Deep Structure and Transformations

The purpose of this paper is to provide a clear account of the concept of deep
structure throughout the whole evolution of transformational generative grammar
proposed by Chomsky. The paper will be divided into four parts. In the first part, the
concept of deep structure will be considered from the perspective of Chomskys
grammatical theory published in his Syntactic Structures. The second section considers
deep structure and its impact but after the publication of Chomskys Aspects of the
Theory of Syntax. The third part is concerned with deep structure within the Government
and Binding theory framework. The last part of the paper considers the status of deep
structure within the minimalist approach.

I. Deep Structure after the Publication of Syntactic Structures

After the publication of Chomskys Syntactic Structures in 1957, the notion of
transformational generative grammar started formally. In Syntactic Structures, Chomsky
did not elaborate on his belief in the existence of a universal grammar. However, the
concept was certainly there since Chomsky argued for the need of developing grammars
of specific languages to be certain that these particular grammars fall within the
framework of a general linguistic theory. In addition, Chomsky was interested in
determining the basic underlying principles of a successful grammar. He also presented a
new formulation: phrase structure rules plus transformation rules. Chomskys generative
grammar was based on rationalism and it was developed as the perfect alternative to
structuralism based on behaviourism. According to Chomsky, a grammar or an adequate
grammar should be capable of producing grammatical sentences and rejecting
1

ungrammatical ones. It must be able to mirror the speakers ability to both produce and
understand an infinite number of grammatical utterances which he has never spoken or
heard before. To illustrate this point, consider this classical pair of sentences:
1) Colorless green ideas sleep furiously.
2) Furiously sleep ideas green colorless.
Despite the fact that neither sentence has meaning, every native speaker of English can
recognize that the first sentence is grammatical, while the second is not. An adequate
grammar of English should be able to account for this ability. Moreover, such examples
suggest another important linguistic property that the native speaker can distinguish
grammatical and ungrammatical sentences even though the utterances themselves are
devoid of semantic meaning. Accordingly, Chomsky concluded that a grammar model
should be based on syntax rather than on semantics. Syntax is an independent component
of a grammar system and it is the primary one.
In Syntactic Structures, Chomsky examined phrase-structure grammar also called
IC grammar. He showed its limitations and for the first time, he proposed a formal,
mathematical phrase-structure grammar model. Chomsky demonstrated that language,
like mathematics, is a system in which a finite number of rules are able to generate an
infinite number of correct results sentences. Chomsky s contributions in Syntactic
Structures can be summarized as follows: 1) he introduced a precise, mathematical
method of writing grammar rules, 2) he added a third level, the transformational level, to
grammar theory. The model generative grammar proposed by him in Syntactic Structures
was a three-level, rule-based system: 1) The level of phrase structure
2) The level of transformational rules
3) The level of morphophonemics

The rules of the phrase-structure level were supposed to generate a finite set of
underlying terminal strings, that is, strings with phrase structure interpretations.
Transformational rules then mapped these underlying terminal strings into other strings,
each derived string is assigned a new constituent structure. At that time, Chomsky had no
term for the output of the transformational rules more technical than a string of words.
The introduction of the terms deep and surface structure had to await until a clear
statement of the competence-performance dichotomy was made. Meaning, in this view,
was part of performance and use; it was not part of any linguistic grammatical model.
Consequently, the undeniable semantic correspondences between declaratives and
interrogatives, affirmatives and negatives, actives and passives, etc. were not admitted as
direct evidence for setting up semantically generative transformations relating them.
Rather, their relationship was explained by syntactically motivated analysis, that is, by
the syntactical transformational rules in question. A weak point noticed about this
transformational grammar model is that sometimes transformations change meaning. For
example, the following active-passive pair has different meanings, with the first
quantifier having wider scope in each case:
1) Everyone in this room speaks two languages.
2) Two languages are spoken by everyone in this room.
To summarize, we can notice that at this stage of the development of transformational
generative grammar, semantics and meaning were neglected terms; the notion of deep
structures has not been introduced yet. However, it was admitted that the underlying
string of a sentence might reveal semantically relevant structures that are obscured in the
terminal sting of words. Consider Chomskys classic pair of sentences:
1) John is easy to please, and
2) John is eager to please.
3

The underlying terminal strings of the first sentence predicts that the sentence should be
interpreted as for someone to please John is easy, while the second has to be interpreted
as John is eager (for John) to please (someone).

II. Deep Structure after the Publication of Aspects of the Theory of Syntax
In the second section of this paper, we will trace the development of deep
structure from Syntactic Structures to Aspects. At this stage, the notion of deep structure
started to emerge into the generative grammar model of transformations. Chomskys aims
in Aspects of the Theory of Syntax (1965) became more ambitious in order to explain all
of the linguistic relationships between the sound system and the meaning system of the
language. Chomskys transformational grammar explained in Aspects had three
components: syntactic, semantic, and phonological; the semantic and phonological
components have access to the output of the syntactic component, but not vice versa. The
syntactic component falls into two major sub-parts: a base containing phrase-structure
rules and a lexicon, and transformational rules. Phrase-structure rules state what basic
combinations of syntactic categories are permissible within the sentence using labels like
S, NP, V, and so on. Into the trees created, words from the appropriate categories are to be
inserted from the lexicon. The resulting trees are the deep structures of English. The
second major type of syntactic component is the transformation, which converts trees or
deep structures produced by the phrase-structure rules into other trees; therefore, all
transformations are structure-dependent. The tree that results from the application of a
transformation is a derived structure and the tree resulting from the application of all
transformations is the surface structure. Thus, the syntactic component produces two
outputs for each sentence: a deep structure, and a surface structure. According to
Chomsky, the syntactic surface structure is relevant to the phonological rules, while the
4

syntactic deep structure is relevant to the semantic interpretation of sentences. At the time
of the publication of Aspects of the Theory of Syntax, it seemed that all of the
semantically relevant parts of the sentence, all the things that determine its meaning, were
contained in the deep structure of the sentence. For example, the sentence I like her
cooking has different meanings because it has different deep structures though only one
surface structure. In contrast, the pair the boy will read the book and the book will be
read by the boy has different surface structures but one and the same deep structure.
This theory of the relation of syntax to semantics and phonology can be shown
graphically as follows:
Base Component
PSR
lexicon
Deep Structures

Transformational Component

Surface Structures

Semantic Component

Semantic representation
of sentences

Phonological Component

Phonological representation of sentences

From the diagram above, we can see that the syntactic component is assigned a
prominent role in this grammar model that became known as The Standard Theory. As we
have mentioned before, this model consists of a base that generates deep structures and
transformational part that maps them into surface structures. The deep structure of a
5

sentence is submitted to the semantic component for semantic interpretation, and its
surface structure enters the phonological component and undergoes phonetic
interpretation. The final effect of a grammar, then, is to relate a semantic interpretation to
a phonetic representation (i.e. to state how a sentence should be interpreted). This relation
is mediated by the syntactic component of the grammar, which constitutes its sole
creative part, while the semantic and phonological components are assigned interpretive
roles. This model differs from the previous one in that Aspects of the Theory of Syntax by
Chomsky included a detailed account of the semantic component (projection rules, global
insertion rules, selection restrictions, rules for the interpretation of subcategorization, and
semantic distinguishers). Chomsky explained that the projection rules of the semantic
component operate on the deep structure generated by the base, assigning a semantic
interpretation (a reading) to each constituent, on the basis of the readings assigned to its
parts (ultimately, the intrinsic semantic properties of the formatives) and the categories
and grammatical relations represented in the deep structure (1965: 144). Formatives
such as boy, the, seem, and so on; they can be subdivided into lexical items (boy, girl,
freedom) and grammatical items (perfect, possessive, pronoun, etc.). In Chomskys
Aspects, it was argued that dictionary entries should contain subcategorization and
selectional information. Radford defined subcategorization restrictions as restrictions on
the range of categories which a given item permits or requires as its Complements. For
example, devour takes an obligatory NP Complement, rely takes an obligatory PP
Complement headed by on, and so on. Selection restrictions are the idiosyncratic
properties of lexical items. According to Radford, selection restrictions are
semantic/pragmatic restrictions on the choice of expressions within a given category
which can occupy a given sentence-position (1988: 370). Thus, it follows that the
selectional properties of items should be specified in their lexical entries. For example,
6

murder is a verb that requires an NP Complement. Moreover, its subject (first argument)
and its object (second argument) must be human. These pieces of information of the
lexical entry can be shown as follows:
murder: CATEGORIAL FEATURES: [+V, -N]
SUBCATEGORIZATION FRAME: [NP]
SELECTION RESTRICTIONS: <HUMAN - HUMAN>
The third line of the entry tells that both the NP preceding murder and that following it
denote a human being.
The above elegant picture of transformational generative grammar impressed
many linguists and scholars for many reasons. First, transformational rules facilitate
economy; for example, we do not have to say, I like that she cooks in a certain way, we
can simply say, I like her cooking. Although we pay a small price for such economies
created by transformations in having ambiguities, they do not hamper communication
because when we talk the context usually resolve the ambiguities. Second,
transformations facilitate communication by enabling us to emphasize certain things at
the expense of others. For instance, we can say only the exam was difficult for Sara to
pass but also it was difficult for Sara to pass the exam and Sara had difficulty passing
the exam. In general, an understanding of transformational rules requires an
understanding of their function in communication since communication is what language
is all about.
As mentioned earlier, Chomskys transformational generative approach to
grammar as claimed in Aspects ran into serious problems. The very appealing view of
transformational generative grammar fell apart when the behaviours of quantifiers and
reflexives were noticed. For example, the sentence: John is eager to win is derived from

the deep structure: John is eager for John to win by means of Chomskys EQUI-NP
transformation:
S.D

S.C

NP

be

ADJ eager

for NP

to VP

obligatory

Condition: 1= 4
This transformation seems valid with the above-mentioned example, whereas it seems
problematic with the deep structure: Everyone is eager for everyone to win supposed to
derive the surface structure: Everyone is eager to win. This deep structure would be better
semantically as (Everyone x) (x is eager for x to win); however, the classical deep
structure developed by Chomsky cannot handle such meanings. Concerning reflexives,
some problems also manifested. For example, the sentence: John voted for himself is
assumed to be derived from the deep structure: John voted for John by means of
reflexivization that can be stated formally as:
S.D

S.C

x NP Y NP Z
1

4
Ref

obligatory

Condition: 2= 4, and both 2 and 4 are in the same sentence

This transformation proves true for the sentence: John voted for himself; however, it is of
a problematic case to account for the sentence: Every candidate voted for himself. This
surface structure cannot be derived from the deep structure: Every candidate voted for
every candidate. The meaning here is not clear; it is ambiguous and illogical.
Such problems of Chomskys classical deep structure led to the development of
two different approaches to semantics within transformational generative syntax. Both of
8

these approaches make no appeal to the level of deep structure as Chomsky defined it.
The first semantic theories designed to be almost compatible with transformational
syntax were interpretive. Interpretive semanticists argued that syntactic rules enumerate a
set of well-formed sentences, each of which is assigned an interpretation by the rules of a
separate semantic theory. This left syntax relatively autonomous with respect to
semantics, and was the approach preferred by Chomsky. According to Smith and Wilson,
interpretive semanticists have successively abandoned the defining assumptions which
underlay the level of deep structure, transferring more and more of the work of the
syntactic rules into the semantic component, so that deep structure has gradually moved
nearer and nearer to the surface syntax of the sentence (1979: 115). The second
approach developed against the existence of the classical deep structure became known as
generative semantics. Generative semanticists argue that there is no clear-cut distinction
between syntactic and semantic rules. Therefore, the level of syntactic deep structure
defined as the input to semantics and syntax cannot hold. Generative semanticists took
Chomskys concept of deep structure and ran with it, assuming that deep structures were
the sole input to semantic representation. This assumption led to the development of more
abstract and complex theories of deep structure than those advocated by Chomsky, and
hence to abandon altogether the notion of deep structure as a locus of lexical insertion.
Generative semanticists suggested as an alternative a new component which contains
both semantic and syntactic rules to replace the transformational component. These rules
take semantic representations as their input and produce surface structures as their output,
with no intervening level of deep structure between semantic representation and surface
structure. The following diagram can illustrate this approach:
Semantic
representatio
n

Semantic &
syntactic rules

Surface
structure
9

On the basis of these two types of argument, we will discuss these alternative
approaches in detail. First, interpretive semanticists presume that deep structures are too
deep for all practical purposes. They claim that deep structures should be seen as lying
much nearer the surface than was thought in Chomskys Aspects in 1965. As we all know,
one of the basic motives behind transformations is to facilitate economy, and hence the
1965 model was developed to assign the same deep structure to different surface
structures generated by transformations. For example, the active and passive sentences:
1) John wrote the book, and 2) The book was written by John are derived from the same
deep structure. The passive sentence is derived from the sentence underlying the active
one by means of the passive transformational rule. However, one argument against this
model is that certain sentences with identical deep structures do not produce identical
surface structures. Kebbe stated, deep structure seems to be unable to account for the
idiosyncratic behaviour of particular words in English (1995: 175). Consider the
following examples given by Wilson and Smith:
1) John caused Bill to die.
2) John killed bill.
3) Mary caused the door to open.
4) Mary opened the door.
5) It is probable that Manchuria will annex Wales.
6) Manchuria will probably annex Wales (1979: 119).
From the sentences given above, it is clear that 2) is derived from 1), 4) from 3), and 6)
from 5). However, the classical deep structure supposed forbids the suggestion of a
transformational relationship between the above-mentioned pairs. In Chomskys standard
theory model, these relationships would be stated, not at the level of deep structure, but in
the lexicon, where information about the idiosyncratic behaviours of lexical items is
10

provided. Therefore, it would be mentioned in the lexicon that for intransitive verbs like
open there was a transitive counterpart, carrying the meaning to make or cause to open.
Similarly, for adjectives such as probable which can occur with a sentential complement,
as in 5), it could be stated in the lexicon that there was a related adverb that could occur
in that sentence which acted as a complement to the adjective, and had the same meaning,
as in 6). There are many reasons for attributing these relationships to the lexicon rather
than to syntactic transformational rules. First, it would be difficult for classical deep
structure to handle such cases without further complicating the syntactic transformational
rules. Second and more important, not every verb or adjective are subject to enter into
such relationships. For instance, it has to be stated for each verb whether it has a
transitive causative form or not. As we have just seen, the verb open has a causative
counterpart, while the verb creak does not. Consider the following pair:
1) Mary caused the door to creak.
2) * Mary creaked the door.
From this pair of sentences, we can see that the second sentence is deviant whereas its
counterpart is well-formed. Similarly, it has to be stated for each adjective whether or not
it has a related adverb. The following sentences illustrate the difference between the
probable/probably stated earlier versus impossible/impossibly structures:
1) It is impossible that Manchuria will annex Wales.
2) * Manchuria will impossibly annex Wales.
From the sentences above, we can see that the first sentence is well-formed, while the
second is deviant.
In addition, it is true that classical deep structure permits a transformational
treatment in the case of passivisation because almost every transitive verb has a related
passive; however, this fact cannot hold for all verbs. It is worth noticing that there are
11

certain English verbs which enter into structures of the type NP V NP, but which do not
have related passives. The following sentences explain this point:
1) The meal costs ten pounds.
2) * Ten pounds are cost by the meal.
3) Sara resembles your sister.
4) * Your sister is resembled by Sara.
5) Fatima became the head of the department.
6) * The head of the department was become by Fatima.
7) The athlete ran two miles.
8) * Two miles were run by the athlete.
The deviant cases in the above list of examples indicate that it is impossible for the
classical deep structure to handle such cases without further complicating the syntactic
component and the lexicon. One way to solve this problem is to state in the lexicon that
for each passive-forming verb if a structure of the form: NP1 V NP2 is well-formed,
so is the corresponding sentence: NP2 be V-ed by NP1. However, when there is no
passive form of a verb, such a statement will simply be omitted. Some linguists objected
that such a treatment would be better if incorporated into the grammar rather than the
lexicon. Any verb which can occur in the active form will also be able to occur in the
passive form, unless it is explicitly marked as being unable to. This would have the effect
of leaving the general case, e.g. write, unmarked, while the exceptions such as run
marked in the lexicon. Consequently, the fact of incorporating information into the
grammar would be considered not only misleading but also redundant, a property against
the ultimate goal of transformations which is economy and simplicity. The same
information of lexical relationships can be captured in the lexicon without any mention of
a level of deep structure. In Wilson and Smith, it was argued that deep structure can be
12

abandoned, not because it is not deep enough to handle all the facts, but because it is
deeper than is necessary (1979: 122).
Moreover, the principle that transformations are meaning-preserving was soon
discovered to be one of the weak points of the Standard Theory model from two different
points of view. First, as we know Passivization as an example of a transformational rule
should not affect the semantic interpretation of the derived passive construction.
However, this principle cannot be true for all sentences. Kebbe (1995) provided two pairs
of examples to illustrate this point and he explained that the sentences: Someone has
brought the flowers, and The flowers have been brought are synonymous, while the
sentences: Beavers build dams, and Dams are built by beavers are not. The latter sentence
can be assigned two interpretations. One is identical to its active counterpart, whereas the
second differs from it. The second reading implies that all dams are built by beavers;
however, this meaning was not obviously suggested by the active form. Cases like this
have proved the inadequacy of the claim that transformations do not change meaning, and
have raised doubts about the validity of deep structure.
The second argument against classical deep structure was based on the belief that
deep structure is not deep enough; this argument is known as generative semantics as
opposed to the argument explained earlier known as interpretive semantics. Generative
semanticists take classical deep structures and carry them a stage further.
It should be recalled that in Chomskys standard theory deep structures contain
actual words inserted from the lexicon, and that once a word has been inserted into a tree,
it cannot be turned into another word. Thus, no deep structure containing the verb give
could be transformed into a surface structure containing the verb receive, for example.
Consider the relationship between the following active and passive pair:
1) Sara gave a book to Lama.
13

2) Lama was given a book by Sara.

These sentences are synonymous because they express the same subject-verb-object
relationships. In both sentences, Sara is the giver, and Lama is the receiver. These facts
are correct because both sentences are derived from the same deep structure; the second
sentence is derived by the passive transformational rule. Similarly, generative
semanticists argue that if transformations are allowed to derive one syntactic structure
from another, there is no reason why we cannot carry the principle a step further to
generate one lexical item from another. To illustrate this point, consider the relationships
between the following sentence pairs:
1) John gave a book to Bill.
2) Bill received a book from John.
3) I consider your suggestion insulting.
4) Your suggestion strikes me as insulting.
5) Maria sold Harry an icebox.
6) Harry bought an icebox from Maria.
7) I liked the play.
8) The play pleased me.
9) John struck Bill.
10) Bill received a blow at the hands of John.
Clearly, there is a meaning relation between each pair to the extent that each of these
pairs may be regarded as synonymous. The first pair of sentences, for example, expresses
the fact that John is the giver and Bill the receiver of the book. Accordingly, these pairs
should be given common deep structures, and their relationships should be expressible in
transformational terms, as is possible, for example, in these cases:
1) i) John is easy for us to please.
14

ii) It is easy for us to please John.

2) i) It was yesterday that he came.
ii) He came yesterday.
In the above examples, the deep structures of the paired sentence are identical in all
respects relevant to semantic interpretation. In these sentences, transformations account
for their synonymous relationships. Therefore, from the way deep structure was defined,
only the above pairs can be given a common deep structure because they contain only the
same words. The sentences from (1) to (10) cannot be dealt with in this way, since they
contain different words. The criticism made by generative semanticists is that Chomskys
classical deep structure misses the obvious relationships between the pairs of (1)-(10).
They claim that deep structure should be more abstract to allow us to derive all
synonymous pairs from a common source, even if this means adopting the claim that
deep structures do not contain actual words but a representation of the meanings of
words.
Furthermore, another argument for having a deeper and more abstract deep
structure may be furnished by the implicit assumption of the classical deep structure
itself. Classical deep structure argues that every sentence with two or more distinct
meanings correspond to two or more distinct deep structures. Given this assumption that
an ambiguous surface structure calls for two different deep structures, we are
immediately committed to finding two deep structures for the sentence: John wants to fly
an aeroplane over the North Pole. This sentence may be assigned two readings, one of
which means that there is a particular aeroplane which John wants to fly over the North
Pole, and no other will satisfy him. The second reading can be interpreted as he wants to
fly an aeroplane over the North Pole, but any one will do. If these two different
interpretations reflect differences in meaning, then, we need two different deep structures
15

to account for this fact. However, classical deep structure analysis gives only one deep
structure for the sentence, and both interpretations seem to be syntactically identical. To
provide different deep structures, we must go deeper than classical deep structure, and
provide such paraphrases to capture the differences in meaning that can be conveyed by
the above sentence:
1) There is an aeroplane which John wants to fly over the North Pole.
2) John wants to fly some aeroplane or other over the North Pole.
Again, the conclusion is that deep structures are much deeper than Chomsky classical
deep structure and that linguistic theory must be revised accordingly.
As a result, the arguments against classical deep structure led most linguists to
abandon it opting for the generative or interpretive approach. However, it would be a
mistake to think that the notion of transformational grammar is a thing of the past and
should be completely abandoned. Classical deep structure has revealed a great wealth of
facts and assumptions for the new theories to come.
By 1972, more revisions in Chomskys Aspects model took place so that to
overcome the problems mentioned above and this led to the development of the extended
standard theory. The extended standard theory is the model of grammar without
autonomous syntax. There were several reasons for modifications. First, what was
referred to as the semantic representation of a sentence was no longer seen as a single,
uniform structure. This theory assumes that the syntactic component interacts with the
semantic component many times during the processing of syntactic structures. Before the
application of transformations, the deep structure enters into the semantic component
where it is interpreted in terms of its functional structures. At this stage, the semantic
component provides information on the interpretation of the various semantic roles used
in the sentence such as agent, patient, goal, experiencer, etc. This deep structure continues
16

to be processed syntactically as it undergoes various cycles of transformations. This

information is again directed to the semantic component for further interpretation. This
time the semantic component provides information on modal structures including the
scope of negation, quantifiers, and a table of coreferences. To explain scope relations,
consider the following VP: not wholly understand it. Here, we can say that not has scope
over the VP: wholly understand it. To shed light on the notion of coreference, the
following sentences can be given a table of coreferences in the following way:
1) Mary saw herself in the mirror. (the pronoun here is co-referential)
2) Mary saw her in the mirror. (the pronoun here is not co-referential)
3) Mary thinks that she is attractive. (the sentence here is ambiguous because there is
no specification of referentiality).
These rules could not be processed at the deep structure level; the process has to be
delayed until certain modifications and rule applications take place within the
transformational component of the language. Finally, at the surface structure level, the
output of the final transformational cycle is sent to the semantic component for further
processing. At this stage, the semantic representation is analyzed for focus,
Base Component
presuppositions,
and topicalization. The presupposition of a declarative sentence, for
Semantic Representation
PS Rules
LexicalisInsertion
example,
concerned with what the speaker assumes the hearer knows. Focus, on the
other hand, has to do with what the speaker assumes that the hearer does not know. Words
Functional Structures
assumedDeep
to beStructure
focused have extra heavy stress. Topicalization differentiates between old
and new information. The following examples illustrate
this point:
Modal
Structures
Scope of Negation
Scope of Quantifiers
Component
Table of Coreferences
TheCyclic
speaker
presupposes that Mary exists, Mary did something, and Mary drank
Transformations
Final Cycle Transformation
something. The focus here is on the MILK in that Mary drank milk and not water.
1) Mary
drank the MILK.
Transformational

Focus
Concerning topicalization, we can say that it is old information
that Mary did something,
Presupposition
Topicalization
17
Surface Structure

while the new information is that Mary drank MILK. The model of the extended standard
theory described as the model of grammar without autonomous syntax can be given the
following diagram:
Base Component
PS Rules
Lexical Insertion

Deep Structure

Transformational
Component
Cyclic Transformations
Final Cycle Transformation

Surface Structure

Semantic Representation

Functional Structures

Modal Structures
Scope of Negation
Scope of Quantifiers
Table of Coreferences

Focus
Presupposition
Topicalization

After the development of the above-mentioned grammatical theory, several attempts have
been made to render syntax as autonomous. Some linguists proposed that full lexical
forms are assumed to exist at the deep structure level. Reflexivization, for example,
requires semantic information on coreferentiality to transform John saw John into John
saw himself. In the case of Equi-NP Deletion, coreferentiality is also needed to transform
Mary expects Mary to win into Mary expects to win. However, it is possible to avoid
these references to the semantic component if pronouns and dummy subjects (gaps)
already exist at the level of deep structure. This introduction of abstract elements and
empty categories into the deep structure of sentences marked an important turning point

18

in linguistic theory. In fact, it led to the emergence of Government and Binding GB

theory.

III. Deep Structure within GB Theories

The development of GB theory as a grammar model began in 1977 when Chomsky
and his colleagues proposed some major revisions in the extended standard theory. Since
the concepts of deep structure and surface structure were substantially revised by the
inclusion of abstract elements, trace elements, and empty categories, they were given new
terms, namely D-structure and S-structure respectively. These are related to each other by
means of movement rules. S-structures are further developed into phonetic forms PFs and
Logical forms LFs. Thus GB-theories identify four levels of grammatical representation:
D-structure (DS), S-structure (SS), Logical form (LF), and Phonetic form (PF). Dstructure can be described as the phrase marker at which pure grammatical functions and
thematic relations may be represented. According to Hornstein, Nunes, and Grohman
DS is where an expressions logical/thematic role perfectly coincides with its
grammatical function GF: logical subjects are DS (grammatical) subjects, and logical
objects are DS (grammatical) objects, etc. (2005: 20). At DS, positions that are
thematically active must be filled and positions with no thematic role must be left empty.
To illustrate this point, consider the verbs in the following sentences:
1) John persuaded Sara to interview Mary.
2) John seems to like Mary.
3) It seems that John likes Mary.
In the first sentence and in terms of thematic relations, persuade requires a persuader
(agent), a persuadee (theme or patient), and a propositional complement, while interview
requires an interviewer and an interviewee. Given that 1) is an acceptable sentence, each
19

of these -roles must correspond to filled positions in its DS representation illustrated as

follows: [ John persuader persuaded Sara persuadee [ec interviewer to interview Mary
interviewee

] proposition ]. We can notice here that once we assume the notion of DS, its

representation must have a filler in the position associated with the interviewer -role,
although it is not phonetically realized. In GB, the empty category realized as ec is an
obligatory controlled PRO, whose antecedent is Sara. In the sentences 2) and 3), the verb
like has two -roles to assign: the liker (agent), and the likee (theme or patient), whereas
seem has only one -role to assign to its propositional complement. Specifically, seem
does not assign a -role to the position occupied by John in 2), as can be seen by the fact
that this position may be filled by an expletive as in 3). This means that John was not
base-generated but rather it must have gotten there transformationally. Thus, the matrix
subject position of the DS representation of 2) is filled by nothing at all, not even a null
expression. This can be shown as follows where represents an empty position:
[ seems [ John liker to like Mary likee ] proposition ]
As regards the functional characterization, DS is defined as the starting point for a
derivation. It is the phrase marker that is the output of phrase-structure operations plus
lexical insertion and the input to transformational operations. Thus, it is the locus of a
grammars recursivity. S-structure is the point at which the derivation splits, sending off
one copy to PF for phonetic interpretation and one copy to LF for semantic interpretation.
It is regarded as the queen of GB-levels. PF and LF are interface levels within GB. They
provide the grammatical information required to assign a phonetic and semantic
interpretation to a sentence. This model of grammar can be given the following diagram:

20

D-Structure
Movement Rules

S-Structure

Phonetic Form
Phonological
Rules

Logical Form
Semantic Rules

GB embodies a very simple transformational component. It includes two rules: Bind and
Move. Bind allows free indexing of DPs determiner phrases, and Move allows anything
to move anywhere anytime. The sentence is now represented as an inflectional phrase IP.
The two very general rules of the transformational component over-generate
unacceptable structures. Therefore, GB-grammars deploy a group of information-specific
modules that interact to exclude unwanted structures and prepare a phrase marker for
interpretation at LF and PF. -roles (Theta theory), phrase-structure (X- theory), case
theory are examples of the most essential modules with GB theory. However, as a deeper
understanding of the structure of specific constituents was achieved, many of these
modules lost empirical motivation. For example, the X-theory at D-structure had to be
revised form a minimalist perspective. As was mentioned, DS within GB is described as
the level where lexical properties meet the grammar. Logical subjects are syntactic
subjects, logical objects are syntactic objects, etc. Two grammatical modules known as
Theta theory and X theory govern lexical properties and phrasal structures at DS. As was
shown in the examples above, Theta theory ensures that only thematic positions are filled.
X-theory ensures that the phrasal organization of all syntactic objects has the same
21

general format. XPs consist of a head element within a phrasal construction. The head is a
lexical category and the phrasal component is called a projection. For example, N is the
head in an N-bar, and N-bar is the intermediate projection of NP or N-double bar. The
diagrams below illustrate this point:
Maximal Projection

N-Double Bar

Intermediate Projection

Spec

Base Category

the

N-Bar
N
president

Many arguments against DS within GB were stated and this led to the revision of
this grammatical level. First, DS within GB is the place where grammatical recursion
occurs. A language universal is that sentences can be of arbitrary length. This potential
within GB is captured at DS by allowing a category A to be embedded within another
category of type A. Consider the following examples to clarify this point:
1) i. [ a tall man]
2) ii. [a tall bearded man]
3) iii. [ a tall bearded man with a red shirt]
In terms of functions, DS within GB is the output of phrase-structure operations and
lexical insertion, and the input to movement operations. However, some arguments
against GBs DS showed that recursion would appropriately occur without any reliance
on DS, as we will see later.
Another weak point about DS relates to the fact that DS within GB is the pure
representation of thematic properties in that all lexical/thematic properties must be
satisfied there. For instance, if we take the control structure whose subject is understood
22

as playing a semantic role with respect to both the control and the embedded predicate
such as Sara hoped to interview Chomsky. Here, the verb hope requires a proposition for
a complement and a hoper for its external argument. By the same token, the embedded
verb interview must discharge its interviewer and interviewee -roles. The subject
position associated with interview must be filled at DS although there is no phonetically
realized element to occupy this position. IN GB, this position should be filled by the
empty category PRO which is coindexed with the matrix subject so that Mary here plays
two different semantic roles. DS can be stated formally as follows:
[ Mary hoper hoped [ PRO interviewer to interview Chomsky interviewee ] proposition]
As regards raising structures whose subjects are interpreted as playing only a role
associated with the embedded predicate, the sentence Sara seemed to interview Chomsky
may illustrate the role of DS within GB. In the example, the verb seem takes a
proposition for a complement, but its subject position in non-thematic. Thus, Sara cannot
occupy this position at DS. However, the embedded verb interview assigns two -roles.
The DS representation of this sentence can be stated formally as follows:
[ seemed[ Sara interviewer to interview Chomsky interviewee] proposition]
These two simple examples argue for the existence of DS to account for the
establishment of thematic relations; however, arguments against DS rely on the fact that
Theta-Criterion is furnished at DS and due to the Projection Principle at SS and LF.
According to Radford, Projection Principle can be defined, as syntactic representations
[i.e. syntactic structures] must be projected from the lexicon, in that they observe the
subcategorization properties of lexical items (1988: 391). Assuming that LF is the input
to rules mapping to the semantic interface; therefore, notions like agent, patient, etc. are
encoded at this level. Moreover, the Projection Principle ensures that some kinds of

23

information are preserved during derivation by inspecting them at subsequent levels of

representation. Thus, the thematic relations encoded at DS are a subset of the ones
encoded at LF, and the Projection Principle makes the system redundant. Minimalists
argue against this redundancy by eliminating DS form grammar, as we will see later.
Third, a serious empirical problem for DS as conceived by GB is posed by toughconstructions such as in the sentence: Moby Dick is hard for Bill to read. If we replace
Moby Dick with these books as in the sentence: These books are hard for Bill to read, the
agreement features of the copula change indicating that these elements occupy the matrix
subject position of their sentences. The synonymous sentence: It is hard for Bill to read
Moby Dick indicates that Moby Dick is related to the embedded object position; that is, it
is understood as the thing being read. The problem here is that it is quite unclear what
sort of movement this could be. Some claim that it is a trace of A-movement that can be
illustrated as follows:
1) [ Moby Dicki is hard [ for Bill to read ti]]
ti is an anaphor and should be bound within the embedded clause in order to comply with
Principle A of Binding Theory. Principle A states that an anaphor (e.g. a reflexive or
reciprocal) must be bound in its domain. Since t i is unbounded in this domain, then it
cannot be A-movement. Chomsky suggest that it is an instance of A-movement with a
null operator OP moving close to the tough-predicate and forming a complex predicate
with it. Its structure can be shown as follows:
2) [ Moby Dick is [ hard [ OPi [ for Bill to read ti ] ] ] ]
Movement of the null operator allows the formation of the complex predicate [ hard [ OP i
[ for Bill to read ti ] ] ], which is predicated by the subject Moby Dick. Therefore, the
matrix subject position in the sentence is a -position because Moby Dick receives a role under predication. Both analyses prove problematic for DS concerning the thematic
24

status of the matrix subject in 2). The sentence: It is hard fro Bill to read Moby Dick
revealed that the matrix subject of a tough-predicate is not inherently a -position since it
can be occupied by an expletive. This implies that the matrix subject position in 2) is only
a -position after A-movement has been applied and the complex predicate has been
formed. If Moby Dick is inserted at DS, then it is not inserted at the point when the
matrix subject is a -position. If after the null operator has moved, the conclusion then is
that we can have insertion into a -position after DS. Both ways are problematic and
require the revision of the status of DS as a grammatical level as we will see within the
minimalist approach.

IV. Deep Structure from the Perspective of Minimalism

As was explained, DS is the place where grammatical recursion occurs. In terms of
functions, DS within GB is the output of phrase-structure operations and lexical insertion,
and the input to movement operations. Although the notion of DS was very important
within GB, it became questionable after the development of minimalism. In 1990s,
Chomsky introduced a new grammar model known as minimalism in which DS and SS
no longer featured and only PF and LF remained as the only levels of grammatical
representation for many reasons.
First, as we have seen DS is the level where recursion occurs. However,
minimalists argue that recursion would not be lost after the elimination of DS. Their
argument was based upon the principle that earlier approaches to UG adequately captured
recursion but did not suggest DS. Within the minimalist approach, the operation merge
can ensure recursion in a system that lacks DS because it takes the lexical items together
and puts them into phrasal structures that comply with X-theory. For example, merge
takes two lexical items saw and Mary to form the VP, this VP is then merged with Infl to
25

form the I. Further applications of merge apply to arrive at the IP: John said that Bill saw
Mary, for example. This sentence is a standard example of grammatical recursion because
it involves a VP embedded within another VP, an I embedded within an I, and an IP
embedded within another IP. Therefore, we can say that recursion can be captured
without any mention of DS.
Second, we have seen that within GB, the Projection Principle ends up rendering
the system intrinsically redundant. From the perspective of minimalism, this redundancy
can be eliminated by assuming that the Theta-Criterion holds at the level of LF. To
explain this point, we need to illustrate that the differences between raising and control
structures can be accounted for, even though their thematic properties are assumed at LF.
For example, the fact that Sara is understood as both hoper and interviewer in the
sentence Sara hoped to interview Chomsky can be captured by the structure in 1), but not
by the one in 2).
1) [ Sarai hoped[ PROi to interview Chomsky]]
2) *[ Sarai hoped [ ti to interview Chomsky ]]
If we argue for the existence of DS and assume that the Theta-Criterion holds at this
level, we are obliged to choose the representation in 1) because in 2) Sara is not at the
matrix subject position at DS and the Theta-Criterion is violated at this level as shown
below:
DS: *[ hoped [ Sara interviewer to interview Chomsky interviewee ] proposition ]
However, according to minimalists the elimination of DS does not interfere with the fact
that 1) still illustrates the adequate representation of the sentence Sara hoped to interview
Chomsky by exploring the different empty categories that each structure employs. The
difference between PRO in 1) and the trace in 2) indicates that -relations must be

26

established by lexical insertion and cannot be established by movement. This assumption

does not necessarily presuppose DS. As was seen earlier, recursion/generativity, within
the minimalist approach, is captured by the operation merge. Moreover, the principle
known as the Theta-Role Assignment Principle (TRAP) states that -roles can only be
assigned under a merge operation (Hornstein, Nunes, and Grohmann, 2005). TRAP is not
specific to any level of representation. According to this principle, the structure in 1) is
well formed because the interviewer -role was assigned to PRO when it was merged
with the embedded I and the hoper -role was assigned to Sara when it merged with the
matrix I. Thus, when the Theta-Criterion applies at LF, the derivation will be judged
convergent. By contrast, although Sara can receive the interviewer -role in the above 2)
when it merges with the embedded I, it cannot receive the hoper -role because it is
connected to the matrix clause by move and not by merge. According to Chomsky
(1995), the movement operation means that a category can be moved to a target position.
He goes on to state that movement of an element always leaves a trace and, in the
simplest case, forms a chain (, t), where , the head of the chain, is the moved element
and t is its trace (1995: 43). Therefore, here once the hoper -role has not been fulfilled,
2) violates the Theta-Criterion at LF and the derivation fails.
Concerning the raising construction, the same reasoning applies. For example,
Sara seemed to interview Chomsky can be given the LF representation in 3), but not the
one in 4). 3) and 4) can be stated formally as follows:
3) LF: [ Sarai seemed [ti to interview Chomsky]]
4) LF: *[ Sarai seemed [ PROi to interview Chomsky]]
3) is well formed because Sara receives its -role when it merges with the embedded I
and moves to a non-thematic position. In contrast, in 4) Sara receives no -role when it
merges with the matrix I, violating the Theta-Criterion and causing the derivation to fail
27

at LF. From the above examples, we can see that the TRAP accounts for the well
formedness of raising and control structures with out assuming that we need a level like
DS.
Third, as was noted earlier DS within GB is problematic in tough-constructions;
however, they become non-problematic if we dispense with DS. For instance, Hornstein,
Nunes, and Grohmann explained this point by showing how the many applications of
Move and Merge operations can account for the sentence: Moby Dick is hard for Bill to
read. They went on to show how the derivation of the latter sentence can be traced as
follows: a. Applications of Merge
[C for Bill to read OP]
b. Move OP
[CP OPi [for Bill to read ti ] ]
c. CP + Merge hard
[AP hard [CP OPi [for Bill to read ti ] ] ]
d. AP + Merge is
[I is [AP hard [CP OPi [for Bill to read ti ] ] ] ]
e. I + Merge Moby Dick
[IP Moby Dick is [AP hard [CP OPi [ for Bill to read ti ] ] ] ] (2005: 67).
In this sentence, read merges with the null operator and after further applications
of Merge, we obtain C in a. In b. the null operator moves to derive the complementizer
phrase CP. Then, CP merges with hard as shown in c. After that, the adjective phrase AP
merges with is to form I forming a complex predicate that can assign a -role to the
28

external argument (Moby Dick). Finally, when Moby Dick merges with I, it becomes the
matrix subject and it will be -marked according to the Theta-Role Assignment Principle
(TRAP) stated earlier. Therefore, it seems that we can provide an adequate account of
simple and complex predicates of tough-constructions if we eliminate DS and this is the
strongest argument against DS.
In conclusion, in the above sections we have examined the evolution of the notion
of deep structure throughout the development of transformational generative grammar.
DS was an essential feature until the development of the minimalist approach. Within this
approach, it has been argued that we can dispense with DS without any loss of
information for the sake of economy and simplicity which constitute the ultimate goals of
minimalism as an adequate grammar model.

29

References:
Chomsky, Noam. (1995). The Minimalist Program. Cambridge, Massachusetts: The MIT
Press
_____________. (1993). Lectures on Government and Binding: The Pisa Lectures.
Berlin, New York: Mouton de Gruyter
_____________. (1965). Aspects of the Theory of Syntax. Cambridge, Massachusetts:
The MIT Press
Hornstein, N., Nunes, J. and Grohmann, K. (2005). Understanding Minimalism.
Cambridge: Cambridge University Press
Kebbe, M. Z. (1995). Lectures in General Linguistics: An Introductory Course.
Radford, Andrew. (1997). Syntax: A Minimalist Introduction. Cambridge: Cambridge
University Press
______________. (1988). Transformational Grammar: A First Course. Cambridge:
Cambridge University Press
Smith, N., and Wilson, D. (1979). Modern Linguistics: The Results of Chomskys
Revolution. England: Penguin Books

30

31