves.

But if we assume that the noun boy underlies that at some level of the grammar，we can immediately tell why all these sentences are ungrammatical，without having new special rules in the grammar to explain them：②-④are ungrammatical for the same reasons as ⑤-⑦are（and the translations of both sets，in any natural language，also are）：

⑤*The boy elapsed yesterday.

⑥*The boy was pregnant.

⑦*The boy laughed at themselves.

These are some of the reasons for proposing a theory with deep structures and transformational rules to change them into the superficial structures.

In addition，transformational grammar assumes that most sentences in a language are formed by combining two or more smaller sentences：sentence①is produced by combining⑧and⑨：

⑧I saw a boy.

⑨A boy hates potato chips.

There will be constraints on what kinds of sentences can be combined if the transformational rule yielding relative sentences is to apply properly.Another obvious fact these sentences illustrate is that it is not possible to construct“the longest sentence”of any language：one could always，for instance，add a new relative clause，or a conjunction，or a complement sentence as in⑩，where that...stupid is a complement embedded in the larger sentence：

⑩I told the boy that he was stupid.

The result is that there is no“longest sentence”of any language.Moreover，it follows that one could not expect to learn a language，native or foreign，by merely memorizing a list of sentences，however long，so the behaviorist theory of language learning is incompatible with this evidence.Moreover，a child at age five is already able to tell whether a given sentence of his language is grammatical or not—though he has never heard that sentence before.As Chomsky points out，on the basis of factssuch as these，language-learning must be viewed as a process depending on reasoning rather than on memorization.

People should perhaps not have made such a categorical statement.Some linguistic phenomena are，of course，not based on logic that they are accidental.If people watch a child learning to speak his own language，or even more clearly，an adult successfully learning a second language，people note that he often has recourse to his memory.The child learns vocabulary by memorization：though he quickly learns how to generalize—to find the rules—or construct new sentences，he soon learns that if he constructs new words by a rule he devises，he will often not be understood，logical as his formulations may be.He learns a set of endings on words—plurals，tenses，diminutives，etc.—as a list that he must memorize.There are no‘reasons’why the plural has an ending and the singular has not.（He does，of course，learn that this is a rule—if he learns ten cases where the plural is formed with-s，he will try out an eleventh case he has never heard before，and generally be correct.）He also must learn that sentence⑩is good，but sentence⑪ which has roughly the same‘logical’structure is not，nor is⑫.These facts—which‘complementizers，’as they are called，go with which verbs—must be memorized，just as vocabulary items are.

⑪*I ordered the boy that he gives me his dictionary.

⑫*I said the boy to be stupid.

And similarly，there are in every language exceptions to rules，and these must be memorized.There is no conceivable‘rule’to tell one why⑬ is a good sentence of English，and⑭ not.There are languages where the translations of both are good.There is a difference of that sort in languages，it is generally the case that we are dealing with something‘illogical’—something that requires some amount of rote memorization：it is not universal.

⑬ John is likely to leave.

⑭ John is probable to leave.

There are also cases where rules can be given to the learner，where all he will need to know to tell whether to apply the rule in a given case is what other words are present in the sentence，and what they mean：these are purely superficial phenomena.So，for instance，in sentence①，a restrictive relative clause，the pronoun that is grammatical.But in sentence⑮，who cannot be replaced with that：⑮ is a non-restrictive relative sentence.

⑮ I saw the boy，who was running fast.

Facts like these can be stated in rules and generalizations learned.In this way they are different from the sorts of facts given earlier，which are idiosyncratic and must be memorized individually.But in both cases，rote learning is possible either of the list of forms or of the rule that generates a set of possible forms.The teacher can give a rule like“That is never used in nonrestrictive relative clauses”and have the students memorize it；and，assuming that the teacher has fully and accurately explained the distinction between restrictive and non-restrictive relative clauses，there will be no mistakes.Pattern-practice drills are of value in these cases.There is nothing wrong with constructing drills to facilitate the memorization of facts about pluralization，complementizer-selection（as in⑬，subject-raising as in⑭），or restrictive vs.non-restrictive relative pronouns.

In all these cases，the speaker or learner or teacher need only know the superficial form of the sentence in question.He need not worry about contextual factors outside the scope of rules：what has been said earlier，what the speaker knows about the topic of conversation，what is common knowledge，or knowledge of the world.But there are cases where such factors are irrelevant.Sometimes—more frequently than has been assumed—to judge whether a sentence is correct in its context，one must know something about the speaker's unstated belief about the world.

Consider the use of the articles in English.In any given sentence，either a definite or indefinite is generally possible.So，judging merely from the immediate environment，both⑯ and⑰ are grammatical：

⑯ The doctor is over there.

⑰ A doctor is over there.

That is，it is impossible to state a rule using as the environment the superficial form of the sentence alone to predict whether⑯ or⑰ will be correct in a given sentence.The problem is even worse in complex sentences.Sentence⑱ is good，but⑲，with only an adjective changed，is not：

⑱ Black spoke with a warmth that was surprising.

⑲*Black spoke with a warmth that was usual.and in⑳ and㉑，only a definite article is good if the adjective is one of the same class as usual in⑲：

⑳ Black spoke with the enthusiasm that was expected.

㉑*Black spoke with an enthusiasm that was expected.

To return to the earlier examples，in ordinary conversation，a speaker may say，“I'm looking for a doctor who was asking for money the day before yesterday.”The conversation may then take a completely different turn，and deal for some time with different topics entirely.But if after a while，the second speaker in the conversation wants to refer again to the boy in question，he can use sentence⑯，but not⑰ to call attention to that fact，even though the doctor may not have been the topic of conversation for some time and even though the first speaker himself used the indefinite article.In the sentences of ⑱-㉑，it is the meaning of the adjective that conditions the choice of article.The meaning has to do with how likely the speaker feels the warmth，or enthusiasm：whether he anticipated it or not.This fact，too，cannot be expressed as an invariable rule，or learned by memorizing sentences that follow the pattern.One must know what is in the speaker's mind，the hearer's mind，and in the previous conversation before one can judge the grammaticality of such sentences.Consider too sentences like㉒ and㉓：

㉒ Albert is a surgeon in my neighborhood.

㉓ Albert is the surgeon in my neighborhood.

The difference between these sentences cannot be found in any context that can be located by a rule.Either could perfectly well begin a conversation between two people，one of whom didn't know Albert；The distinction lies in whether the speaker feels that it is normal，or necessary，for every neighborhood to have a doctor，or whether this is merely incidental，and Albert is a surgeon who happens to live or work in the speaker's neighborhood.The choice of article thus depends on the speaker's feelings or beliefs about the world and how he sees its organization.No rule can give the learner this information.One can give general rules for article choice：use the definite article for something that is already known，or mentioned；otherwise use the indefinite.The non-native speaker，unless he has learned the language extremely well，or has as his native language one in which there is an article system similar to that of English，will make mistakes，some more frequently than others.The task of the serious teacher，then，is to teach the non-native speaker what presuppositions go with what use of the articles.He must do this by identifying，in the learner's native language，where similar presuppositions have overt counterparts，and matching these language-specific superficial structures with those of the learned language.In practice，of course，it is extremely difficult to do this；Then the teacher must simply give various situations in which one article would be used rather than another，or none，and explain as well as possible why the choice must be this way：the teacher must give the learner a boost to making his own generalizations，to learning how the native speaker understands and intuitively uses these sentences.This necessarily implies that it is essential to give the learner ungrammatical sentences，so that he can study these along with the grammatical ones to decide for himself what the difference is，so that when he is on his own and has to make a decision for himself，he can rely on his own new generalizing ability in this sphere to make the right generalization.With rote pattern practice alone，he would either be helplessly presented with a situation that felt outside of the patterns he had studied（which would，of course，be extremely frequent）or he would overgeneralize，applying a pattern where it did not fit，since he would not know the reason why that pattern took that form.

A second case from English involves the distinction between the past and perfect tenses.This distinction is very difficult for anyone whose tense-system is not as complex as that of English，or where the meanings of the tenses are differently divided.Compare，for example，sentences㉔ and㉕ pointed out by Jespersen.Both are grammatical，but they are used in different situations：

㉔ The patient has gradually grown weaker.

㉕ The patient gradually grew weaker.or㉖ and㉗：

㉖ I saw John every day for twenty years.

㉗ I have seen John every day for twenty years.

If㉔ is used，the assumption must be that the patient is still alive.If㉕ is used，there is no such assumption.That is，to tell which of two tenses to use in referring to something that happened in the past，the speaker must have access to information about what is true in the present，information that is nowhere overtly stated.No list could be used to predict accurately which of these sentences could be used in a given conversation.The use of㉔ instead of㉕，though not overtly carrying information that the patient was alive，could mislead a hearer.Similarly in㉖ and㉗ if there is a gap in time between the years in which the speaker saw John every day，and the time at which he is speaking，only㉖ is possible.The use of㉗ for㉖ creates confusion if the speaker has not seen John for some time.The sentence is not ungrammatical，as*I saw the boy that elapsed，or*I said John to be a fool；but its use in the wrong context stamps one as a non-native speaker of English，just as certainly as the others would.But no rule can be given for the learner to follow in any situation.

Consider one last case from English.It is frequently stated that if in a positive sentence some can occur，in the corresponding negative，interrogative，or conditional sentence any is found instead.As that statement stands，it sounds as though the someany distinction would be a very good candidate for rote learning and pattern practice.But the following sentences are not amenable to any imaginable form of pattern-practice drill：

㉘ Does someone want these eggs？

㉙ Does anyone want these eggs？

㉚ If he eats some candy，let me know.

㉛ If he eats any candy，let me know.

All these sentences are grammatical.Each pair is identical superficially except for the presence of some or any.If the rule above is correct，the pair of sentences should be synonymous.But it is clear that there are situations where㉘ is appropriate，but not㉙，and㉚ but not㉛，or vice versa.For example，㉘ might be used if the speaker expected someone to ask for the eggs and㉙ if he really didn't think anyone would want them.The speaker might use㉚ if he secretly hoped the person addressed would eat the candy but㉛ if he hoped he wouldn't.Thus，we can explain the strangeness of㉜ and㉝，where these presuppositions are combined with overt statements that directly contradict them：

㉜？If he eats some candy，I'll punish him severely.

㉝？If he eats any candy，I'll give him ten dollars as a reward.

（In all these cases，the any is unstressed，not the stressed any（at all））.The oddness of these sentences cannot be ascribed to anything present overtly in the sentences themselves，not even necessarily to anything in the speaker's or the hearer's knowledge of the world.The presence of some or any can be predicted only if you know what is going on in the speaker's mind—that is，if you are the speaker.Hence，one cannot give a rule for the distribution of some and any in sentences of this type：rather，the learner must be informed as to which to use according to his state of mind，or his beliefs about things.

The theory of transformational grammar has built into it various self-policing mechanisms in the form of principles by which the linguist can judge whether an analysis is rigorous or ad hoc；whether it is a complete analysis，or only scratches the surface of the problem，whether a formulation is precise or vague.These principles will enable the linguist to know，for instance，whether to posit an analysis in which two superficially different sentences are identical at a deeper level；whether two superficially identical sentences are different at a deeper level one，the principle of recoverability of deletion mentioned already，allows deletion to take place only in case there is an element identical to the element to be deleted，which will remain in the surface structure so that its meaning can be recovered，or there is an abstract element that has left behind it syntactic markers from which its original presence can be deduced.It's allowed that two sentences to be derived from the same deep structure only in case the selectional restrictions in one are the same as those in the other：the same kinds of nouns can be subjects or objects，the same classes of verbs occur.It is assumed that two superficially identical sentences are not really identical if it can be shown that，by substituting one word for another，one of the meanings is made impossible.Tests of this kind allow us to propose deep structures in a responsible way，and relate these deep structures to the proper surface structures.This is of use in language teaching in a number of ways，none direct in the sense that writing transformational rules is direct.First，we ourselves become more sensitive to language through applying these tests and demanding proof of every claim.This enables us，hopefully，to see better than someone who has not been trained the relationships among sentences in English，and their relationship to universal facts and language-particular rules of the learner's native language.Then，when we say that a sentence of a certain type in English is related to，or obeys some of the constraints of，a sentence of maybe a quite-different-looking type in the learner's language，we have a reasonable idea that we are basing our conclusions on more than personal caprice.The danger of the latter is that，if the relationship is not real，or is only partial，the speaker may wrongly generalize.So，for instance，let us say that we have taught the speaker sentence㉞：

㉞ John couldn't lift 500 Ibs and we point out to him that it is ambiguous because it can be paraphrased either by㉟ or㊱：

㉟ John was physically unable to lift 500 Ibs.

㊱ It is impossible that John lifted 500 Ibs.

Now if we look only this far into the language，and don't do any testing，we may give the speaker a rule saying：can have either of these two meanings.And，if，as is frequent，there are in the learner's language two verbs translatable as can，one with one sense and one with another，if we're not careful，we may say that can will translate either of these，freely，and that the synonymy is total.

But now let us look at sentence㊲.Here something surprising happens.Only the second interpretation，that of“is possible that，”is found：

㊲ John couldn't be as stupid as Harry！

The reason is that the verb following the modal in㉞ is active，or voluntary，while that of㊲ is stative，or involuntary.Physical inability is not a factor in the meaning of stative verbs.If we know this fact，we are better able to explain the use of the modal can to non-native speakers，and，having tested and looked at various facts，we can avoid overgeneralizations，such as assuming（since when there are two identical verbs conjoined in a sentence conjunction reduction is possible）that，if the verbs can underlying㉞ and two meanings are identical，we could get sentences like㊳：

㊳ John couldn't lift 500 Ibs，or be as stupid as Harry.

With the interpretation of the first can as“be physically able.”Knowing there is a distinction and pointing it out avoids this danger.

We can also use transformational analyses when two sentences that look quite different share similar deep structures.This can be shown to be true for㊴ and㊵：

㊴ Bill cut the sausage with a knife.

㊵ Bill used a knife to slice the sausage.

It would take too long to go through the proofs that these sentences do，in fact，come from similar or identical deep structures.But if we can make that assumption，we can show that this similarity of underlying structure can be put to use in teaching English as a foreign language.Suppose we are dealing with a language in which either the analog of㊴ or of㊵ did not occur，or where one or both looked quite different from either㊴ or㊵，and we wanted to teach how both were used.It could be pointed out that they shared a common meaning；and，further，that just as the ungrammaticality of㊶ can be expressed in terms of selectional restrictions between a verb and its subject.㊷，is ungrammatical in the same way.If one of these constructions exists in the learner's language，he can make the generalization and know at once which sentences will be grammatical，which will not，though no rule can be stated（since with occurs with other meanings and restrictions）in English.

㊶ The book used a knife to cut the sausage.

㊷ The book cut the sausage with a knife.

Using insights such as these，made possible by a knowledge of transformational grammar，but not its formal devices，we can，we hope，teach languages better.

As has been discussed early in this chapter，it'sstated that the structuralists resorted to rote learning and pattern practice in the belief that they would thus recapitulate in second-language learning the processes of first-language learning as seen by behavioral psychologists.First-language learning clearly involved rule-formation，or generalization from raw data.Are we then trying to reproduce the process of first-language learning in second-language teaching？It should be noted that in the other transformational approaches，the answer would have to be‘no.’It is never assumed that the child memorizes rules such as have been given.

Despite numerous psycholinguistic experiments，which show the order in which rules are learned，and mistakes made in learning a few isolated syntactic phenomena，mostly in English，practically nothing is known about first-language learning.No one knows how the child shifts out the rules from the huge mass of data，how he decides what is grammatical from the semigrammatical and ungrammatical strings he hears along with the fully-grammatical ones.Also unknown is how he tells what rules are universal，what are not.It is not likely that we will know the answers to these questions for a long time.The question of the nature of first-language acquisition is just as dark，in fact，as second-language acquisition.

We can say a little，and that little enables us to give the answer to our question：are we attempting to recapitulate firstlanguage learning？The answer is both yes and no.No，first，because certain abilities the child has are lost，and we cannot hope to use them again after he is ten years old or so.We do not even know what these abilities are.We know the child can，given raw data，derive the rules with no help from anyone.He can learn and memorize astounding amounts of vocabulary，including lists of exceptions.And he does all this，or most of it，unconsciously.You never hear him muttering，‘Let's see—is there a variable in that relative-clause formation rule or not？’or‘Hmmm...I wonder if equi-NP-deletion is governed，and what the exceptions to it are？’But somehow he knows.We have tried to enable the second-language learner to recapture some of his old ability by providing him with lists of things to be memorized（as is usual）and with the generalizations that the child would make himself.We have not yet tested our grammar：it is not even written.We would hope that，provided artificially with what the child has naturally，the second-language learner would go about learning his second language rather in the way he learned his first.But we must remember that artificial devices are seldom as good as the real thing.Probably too the fact that the generalizations must be consciously articulated will make a difference in how they are learned.But the similarity lies in appealing to the learner's ability to reason，compare data，and generalize.In this way our secondlanguage teaching is like the process of first-language learning as transformational theory views it.

Some interesting similarities and differences have been noticed between the two types of learning.First，some work of Carol Chomsky's seems to indicate，if we can give this interpretation to her findings，as she did not，that universals are learned in a different way and at a different time from languageparticular facts，in first-language learning.According to a universal constraint on pronominalization，in no language is it possible to say sentences like㊸：

㊸*He said that John was here.

Here he and John refer to the same person.But some pronominalization-related facts are dialectal，or at least not universal.Thus，for instance，in a sentence like㊹：

㊹ John asked Bill when to leave.

In most dialects of English，the understood subject of leave is John.But dialects have been found where it can also be Bill.So this is not universal.It has been found that，with facts that are universal，like those in sentence㊸，a child learns not to make mistakes in them very quickly，over a short time.But with cases like㊹，he goes through a period of fluctuation—from one interpretation to the other，it takes a good deal longer for him to master the rule completely，and he makes many more mistakes.This is of interest in teaching second languages，because，it gives us a bit of a clue as to how to integrate universals in our texts.There are perhaps，some evidence from transformational grammar，that universals are kept apart from language-specific facts by the child，in learning a first language.Moreover，he keeps them apart：one never hears a question in a language classas to the grammaticality of㊸，while one might get questions on the meaning of sentences like㊹.Therefore，people can assume the speaker is probably unconsciously aware that the universals are universal，and there's no need to talk about them，unless they are of use in explaining language-specific facts.

Language-systems are productive，in the sense that they allow for the construction and comprehension of indefinitely many utterances that have never previously occurred in the experience of any of their users.In fact，from the assumption that human languages have the property of recursiveness—and this appears to be a valid assumption—it follows that the set of potential utterance works.Chomsky drew attention to this fact in his earliest work，in his criticism of the widely held view that children learn their native language by producing，in whole or in part，the utterances of adult speakers.Obviously，if children，from a fairly early age，are able to produce novel utterances that a competent speaker of the language will recognize as grammatically wellformed，there must be something other than imitation involved.They must have inferred，learned，or otherwise acquired the grammatical rules by virtue of which the utterances that they produce are judged to be well-formed.Here，it is sufficient to note that，whether Chomsky is right or wrong about other issues that he has raised in this connection，there can be no doubt that children do not learn language-utterances by rote and then simply reproduce them in response to environmental stimuli.

‘Stimulus’and‘response’are key-terms of the school of psychology known as behaviorism，which was very influential in America before and after the Second World War.According to the behaviorists everything that is commonly described as being the product of the human mind—including language—can be satisfactorily accounted for in terms of the reinforcement and conditioning of purely physiological reflexes：ultimately，in terms of habits，or stimulus-response patterns，built up by the same kind of conditioning as that which enables experimental psychologists to train laboratory rats to run through a maze.Since Bloomfield himself had come to accept the principles of behaviorism and had explicitly advocated them as a basis for the scientific study of language in his classic textbook（1935），these principles were widely accepted in America，not only by psychologists，but also by linguists，throughout the so-called post-Bloomfieldian period.

Chomsky has done more than anyone else to demonstrate the sterility of the behaviorists'theory of language.He has pointed out that much of the technical vocabulary of behaviorism（‘stimulus’，‘response’，‘conditioning’，‘reinforcement’，etc.），if taken seriously，cannot be shown to have any relevance to the acquisition and the use of human language.He has shown that the behaviorists'refusal to countenance the existence of anything other than observable physical objects and processes is based on an outdated pseudo-scientific prejudice.He has asserted and，as far as the evidence goes，correctly that language is free from stimulus-control.This is what he means when he talks of creativity：the utterance that someone produces on any particular occasion is，in principle，unpredictable and cannot be properly described，in the technical sense of these terms，as a response to some identifiable linguistic or non-linguistic stimulus.

Creativity is，in Chomsky's view，a peculiarly human attribute，which distinguishes men from machines and，as far as we know，from other animals.But it is rule-governed creativity.And this is where generative grammar comes into its own.The utterances that we produce have a certain grammatical structure：they conform to identifiable rules of well-formedness.To the extent that we succeed in specifying these rules of wellformedness，or grammaticality，we shall have provided a scientifically satisfying account of that property of language—its productivity—which makes possible the exercise of creativity.Productivity，it should be noted，is not to be identified with creativity：but there is an intrinsic connection between them.Our creativity in the use of language—our freedom from stimulus—control manifests itself within the limits set by the productivity of the language-system.Furthermore，it is Chomsky's view and this is a very central component in Chomskyan generativism that the rules that determine the productivity of human languages have the formal properties that they do have by virtue of the structure of the human mind.

Crucial insight into language is that sentences have more than one level of structure.In transformational grammar，this insight is captured in the distinction between deep structure and surface structure.These are both tree structures，differing in emphasis.Deep structure is the underlying structure of a sentence that conveys the meaning of a sentence.Surface structure refers to the superficial arrangement of constituents and is closer to how the sentence is actually pronounced.Three arguments can be made for the usefulness of this distinction.First，consider sentence㊺：

㊺ Flying planes can be dangerous.

This sentence is ambiguous，but not in the sense that the constituents may be grouped in more than one way.Here the ambiguity comes from the（optional）deletion of certain elements of the sentence（or，more precisely，the deep structure of the sentence）.The sentence may be paraphrased roughly as The act of flying planes can be dangerous or Planes that are flying can be dangerous.This type of ambiguity，called deep-structure ambiguity，comes from a single surface structure that is derived from two distinct deep structures.

A second reason for the distinction is that some pairs of sentences are similar in their phrase structure but not in their underlying structure.Consider，for example，sentences㊻ and㊼：

㊻ John is easy to please.

㊼ John is eager to please.

These sentences are apparently similar，but their paraphrases reveal their dissimilarity.We can explain this by observing that John is the object of the deep structure in㊻ and the deepstructure subject in㊼.

Third，other pairs are quite distinct in their surface arrangement but similar in their deep structure，such as the following sentences in active㊽ and passive voice㊾：

㊽ Jack played the tuba.

㊾ The tuba was played by Jack.

In this case，the active and passive sentences are considered two manifestations of the same deep structure.

Another way of putting these points is to say that a grammar that includes only one level of structure is not descriptively adequate.To fully capture these grammatical relationships，we need to posit a second level of structure，which in turn brings into play a new set of rules called transformational rules.

Within transformational grammar，the entire derivation of a sentence is a two-part process.First，phrase-structure rules are used to generate the underlying tree structure.Second，a sequence of transformational rules（sometimes simply called transformations）is applied to the deep structure and the intermediate structures（those between the deep and surface structure），ultimately generating the surface structure of the sentence.Unlike phrasestructure rules，which apply to only one constituent at a time，transformations apply to entire strings of constituents.They transform them by adding，deleting，or moving constituents.

Let us look at a few transformations and see how they work.One is called the particle-movement transformation.We know that the following two sentences mean the same thing：

Johnson phoned up the woman.

Johnson phoned the woman up.

The concern is with the placement of the particle up；in these sentences，the particle may occur either just before or just after the noun phrase.Accordingly，we might write two different phrase-structure rules for the two instances，the first conforming to

（PS1）VP→V+（part）+NP

and the second to

（PS2）VP→V+NP+（part）

The problem with this approach is that it lacks descriptive adequacy；it does not reveal the similarity of the two sentences.In this approach，the two sentences are derived from different phrase-structure rules.An alternative approach is to assume that the two sentences have the same deep structure and to apply the particle-movement transformation to.The transformational rule looks like this：

（T1）V+（part）+NP→V+NP+（part）

Notice that the transformational rule simply moves the last two constituents of the verb phrase.Unlike phrase-structure rules that rewrite one constituent into a series of constituents，transformational rules begin with a series of constituents and transform them.

Consider now the following sentences：

Johnson phoned up the interesting woman.

Johnson phoned the interesting woman up.

Johnson phoned up the woman with the curly hair.

Johnson phoned the woman with the curly hair up.

Notice that in each case the particle is shifted around the entire NP-two words in，three in，and six in.The point is that the particle movement is defined in terms of constituents，not words.This condition gives transformational grammar tremendous power to apply to an infinite number of NPs，instead of stating the rule in terms of the number of words，which will vary from sentence to sentence，state it in terms of grammatical structures such as NPs，Because the movement is dependent upon the grammatical structure，rules such as this are said to be structuredependent.

A second example is passive transformation.Simplified rule is as follows：

（T2）NP 1+V+NP 2NP 2+be+V+-en+by+NP 1

This complex transformation，which might be involved in the derivation of sentences such as，contains several elementary operations.Let us begin with the active sentence and then add the transformations needed to produce the passive sentence.First we invert subject and object，a transformation that produces.Sequences that are not grammatically acceptable are，by convention，marked with an asterisk.Then we insert the preposition by in.Finally，we add a form of the auxiliary verb be to：

Jack played the tuba.

The tuba played Jack.

The tuba played by Jack.

The tuba was played by Jack.

One final property of transformational rules deserves mention.These rules may be blocked under certain circumstances.For example，the particle-movement transformation does not work with pronouns：

Johnson called them up，

Johnson called up them，

These restrictions on transformations would be specified in the description of the rule.The rule would operate under specified conditions but would be blocked when these conditions did not apply.In conclusion，transformational grammar assumes that sentences have a deep structure and surface structure.The deep structure is derived by a series of phrase-structure rules and the surface structure is derived from the deep structure by a series of transformational rules.Transformational grammar can explain certain aspects of language.