Upload
lenga
View
226
Download
0
Embed Size (px)
Citation preview
明治大学教養論集 通巻255号 外国語・外国文学(1993)pp.15-31
DEEP STRUCTURE AND
SYNTACTIC THEORY
James R. Bowers
DEEP STRUCTURE AND SYNTACTIC THEORY………・………・…・15
1NTRODUCTION・…・……………・……………・・……………………・16
PRECURSORS TO CONTEMPORARY NOTIONS OF DEEP
STRUCTURE……………・…・…………・・………………・…・・…・…・…17
The Port Royal Grammarians・………・……・………………17
Wilhelm von Humbolt and innere Sprachform………・-18
Ferdinand de Saussure…………・……・……………………・…19
GENERATIVE SYNTAX AND THE NOTION OF DEEP
STRUCTURE・………………・………・…………………………・・……・20
The助π航’ゴ6 Structures Model………………・…・………・…20
Underlying Trigger Morphemes…………・……・……・………22
The Katz-Postal Hypothesis…・…・…・………・………・・……23
The AsPects Model・…・………・…・…一………・…・・………・…24
Abstractness ………・……………・…・・…・・……………・・……・25
Generative Semantics and Deep Structure………・………・・26
The Lexicalist Hypothesis……・…・・…………・・……・………28
Contemporary Theories…………・…………………・・………29
REFERENCES ………………・…・……………・・…・…………・………30
一15一
INTRODUCTION
Crystal (1991; 94-95) asserts that deep structure is a ‘‘central
theoretical term in transformationql grammar.” @(Crystal uses the term
transformational grammar as synonymous with generative grammar・I
will use the latter term in this paper because of the controversy concern-
ing the transformational notion that characterizes some of the discussion
of deep structure.)He goes on to define deep structure as “...the
abstract syntactic representation of a sentence--an underlying level of
structural organization which specifies all the factors governing the way
the sentence should be interpreted.” @ 一
Crystal’s definition captures some commonalities of the idea of deep
structure which can be traced from the Port Royal grammarians, through
Wilhelm von Humbolt and Ferdinand de Saussure to the generative
syntactics of the latter half of the twentieth century. His definition also
refers to the two key controversies concerning deep structure that have
characterized the evolution of syntactic theory since Chomsky’s 1957
1andmark work:Syntactic Structures. They are the question of degree of
abstractness and the question of whether deep structure ‘‘specifies all the
factors governing the way the sentence should be interpreted。” @This
paper will look briefly at the history of the notion of modern precursors
to deep structure。 It will then examine the evolution of that concept from
the Syntactic Structures model through contemporary models of genera-
t1Ve grammar・
一16一
PRECURSORS TO CONTEMPORARY NOTIONS OF DEEP STRUCTURE
The Port Royal Grammarians
Robins(1990:137-138)points out that the structural interpretations
that the Port Royal grammarians gave to the functions of certain classes
of words was “the positing, in modern terms, at a deeper structural
level, of elements that in actual sentences were represented conjointly
with other elements.”This notion is similar to the generative semanticist
concept of lexical decomposition that will be discussed below. Specifi-
cally, Robins reports the Port Royal grammarians considered adverbs to
be abbreviations for a prepositional phrase, an analysis that can be found
in Syntactic Structures.
The abstract analysis of verbs received much attention as well.
Reviving an analysis that originated with Aristotle, the Port Royal
grammarians viewed all verbs other than the copula as consisting of a
verb and a particle. Robins’example is, Peter lives viewed as structurally
equivalent to Peter is living, another analysis similar to early Chomsky
and, later, the Generative Semanticists. Finally, the Port Royal gram-
marians saw such aspect features as passive and transitive as associated
with the adjectival nature of verbs rather than as a property of verbs
themselves.
一17一
In Cartesinn LinguiStics, Chomsky explicitly attributes the idea of
deep structure to the Port Royal grammarians.“ln short language has
an inner and outer aspect. A sentence can be studied from the point of
view of how it expresses a thought or from the point of view of its
physical shape, that is from the point of view of either semantic interpre-
tation or phonetic interpretation.” @(1966:32-33) ‘‘Using some recent
terminology we can distinguish the‘deep structure’ of a sentence from its
‘surface structure.’(1966:33) “This point is brought out with particular
clarity in the Port Royal Grammar, in which a Cartesian approach to
language is developed, for the first time, with considerable insight and
subtlety.”@ (1966:33)
●
Wilhelm von Humbolt and innere Sprachform
As we will see below, one of the facts of human language that led
Chomsky to postulate a level of deep structure in syntax in his Syntactic
Stractures was the‘creative’ability of human beings to generate an
infinite set of sentences from a necessarily finite syntactic system. Such
amodel needed to account for ambiguous sentences on the one hand and
synonymous sentences on the othe士. Assuming a level of abstract struc-
ture independent of surface representation accounted for these facts
handily at the time. However, Chomsky was not the first to suggest that
this fact of human linguistic creativity was central to a theory of lan-
guage・
Wilhelm von Humbolt(1767-1835)is reported by Robins to have
asserted that,“A language is to be identified with the living capability
by which speakers produce and understand utterances...” @(1990:193)
一18一
This living capability was, in von Humbolt’s theory of language, for-
mally constituted as “the semantic and grammatical structure of a
Ianguage, embodying elements, patterns and rules imposed on the raw
material of speech.” @(Robins,1990二193)Von Humbolt’s term for this
formal structure was innere Sprachform.In that it postulates an abstract
level of language separate from the physical representation of the ele-
ments of language, it can be said to anticipate the contemporary concept
of deep structure. In particular, note the parallel with Chomsky’s state・
ment regarding the inner and outer aspect of language quoted from
Cartesinn Linguistics above.
Ferdinand de Saussure
The view that“...language can be studied from the point of view
of how it expresses a thought or from the point of view of its physicaI
shape...” @quoted from Cartesian Linguistics was also stated by
Ferdinand de Saussure. As Robins reports, de Saussure expressed this as
langue空.forme, non subsinnce.(Robins,1990:221)Language is form
without substance. Form on the content plane deals with semantics and
grammar. As Anderson points out,“De Saussure, himself, apparently
held that the domain of the sign relation(the minimal scope within which
phonological form is consistently associated with its semantic content)
was the word or complex form, not the morpheme or simple form.”
(1988:152)This view of the word seems to presage the lexical de・
composition position of the generative semanticists as well as Chomsky’s
own early stand in Syntactic Structures. The distinctions of form from
substance and complex structure separate from phonological form
require the postulation of an abstract layer of representation, which is
our definition of deep structure.
一19一
GENERATIVE SYNTAX AND THE NOTION OF DEEP STRUCTURE
The Syntactic Stntctzares Model
De Saussure’s theorizing concerning the langue/parole distinction is
not exactly equivalent to Chomsky’s competence/performance distinc・
tion. Much of what generative grammarians would today call syntax was
assigned by de Saussure to parole. Chomsky’s contribution was to place
syntax at the core of theorizing about human language, a step necessary
to account for the creativity that is its distinctive feature.
Chomsky viewed syntax in formal terms and established as criteria
for an adequate grammar that it be able to generate all and only the
sentences of a natural language, and be able to assign structural descrip-
tions to them. Generate is a mathematical term. It is used in a technical
sense of mapping;not in the everyday notion of actually making sen-
tences.
In Syntactic Structures, Chomsky postulated three levels to generate
sentences that would meet his criteria of obse】rvational adequacy. The
levels were a phrase structure level, a transformational level, and a
morphophonemic level.
The phrase structure level was a direct continuation of Zellig
Harris’s concept of kernel sentences. This component generated a finite
-20 一
set of strings with structural interpretations. The phrase structure com-
ponent was finite in the sense that it contained no rules for handling the
recursiveness of natural languages.
It was the transformational level of the theory of grammar proposed
by S.・ソntactic Strzactures that introduced the concept of an abstract‘deep
structure.’Transformational rules operated on the strings generated by
the phrase structure Ievel to account for such properties as infinite
recursiveness, embedding, ambiguity, synonymy, and the intuitive
relatedness of such surface structures as statements and yes/no questions
and the active/passive voice distinction.
The morphophonemic level assigned a phon610gical representation to
the output of the transformational leve1. As an example, in a sentence
such as John十take十PAST十a十cookie the rules of this level would
operate to yield the surface string:John took a cookie.
From Roberts(1964)we have the following examples of rules from each
level:
S-一一>NP十VP
T-do;tense >do十tense
T-conji insert:X(1)+A(2)+Y(3)
matrix;X(4)十B(5)十Y(6)
result:4十2十Conj十5十6
take十(PAST)一一一>took
(Phrase structure level)
(Single base transformation)
(Double base transformation)
(Morphophonemic level)
一21一
In the Syn tactic Structures model, the role of the transformational
level is to link the relatively abstract phrase structure level to the much
less abstract morphophonemic level. Transformations allow generaliza・
tions to be made about a number of relationships that obtain among
strings generated by the phrase structure level. The model does not deal
directly with the relationship between meaning and structure. In fact,
Chomsky argues vigorously for an autonomous syntax.
Underlying Trigger Morphemes
One development of the Syntactic Structu.res model of syntax was a
belief that deep structure, the undβrlying levels of phrase structure and
transformational structure, could account for “..,all the factors
governing the way the sentence should be interpreted.” @(Crystal 1990:
94)However, if transformations such as the passive or yes/no question
formation were considered optional as in the original conceptualization,
this was clearly not the case, In the early’60’s investigation of co・
リ コ ロ
occurrence restrlctlons on negatlves led to the conclusion that such
transformations had to be obligatory. If they were obligatory and if the
deep structure contained all the faごtors governing how a sentence was to
be interpreted, then there had to be some element in the deep structure
representing such necessary meanings as questioning, passive voice,
negat10n, etC・
The solution was to postulate abstract triggering morphemes such as
Q(Question)or N EG(negation).This was an initial step toward a more
abstract conception of deep structure. It also was a commitment to the
idea that transformations were meaning preserving.
一22一
The Katz・Postal Hypothesis
Syntactic Stntctures in promoting an autonomous syntax ignored the
question of the relationship between syntax and semantics. However,
semantics became a concern with the postulation of underlying trigger
morphemes. Katz and Postal in their 1964 book,An。lntegrated 7劾oη(ゾ
Linguistic Descn’垂狽奄盾獅刀C carried the process a step further by explicitly
stating that transformations did not affect meaning. They argued that
the concept of abstract, meaning-bearing morphemes allowed nominal-
izations to be analyzed as structures with abstract noun heads represent-
ing each nominal’s abstract qualities. One way of discovering such
abstract heads was to employ a heuristic. This heuristic is known as the
Katz-Postal hypothesis, and its effect was to increase the abstractness of
the deep structure level by requiring additional levels of derivation before
the output of deep structure became available to the morphophonemic
level. The Katz-Postal hypothesis also paved the way for the advent of
generative semantics. Newmeyer(1986:71)summarized the hypothesis
as follows:
Given a sentence for which a syntactic derivation is needed;look
for simple paraphrases of the sentence which are not paraphrases by
virtue of synonymous expressions;on finding them, construct gram-
matic41 rules that relate the original sentence and its paraphrases in
such a way that each of these sentences has the same sequence of
underlying P-markers. Of course, having constructed such rules, it is
still necessary to find INDEPENDENT SYNTACTIC JUSTIFICATION for
them.(Katz and Postal 1964:157, emphasis in original.)
一23一
、
Although the motivation for the Katz-Postal hypothesis is based on
an analysis of English nominals, there is a clear parallel with the Port
Royal Grammar’s analysis of verbs.
The AsPects ModeI
It has been noted that the phrase structure level of the model of
grammar proposed by Chomsky in Syntactic Structures had no provision
for recursion. That was handled by the transformational level. In 1965,
Chomsky introduced an essentially, completely new model in、AspectS on
the Theory of Syntczx . The new-model postulated three sets of rules that
comprised deep structure:phrase.structure rules, subcategorization
rules, and rules of lexical insertion.(Newmeyer,1986:75)Recursion
was now handled by the phrase structure rules, Subcategorization rules
and lexical insertion rules were completely new.
The phrase structure rules and the subcategorization rules
constituted base rules. The former generated phrase markers that pro・
vided information on categories such as noun phrase, verb phrase, etc.
and defined grammatical relationships such as subject of sentence, direct
object of verb, etc. The subcategorization rules were composed of
context free subcategorization rules such as those that distinguish proper
nouns from common nouns in English, context sensitive subcategoriza-
tion rules which defined lexical categories in terms of syntactic frames
such as whether a verb required an object or not, and selectional restric-
tions which subcategorized verbs on the basis of such criteria as whether
they required a human subject or not. The lexical insertion rule inserted
一24 一
lexical items into the phrase structure generated by the base rules on the
basis of whether their structural descriptions matched those on the nodes
of the phrase structure.
There was also a transformational component. It applied to the
output of the base and lexical insertion rules. Transformations were now
obligatory and applied cyclically in order from the most basic embedded
sentence to the highest. In fact, the primary role of the transformational
component was to deal with embedding. Relatedness between sentences
was handled by elements in the base rules. Idiosyncratic properties were
relegated to the lexicon・
Abstractness
Before turning to further developments of syntactic theory as related
to the notion of deep structure, it is necessary to discuss the question of
abstractness. In the AspectS model of,syntax compound or complex
sentences were the result of transformations operating on structures
generated by the base rul〔?s and the lexical insertion rules・
Generally, the more complex a sentence was, the more embedded
sentences it had in deep structure, and the more cycles of transformations
it had. to undergo to produce a surface structure. Thus, it became
possible for surface structure to be radically different from the form of
deep structures. This degree of difference as, defined by distance in
terms of applications of transformations, is the syntactic notion of
abstractness.
一25一
The merit of abstractness in linguistic theory is to enable generaliza・
tions to be made which reveal underlying regularity in apparently quite
diverse structures. For example, if we observe in Japanese that the
syllables‘ta’‘chi’‘tsu’‘te’‘to’are actually combinations of a single sound
[t]with a vowel and with accompanying phonetic modification, we not
only establish an underlying commonality, but we are able to observe a
regularity that, in fact, corresponds to a general process in the language,
ka, ki, ku, ke, ko;pa, pi, pu, pe, po;ba, bi、 bu, be, bo, etc.
The downside of abstractness is that it may only be a convenient
fiction with no correspondence to reality. At its worst it may obscure the
very generalizations and facts about language that we wish to investi-
gate. Halle’s(1957)dissertation on Russian, for example, demonstrated
that the classic structuralist formulation of the phoneme obscured the
.generalization that obstruents are voiced uniformly in that language. The
intermediate level of abstractness required by the phonemic level’s inter・
vention between surface represelltations and underlying forms forced an
arbitrary split of that process intO two iden声ical rules:one describing
allophonic variation, and the other describing morphophonemic varia-
tion.(adapted from Durand,1990:23-24)This issue of levels of represen-
tation and abstractness in syntax was to become the driving force in the
dispute between Chomsky and those linguists who subscribed to the tenets
of generative semantics during the decade of the 1970s.
Generative Semantics and Deep Structure
For Chomsky in the AspectS model, deep structure was a kind of
cover term for the interlocking set of rules necessary to produce a
一26一
grammar that was both observationally and descriptively adequate. The
latter criteria was new to the.AspectS model, and was used to distinguish
among rival grammars that met the criteria of observational adequacy.
Halle’s argument with regard to the level of phonemic representation is
one that revolves on explanatory adequacy・
However, in the mid-60s, other syntacticians influenced by the
Katz-Postal hypothesis began to look on deep structure as a way of
linking semantics to syntax, something Chomsky has dealt with only
lightly. They began to posit deep structures that represented every aspect
of the meaning of a sentence under investigation, and this led to abstract
descriptions. One element of this process was the decomposition of
lexical items into highly abstract semantic notions such as‘kill’=CAUSE
TO D l E, and embedding of sentences to a very great depth. For example,
in McCawley(1988:241),the sentence:“Tom hasn’t been watching
television recently,”is analyzed with a deep structure composed of five
embedded sentences, the four topmost of which serve only to represent
negation and aspect.
The most extreme form of the generative semanticist’s view of deep
structure is set forth by Lakoff and Ross(1976).Deep structure(quoting
from Newmeyer 1986:92)was:
12
3.
the base of the simplest syntactic component
the place where subcategorization and selectional
restrictions were defined
the place where basic grammatical relations were defined
一27一
4.the place where lexical items were inserted from the
Iexicon
However, observations 1-3 c6uld be argued to be properties of
semantic representation, and, therefore, no independent level of deep
structure was needed. The concept of an independent level of deep
structure was abandoned altogether.
The Lexicalist Hypothesis
On the basis of his work on nominalization, Chomsky challenged
abstract syntax by means of a concept called the lexicalist hypothesis.
The hypothesis eliminates category changing transformational rules from
the grammar. A category changing transformation would be one that
turned a verb into a noun, for example. The three arguments Chomsky
presented in favor of the lexicalist hypothesis argued for a‘shallow’deep
structure. The first argument asserted that derived nominals occurred in
sentences corresponding to base structures, but never transformationally
to derive structures. This challenged the generative analysis of such
nominals.(Newmeyer ,1986:107)The second argument was that a
transformational rule should capture a regular productive relationship.
This led to the presentation of the X-bar convention which is now stan-
dard with all current theories of syntax.(Newmeyer,1986:108)The
third argument asserted that the structures in which derived nominals
occur resemble noun phrases in every way . This would follow automati-
cally from the postulation that derived nouns are nouns in the lexicon and
are inserted as such in deep structure. This argues against the abandon-
ment of an independent deep structure.(Newmeyer,1986:109)
一28一
Eventually, the very arguments generative semanticists used to
support the abandonment of deep structure resulted in its rehabilitation.
Lexical decomposition was attacked on two grounds. The first was that
there were many instances where the semantic and syntactic behavior of
lexical items was at variance with the behavior of their supposed sources.
The second was that lexical decomposition predicted more ambiguities
than actually occurred.
The strongest argument in favor of rehabilitating deep structure was
that semantic specification could not deal with specifying morpheme
order. Ordering relationships for most classes of morphemes are simply
not predictable on semantic criteria alone・
Contemporary Theories
In the end, the most telling argument against generative semantics
was that its constructs had become too powerful. They were so abstract
that anything could be potentially explained by their theory. Unfortunate-
ly, such a strong theory might have very little insight to offer into the
actual working of natural languages. In the 1980s a movement was made
toward strongly constrained theories. All retain the concept of deep
structure but it is generally a modest concept. In Chomsky’s government
and binding theory deep structure is where movement alpha, the only
remnant of a‘transformation’takes place. In generalized phrase struc-
ture grammar, deep structure is the locus of the phrase structure compo-
nent. In lexical-functional grammar, deep structure is where the lexicon
projects structure・
一29一
Deep structure is no longer the complex region where transforma-
tions operated under th6∠AspectS and generative semantic models. Yet, it
has a role to play in accounting for a sufficient level of abstractness to
reveal actual regularities in language rather than the artificial constructs
of linguists. Just as the phoneme has again found a role as a psychological
construct in natural phonology as oPPosed to the structuralists’analytical
role, so too has deep structure returned to the fold as a central tenet of
generat1Ve grammar・
REFERENCES
Anderson, S.(1988)Morphologi¢al theory. in Newmeyer(1988).146-
191.
Andrews, A.(1988)Lexical structure. in Newmeyer(1988).60-88.
Botha, R.(1989)Challenging Chomskγ. Oxford:Basil Blackwel1.
Chomsky, N.(1957)Syntactic Stractures. The Hague:Mouton.
(1965)AspectS of the Theory Of 5吻燃. Cambridge , MA:The
MIT Press.
(1966)Carteslan Linguistics .()ambn’dge, MA:The MIT Press.
Crystal, D.(1991)ノ1 Dictio na7 y(ゾLingudStics and、Phonetics(3rd ed.)
Oxford:Basil Blackwell
一30 一
Durand, J・ (1990) Generative and .〈1∂n-Linear 、Phonology. London:
Longman.
Katz, J・and P・Postal(1964)An Integrated 7物60η(ゾLingudStic
Descriptions.Cambridge, MA:The MIT Press.
McCawley, J.(1988)The Syn tactic Phenomena(ゾEnglish Vol.1.
Chicago:The University of Chicago Press.
Newmeyer, F.(1986)LingudStic Theo2ry勿、4 merica(2nd ed.).New
York:Academic Press.
(Ed.)(1988)LinguiStics, The Cambri°dge Surveyγ∂1.1∴乙inguts-
tic Theo7 y: Foundations・Cambridge University Press.
Roberts, P.(1964)English Syntax. New York:Harcourt, Brace&
World.
Robins, R.(1990)ノ1 Short、Histoiry(ゾ、乙inguiStics(3rd ed.).London:
Longman.
Sells, P.(1985)Lectures on Contemψora7Zy Synltzctic Theories.Stanford,
CA: DCLSI
一31一