Seeking a suitable framework

This chapter considers the reasons why the linguist Noam Chomsky set up a transformational grammar, one with two levels of structure, deep and surface, and explores the basic characteristics of such a grammar.

Noam Chomsky has been perhaps the most influential figure in twentieth-century linguistics. His contribution has been twofold, as we noted in Chapter 3. On the one hand, he initiated the era of generative linguistics, in that he directed attention towards the rules which underlie a person’s knowledge of their language. Someone who knows a language is somewhat like a chess-player who, in order to play the game, has had to learn rules that, specify which moves are possible, and which not. These rules crystallize the essence of the game. Similarly, the set of rules or ‘grammar’ underlying a language was, in Chomsky’s view, of greater interest than any actual utterances a speaker happened to make.

[]()Insight

Chomsky initiated the era of generative linguistics, which pays attention to the rules which underlie language, not in the actual utterances themselves.

On the other hand, Chomsky renewed people’s interest in language universals. This topic was somewhat unfashionable in the early part of the twentieth century, when it was commonly assumed that ‘Languages differ without limit and in unpredictable ways.’ Chomsky argued that linguists should concentrate not so much on finding out components that []()are common to all languages, which may well be few in number, but on discovering the bounds or constraints within which language operates.

Chomsky did not simply make vague statements about the need for generative grammars and universal constraints; he put forward a number of detailed proposals for a universal framework. Unfortunately for those trying to come to grips with his ideas, he has changed his mind over many facets of his theory since it was first proposed in the 1950s. It started as a transformational grammar. In this chapter, we will explain how he arrived at this particular type of grammar in the first place, and sketch out its main characteristics. Then we will consider why he emended his original ideas and also outline some of his more recent proposals.

[]() Simple models of grammar

Let us now assume that we are in the position that Chomsky was in over half a century ago – that of a linguist trying to set up a universal grammatical framework. Where should we begin? One fairly obvious way to get going is to write a grammar of a language we know, say, English. If we managed to do this adequately, we could then see to what extent the framework might be used for other languages also.

In writing a grammar for English we would adopt the procedure used by all social scientists: we would make a guess or hypothesis, in this case about the rules internalized by someone who knows English. Then we would test the validity of this hypothesis by checking it against some raw data – the sentences of English. If the rules we hypothesized did not lead to good English sentences, they would have to be discarded or amended.

In doing this, we are not trying to describe the way in which humans prepare a sentence for utterance. A grammar is above all a device which specifies what is, and what is not, a well-formed sentence.

It encapsulates rules which define []()possible sentences, but it is not concerned with how these possibilities are assembled.

[]()Insight

Chomsky was not trying to describe how humans prepare sentences for utterance; his grammar was one which defined what was, and what was not, a possible sentence in the language.

The main task, therefore, is to write a grammar which has the same output as a human being – though there is no guarantee that it will replicate the rules in a person’s mind. There may be some overlap between a linguist’s rules and those actually internalized by human beings, but the mechanisms are unlikely to be identical.

Let us now consider how we might go about forming a hypothesis which would account for the grammar of English. Perhaps the best way is to start with a very simple hypothesis – possibly an over-simple one – and see what flaws it contains. Then, in the light of what we have learned, we can proceed to a second, more complex hypothesis. And so on.

The simplest possible hypothesis might be to suggest that words are linked together in long chains, with each word attached to the next. For example, the determiners the and a might be linked to a set of nouns such as camel, elephant, which in turn might be linked to a set of verbs such as swallowed, ate, and so on ().

[]()

Figure 16.1.

But we would very quickly have to abandon such a simple model. Neither English, nor any other language, works in this fashion. A word is not necessarily dependent on adjacent words. Often, it depends on another word []()that is some distance away, as in the sentences:

[]()Either learn to play the trumpet properly or take up yoga. Petronella fell and hurt herself.

The or which is intrinsically connected to either appears several words away, not directly after it. Similarly, herself which is dependent on Petronella is some distance away. Another problem with the ‘chain’ model above is that it wrongly regards each word as attached to the next by an equal bond. The model fails to show that, in the sentence:

[]()The camel swallowed an apple,

the words the and camel are more closely related to one another than swallowed and an. So this simple model must be abandoned.

A somewhat more satisfactory model might be one which treats sentences as if they had a ‘layered’ structure, as represented in the tree diagrams discussed in Chapter 7. This assumes that languages have several basic sentence patterns, each with a number of different ‘slots’ which can be expanded in various ways. A noun phrase (NP) followed by a verb phrase (VP) is a basic English sentence type, and this pattern can be expanded in various ways ().


NP	VP
Ducks	bite
Ducks	bite burglars
The duck	bit the burglar

[]()Figure 16.2.

Such a grammar (often called a phrase structure grammar) contains a series of phrase structure rules, normally in the form of rewrite rules which show the progressive expansions as in (see also Chapter 7).

S → NP VP

VP → V (NP)

NP → D N

[]()

Figure 16.3.

There seems little doubt that some []()such expansion mechanism must be built into any grammar. However, as Chomsky pointed out, such a model is incomplete. It contains at least two serious flaws. First, we require an enormous number of rules in order to generate all the sentences of English. Second, it groups together sentences which are dissimilar, and separates others which are similar. Take the sentences:

[]()Hezekiah is anxious to help.
[]()Hezekiah is difficult to help.

To someone who knows English, these sentences are radically different. In the first one, Hezekiah is planning to do the helping, and in the other, he is the one liable to be helped. Yet the ‘slot’ pattern of both is identical ().


NP	V	ADJ	INF
Hezekiah	is	anxious	to help
Hezekiah	is	difficult	to help

[]()Figure 16.4.

A similar problem occurs with the sentence:

[]()Hezekiah is ready to eat.

Any English speaker could []()(with a bit of thought) interpret this sentence in two ways: Hezekiah is hungry, and wants to have his dinner. Or Hezekiah has perhaps fallen into the hands of cannibals, and has been trussed up and seasoned ready for consumption. The slot model, however, cannot easily show the two radically different interpretations.

The reverse problem occurs with pairs of sentences such as:

[]()To swallow safety pins is quite stupid.
[]()It is quite stupid to swallow safety pins.
[]()Yesterday it snowed.
[]()It snowed yesterday.

The sentences in each pair would be regarded as very similar by English speakers, yet this similarity cannot be captured by the model of grammar outlined above, since each sentence requires a different slot pattern.

Chomsky argued that a grammar which provides only one structure for sentences which are felt to be different by native speakers, and different structures for sentences which are felt to be similar, was a bad grammar. A transformational model, he claimed, overcame these problems.

[]() Deep and surface structures

Chomsky’s solution to the problem was to suggest that every sentence had two levels of structure, one which was obvious on the surface, and another which was deep and abstract. Let us see how this works in connection with the sentences discussed on the previous page. Chomsky accounted for the difference between:

[]()Hezekiah is anxious to help.
[]()Hezekiah is difficult to help.

by suggesting that these sentences []()had a similar surface structure, but different deep structures (, where PRES means ‘present tense’).

[]()

Figure 16.5.

The deep structures discussed in this chapter are simplified versions of those proposed by Chomsky in his so-called Standard Theory of transformational grammar, outlined in his book Aspects of the Theory of Syntax (1965).

A similar deep-surface explanation accounted for the ambiguity in:

[]()Hezekiah is ready to eat.

where two different deep structures were realized by a single surface structure. But the situation would be reversed for pairs such as:

[]()Yesterday it snowed.
[]()It snowed yesterday.

Here two different surface structures shared a common deep structure ().

[]()

Figure 16.6.

If every sentence had two levels of structure, then it was clearly necessary to link these two levels in some way. Chomsky suggested that deep structures were related to surface structures by processes called transformations. A deep structure was transformed into its related surface structure by the application of one or more transformations. For example, the sentence:

[]()It snowed yesterday.

required only one transformation – the attachment of the tense to the end of the verb. But the sentence:

[]()Yesterday it snowed.

required a second one also, one which moved the adverb yesterday from the end of the sentence to the beginning.

[]() Transformational grammar

[]()Insight

We are now able to give a definition of a transformational grammar. It was a grammar which set up two levels of structure, and related these levels by means of operations known as transformations.

[]()A transformational grammar had (like most other types of grammar) three major components: a syntactic component (dealing with syntax), a phonological component (dealing with sounds) and a semantic component (dealing with meaning). However, it differed from other grammars in that the syntactic component was split into two components: the base, and the transformational rules ().

[]()

Figure 16.7.

In the Standard Theory, the base contained phrase structure (PS) rules for the formation of deep structures, and also a lexicon, from which words were slotted into the output of the PS rules ().

The deep structures then passed to the transformational rules in order to be converted into the surface structure. At this point, the surface structure of a sentence was still abstract: it did not yet have a phonetic form. This was coped with by the phonological component, which converted each surface structure into a phonetic representation. Meanwhile, transformations could not change meaning, so the deep structures were fed directly into the semantic component, which gave a semantic interpretation of each ().

[]()

[]()Figure 16.8.

[]() Deep structure

Chomsky did not base his claim that there are two levels of structure purely on the flimsy notion of a native speaker’s intuitions.

There were other, more technical reasons. The most important arguments were based on movement, cases in which sentence constituents appear to have been moved out of their ‘proper’ place. Consider the sentence:

[]()Petronella put the parrot in a drawer.

[]()This sentence contains the verb put which, as we saw in Chapter 7, has to be followed by both an NP and a PP ():


NP	V	NP	PP
Petronella	put	the parrot	in a drawer

[]()Figure 16.9.

We cannot say:

[]()\*Petronella put in the drawer.
[]()\*Petronella put the parrot.

Now look at the following sentences:

[]()What did Petronella put \[—] in the drawer?
[]()What did Petronella put the parrot in \[—]?

These sentences appear to have broken the requirements that put must be followed by an NP and a PP. Instead, what appears at the beginning of the sentence, and there is a gap in the place where one might have expected a word such as parrot, drawer to occur. How are we to deal with these sentences? One possibility is to complicate the lexical entry for put, and to say that put allows several alternatives:

[]()NP put NP PP (Petronella put the parrot in the drawer.)
[]()What NP put PP (What did Petronella put in the drawer?)
[]()What NP put NP P (What did Petronella put the parrot in?)

Such extra additions to the lexical entry would eventually get extremely complicated, as they would have to take into account cases in which extra sub-sentences had been added between what and the original sentence, as in:

[]()What \[did you say] \[the police alleged] Petronella had put in the drawer?

Of course, if the verb put was the only lexical item which allowed such manipulations, one might simply put up with this []()one huge and messy lexical entry. But every verb that is normally followed by an NP allows similar contortions.

[]()Felix grabbed the canary.
[]()What did Felix grab?
[]()What \[did Angela claim] Felix grabbed?

Because of the generality of this occurrence – leaving a ‘gap’ where an NP was expected, and putting what in front of the sentence – it seemed more plausible to conclude that it was a general syntactic rule, which said: ‘In order to form one common kind of question, substitute what in place of an NP, and move it to the front of the sentence.’

[]()

Figure 16.10.

If this solution was adopted, []()one would then hypothesize that the deep structure of the sentences was something like (assuming Q is ‘question’) ():


Q	Petronella put what in the drawer
Q	Petronella put the parrot in what

A transformation then brought what to the front, and the sentences ultimately surfaced as:

[]()What did Petronella put in the drawer?
[]()What did Petronella put the parrot in?

Such arguments convinced many people that sentences did indeed have two levels of structure: a deep structure and a surface structure linked by transformations.

[]()

Figure 16.11.

[]() Transformations

Let us now look more closely at the form []()which transformations, also known as T-rules, took in the (1965) Standard Theory of transformational grammar. Unlike the rewrite rules discussed in Chapter 7, each rule had two parts to it. First, an applicability check (usually called the structural analysis (SA)) that stated the structure to which the rule could be applied, and second, instructions concerning the change it brought about in this structure (called the structural change (SC)) ():


1	STRUCTURAL ANALYSIS (SA) (applicability check)
2	STRUCTURAL CHANGE (SC) (change brought about)

[]()Figure 16.12.

[]()Insight

In the Standard Theory (1965) version of transformational grammar, each transformation involved an applicability check (SA = structural analysis) followed by the transformational alteration required (SC = structural change).

Consider, for example, the transformation which moved adverbs to the front of a sentence, T-adverb preposing, as in:

[]()Bill shrieked suddenly. → Suddenly Bill shrieked.

The structural analysis (applicability check) was needed in order to ensure that the sentence contained an adverb. It said (in more formal terms): ‘Check that the sentence contains an adverb.’ Once this had been assured, the structural change could be specified. This part of the transformation said (again in more formal terms): ‘Move the adverb to the front.’

A formal way of expressing this would be to say:


SA	X – ADV
SC	X – ADV → ADV – X

Here, X is a ‘variable’. This means that its composition []()can vary. In other words, the structural analysis says: ‘The sentence can contain anything you like, as long as it ends in an adverb.’ The structural change says: ‘X followed by an adverb changes into an adverb followed by X.’

In the Standard Theory of transformational grammar, there were maybe two dozen of these transformations, each applying to a specific structure. In addition to moving things around, as in the examples so far, others deleted items. For example, a command such as Come! was assumed to reflect a deeper:


IMP	You will come

(where IMP stood for ‘imperative’). A T-rule (T-imperative) deleted the words you and will, and the instruction IMP. Other transformations added items. A sentence such as There is a dodo in the garden was assumed to reflect a more basic:

[]()A dodo be + PRES in the garden.

A T-there-insertion transformation added there.

In the 1960s, confident researchers thought that, sooner or later, we would compile a definitive list of all the transformations of English, and a complete specification of how they worked. Unfortunately, however, this ambitious programme was never fulfilled, for reasons which will be discussed in the next chapter.

[]()[]()THINGS TO REMEMBER

[]()Chomsky initiated the era of generative linguistics.
[]()A generative grammar is one that is interested in the rules which underlie a language, rather than the utterances themselves.
[]()Chomsky’s small book Syntactic Structures (1957) revolutionized linguistics. It discussed possible grammar models.
[]()Chomsky pointed out that a grammar in which each word triggered off the next could not handle discontinuous constructions.
[]()Chomsky also noted that a phrase structure grammar grouped together sentences that were dissimilar, and did not group together sentences which were felt by speakers to be stylistic alternatives.
[]()Chomsky proposed a transformational grammar.
[]()A transformational grammar has two levels of structure, a deep and a surface level, linked by transformations.
[]()Chomsky’s ‘classic’ (Standard Theory) version of transformational grammar was expounded in his book Aspects of the Theory of Syntax (1965).
[]()This version of transformational grammar had multiple transformations.
[]()The transformations could move things about, could add items, and could delete them.