The Generative Enterprise

THE GENERATIVE ENTERPRISE1
José Borges Neto2

(UFPR/CNPq)
In 1982, Riny Huybregts and Henk van Riemsdijk published a long interview
with Noam Chomsky, organized as a discussion of different issues in the history and
structure of the linguistic theory. The book containing that interview was given an
extremely adequate title: ‘The Generative Enterprise’ (see Chomsky 1982a). I share the
interviewees’ opinion that the linguistic theory, proposed, developed and permanently
‘protected’ by Chomsky’s extraordinary presence can in fact be understood as an
‘enterprise’ – a collective one - having Chomsky as its unequivocal leader.
My aim here is to demonstrate that the linguistic theory widely known as
Generative Grammar (among other names it has been given at different times) is an
extremely coherent Scientific Research Program that starts to be built around the mid
XXth century and that has become, from its early days, a way of understanding the
structure of the human language that can be questioned – an appropriate procedure in
the development of scientific theories – but that cannot be ignored.
I shall consider the roughly 50 years of Generative Grammar (hereafter GG) as
a period of time in which a Scientific Research Program is constructed, with roughly
the same meaning the term has in the proposal presented by the Hungarian philosopher
Imre Lakatos for approaching the history of sciences in general (see Lakatos 1978).
Although I will not adopt an orthodox lakatosian approach, I believe Lakatos’s ideas
allow for a better quality understanding of what actually happened to GG during the last
50 years.
Therefore, the present work is not a strictly linguistic work. Better, it is a work
that fits the areas of History and Philosophy of Science3, directed to the analysis of a
specific case: the ‘internal’ history of GG.
1. Lakatos’s methodology.
I shall start the presentation of Lakatos’s methodology quoting Feyerabend:
Let me now present in its entirety the picture of science which I think
should replace Kuhn’s account.
This picture is the synthesis of the following two discoveries. Fist, it
contains Popper’s discovery that science is advanced by a critical
discussion of alternative views. Secondly, it contains Kuhn’s discovery of
the function of tenacity which he has expresses, mistakenly I think, by
postulating tenacious periods. The synthesis consists in Lakatos’
assertion … that proliferation and tenacity do not belong to successive
periods of the history of science, but are always copresent.
(Feyerabend 1974 p. 211)
1
Published in Portuguese as “O empreendimento gerativo” In Mussalim, F. & Bentes, A.C. (eds) 2004. Introdução à
Lingüística, vol 3. São Paulo: Cortez.
2
I would like to thank my colleagues Maria Cristina de Figueiredo Silva and Evani Viotti. for the comments, ever
pertinent, they made on a first version of the present work; and also the editors Fernanda Mussalim and Anna
Christina Bentes for their welcome suggestions and comments. Of course, the responsibility for the final results is
mine.
3
As Lakatos (1971:102) says, paraphrasing Kant: ‘Philosophy of Science without history of science is empty; history
of science without philosophy of science is blind’.
The Generative Enterprise 2
Let us start with the concepts of proliferation and tenacity.

In Lakatos, Proliferation means that it is desirable to have theories in
competition. According to him, the history of science has been – and so it should be– a
history of scientific research programs in competition.
Tenacity, in Lakatos, means that the scientist does not abandon a theory that
has been refuted, as Popper would like it to happen but, on the contrary, he does all he
can to maintain it either ignoring the counter-examples or reanalyzing them in such a
way as to transform them into validating evidences of the theory. In his own words:
Nature may shout no, but human ingenuity – contrary to Weyl
and Popper – may always be able to shout louder. With
sufficient resourcefulness and some luck, any theory can be
defended “progressively” for a long time, even it is false
(Lakatos 1971, p. 111)
According to Lakatos, the best way of starting out the ‘science game’ is not
with a falsifiable hypothesis, but with a Scientific Research Program (SRP) consisting
of basically a nucleus and a heuristics.
The nucleus of a SRP is a set of propositions that, based on methodological
decisions, are taken as ‘untestable’, that is, propositions that are sometimes said to be
‘metaphysical’ and that reveal the point of view that will orient the approach to the
subject, the very definition of the object of study, etc.
The heuristics of a SRP is a set of methodological rules that tells us which
directions must be taken in searching for scientific ‘explanations’. The heuristics is a
kind of ‘development policy’ of the program, that is, a selection and ordering of
problems, a plan that leads to a progressive sophistication of the explanatory models. It
is a plan that establishes a sequence of reality simulation models, ever more and more
complex, deep, and comprehensive.
Lakatos offers as an example of the action of such heuristics the developing
process of the Newtonian program. Newton initially developed a model for a planetary
system having only one planet gravitating around the sun in which both the sun and the
planet were referred to as points. In that model, he managed to obtain the inverse square
law for Kepler’s ellipse. The third law of dynamics, however, forbade such extremely
simple model and Newton replaced it by another model in which both the sun and the
planet rotated around the gravity center of a system formed by both. Next, Newton
adapted the model in order to include more planets, admitting, however only
heliocentric forces, but not interplanetary forces. Subsequently, he worked on the
possibility of the sun and the planets being spheres and not points. This stage of
development of the program demanded overcoming enormous mathematical difficulties.
Problems solved, Newton started to work with rotating spheres and their oscillations.
He accepted the interplanetary forces and started to work with perturbations. Later on,
he worked with irregular planets instead of spherical planets, coming closer and closer
to the real planetary systems4.
For Lakatos, then, the program progresses through the development of a series
of models that result in creative shifts in the heuristics, in other words, in revisions
made in the program’s ‘developing plan’.
4
For a more detailed presentation of the present case, see Lakatos 1970, p. 50-51.
2. GG: a Scientific Research Program.
The idea I have defended for some time now is that GG is an SRP and not a
linguistic ‘theory’. I am not going to justify this position here (see Borges Neto 1991 for
details). On this line of thought, instead of discussing alternative proposals developed
by such and such linguists, it is more interesting to follow the major directions the
program assumes, through its rules and heuristics. It is also interesting to see the
creative shifts that from time to time redirect the efforts made by the scientists linked to
the SRP.
Such is the investigation we intend do carry out within the scope of the present
work.
2.1. The chomskian SRP: nucleus and heuristics.
Without further discussion, I would like to propose that the nucleus of the GG
consists of the following statements:
(1) The effective linguistic behavior (utterances) is, at least partially,
determined by states of the mind/brain.
(2) The nature of the states of the mind /brain, partially responsible for the
linguistic behavior, can be captured by the computing systems that form
and modify representations5.
I believe these two statements adequately summarize the concept of human
language that has presided the fifty years of chomskian thought.
The GG program’s heuristics determines that the linguist’s basic task is to
create computing systems that may serve as model for the speaker/hearers’ linguistic
knowledge of the language. These computing systems must be understood as
explanatory hypotheses, and their empirical consequences must be evaluated in a
deductive system.
Despite the enormous difference between the analyses actually proposed for
the natural languages phenomena at different points in time in the history of GG (the
proposed computing systems), the general aim of chomskian linguistics has been
remarkably consistent during all these years. We can say, in general terms, that
Chomsky has obsessively pursued the same objective for 50 years, although from time
to time he replaces the conceived theoretical device in order to attain the major task of
his linguistics conception.
Deep down, what generative grammar intends to do is to construct a computing
device, capable of forming and transforming representations, that can ‘simulate’ the
linguistic knowledge a speaker of a natural language has ‘registered’ in his mind/brain.
It is this ‘nucleus’, constantly present in the 50 years of generative grammar
history, that allows us to say we are before one and the same research program, in spite
of the various deep changes the theoretical device (the computing system) has
undergone.
2.2 The first proposal of generative grammar: LST and SS.
It is largely accepted that the GG history starts in 1957, with the publication of
5
The term ‘representation’ assumes many meanings in the linguistic literature. I am using it here to refer to formal
objects of the theoretical construct that correspond to the things that belong to the modeled ‘reality’. In other words,
the states of the mind/brain are, in our case, represented by expressions of a formal language.
Syntactic Structures (Chomsky 1957 – hereafter SS). Chomsky’s previous works –

either because they circulate only among non-linguists, or because they had a limited
circulation – exerted little influence in the development of the chomskian program.
Chomsky’s M.A. dissertation (1951), for instance, was almost completely ignored by
the linguistic community, although it has called some attention outside that area6. Other
articles prior to that (Chomsky 1953 and 1955a) were more oriented to logicians and
philosophers than to linguists: the former was published in the Journal of Symbolic
Logic, a periodical rarely read by linguists; and the latter, although published in
Language, supported a dispute with the Israeli logician Yehoshua Bar-Hillel about the
applicability of symbolic logic developments to linguistic studies.
Around the mid-fifties, Chomsky finished writing an extremely pretentious
book – The Logical Structure of Linguistic Theory (Chomsky 1955c – hereafter LSLT)
– in which he put together - though as separate chapters - his Master’s dissertation and
his PhD thesis (Chomsky 1955b) and cast the basis of a ‘new linguistics’. The book did
not succeed in raising the editors’ interests and remained filed in microfilm form till its
publication in 1975, as if it had just a historical value. In Chomsky’s words:
As for how LSLT was received, there is not much I can say about it. I
have already mentioned that I was not aware I was doing linguistics.
Therefore, the lack of interest or reaction from linguists was not
surprising to me. I put LSLT forth to the MIT Press for consideration…
but they refused it. They were right, I believe, because the contextual
forces at the time were rather unfavorable for a general book on a
subject like that. I also submitted a technical article on a limited part of
the question to the Word magazine - following Roman Jakobson’s
suggestion - but it was refused and sent back to me by mail. At the time
I had little hope about having such work published, at least in a
linguistics magazine, but quite frankly, I did not care much about that.
(Chomsky 1977, p. 121)7
Then we come to 1957 and to Syntactic Structures. This book is a collection of
notes from an undergraduate course Chomsky taught at the MIT and, according to
Chomsky himself, does not reflect fairly his linguistic thoughts at that time. In his
words:
You know what Syntactic Structures was. It was course notes for an
undergraduated course at MIT. Van S showed up here once and took a
look at some of my course notes for the undergraduate course I was
teaching and said I ought to publish it. Since I had not been published
anywhere, I said, why not, and that is what Syntactic Structures was. In
fact, Syntactic Structures is a very misleading book.
(Chomsky 1982a, p. 63)
Chomsky’s ideas only start to spread among linguists after the publication of
Syntactic Structures (SS hereafter) and, mainly, after a long review that Robert Lees - a
linguist who had a certain prestige in the community - published in Language (Lees
1957). So, the theory of this first period came to be known as ‘Syntactic Structure
Theory’.
6
And this caused Chomsky to open his heart and say that his work called Bar-Hillel’s attention, for example, who
said he was extremely interested in it. Also Quine were interested in the methodological aspect. But among linguists,
no one showed any interest in that type of work. (Chomsky 1977, p. 120).
7
Translated from the Portuguese version by the author.
At this point, the nucleus of the chomskian program has already been fairly
established, although not yet fully explained. The best way of capturing this nucleus is
by investigating the main disagreements that Chomsky and some of his disciples
understood to exist between their proposals and the proposals of the mainstream
linguistics at the time: the American structuralism from a bloomfieldian lineage.
Chomsky’s proposal, embodied in the SS theory, differ from structuralism in
some important points:
First, the object of study of structuralism was language, understood as ‘the
totality of statements that can be uttered in a linguistic community’, according to
Bloomfield (1926, p.47). It was the linguist’s task to describe that language, and that
would be done from the starting point of a ‘representative corpus’ of raw language data,
which was described in detail with the tools offered by the ‘discovery procedures’.
What calls Chomsky’s attention is the need to assume the existence of something prior
to the structuralists’ language: the capacity the speakers have to produce exactly the
statements that can be produced. In other words, Chomsky moves the fundamental
question of linguistic theory to determining the rules that govern those ‘representative
corpora’, which lose their status as starting point of the linguistic theory and become its
target. For Chomsky, the linguistic community has a shared knowledge about those
utterances that can and those that cannot be produced, and it is exactly this knowledge
that must be described and explained by the linguistic theory. The ‘representative
corpus’ is the result of this knowledge and to take it as a starting point is
methodologically uninteresting. According to Chomsky, a good indication of the
existence of such knowledge lies in linguistics creativity: the speakers’ ability of
producing and understanding sentences to which they have not been exposed before.
We can then say that one could devise in Chomsky’s initial works a
psychological object for linguistic studies, although Chomsky did not put it explicitly
like that. In other words, ‘the state of mind/brain’ mentioned above had already been
glimpsed.
Another point of disagreement between American structuralism and the SS
theory concerns the definition of the aims of those theories. While the structuralist
theories were, generally speaking, explicitly descriptive, the SS theory intended to be
explanatory, in the sense that the phenomenon should be deduced from a set of general
principles8. The adoption by Chomsky of a hypothetical-deductive model of science
presents deep implications to his program’s procedures. It is no longer a question – as in
structuralism – of describing data that reveal themselves to the linguists’ perception, but
a question of finding the general principles from which the descriptions of observable
data can be logically derived. With Chomsky, the theoretical aspect takes precedence
over the empirical aspect. It is not surprising, however, that a large amount of
Chomsky’s work, at that time, discusses formal languages and tries to define the formal
nature of natural languages as compared to the languages used by logicians and
mathematicians9 . It is the program’s heuristics determining the priorities and, clearly
enough, Chomsky realizes that what is fundamental at that moment is the definition of a
8
Chomsky explicitly assumes a deductive-monological perspective in his characterization of what might be an
explanatory theory (see, for example, Chomsky 1997, p.106).
9
It is in this period that Chomsky develops a classification and a typology of languages that is still used by logicians,
mathematicians and computer scientists – the so-called Chomsky’s Hierarchy. With his hierarchy, Chomsky intends
to show that natural languages present properties that cannot be represented by grammars that would be perfectly
adequate to account for the languages used by logicians. For example, while logicians’ languages can be represented
by formal systems (free-context grammars) that construct expressions without taking the context into account (the
adjacent expressions), natural languages allow to construct expressions that can only be represented by more
powerful formal devices - the transformational grammars.
formal notion of grammar, understood as a computing system (a generative grammar,

in the technical sense of the word), which, like a kuhnian paradigmatic sample, could
be used as a tool for describing natural languages phenomena. In other words, to effect a
proposal of describing the speakers’ implicit knowledge and do so in the frame of an
explanatory theory, Chomsky needs to construct a formal device (a generative
grammar) capable of accounting for the well-formation rules of any language L and of
relating this formal device to some set of general principles (that may determine what
can stand for a ‘generative grammar’ of languages in general).
Then a division emerges classifying the tasks in linguistics into two groups: the
construction of grammar for particular languages, and the construction of general
principles for language competence (‘universals’). The process of constructing
particular grammars requires formal devices that are powerful enough to account for
details and specificities of languages: the process of construction of general principles
must ignore the specificities and search for the ‘universals’. To a certain extent, the
tension between these two processes will be responsible for the changes in the models
of analysis that will recur in time during the last 50 years.
The first model of analysis proposed by Chomsky is presented at length in
LSLT and SS and consists basically of a sophisticated form of immediate constituent
grammar plus a transformational component. This model of analysis presents two main
components: one forming expressions, which is a generative version of the immediate
constituent grammar, developed and presented by Rulon Wells (1947), among others;
and another which transforms expressions, and that, at least in part, stands on the notion
of ‘transformation’ developed by Zellig Harris, who was Chomsky’s tutor for his
doctorate degree10. Besides these basically syntactic components, there is a
morphophonemic component, which attributes phonological readings to the output of
the transformational component.
The basic notion is that of linguistic level.
A language is an enourmously complex system. Linguistic theory
attempts to reduce this immense complexity to maneageable
proportions by the construction of a system of linguistic levels, each of
which makes a certain descriptive apparatus available for the
characterization of linguistic structure. A grammar reconstructs the
total complexity of a language stepwise, separating out the contribution
of each linguistic level.
(Chomsky 1955c, p. 63)
According to Chomsky in LSLT, a linguistic level is an L-system in which
unidimensional representations of utterances are constructed. Each level presents a fixed
and finite ‘alphabet’ of primitive elements. Through a concatenation operation, we can
obtain sequences of elements that will be called chains in L. In the process of linguistic
analysis, a set of chains called L-markers is constructed at each L-level, which will be
attributed to the sentences of the language under analysis. The L-marker of a given
sentence S must contain all the structural information referring to S at L-level. The
relationship between various levels L1, L2, …, Ln is established through mapping
operations, which associate the elements on a level to the elements on other levels. If we
organize the various levels into a hierarchy, we can think of a sequence of mappings
(from L1 to L2, from L2 to L3, and so on so forth up to Ln) till we get to a last level,
which associates L-markers to the sentences of the language.
10
See, for example, Harris 1957.
We will find it necessary to distinguish at least the following levels for

linguistic description: phonemes (Pm), morphemes (M), words (W),
syntactic categories (C), phrase structure (P), and transformations (T).
It is then within the scope of grammar to attribute at least six representations –
one for each level – to each sentence of the language. In other words, the grammar must
assign to each sentence a formal representation (an ‘expression’ of language in which
the computing system is written), on each of the linguistic levels, which may represent
(simulate, model) the properties of the sentence referring to that given level.
A grammar of a language must tell us exactly what are the
grammatical sentence tokens, and exactly how these are represented on
each level.
Let us now look at the form a grammar must take in order to carry out these
tasks. First, it is necessary that the grammar establishes for each sentence-token a
sequence of representations <R1, ..., Rn>, where R1 is a representation of Sentence, Rn is
a phonetic representation and R2, ..., Rn-1 are intermediate representations.
We can generate these representation sequences by rules of the form
(1) X → Y
interpreted as the instruction “rewrite X as Y.” We call each such rule
a conversion. (…) We say that the sequence <R1,…,Rn> is a derivation
of Rn, generated by a set C of conversions, if R1 is Sentence and for
each i (1≤ i ≤ n), Ri+1 follows from R1 by one of the conversions of C.
(Chomsky 1955c, p. 114)
The basic idea is that a sentence such as Pedro viu Maria ‘Peter saw Mary’,
like any other sentence in the Portuguese language, is assigned a set of representations
and that these representations may be constructed in a sequence, one after the
immediately previous one. Let us suppose that the system starts saying that Pedro viu
Maria is a sentence. From that point, the system should say how the sentence consists of
syntactic categories, showing its structure. Our sentence, for example, consists of a
noun phrase (NP) followed by a verb phrase (VP); the noun phrase on its turn consists
of a verb (V) followed by another NP. The NP’s consist of names (N). Thus, at phrase
structure level, the sentence Pedro viu Maria will be assigned a representation that may
have one of the following two forms (which are absolutely equivalent):
(3)a. S
SV
SN SN
 
N V N
  
Pedro Viu Maria
‘Peter’ ‘saw’ ‘Mary’
(3)b. (((Pedro)N )SN ((viu)V ((Maria)N)SN)SV)S

‘Peter’ ‘saw’ ‘Mary’
Each of the words Pedro, viu, and Maria will be assigned a representation at
the morphological and phonological levels, which will allow to obtain a morphological
representation for the sentence. And so on so forth, till we end up with a fully
represented sentence in all six linguistic levels.
Each conversion must be marked somehow in order to indicate which linguistic
level it belongs to, and this can be done by grouping the conversions of the same level
and by establishing a certain order of application.
Thus we come to the well known grammar of Syntactic Structures:
E: Sentence
F: X1  Y1
⋅ ⋅
⋅ ⋅ Phrase-structure
⋅ ⋅
Xn  Yn
T1
⋅
⋅ Transformations
⋅
Tj
Z1  W1
⋅ ⋅
⋅ ⋅ Morphophonemics
⋅ ⋅
Zm  Wm
To produce a sentence from such a grammar we construct an

extended derivation beginning with Sentence. Running through
the rules of F we construct a terminal string that will be a
sequence of morphemes, though not necessarily in the correct
order. We then run through the sequence of transformations
T1,…,Tj, applying each obligatory one and perhaps certain
optional ones. These transformations may rearrange strings or
may add or delete morphemes. As a result they yield a string of
words. We then run through the morphophonemics rules, thereby
converting this string of words into a string of phonemes.
(Chomsky 1957, p. 46)
This, in short, is the model of analysis Chomsky proposed in the late fifties.
The hypothesis put forward then is that this linguistic analysis device would be
an adequate representation of the computing system, present in the speakers’
mind/brain, able to determine, at least partially, its linguistic components. That is,
assuming the affirmations of the nucleus, the heuristics determined the construction of a
representation hypothesis of the computing system, which, until disproved, was
considered the grammar present in the speakers’ main/brain.

It is easy to realize that this model is not far away from the models resulting
from the American structuralism. Chomsky himself does not see any fundamental
differences between his model and Bloomfield’s and Harris’s distributional model: at
the end of chapter IV in LSLT, after presenting his model of grammar, Chomsky states:
We will refer to linguistic analysis carried out in these terms as
“distributional analysis.” This usage seems to me to correspond to the
practice of what has been called distributional analysis.
(Chomsky 1955c, p. 127)
In what concerns the notion of linguistic level, the chomskian approach is not
markedly different from the structuralist approaches: Chomsky refers to Hockett (1955)
‘for a similar approach to linguistic levels’ (1955c, p.97). The only point in which
Chomsky’s proposal seems to fall far apart from the structuralist approaches is in what
concerns the number of levels necessary for a linguistic description.
Our main conclusion will be that familiar linguistic theory has only a
limited adequacy – i.e., that it is attempting to do too much with too
little theoretical equipment. (...) We will argue that the remedy for
these deficiencies is not to be found in the extension of the
distributional basis for linguistic theory to include meaning, situational
context, etc., nor, apparently in the introduction of probabilistic and
statistical conceptions. Instead, a new level of transformational
analysis is proposed as a higher level of linguistic structure. It will be
shown that the theory of transformational analysis can be formulated in
the same completely distributional terms that are required anyway for
lower levels, and that a large and important class of problems that
arise in the rigorous application of familiar linguistic theory
disappears when it is extended to include transformational analysis.
Apparently, according to Chomsky, the addition of a transformational level to
the ‘known linguistic theory’ (in other words, American structuralism) is enough to turn
it adequate. Lees, by the way, assume the same position, with an exaggerated optimism,
He declares:
Chomsky...has been led to set up a whole level of grammatical
transformations to deal with all the difficulties encountered in trying to
state explicitly a complete and simple immediate-constituent grammar.
(Lees 1957, p. 52 – bold type added by the author)
However, it is somehow strange to consider that Chomsky constructs a set of
transformational rules aiming at overcoming the deficiencies of the AS models since in
Harris (1952) transformational rules were already used for the analysis of natural
languages, and Harris is one of the most characteristic representatives of AS. We must
make it clear then how Chomsky’s notion of transformation differs from Harris’s
transformation.
Harris’s notion of transformation lies in the notion of sentence form. Harris
arrives at this notion by the definition of variables having word classes as domains
(variable N has as its domain the class of name, for example). From there on, it is
possible to define sequences of well-formed variables11. Harris calls these sequences of

well-formed variables sentence forms.
It happens, however, that not all sentences obtained by attributing values to the
variables of a sentence form are equally acceptable. For example, the sentence o homem
pensa ‘men think’ is more acceptable than a pedra pensa ‘stones think’, although both
belong to the same sentence form. Each sentence form, then, has a gradation of
acceptability for the values of its variables and such gradation in fact characterizes the
sentence form.
Harris defines, then, pairs of sentence forms, made up exactly of the same
variables, differing only by some univocally determinable characteristics (difference of
order between variables; constant presence or absence of a certain element; etc.). If two
sentence forms belong to a pair thus defined and, in addition, present the same
acceptability gradation, Harris will say that they are in transformation relation. For
example, the sentence forms N1+V+N2 and N2+ser+V-do por+N1 constitute a pair of
forms that (i) present the same variables; (ii) present constant elements in one of them
(ser, -do, por ‘be, -ed, by’), and (iii) present a change in order between N 1 and N2. In as
much the acceptability gradation of both sentence forms is the same, Harris will say that
these two forms are in a transformation relation named active/passive.
According to Harris, then, a transformation is nothing more than a class of
pairs of sentences. As Milner (1973, p. 191) points out, Harris’s transformations are
relations that can be expressed in a class language and can be labeled relations-in-
extension.
Let us now see what transformations consist of in Chomsky. He starts from the
definition of ‘analyzable’ predicate. Let us imagine a sentence consisting of a sequence
of elements t; let us imagine that Q is a syntagmatic indicator (a ‘tree’, as the one in
(3b)) representing structure t; let us now suppose that t can be subdivided into
successive segments t1, ..., tn so that each ti is linked to a node Ai in Q. Given these
conditions, we can say that t is analyzable in <t1, ..., tn; A1, ..., An> from the point of
view of Q.
Let us take the sentence Pedro viu Maria ‘Peter saw Mary’, for example. We
can represent its structure by means of the syntagmatic indicator in (3b). The sentence is
divided into three successive segments (Pedro, viu, and Maria) and for each segment
we can find in the syntagmatic indicator a node to which it is linked: Pedro is linked to
N, viu is linked to V, and Maria is linked to N. We can then say that the sentence Pedro
viu Maria is analyzable into <Pedro, viu, Maria; N, V, N>12.
According to Chomsky, a transformation has a domain (or a ‘structural
condition’) and an effect (or ‘structural change’). The domain indicates the class of
linguistic expressions that can undergo transformation and is specified by a sequence of
symbols <A1, ..., An>, which are symbols of nodes of a syntagmatic indicator. In order
for a linguistic expression to be in the domain of transformation it is necessary that it be
analyzable into <t1, ..., tn; A1, ..., An>. The transformation effect, on its turn, is described
by a rule that projects the starting sentence (the set t1, ..., tn of the sentence that will
undergo transformation) into the target sentence (i.e., the already transformed sentence).
Let us take a passive transformation for example. Its form in Portuguese would be
roughly the following13
11
In Portuguese, for instance, the sequence N+V+N is well formed, whereas the sequence V+N+V is not.
12
The sentence Pedro viu Maria is analyzable, from the point of view of the same syntagmatic indicator, into
<Pedro, viu Maria; N, SV>, provided we consider the SV internal structure.
13
This is just an outline of the passive transformation in Portuguese. It is necessary to underline that this type of
transformation was already abandoned in the initial revisions of the GG and, as much as I know, it has never been
Domain: N1 Aux V N2 ⇒
Effect: N2 Aux+ser V+-do por N1
‘be’ ‘-ed’ ‘by’
The domain tells us that expressions can undergo transformation (the sentence
Pedro viu Maria, for example, is within the transformation domain 14); the effect tells us
which are the changes the starting sentence (let us say, Pedro viu Maria) must undergo
in order that the target sentence can be obtained. Basically, the changes consist of the
permutation of subject (N1) and direct object (N2); the addition of verb ser ‘to be’, as
‘tense bearer’ auxiliary; the addition of the past participle ending to V; and the addition
of the preposition por ‘by’ before N1. After applying the rule to the sentence Pedro viu
Maria, we will obtain the target sentence Maria foi vista por Pedro ‘Mary was seen by
Peter’.
According to Milner, what Chomsky defines with his transformation notion…
c'est une entité spécifique, la règle de transformation et non une
classe de paires de phrases; le fait que deux phrases soient en
relation est envisagé comme une propriété de la paire, distincte
de la paire elle-même, et dont la règle prise dans son ensemble
est le symbole.
(Milner 1973, p. 104-105)
For Milner, chomskian transformations cannot be dealt with in a class language
and characterize relations-in-intention.
It is easy to see that there are many similarities between the two concepts of
transformation. According to Milner, what make them different are the approaches:
extensional in Harris and intentional in Chomsky.
In extensional terms, a transformation relation is totally determined by the pair
of sentences it relates and the task of the grammar is simply to establish those pairs.
Harris states quite clearly that, basically, transformational analysis is not a means of
determining the structure of each sentence considered separately, but rather a way of
grouping the sets of sentences into pairs (Cf. Harris 1968, p. 68).
From an intentional perspective, on the other hand, given a pair of sentences,
we can imagine various relations (rules) between them, specified differently by the
analyzable predicate. The scientist’s task is much more complex than it would be from
an extensional perspective.
We could always argue that the extensional and intentional theories are
equivalent, since it seems possible to establish corresponding tables that relate
properties and classes, relations-in-extension and relations-in-intention, etc. For
example, the statement ‘sequence t is analyzable into...’ is nothing more than the
intentional version of the statement ‘sequence t is a member of the sentence form...’.
However,
Il exist bien des cas où les paires sont parfaitement connues, mais où la
règle de transfomation intensionnelle ne peut être formulée parce que
le prédicat « analisable » ne peut être prècisé à coup sûr.
Les examples réels abondent et certains d’entre eux sont très connus :
ainsi les paires phrases actives/phrases passives sont parmi les mieux
studied seriously in relation to Portuguese data.

14
The node Aux (auxiliary) serves, basically, to ‘transport’ verb tense. In the case of Pedro viu Maria the Aux is
[+pass].
attestées de la tradition grammaticale ; en termes extensionnelles, la

transformation passive est donc une de plus certaines et des mieux
décrites. En revanche, si on envisage les données du point de vue du
prédicat « analysable », de graves obscurités appairaissent : quelle est
la nature exacte du complément d’agent, quelle est la condition exacte
qui spécifie le domaine de la transfomation (la nature transitive du
verb est-elle suffisante ? D’autres facteurs interviennent-ils : par
exemple, un Adverbe de Manière fictif ? cf. Aspects, p. 145-150), etc.
Bref, la règle de transformation passive est une des plus mal connues
de la grammaire intensionnelle.
(Milner 1973, pp. 198-199)
We cannot then speak about equivalence between the two perspectives because
the relationship between them is asymmetrical: given a well-formed intentional theory,
we can easily obtain an extensional counterpart, but the opposite is not true.
There is still another difference between the two notions of transformation that
is fundamental for its comprehension (and that is ignored by Milner). It is the fact that
not everything Chomsky lists as transformations are sentences. The harrisian notion of
transformation pairs sets of sentences (sentence forms); the chomskian notion of
transformation maps a structure of a sentence onto a given derivational stage in another
structure of the same sentence obtaining another derivational stage. Chomskian
transformations definitively are not devices constructed for obtaining pairs of sentences
although, through them, we can justify the pairs of sentences our intuition recognizes in
the language.
It is then clear that Chomsky really innovates when he proposes his intentional
theory of transformations. The notion of transformation – and the role this notion plays
inside the model – in fact opposes Chomsky to AS. It thus justifies the emphasis
Chomsky and those who publicize his theories (such as Lees) give to this notion.
In spite of the innovation represented by the introduction of the
‘transformational level’ into the linguistic analysis, generally speaking, the chomskian
descriptive model does not deviate much from the structuralist tradition.
Thus, we find GG in its initial years having to face an inconsistency that,
although not perceived as such at that time, will require deep changes in the form of the
theory: the program’s proposal is different from that of the structuralist program, but the
available analytical tools are basically the same. That is, there is a mismatch between
what is intended to do and what, in fact, ends up being done.
So, it is not surprising to see that alterations are soon included in the
descriptive model by assuming new auxiliary theories.
2.3. Generative Grammar in the sixties: the standard theory.
The first ten years of generative grammar were its ‘heroic years’ in which the
combat with the forces of American structuralism dominated the scene15. The SS theory
proceeds having its presuppositions explicited and some of its theoretical devices
changed or replaced in order that it could manage - in better conditions - to carry out the
task the theory imposed, namely the description of a computing system able to define,
15
Despite being very similar, as we saw, GG and the American structuralism soon start out a dispute for space and
prestige inside the North-American academic institutions. As this dispute is a matter of sociology of science, rather
than of philosophy, and the present work intends to have a basically philosophical character, I am not making any
attempt to explore here this bias of the history of science.
generatively, natural languages. In the mid sixties one could already consider that the
battle had been won: the generative program was clearly dominating among the North-
American linguistics and began its expansion to other places, by conquering new
followers outside the USA.
Aspects of the Theory of Syntax (Chomsky 1965) is a great synthesis of the
changes that were made in the SS theory. Besides the first great clarification of the
program’s postulates - presented in the ‘Methodological Preliminaries’ 16 - Chomsky
develops a new configuration for his grammar: a model that came to be known as the
‘Standard Theory’.
This new theory changes significantly the descriptive model and clarifies a
series of postulates that, in the previous theory, had been presupposed in a rather
obscure way or, though resulting from those presuppositions, were not presented
clearly. It is during this period, for example, that the question of innatism is raised as a
working hypothesis, with the resulting strong psychologization of the grammar 17. It is
also in the same period that powerful auxiliary theories were developed, which will
allow for a better descriptive and/or explanatory performance of the program: the
lexicon becomes relevant and receives its first consistent theoretical formulation; the
‘deep structure’ appears, having as its strongest consequence the outset of a deeper
concern about semantics (concern that provided for the emergence of a series of more or
less heterodox alternative theoretical formulations)18.
It is also not surprising that a high number of discussions focused the
transformational component of the model of analysis. Given a set of linguistic
phenomena, if what differentiate a generative analysis from the one carried out by a
structuralist is the presence of transformations, it is obvious that the heuristics should
determine a more careful exam of that notion aiming at the solution of formal problems
that might emerge there.
In the standard theory, the form of grammar - which in a way represents the
image that one had at that time about the functioning of the linguistic knowledge the
speakers have registered in their mind/brain - is organized into three major components:
a syntactic component, which is generative, in as much it is the only component that
constructs representations, and two interpretive components; the semantic component;
and the phonological component19.
The process of generating sentence starts from the syntactic component, which
has the following basic structure: a base subcomponent (or simply BASE), which is the
responsible for generating the deep structures (DS); and a transformational
subcomponent, which converts the DS’s into surface structures (SS). The base
subcomponent contains (i) a set of rewriting rules (sometimes called categorical
component), which, applied to the initial axiom S, generates tree structures ‘labeled’
with the symbols of the categories whose terminal nodes are not filled in; and (ii) a
lexicon, which inserts lexical items into the terminal nodes of the tree. The base input is
the axiom S and the deep structures are the output. The transformational component is
assigned deep structures, as input, and, through transformational rules, converts them
into surface structures. In a diagrammatical form, we would have:
16
First chapter in Chomsky 1965.
17
Linguists spent a lot of time discussing the Language Acquisition Device (LAD).
18
As Generative Semantics, for example (see Lakoff 1971, Kato 1974, Dascal 1978, and Galmiche 1979, among
others).
19
The difference between the generative component and an interpretive component lies in the property that the
generative component has of creating new representations, whereas the interpretive components only associate (pair,
relate) representations among themselves.
Syntactic component
BASE
Categorial  Lexicon  DS  Transformational  SS
component component
The syntactic component generates ordered pairs <DS, SS> and the two
interpretive components associate representations to the elements of the pairs: the
semantic component associates semantic interpretations to DS’s and the phonological
component associates phonetic interpretations to SS’s. A complete diagram of it would
be the following:
BASE: Categorial c.
Lexicon  DS  Semantic interpretation  Semantic c.

Transformational c. 

SS  Phonetic interpretation  Phonological c.

The DS must contain all the necessary elements for the semantic interpretation
of the sentence while the SS must contain the information for its phonetic reading. A
grammar is understood as a ‘system of rules that link the phonetic signs to the semantic
interpretations’ (Chomsky 1966, p. 12) or - as Chomsky will reiterate in other places –
‘a system of linking sounds to meanings’.
Since the only generative component is the syntactic component, this will be
the central component in the grammar, in the sense that it is the component that allows
establishing the relationship between the semantic content and the phonetic form of the
linguistic expressions.
With the standard theory, the GG program comes to the end of a stage.
Apparently, there exists a good model of linguistic analysis to support the requirements
of the heuristics. The syntactic devices seem to be powerful enough to provide an
adequate description of the linguistic structures; the auxiliary theories and a general
theory of generative devices (Formal Theory of Grammar) seem to provide enough
support to the descriptions and explanations obtained by GG. The analyses of new facts
in the English language multiply, as do the analyses of facts from other languages. The
success of these analyses reinforces the feeling that one has managed to obtain an
adequate theory of the speakers’ linguistic competence. So, the major aim of the
Program, if not yet reached, seemed to be very close to being so.
At the end of 1965, the first criticisms to Chomsky’s ideas begin to be raised,
within the very generative circle. The main conflict area, at that time, was the degree of
abstraction of the underlying linguistic structures. The center of dispute was the
distance between DS’s and SS’s and the distance between DS’s and semantic
representations. While the standard theory tried to maintain the DS and the SS very
close together, the ‘dissenters’ proposed to increase the distance between DS’s and SS’s
and decrease the distance between the DS’s and the semantic structures.
Chomsky’s statement saying that
The syntactic component specifies an infinite set of abstracts formal

objects, each of which incorporates all information relevant to a single
interpretation of a particular sentence.
(Chomsky 1965, p.16)
induced the generativists to search for syntactic solutions to semantic problems and to
look for DS’s that could represent all aspects of meaning in the sentences under
analysis. This procedure led to the postulation of DS’s that were more and more abstract
and closer and closer to the semantic representations20. The commitment with the
hypothesis that the semantic interpretation occurs at DS level leads many linguists to
conclude that everything one can consider as being part of the meaning of the sentence
must be included in the DS. Thus, for example, all ambiguities observed in the
sentences should be solved by postulating different DS’s, not mentioning the
phonological phenomena with semantic consequences, as focalization, presuppositions,
and performatives, for example.
A series of analyses and a series of empirical and theoretical arguments were
raised by the ‘abstractionists’, all leading to the same conclusion: it is necessary to
postulate more abstract DS’s that may represent more directly the semantic
representations present in the sentences.
It is important to point out that the ‘abstractionists’ remained strictly inside the
standard theory and, consequently, inside the GG program, what can explain the almost
general acceptance of their analyses by the generative community, at least for a while.
With the expansion of the abstractionist posture, one comes to an almost
complete decharacterization of the DS notion - as Chomsky imagined it - and it became
meaningless to try and differentiate them from the semantic representations.
The ‘abstractionists’ tried – in an ordered way – to gather their ideas in what
was considered at the time a new ‘paradigm’, which was labeled Generative
Semantics21.
Chomsky’s reaction did not take long to manifest and, in 1967, he proposed
some changes in the standard theory in order to avoid the uncontrolled abstraction.
Chomsky’s reaction had some consequences. First, the breaking up with the
‘abstractionists’ is inevitable and a first group of linguists ‘generated’ inside the GG
program emerges and they plunge into the task of constructing new research programs.
Second, a new model of linguistic analysis is established. This new model came to be
known as the Extended Standard Theory (EST)22.
As the name reveals, the EST is not understood as a new theory. It is just the
old standard theory that received some new auxiliary theories; the most important of
them is the X-bar Theory. The role of the lexicon is also changed; the lexical items, for
example, start to be considered a bundle of traces, and many phenomena that were
approached via transformations began to be approached via lexical relations. In all,
however, the theory remains de same.
2.4 From rules to principles.
After the heat of the famous sequence of clashes with the Generative
20
That is, the formulae (expressions of the formal system language) which were supposed to represent adequately the
meanings of natural language expressions.
21
As I cannot present a more detailed analysis of this period within the scope of the present work, I suggest the
interested reader to refer specially to Newmeyer 1980, Harris 1993 (and Borges Neto 1991) for details.
22
EST is initially proposed by Chomsky 1967 and developed in Chomsky 1968 and 1971.
Semantics cooled down, Chomsky and his associates could go back and think about the
development of the GG program. The greatest legacy of Generative Semantics was the
proliferation of theoretical devices and of the types of available rules. The immediate
task the chomskians had to face then was that of restricting the descriptive power of the
grammar in order to increase its explanatory power.
A certain ‘tension’ is felt in the GG program between the requirement of
descriptive adequacy and the requirement of explanatory adequacy. Chomsky says that
a theory is explanatorily adequate when it manages to successfully select a
descriptively adequate grammar from a set of possible grammars (defined by the
general theory) and from the primary linguistic data. In other words, the theory is
explanatorily adequate when it reproduces the behavior of children acquiring language:
from the raw linguistic data, they select a grammar among the possible grammars
admitted by the innate component of their linguistic competence23. Thus, in order to
attain the explanatory adequacy, the available theoretical devices in the general theory
(the grammar innate component theory) must be restricted, so that few grammars can be
obtained and one is able to understand how a child quickly selects an adequate grammar
for the data available to him/her.
On the other hand, in order to attain the descriptive adequacy, that is, in order
to construct grammars to all natural languages, the available theoretical devices must be
rich and varied enough to cover all the richness and diversity of natural languages.
The conflict between these two requisites of adequacy is obvious and the
search must go for a theory that is, at the same time, rich enough to account for the
variety of languages and restrictive enough to provide a small number of possible
grammars.
In the mid sixties it is already possible to find restrictive proposals to the
descriptive power of transformational rules. John Robert Ross’s PhD Dissertation (Ross
1967), postulating the ‘islands’ (syntactic configurations that prevented the extraction of
elements), is a good example of this concern. But it is with Peters and Ritchie’s works,
in the early seventies, in the heat of the debates between the GG and Generative
Semantics, that the need to restrict the power of those rules becomes urgent.
Peters and Ritchie (1969, 1971, 1973) demonstrate that de weak generative
capacity of a grammar that includes transformational rules, as those proposed at that
time, is equivalent to an unrestricted rewriting system (a Turing machine). This means
that a transformational grammar of any natural languages do not reveal anything about
the structural characteristics of that language, but only states that that language -
understood as a set of sentences - is recursively enumerable (and, consequently, can be
generated by calculus). Peters and Ritchie’s works show that the major problem of
transformational grammars was not the proliferation of rules or categories but the lack
of strong restrictions on the functioning of those rules. Without restrictions, the
transformations – that could eliminate, create, exchange, move, or change elements –
were useless as exposing devices of linguistic structures. Applying the adequate
deletions, movements, and additions, one could start from any sentence and arrive to
any other sentence.
The task of proposing restrictive conditions on the functioning of the rules
becomes the priority of the program. It seems clear that here we have a case of creative
change in the heuristics. The focus is not placed on the descriptive adequacy any longer
23
The innate component – Language Competence – is a set of general principles about the nature of the linguistic
representations, principles that can be manifested in a whole range of alternative realizations (the parameters).
According to Chomsky, children apply these ‘parameterized’ principles to construct the grammar of their language,
provided they are exposed to the data.
but on the explanatory adequacy. Instead of having a proposition of computing systems

as the most important task, the restrictions of the systems previously proposed become
the priority. Obviously, the orientation of the theory - its ‘plan of development’-
undergoes a substantial change.
It is possible to see two main tendencies in this process of restricting the power
of rules. The first tendency is to impose general conditions to the application of rules:
Ross’s Island Conditions (1967), Emonds’s structure-preservation restriction (1979),
and Chomsky’s conditions (condition over specified subject, condition over sentence
with tense and subjacency condition) are all conditions over the application of
transformational rules. In other words, they tell us that transformations only apply if
certain conditions are present. For example, Emonds’s structure-preserving restriction
tells us, among other things, that one cannot delete anything in the structure that cannot
be structurally recovered: it is only possible to delete the subject of the sentence in
Portuguese because the verbal inflection allows to recover structurally the subject
position; in a language without verb inflection, such as English, subject deletion is
hindered.
Another tendency, besides restricting the application of rules, proposes a strong
restriction onto the number of available rules. This tendency emerges with Chomsky
(1976a) and becomes a dominating tendency in GG from then on. In Lobato’s words:
Restriction to the number of transformations has been a
characteristic of the chomskian generative theory since 1976
(‘Conditions on rules of grammar’), when the transformational
component was said to consist of just two rules
NP- move,
Wh- move.
(Lobato 1986, p.337)24
This second tendency is interesting and deserves some attention.
When Chomsky proposes that the NP postposition and NP anteposition
transformational rules be replaced by only one rule of NP-move, on the one hand, he
manages to reduce the number of available rules, but, on the other hand, he obtains a
rule that is so general that any NP is possible to be moved from anywhere to anywhere
else. How is it then possible to maintain the reduction in the number of rules without
losing sight of the general need for restricting the grammar as a whole? Chomsky’s
solution is to propose an interaction of rules with a set of general principles about
grammar. The solution proves to be so operational that Chomsky goes ahead and
reduces even more the transformational component, maintaining only one rule: the
MOVE ALPHA rule.
As the transformational component is reduced to only one rule, and
considering that this rule is optional, in order to permit the generation of alternative
surface forms for one and the same underlying structure, it is necessary to find out
devices that not only hinder undesirable moves but also devices that force the move in
cases in which it should be obligatory.
Chomsky’s solution lies in the establishment of some new auxiliary theories:
the Case theory, the Thematic Role theory (Theta theory), the trace theory, the empty
category theory and the Binding theory besides the X-bar theory, already on. All these
theories impose conditions onto the possible representations and, consequently, both
force and hinder moves.
24
Translated from the Portuguese original by the author.
I am not going into long analyses within the scope of the present work. But I
would like to show just how these auxiliary theories work.
The trace theory, for example, developed in Chomsky (1976a), postulates that
all moved element leaves a trace in the place it left, and this trace acts as a full element
for the purposes of syntactic rules. For example, the analysis of sentence (4) proposes it
be derived by transformation from structure (5) through the move of the clictic nos to a
position next to the verb of the main clause25.
(4) Paulo nos viu examinar a garota
‘Paul saw us examine the girl’
(5) Paulo viu [nos examinar a garota]
‘Paul saw [us examine the girl]’
It is not possible, however, to obtain (6) from (7) because the clictic a would
have to ‘fly over’ the subject of the subordinate and this is not allowed by the
Conditions over Specified Subject26.
(6) * Paulo a viu nós examinar
‘Paul saw us examine her’
(7) Paulo viu [nós examinar a]
‘Paul saw [us examine her]’
Let us consider a fairly more complex case. Sentence (8) would be obtained
from the structure present in (9) by the move of the subordinate subject to the position
of subject of the main sentence and by the anteposition move of the clictic os.
(8) Paulo parece os ter examinado
‘Paul seems to have examined them’
(9) ∆ parece [Paulo ter examinado os]
‘∆ seems [Paul to have examined them]’
Sentence (10), however, which should behave in the same way, is not
grammatical, although there is apparently no other reasons for the application of the
specified subject condition.
(10) * Paulo os parece ter examinado.
‘*Paul seems to have examined them’
The solution to the problem brought by the trace theory suggests that between
sentence (9) and sentence (10), we should have an intermediate structure (11) that,
having its trace in subject position, hinders the move of the clictic outside the
subordinate sentence.
(11) Paulo parece [t ter examinado os]
‘Paul seems [t to have examined them]’
Clictic moves are allowed inside the subordinate sentence, and so, from
structure (9) we could obtain structures (12) and (13) with no difficulties.
(12) Paulo parece [t tê-los examinado]
‘Paul seems [t to have them examined]’
25
These analyses are borrowed from Quícoli 1976.
26
Which is one of the conditions on transformations proposed by Chomsky.
(13) Paulo parece [t os ter examinado]

‘Paul seems [t to have examined them]’
The trace hypothesis hinders the undesirable move maintaining the generality
of the move rule.
The Case theory, on its turn, determines that all full NP (i.e., a morphologically
realized NP) receives a CASE27; and, in addition, establishes the contexts28 in the
structure in which an NP can receive a CASE. Thus, an NP generated in a position that
does not allow Case attribution must necessarily be moved to a position in which Case
may be assigned. For example, in (14) the subject position of the main sentence is
empty in the underlying sentence and, in order that a well-formed sentence can be
obtained, it is necessary that the subject of the subordinate occupies the subject position
in the main sentence, producing (15).
(14) ∆ parece [João estar alegre]
‘∆ seems [John be happy]’
(15) João parece [t estar alegre]
‘John seems [t be happy]’
Now, that move is not obligatory in all cases, since we could obtain (16) from
(17), for example.
(16) CV parece [que João está alegre]
‘CV seems [that John is happy]’
(17) ∆ parece [João está alegre]
‘∆ seems [John is happy]’
Instead of proposing, in an ad hoc way, that certain moves are obligatory, it is
preferable to establish a general principle, such as the attribution of Case to full NP’s -
like João - and determine the contexts in which these attributions will take place. We
can see in our examples that in (17) the NP João is assigned Case because it is the
subject of a sentence with a finite verb (the subordinate sentence) whereas in (14), since
the verb in the subordinate sentence is an infinitive verb, the NP João is not assigned
Case, requiring, however, to be moved to the subject position in the main sentence in
order to be assigned the nominative Case (the main verb parece is a finite verb and,
consequently, allows the attribution of Case). Notice that the move remains optional.
However, the absence of move, in (14), turns the structure agrammatical.
Summing up, one of the theories – the trace theory – restricts the number of places
to which the moved element may go to, whereas the other – the Case theory – turns
certain moves obligatory.
Chomsky (1976a and 1976b) notices that these conditions produce a more general
effect than simply that of regulating moves: the conditions also hinder certain
relationships between elements in cases in which there was no move. So, the proposal is
that the conditions work also in the regulation of the interpretive relations. This way, it
is possible to show a structural parallelism between the relationship between the trace
and its antecedent (consequence of the move rule) and the relationship between certain
anaphoric pronouns (reflexive, reciprocal and PRO) and its antecedents (consequence of
interpretive rules).
27
In some of the latest versions of GG, it is also possible to attribute cases to NP’s not realized phonetically.
28
The attribution of Case happens as follows: (i) the sentence with a finite verb attributes a nominative case to its
subject: (ii) the verb attributes an objective case to its complement; and (iii) the preposition attributes an objective
case to its complement.
What can be perceived hereon is that it becomes impossible to admit that the
move rules are free, that is, that they are not submitted to conditions. The conditions on
moves cease to exist, and all their effects turn to be obtained through interpretive
conditions.
The main consequence of this is the substitution of what we can call derivational
perspective by a representational perspective. We are facing a new creative change in
the program’s heuristics.
From a derivational perspective, the various representations of the linguistic levels
(they may be ‘phonemes (Pm), morphemes (M), words (W), syntactic categories (C),
phrase structures (P), and transformations (T)’, as Chomsky intended in LSLT; they
may be ‘Deep structure, surface structure, phonetic form and semantic representation’,
as in the standard theory) are derived from one another through rules. The grammar is
strictly directional, that is, the various levels of linguistic analysis are approached – and
receive representations – in a given order. Thus, until the mid seventies, all proposals of
grammars made inside the GG were derivational.
From the representational perspective, on the other hand, the various
representations are not related by derivation: they are just representations of structural
properties resulting from the theories restricting grammar. P-structures, for example,
can turn to be understood as a ‘pure’ representation of the grammatical functions
relevant to the attribution of thematic roles and, in this sense, as an ‘abstraction’ of the
S-structures. Grammar is not directional. As Lobato states:
This change of approach leads to a switch in the interpretation of
what may be ‘generated by the base’. In the previous versions of the
theory, this expression meant ‘derived from S by successive
applications of syntagmatic rules e by using the lexical substitution
rule’. Now it means ‘be projected from the lexicon, from X,
according to the UG [Universal Grammar] principles and the
parameters established by a certain language’. This new perspective
allows then to consider an S-structure generated by the base, and the
move-ALPHA is a property of the S-Structures and not, from this point
of view, a rule that converts P-structures into S-structures (cf.
Chomsky 1982b: 33). Likewise, any other level of representation can
be seen as ‘derived by the base’, since any level of representation is
determined by the establishment of the UG parameters (Chomsky
1982b: 14).
(Lobato 1986, pp. 403-404)29
Having completed the cycle of substitution of rules by principles, Chomsky
finds himself involved by a grammar theory which is different enough from the
previous ones to justify a new label: Principles and Parameters Theory.
I am not going into details about the Principles and Parameters Theory (also
called, for some time, Government and Binding Theory). It suffices to say that this is the
theory used at this beginning of the XXIst century for the study of the syntax of natural
languages. However, it is worth mentioning some movements that have been observed
inside the GG program – especially the so-called Minimalist Program – the statute of
which is not yet clear to me (and not even to the generativists themselves, so it seems to
me).
29
2.5. The Minimalist Program

I will start quoting an extract of the Presentation Eduardo Raposo wrote for the
Portuguese translation of the book The Minimalist Program (Chomsky 1995):
It is important to point out that the MP [Minimalist Program] is not
a new theoretical framework of the transformational-generative
grammar, in the sense the model P&P [Principle and Parameters],
or the Extended Standard Theory, or the Standard Theory are. In
this sense, the MP does not replace the P&P model. Quite the
opposite, MP stands crucially on the P&P model, and starts from it
to propose some new questions that could not, in fact, be conceived
outside that model. In a way, MP is a set of ‘guidelines’ oriented by
the intuitive idea of avoiding the postulation of theoretical entities
that are not conceptually necessary inside the theory’s logic.
(Raposo 1999, pp. 15-16)30
In Raposo’s words, MP should be considered not only part of the GG Program,
like all other models we have seen so far, but also as part of the model that resulted
from the last major elaboration of the program: the Principle and Parameters Theory.
Actually, MP would consist of just ‘guidelines’ of methodological nature to
help the linguists to apply the ‘Ockham's razor’ to the Principles and Parameters Theory
(P&P), eliminating what was unnecessary, basically for theoretical economy reasons31.
Being so, MP would not raise much interest in our investigation. We have been
dealing with the major ‘changes of course’ of the Program (from LSLT to the Standard
Theory and from the Standard Theory to P&P) as creative changes in the heuristics,
resulting in a new theory. Considering that MP is not accepted as a creative change in
the heuristics, resulting in a new theory, basically we have nothing to say about it.
However, the framework is not so clear and I believe we have to investigate
more carefully the nature and function of the MP in the framework of the GG Program.
In order to understand the MP role, we must understand better the working
hypothesis of human languages proposed by the GG program32.
According to Chomsky, languages are biological systems that men use to speak
about the world (or about the mental representation they have about it), describe, refer
to, ask, communicate with one another, articulate thoughts, talk to themselves, etc.
Those ‘things’ we do with language constitute what Chomsky calls the conceptual-
intentional system. On the other hand, as an ‘expressive’ medium, language must be
associated to a production and reception system, of motor-sensorial nature, capable of
allowing for the production and reception of sounds that constitute the linguistic
expressions. Chomsky labels this second system articulatory-perceptual system.
Thus, the human language must be able to contact (be an interface of) not only
the conceptual-intentional system (C-I) but also the articulatory-perceptual system (A-
P)33.
30
31
There is a concern about showing that the human languages are themselves economical and perfect (I thank Evani
Viotti for this observation).
32
In what follows, I make strong use of Raposo 1999.
33
The question of interfaces is an old question. In the standard theory, it was up to the deep structure to make an
interface with the C-I system, and to the surface structure to make the interface with the A-P system. In the
government and binding theory the interface with the C-I system is done in Logical Form and it is up to the
Phonological Form to make the interface with the A-P system. Certainly, although the idea is very old, the devices to
actualize these interfaces have become more and more sophisticated as the time passes by.
According to Chomsky, the C-I and A-P systems have their own structure and
are independent from the human language. It is possible to assume that they impose
conditions over language. It is reasonable to think that human languages have the
articulatory and hearing capacity of human beings as their limitations, for instance34.
So, for us to use languages it is necessary that the linguistic expressions satisfy
certain conditions imposed by these two outside systems. Thus, the MP fundamental
question is the establishment of the ‘measure’ that allows the evaluation of the
‘optimality’ of the structures in satisfying the conditions imposed by the outside
systems. In other words, it will be considered ‘good’, ‘grammatical’, ‘acceptable’ the
structure that fully satisfies the interface conditions.
Now, though it can sound like big news, it seems that we are facing the same
‘movement’ that led GG to switch from rules to principles: that of obtaining the
maximum generality with the least resources. In a way, following the strong assumption
of ‘psychological reality’ of the computing systems – assumption that, in a higher or
lower degree, has been guiding the generative analysis from their first formulations –
what is being assumed is that the conditions over structures, realized in the auxiliary
theories, are imposed by performance, by ‘pragmatics’, by the ‘use’ we make of them.
And this ‘use’ involves not only the elements of comprehension of the world (the C-I
system) but also sensory-motor elements (the A-P system). The bet made by the MP is
that these general conditions, coming from performance, are capable of imposing proper
conditions to the computing system in order that it works in an ‘optimal’ manner and be
able to perform the task of the theories of the government and binding model with more
economy, considering that it does not need to postulate anything but the interfaces:
syntax is reduced to the minimum.
It is no longer necessary to postulate restrictions on structures. The ‘well-
formed conditions’ of the structures, essential in the other stages of the program,
disappear, and the guarantee that a structure is well formed (grammatical) will depend
on the degree of satisfaction of the conditions imposed by the outside systems (C-I and
A-P) that the structure presents. In other words, it will be considered more ‘adequate’,
‘acceptable’, ‘well-formed’, ‘grammatical’ that structure that best satisfies the phonetic
production/reception and semantic conditions.
From this point of view, MP is nothing more that the radicalization of the
movement that led the GG program to replace rules by principles.
3. Conclusion.
The history of GG shows three major ‘strategies’ in the delimitation of the
language competence present in the speaker’s mind/brain. At time 1 (the SS Theory),
grammar should generate the sentences of the language directly (into their surface
structures). They dealt exclusively with syntax (maybe phono-morpho-syntax), and the
notion of generative grammar was similar to the current notion in logics and
mathematics.
At time 2 (the Standard Theory), grammar begins to generate abstract objects
that are interpreted in the sentences of the language (in its phonetic form and in its
meaning), that is, the set of abstract objects generated by the grammar is projected into
the language, describing it as a set of possible signifiers linked to a set of possible
meanings (pairs <s,m>, where s is a signifier and m is a meaning). Here the notion of
generative grammar undergoes a slight modification in relation to its previous meaning:
34
The limit can also be visual, as long as we consider that sign languages of the deaf are also natural languages and
actualize from the same biological matrix than the, say, ‘audio-oral’ languages.
it does not generate the sentences of the language directly. However, the commitment
with the notion of language remains the same, since grammar will generate as many
abstract objects as there are sentences in the language and no more. This makes it
possible to go on considering grammar as a ‘system of rules that generate the sentences
- and only the sentences - of the language’35.
At time 3 (P&P), grammar generates abstract objects that explicit the
properties that the speakers take into account when they issue evaluations on the
grammaticality of linguistic objects. The sentences of any language constitute only a
sub-set of this set of linguistic objects and so it is never - and under no criterion –
possible to say that grammar generates the sentences of the language – at the most it is
possible to say that grammar allows (licensees), among other things, the sentences of a
given language36.
Using Lakatos’s terminology, we can say that such ‘strategies’ characterize
different heuristics and that the GG program has experienced two major creative shifts:
the first involving SS Theory and the Standard Theory, the second involving the
Standard Theory and P&P. We can also consider that the best periodization of the
development of the program establishes three periods: the SS period, which goes from
Chomsky’s work (around 1954) to the publication of ‘Aspects’ (1965); the Standard
Theory period, which starts with the publication of ‘Aspects’ and goes till ‘Conditions
on Rules of Grammar’ (1976a)37; and the P&P period, starting with Chomsky 1976a and
continuing till today. In its first period, the theory is still very tied to the structuralist
ways of doing linguistics and, consequently, there is a certain conflict between the
requirements of the program and its theoretical availability: it is an unstable period in
the theory. In the second period, it is the very program that faces difficulties: there is a
proliferation of alternative heuristic proposals and there are dissenters. This second
period is characterizes by disputes and by a great theoretical ebullience. The third period
is a period marked by a great development in the expansion of the empirical content of
the theory: a large number of languages are analyzed to a satisfactory degree and the
principles are established in a very consistent form.
Finally, I believe that a few words about Noam Chomsky’s role in this story are
in place.
Chomsky has always been the great leader of the Generative Community,
imposing advances, rewriting the program, rejecting and/or supporting proposals.
Chomsky acts – and he is seen like that by the community – as the ‘owner’ of the
program, the person who utters the last words about the validity of the research lines
proposed by his associates, the person who says what must and what must not be
researched, the person who from time to time carries out a ‘balance’ of profits and
losses (conquests and theoretical costs) of the theory and proposes the major syntheses
35
The SS grammar had as output of the set of conversions the very sentences of the language; the standard theory had as
its output of the set of conversions a set of structures (syntagmatic indicators) that were interpreted in the sentences of
the language.
36
In Chomsky’s words: ‘We may perfectly well think of the grammar of, say, English, as assigning a structural
description to every possible sound. Some will be characterized simply as noise, others as sounds of perhaps some
language (but not mine), others as expressions of my language with some figurative interpretation, others as paired with
strict “literal interpretations,” and so one.’ (1981, p. 5)
37
Usually we find references to the text ‘Conditions on Transformations’ (Chomsky 1973) as being the text that allows a
‘quality leap’ in the GG program. However, I believe that no matter how important this text might be in the process of
change from a system of rules to a system of principles, in fact, it does not modify any of the fundamental concepts of
the program. The text ‘Conditions on Rules of Grammar’ (Chomsky 1976a), on the other hand, because it releases the
syntactic component, allowing for the ‘overgeneration’, forces Chomsky to recognize that the program will only have a
chance if its notion of language is abandoned. In my opinion, it is this second text that opens the third period of the GG
program. The fact the critics/commentators of Chomsky’s work do not recognize in ‘Conditions on Rules of Grammar’
the importance I do, make me feel uncomfortable, but I cannot be unfaithful to my convictions.
which will provide the new directions. Without much exaggeration, we could say that
GG has always been an essentially Chomsky’s creation. All those who did not agree
with Chomsky, at some point in the history of the program, either surrendered to the
power of the ‘master’, rejoining the ‘good path’, or became dissenters, remaining
marginal to the program. No matter how interesting the proposals presented by
Chomsky’s associates were, they are only incorporated effectively to the program’s
theoretical arsenal after having been supported by Chomsky. Chomsky’s centralizing
power is so strong that it is possible to find a book of about 250 pages devoted
exclusively to a survey of the destiny – invariably unhappy – of those who dared to defy
him (Botha 1989).
Nevertheless, an interesting fact starts to unfold. The ‘cold’ reaction from the
community to the Minimalist Program proposed by Chomsky seems to indicate that we
are living a moment in which the program is reaching a certain maturity and that the
community can already walk on their own feet, dispensing with Chomsky’s custody.
Any way, it is too early to make any reliable evaluation of the course the generative
enterprise will take from here.
4. References:
Anderson, S.; Kiparsky, P. (eds.) 1973. A Festschrift for Morris Halle. New York: Holt,
Rinehart and Winston.
Bloomfield, L. 1926. A set of postulates for the science of language. Language 2, pp.
153-164.
Borges Neto, J. 1991. A Gramática Gerativa Transformacional: um ensaio de Filosofia
da Lingüística. Tese de doutorado, IEL-UNICAMP.
Botha, R. 1989. Challenging Chomsky: The Generative Garden Game. New York: Basil
Blackwell.
Chomsky, N. 1951. Morphophonemics of Modern Hebrew. University of Pennsylvania
(unpublished M.A. dissertation).
Chomsky, N. 1953. Systems of syntactic analysis. Journal of Symbolic Logic 18 (3), pp.
242-256.
Chomsky, N. 1955a. Logical syntax and semantics: their linguistic relevance. Language 31,
pp. 36-45.
Chomsky, N. 1955b. Transformational Analysis. University of Pennsylvania
(unpbublished Ph.D. thesis).
Chomsky, N. 1955c [1975]. The Logical Structure of Linguistic Theory. New York:
Plenum.
Chomsky, N. 1957. Syntactic Structures. The Hague: Mouton.
Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press.
Chomsky, N. 1966. Topics in the Theory of Generative Grammar. The Hague: Mouton.
Chomsky, N. 1967. Remarks on Nominalization. In Chomsky 1972 (Spanish translation
in Sanchez de Zavala (comp.) 1974, pp. 133-187).
Chomsky, N. 1968. Deep structure, surface structure and semantic interpretation. In
Chomsky 1972 (Spanish translation in Sanchez de Zavala (comp.) 1974, pp. 276-
334).
Chomsky, N. 1971. Some empirical issues in the theory of transformational grammar.
In Chomsky 1972 (Spanish translation in Sanchez de Zavala (comp.) 1974, pp. 444-
524).
Chomsky, N. 1972. Studies on Semantics in Generative Grammar. The Hague: Mouton.
Chomsky, N. 1973. Conditions on Transformations. In Anderson & Kiparsky (eds.) 1973,
pp. 232-286.
Chomsky, N. 1976a. Conditions on rules of grammar. Linguistic Analysis 2, pp. 303-
351.
Chomsky, N. 1976b. On Wh-movement. In Culicover et. al. (eds.) 1977, pp. 71-132.
Chomsky, N. 1977. Dialogues avec Mitsou Ronat. Paris: Flammarion (tradução para o
português: Diálogos com Mitsou Ronat, São Paulo: Cultrix, s/d).
Chomsky, N. 1981. On the representation of form and function. The Linguistic Review 1
(1), pp. 3-40.
Chomsky, N. 1982a. The Generative Enterprise. Dordrecht: Foris.
Chomsky, N. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press.
(Portuguese translation: O Programa Minimalista, Lisboa: Caminho, 1999)
Culicover, P; Wasow, T.; Akmajian, A. (eds.) 1977. Formal Syntax. New York:
Academic Press.
Dascal, M. (org.) 1978-1982. Fundamentos Metodológicos da Lingüística, 4 volumes,
São Paulo: Global (vol 1); Campinas: organizer’s edition (vol. 2, 3 e 4).
Dascal, M. 1978. As convulsões metodológicas da lingüística contemporânea. In Dascal
(org.) 1978, vol. 1, pp. 15-41.
Emonds, J. 1970. Root and Structure Preserving Transformations. Bloomington:
Indiana University Linguistic Club (IULC).
Feyerabend, P. 1974. “Consolations for the Specialist” in Lakatos & Musgrave (eds), p.
197-230.
Galmiche, M. 1979. Semântica Gerativa. Lisboa: Presença.
Harris, R.A. 1993. The Linguistic Wars. Oxford: Oxford University Press.
Harris, Z.S. 1952. Discourse Analysis. Language 28, pp. 1-30 e 474-494.
Harris, Z.S. 1957. Coocurrence and transformation in linguistic structure. Language 33,
pp. 293-340.
Harris, Z.S. 1965. Transformational Theory. Language 41, pp. 363-401.
Harris, Z.S. 1968. Mathematical Structure of Language. New York: Intersciences.
Hockett, C.F. 1955. A manual of phonology. In Antropology and Linguistics: Memoire
11. Baltimore: Indiana University Publications.
Kato, M. 1974. A Semântica Gerativa e o Artigo Definido. São Paulo: Ática.
Lakatos, I. 1970 Falsification and the methodology of scientific research programmes.
In Lakatos & Musgrave (eds) 1974, p. 91-196 (republished in Lakatos 1978 p. 8-
101).
Lakatos, I. 1971. History of Science and its Rational Reconstructions. In Lakatos 1978,
pp. 102-138.
Lakatos, I. 1978. The Methodology of Scientific Research Programmes (Philosophical
Papers vol 1). Edited by John Worrall & Gregory Currie. Cambridge: Cambridge
University Press.
Lakatos, I. & Musgrave, A. (eds) 1974. Criticism and the Growth of Knowledge. (3rd
impression) London: Cambridge University Press.
Lakoff, G. 1971. On generative semantics. In Steinberg & Jakobovits (eds.) 1971, pp.
232-296.
Lees, R. 1957. Review of Noam Chomsky, Syntactic Structures. Language 33, pp. 375-
408 (reprinted in Otero, C. (ed) 1994. Noam Chomsky: critical assesment. Vol 1
London; Routledge).
Lobato, L.M.P. 1986. Sintaxe Gerativa do Português: da teoria-padrão à teoria de
regência e ligação. Belo Horizonte: Vigília.
Milner, J-C. 1973. Écoles de Cambridge et de Pennsylvanie: deux théories de la
transformation. Langages 29, pp. 98-117.
Newmeyer, F.J. 1980. Linguistic Theory in America: the first quarter century of
transformational generative grammar. New York: Academic Press
Peters, S.; Ritchie, R. 1969. A note on the universal base hypothesis. Journal of
linguistics 5, pp. 150-152.
Peters, S.; Ritchie, R. 1971. On restricting the base component of transformational
grammars. Information and Control 18, pp. 483-501.
Peters, S.; Ritchie, R. 1973. On the generative power of transformational grammars.
Information Sciences 6, pp. 49-83.
Quícoli, A.C. 1976. Conditions on clitic movement in Portuguese. Linguistic Analysis 2
(3), pp. 199-223.
Raposo, E.P. 1999. Da teoria de Princípios e Parâmetros ao Programa Minimalista:
algumas idéias-chave. Apresentação da tradução portuguesa de Chomsky 1995, pp.
15-37.
Ross, J.R. 1967. Constraints on variables in syntax. Bloomington: Indiana University
Linguistic Club (IULC).
Sanchez de Zavala, V. (comp.) 1974. Semántica y sintaxis en la lingüística
transformatoria I. Madri: Alianza.
Wells, R. 1947. Immediate Constituents. Language 36, pp. 360-367.

The Generative Enterprise

Uploaded by

Copyright:

Available Formats

The Generative Enterprise

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Generative Enterprise

Uploaded by

Copyright:

Available Formats

THE GENERATIVE ENTERPRISE1

José Borges Neto2

Let us start with the concepts of proliferation and tenacity.

2. GG: a Scientific Research Program.

2.1. The chomskian SRP: nucleus and heuristics.

2.2 The first proposal of generative grammar: LST and SS.

Syntactic Structures (Chomsky 1957 – hereafter SS). Chomsky’s previous works –

formal notion of grammar, understood as a computing system (a generative grammar,

We will find it necessary to distinguish at least the following levels for

(3)b. (((Pedro)N )SN ((viu)V ((Maria)N)SN)SV)S

To produce a sentence from such a grammar we construct an

considered the grammar present in the speakers’ main/brain.

possible to define sequences of well-formed variables11. Harris calls these sequences of

studied seriously in relation to Portuguese data.

attestées de la tradition grammaticale ; en termes extensionnelles, la

2.3. Generative Grammar in the sixties: the standard theory.

SS  Phonetic interpretation  Phonological c.

The syntactic component specifies an infinite set of abstracts formal

2.4 From rules to principles.

but on the explanatory adequacy. Instead of having a proposition of computing systems

(13) Paulo parece [t os ter examinado]

2.5. The Minimalist Program

You might also like