110 n 1 1 P4RNT RP.SIIM ED 030 113 AL 001 969 By -Schank. Roger C. Linguistics from a Conceptual Viewpoint(Aspects of Aspects of a Theory of Syqtax). Stanford'Univ.. Calif. Spons Agency -Department of Defense, Washington. D.C. Advanced Research-Pro;ectsAgency.; National Inst. of Mental Health (DHEW), Bethesda, Md. Report No-MEMO -AI -88 Pub Date 21 Apr 69 Note -23p. EDRS Price MF -$0.25 HC -$1.25 Descriptors-Child Language. Cognitive ProCesses, Computational Linguistics,*Concept Formation, *Language Universals. *Linguistic Competence, Linguistic Performance. *LinguisticTheory. Models. Semantics, Syntax, *Transformation Generative Grammar Identifiers-*Aspects Of The Theory 'Of *Syntax. Chomsky Noam. ConceptualParser. Stanford Artificial Inteliigence Project . Some of the assertions made by Chomsky in "Aspectsof the Theory of Syntax" are considered.In particular. the" ..notion of a "competence:" model inlinguistics is criticized. Formal postulates for a *Conceptually-basedlinguistic theory are presented. (Author/JD) April 21, 1969 STANFORD ARTIFICIALINTELLIGENCE PROJECT MEMO AI-88
WELFARE U.S. DEPARTMENT OFHEALTH, EDUCATION & OFFICE OF EDUCATION
RECEIVED FROM THE THIS DOCUMENT HAS BEENREPRODUCED EXACTLY AS POINTS OF VIEW OROPINIONS PERSON OR ORGANIZATIONORIGINATING IT. REPRESENT OFFICIM. OFFICEOF EDUCATION STATED DO NOT NECESSARILY POSITION OR POLICY.
Linguistics from a Conceptual Viewpoint
(Aspects of Apects of a Theory ofSyntax)
by
Roger C. Schank
Theory ABSTRACT: Some of the assertions madeby Chomsky in AspecIs_of the of Syntax are considered. In particular, the notion of a competence' model in linguistics iscriticized. Formal pos- tulates for a conceptually-basedlinguistic theory are presented.
This research is supported by GrantPHS MH 06645-07 from the National Institute of Mental Health, and(in part) by the Advanced Research Projects Agency of the Office ofthe Secretary of Defense(SD-183).
AL001969 Linguistics from a Conceptual Viewpoint
(Aspects of Aspects of a Theory of Syntax)
1. Although linguistics is often referred to as the 'science of language', it has been usual for linguists to resist this label and its concomitant expectations.With the exception of the attempts of Bloch (19)48) and Bloomfield (1926) at establishing postulates for language, the formaliza- tion that is often present in a science was difficult to find before 1957.
With the advent of transformational generative grammar however, the direction of the bulk of linguistic research changed considerably. Linguists who are in the transformational camp tend to be more concerned with precision and formalization than was common in the past.
It is perhaps enlightening to attempt to discover just how valid the assumptions that have led to the present conception of transformational grammar are; and, if they are not valid, agood question is just what assump- tions might be considered valid.
Initially, transformational theory dealt only with kernelsand transforms. The kernels were the output of the PSG and certain obligatory transformations. Thus a transformational grammar was simply a specification of the permutations involved in correcting the output of a PSG. This system was still used until itbecame clear that it was insufficient to handle all
the complexities involved in natural language.
Bach (1964) stated that 'The minimal requirement that we can place
on a theory about a particularlanguage is that it specify or predict all
and only the sentences of the language'. Until this time, these were precisely
the requirements that were placed on a transformational grammar. Hawever,
two important problems resulted from this view ofwhat a linguistic theory was to accomplish. The first was that the systom of phrase structure gram- mar and the transformationsthat were added simply did not account for all
the sentences of English. The second problem is that the system As it stood
could not be extended to other languages without completel.y revampingboth
components in such a way that the two grammarswould be very dissimilar.
This clearly is in contradiction to a fact of language that we know verywell..
Namely, than any baby possesses the mechanisms to learn any language andthat
therefore, there must be some similarity between languages. Furthermore,
this similarity should pervade the grammars of different languagessuch that
things like translatability can be explained.
Thus it appeared obvious that an important step towards solving
both the first and second problems would be the positing of aUniversal
grammar that would hold forall languages and a particular grammar for each
language. Therefore, since the framework was alraady available, the phrase
.01.1.04601* r s I
structure component began tobe utilized as the basis for auniversal base
component. (As Robinson [1968] points out this was arelatively arbitrary
decision.) At this pointChomsky's AsEstsof the_l_y_lheor c_layr._IS.2.x appeared.
,
L - ...... MW down the 'mainbackground 2. In Chapter I ofAspects, Chomsky sets examine some assumptions' of transformationaltheory. It is worthwhile to
of these assumptionsin detail. assumption that hemakes is The first and perhapsmost important
(the speaker-hearer'sknow- the 'fundamentaldistinction between competence (the actual use of languagein con- ledge of hislanguage) and performance
cretesituations) [p.4] that he sets up. performance provides for This distinction betweencompetence and make their statementsabout transformationalists theplatform from which to discussion of what a'performance' transformations. Chomsky includes in his limitations, inattention,distraction model should do,factors such as memory 'competence' the formal- and non-linguisticknowledge. He thus leaves for speaker-hearer's know- ization of linguistic processesrepresentative of the
ledge of the language. however. It is This relegation ofcompetence makes abasic mistake of linguistic know- necessary todifferentiate theproblem of formalization from the simulationof linguistic ledge and processes,i.e., competence, and cannot be labeled as knowledge and processes. This simulation is not There is a differencebetween the what Chomsky wouldcall 'performance'. simulation of actualverbal simulation of knowledgeand processes and the Chomsky does, of theideal behavior. It is here that wemust speak, as speaker-hearer is notinattentive or dis- speaker-hearer. Clearly the ideal and non-linguisticknow- tracted. He does howeverhave memory limitations what must be simulated as aninclusive part of ledge. This is certainly which Chomsky linguistic theory. The kind of theoryof 'performance' of future to which Chomskyrelegates it. speaks may well be inthe far distant is not so far off. It would However, a theory ofsimulative performance seem very reasonablethat the possibility of the construction of alinguistic theory that both accounts for the dataand does this in such a way as to appear to be consonantwith the human method for doing so, is not so remote.
Clearly, such a theory must deal withnon-linguistic knowledge and problems
of human memory as well as theprdblems that Chomsky designates as'competence'.
Thus, it seems that the distinctionbetween competence and performance is a
contrived one at best. In particular, after elimination of someGI the be-
havioristic problems such as distraction, we canexpect to find a linguistic
theory that is neither one of'competence' or 'performance' but something in
between and therefore partiallyinclusive of both.
On Page 139 Chomsky states:
Thus it seems absurd to suppose thatthe speaker first forms
a generalizedPhrase-marker by base rules and then tests itfor
well-formedness by applying transformationalrules to see if it
gives, finally, a well-formed sentence. But this absurdity is
simply a corollary to the deeperabsurdity of regarding the sys-
tem of generative rules as apoint-by-point model for the actual
construction of a sentence by aspeaker.
The system presented in a theoryof simulativr, performance, on
the other hand, would be an attempt atformulating a system of rules that are
of a sentence by a speaker. a point-by-pointmodel for the actual construction
Furthermore, this system should be apoint-by-point model for theactual
analysis of a sentence by a hearer.These aims even if they areattainable
would be largely unverifiable exceptby the use of computers assimulative
devices.
On page 141 Chomsky.further statesthat: 6
The grammar does not, initself, provide any sensible
procedure for finding the deepstructure of a given sentence, or
for producing a given sentence,just as it provides nosensible
procedure for finding aparaphrase to a given sentence. It merely
defines these tasks in a precise way. A performance model must
certainly incorporate a grammar;it is not to be confusedwith a
grammar.
Thus it would be wise totake the notion of asimulative perfor-
and mance model asbeing somewhere betweenChomsky's notion of competence performance. Perhaps the notionof grammar too, shouldlie somewhere between
Chomsky's notion of a grammar andthe incorporation of a grammar.
Chomsky describes a grammarof a language as'a description of the
and calls this a'genera- ideal speaker-hearer'sintrinsic competencev[p.4] 'generative grammar' tive grammar'. It is reasonable toconsider whether a describing what Chomsky in a large sense mightalso concern itself with calls 'competence', but in a wayin which the idealspeaker-hearer does so;
considerations as the speaker- in particular withoutexcluding such important
hearer's knowledge of theworld.
Now it is possible toredefine the notion of a'fully adequate
adequate grammar 'assign toeach of an grammar'. Chomsky would have a fully how this infinite range of sentences astructural description indicating However, the sentence is understoodby .the ideal speaker-hearer[p.5]. therefore by transformational structural descriptionemployed by Chomsky and question as to whether syn- grammar is syntacticin nature and there is some by the speaker- tactic descriptions reflectat all how a sentenceis understood 'fully adequate grammar' hearer. It would seem thatalthough the goal of a arrives at this grammar,and is to the point, themethod by which Chomsky therefore the grammar itself, isclearly not. That is, when Chomsky states that 'a grammar is descriptivelyadequate if it generates the correct setof structural descriptions' [p.60] it is reasonable to ask what makes aset of structural descriptions'correct', Since a transformational grammarat- tempts to be justified atthe level of descriptive adequacy, we canassume that to Chomsky a 'correct'structural description includesthe notion of what the Noun Phrase, VerbPhrase, and for that matter, thedummy symbol to be found in the structuraldescription are. If 'correct' for Chomsky does include only a competence grammarthen it must be clear thatthe struc- tural description provided by itdoes not explain how aspeaker-hearer goes about understanding a sentenceand that this problem does notbelong to a performance grammar. The needed grammar must be inbetween Chomsky's
extremes and thestructural descriptions that itprovides must be in terms
of items more familiar tothe speaker-hearer than'dummy symbol'.
Thus, Chomskys 'fundamentaldistinction' between 'competence' and
1 performance' is fundamental only insofar as one would want todevelop a
adequate job of pro- competence grammar. Although Chomsky may have done an
7iding the basis of a competence grammar,the question remains as to what
the point of such a grammaris. Chomsky states that linguistictheory is
mentalistic in that it isconcerned with discovering amental reality under-
lying actual behavior[p.4]. However, results have beenlargely negative
when attempts to provethe.psychological validity of this competence grammar
have been made. Fodor and Garrett(1966;151) comment:
What is one t, make of such negativefindings? The simplest
move would be todeny the validity of theexperimental procedures...
If one is to deny the validityof suchprocedtres in cases where
they appear to fail, it seems onewill equally have to denytheir
41% 8
It is in any validity in the caseswhere they appear tosucceed.
negative data willeventually event nowconceivable that enough the theory that is at accumulate to make onewonder whether it is
...it is a mistake toclaim fault rather thanthe experiments generate psychological reality forthe operations whereby grannnars
structural descriptions.
that their theorydoes not make Thus, eventransformationalists recognize anyverifiable psychologicalclaims. explicating psychological If a competence grammaris not useful for It seems to methat the purpose processes whatother purpose couldit have? than just explainingthings about of a linguistictheory is somehow more 'traditional linguistics'explained language. What Chomsky refers to as but Chomsky and hisfollowers have deemedthis. a greatdeal about language, how formalizationfor its awn sake insufficient. It is hard for me to see A generative grammarshould account would make theseexplanations sufficient. is unverifiable,the formal- for psychological processesand as long as this generative grammarshould be useful in ization techniquespresented by a Clearly they are neededin computational any sensethat they might beneeded. have consistentlybeen very difficultto linguistics. But transformations reversable the outputwould be reverse forparsings. Even if they were of a transformationalgrammar syntactic, since jt isthe syntactic component The question remains asto whether nhat accounts forgenerative capacity. information, such as syntactic information, evenmeaning-oriented syntactic That is, is underlying com- deep structures, canaccount forunderstanding. would a model ofsimulative performance petence significant? In particular, from a psychologicaland compu- not be somewhat moremeaningful, at least
tational point of view? 9
On page 24 Chomsky notes that'we must be careful not to overlook 3.
the fact that surface similarities mayhide underlying distinctions of a
fundamental nature, and that it may be necessaryto guide and draw out the
speaker's intuition in perhaps fairlysubtle ways before we can determine what is the actual character ofhis knawledge of his language orof anything
else. He notes that 'the latter(point) is as old as Plato's Meno'.
Clearly, the method employed by Socratesin The Meno for drawing
out one's actual knowledgeis basically fallaciousboth with respect to the
problem of geometry in The Menoand with respect to language. The slave boy
questioned by Socrates is clearlybeing taught. It is hard to believe that
the knowledge eventuallydisplayed by the slave boy was therefrom another
life, but rather it is easy to detectthe teaching method that Socrates em-
psychiatric ploys. For determing the actualknowledge of a patient in a
interview the Socratic method iscertainly worthwhile.This is so precisely
because the information beingsought is definitely present. Clearly, this
is not the case in questioning a youngboy about geometry. What actually
does happen is that certainimplications of already presentknowledge are
brought to the surface.
Chomsky would have us believethat this process would bevalid in
conversation that attempts to linguistics. One can imagine a Socratic
bring to the surface thesupposedly already presentknowledge that sentences It is necessary to are constructedwith noun phrases as component parts.
distinguish here between the actualknowledge and a description of that
knowledge. Clearly, sentences can be describedin terms of noun phrases.
However, it is difficult toconvince a speaker that he actuallyrewrites an
S as an NP VP. From a descriptive point of viewthese NP's and VP's may
be considered to be present. They are not, however, verylikely to be part 10
of the actual knowledgeof the speaker. Thus, the notion of thespeaker's intuition is a somewhat tenuous one. As was shown bySocrates' attempts, wishes it to the speaker's intuition isexactly that which the questioner
subtle' be.:thus, if we draw out thespeaker's intuition in a 'fairly enough way such as to elicitthe fact that there is an NP presentin all of the sentences that he utters,it is unclear exactly what wehave accomplished.
Certainly this cannot be thebasis for a linguistic theorythat does any more.than describe the data. This description may be nice tohave but it does not tell us anythingabout the actual process oflanguage.
At this point numerous transformationalists are liable toobject
that all a transformational grammaris supposed to do is describethe data.
If this is the case, one is proneto ask why one isconcerned with the
dpeaker's intuition at all, and inparticular haw is it possible tojustify
Chomsky's Beckman Lectures at a statementsuch as the following from
Berkeley (1967:111:9):
We cannot avoid beingstruck by the enormous disparitybetween
knowledge and experience - inthe case of language, betweenthe
generative grammar that expressesthe linguistic competence ofthe
native speaker, and the meagerand degenerate data on thebasis
of which he has constructedthis grammar for himself. Thus Here Chomsky equates'knowledge' with a generative grammar.
it would appear to be the casethat despite protestations tothe contrary,
fact a generative Chomsky believes thatpeople's knowledge of language is in
grammar. That is, a generative grammaris not intended to be a description
of his kpawledge but hisactual knowledge.Yet the information forthis
grammar can bederived from the somewhat tenuousframework of the Socratic
further lauded when Chomsky notes method! The results of this approach are (1967:111:5)
that if success is achieved in constructing empirically
adequate generative grammars and determining the universal prin-
ciples that govern their structure and organization, then this
will be an important contribution to human psychology.
One wonders if the method for determining universals is akin to somelarge scale drawing out of the speaker's intuition. Certainly it is doubtful if the process used by Socrates in The Meno was an important contribution to human psychology (although it certainly may have been to pedagogyand philosophy). One wonders, how a supposedly descriptive device such as a generative grammar formed in the above ways can contribute atall to des- cribing how a human does what he does when he speaks language. intended to be The notion of adescriptive grammar, if it were 4. However, it would appear only that, would not beparticularly bothersome. descriptive grammar,he that although Chomskycontinually refers to just a than this. Consider is just ascontinually implying that heis seeking more from his discussionof univer- for example, a statementof Chomsky's taken .
sals. (p. 30) severely Formal constraints...on asystem of concepts may linguist) of a descriptive limit the choice(by the child, or the
grammar, givenprimary linguistic data. Chomsky seems to beimplying The pareAtheticalremark is most revealing. selection of a grammar bythe that there is somecorrelation between the linguist is supposed tobe construc- child and by thelinguist. Yet, if the only describes the dataof ting a 'descriptivegrammar', that is, one that PK, This seems the language in some way,the child must bedoing this also. description hard to believe. Can the child besaid to be constructing a Somehow this notion seemsunlikely. In of the data thathe has gathered? other notion of fact, if this werethe case, there wouldbe no need for any Yet, Chomskyconsistently maintains a grammarbesides a descriptive one. intended to be 'apoint-by-poin that a transformationalgrammar is not by a speaker'(p. 139). model for the actualconstruction of a sentence which is But it is clearlythe case that achild is constructing a grammar construction of a sentencein exactly a procedure forthe point-by-point procedure that is thechild's grammar. his awn head. Certainly, it is this intends it. This is not a descriptivegrammar inthe sense that Chomsky is possible for Chomskyto make Thus, there is somequestion as to how it
the statement thathe does make on page30. led to a great many The confusion thatChomsky allows here has problems with respect to just what transformational grammars are and what
it is they can do. This confusion muddles not only his followers but
Chomsky himself. Consider for example, the paragraph following the quote
on page 30.
The existence of deep-seated formal universals, in the sense
suggested by such examples as these, implies that all languages.
are cut to the same pattern, but does notimply that there is any
point-by-point correspondence between particular languages. It
does not, for example, imply that there must be some reasonable
procedure for translating between languages.
Certainly there is a reasonable procedure for translating between
languages since people do it all the time. What Chomsky means here is that
the descriptive grammar that he is constructing may not provide that proce-
dure. In fact, after many attempts it has become clear that a transforma-
tional grammar will not provide that procedure. But certainly that pro-
cedure exists. It is necessarily implicit in'the point-by-point procedure
that the child.discovers for construction of a sentence. Thus, a procedure
for translating between languages should be available as soon as a grammar
of the point-by-point model that Chomsky seems so willing to disregard is
available. This model would necessarily be founded on what Chomsky chooses
to call substantive universals. That is, whatever is present in all lan-
guages, or underlying all.languages,should form the basis (and in fact does)
for translating between languages. However, the presence of formal uni-
versals in a linguistic description implies very little. It is a formal
universal of the room in which I am sitting and therefore all rooms that
if I paint everything in the room purple, all things in the roomwillbe purple.
It may be a formal universal of language that if one generates the basis ofa sentence and then permutesthe output with enough special transformations and ad hoc constraints that alanguage can be described by this process.
The question is whether thig is atall meaningful. op, of this kind, to harp on 5 Perhaps it is unfair, in a discussion
what is wrong with a theorywithout explaining what might beright. It
should be clear from what has been said sofar, that it is a theory ofsimu-
lative performance that would satisfy myaims for linguistics. It would
seem to me thatthis theory would have to besemantics-based. That is, it
would, as do humans, start withbeliefs (or meaning) inherent in adiscourse,
and then try to generate thed-iscourse. To do this, it wouldprobably be
necessary to passthrough sorli vetsion of syntax along the way.
This generative the 'Ad also employ a Universal andParticu-
lar grarmnar, but one that Ic ec. phase separate from the other. For
example, Chomsky states that the sentence'The harvest was clever to agree'
violates certain 'rules ofEnglish' (p. 76).We would require that a theory
of simulative performance findthis sentence in perfect accordwith the rules
of English but discordant withthe conceptual rules of theUniversal com-
ponent. 16
6. It is my purpose here to explicate what the point of alinguistic formalization might be, and to give some definitions of theelements needed in that formalization.
My underlying assumption here is that language isconceptually- based. By that I merely intend to emphasize the fact thatlanguage and throught are related, very much so in fact. Thus, it is possible to view the problem of formalization of language as the problemof the formalization of the underlying conceptual basis of language. Therefore, the rules that would be used for this formalization wouldexplicate the elements of the con- ceptual process and then relate the output of that process tolanguage.
Of course, the main problems in an attempt of thiskind are de-
termining what is meant by the conceptual processand how one can be able
to view that process.The answer is that we cannot know for certain that
our formalization isrepresentative of the method that humans employ, hut we can know ifthe output of the conceptual apparatus that humansemploy
is the same as the output of theformalization.
The following, then, are some basic definitions necessary tothe
formalization of the conceptual basis forlanguage. They are the defini-
tions that are used as the basis of acomputational parsing system for
natural language that extracts the language-free conceptualizations inherent in a
piece of discourse. (This 'conceptual parser' is currently operating at
the Stanford Artificial Intelligence Project.[see Schank and Tesler (1969)]) 17
Formal Postulates and Definitions
The Conceptual Basis f Language
1) The basic unit of language is the conceptualization.
a) A conceptualization is a language-free proposition expressing a relation
between a linguistic concept that is an actor or a conceptualization,
and an action and the ob'ect of that action.
b) A linguistic concept with respect to an actor or object, consists
of a bundle of descriptive attributes that define at what point the
notion of the concept of that actor or object is complete in a
given language. Linguistic concepts are language-free insofar as
they have conceptual referents. Each language may define a lin-
guistic concept at any point that it chooses and translatability
between linguistic concepts indifferent languages can be said to
be a function of the number of descriptive attributes needed to
be added in order to have equivalence.
c) An actor (denoted PP) is a category of linguistic concept that
performs certain actions. In a conceptualization an actor may have
a relation to another actor or to a descriptor instead of performing
an action.
d) An object is an actor that is not acting but rather is being acted
upon in a given situation. Thus an object is also denoted PP and
is capable of performing an action simultaneously.
e) An action is a linguistic concept that corresponds to any real
world action.
f) A relation is defined as a dependence between two linguistic concepts.
A dependence denotes that the dependent is adding information to 18
the governor. There are 8 types ofdependence. dependence is necessaryin 1) Two-way dependence ( 04. ) a two-way
order to.havea'conceptualization. This dependency denotes that that the governor-dependentrelationship is mutual and each other. the connected concepts areadding information to
This dependency isreserved for 2) Objective dependency ( ).
objects of actions. The object may be aPP or a conceptualization. of the governor 3) Attributive dependency(0. Here an attribute
is explicated. This dependency givesinfor- 4) Prepositional dependency(44).
mation as to therecipient of an action.
dependency (ft)..This dependency 5) Attributive prepositional those parts of explicates attributesabout a PP as opposed to
a PP. two conceptdaliza- 6) Causal dependency. The dependency between
tions when one isthe cause of the other. two conceptual- 7) Conditional dependency. The dependency between of the izations when thetruth of one isdependent on the truth
other. An entireconceptualization 8) Attributive two-waydependency.
servesattributively. relation Within aconceptual- The Syntax of languageconsists of rules of ization. the syntax of alanguage) is a) The syntax oflanguage (as opposed to but not an expressionof the so-callednatural order of ideas, relations between classes rather is an expressionof the possible all languages is the sameat the of concepts. Thus, the syntax of 19
level of conceptualization. The syntax consists of conceptual
categories and their possible relations.
b) The categories of a linguistic system are(parenthesized expression
corresponds to section numberabove):
1) (lc) PP
2) (1d) PP
3) (le) ACT
4) PA, a descriptor of PP's
5) AA:a descriptor of ACT's
6) LOC, the location of a conceptualization
7) T, the time of conceptualization
c) The possible relaticns between these categories are: PA 1) (lf1) PP a ACT; PP a PP; PP a PA; fACT1
2) (1f2) ACT .-- PP ACT
ACT a a PA PA AA 3) (1f3) PP ttfttt PA AA LOC T PA AAAA
4) (.1t14.) ACT PP + (indicates that this dependence can bemultiple)
5) (1f5) PP
PP
3) The Semantics of language is defined as the rules that permit arelation
to exist. Thus, although the universal syntax may permit a given de-
pendency, that dependency cannot exist unless the universal semantics
also permits it. The semantics consists of semantic files associated
with each possible conceptual governor in a ^onceptual relation for
specific concepts and for ApJAA.qia_sair.A.
a) A semantic category represents a node in a hierarchical treethat
divides the universe of physical objects. The tree branches every 20
of a semantic time a descriptiveattribute can modify a part
category without beingable to modify thecomplementary part
of that semantic category.
b) A semantic file consists ofthe list of descriptiveattributes
obtained by starting at a nodein the hierarchical treeand
tracing up. A semantic fileexisti for each possible dependency.
c) A specific concept is definedby the possible dependent concep-
tual categories and theirassociated semantic files. In each
of the associated semanticfiles specific attributes arefilled
with respect to eachdescriptive attribute for eachspecific
concept. language specific rules of 4) The RealizationRules of a language are the
the linguistic system. These rules map parts of aconceptualization
into an acceptablelinguistic construct. The Realization Rules serve
to define what anacceptable linguistic constructis, in the concept-to-
word phase; and serve torecognize a linguistic constructin terms of The Realization its conceptual realizatein the word-to-conceptPhase.
Rules are reversible andthus are the same inanalysis and synthesis.
However, in the actualuse,.the Realization Rules areonly partially they are used used for understanding(in order to disambiguate), while
in toto for generation. 7. The definitions presented here represent assertionsabout the character of language. They are also the basicsof a computationallinguistic system described in Schank (1969).
I share Chomsky's belief that 'the realproblem for tomorrow is that
of discovering an assumption regardinginnate structure that is sufficiently
rich' (1967:111:10). Many of the notions thatare offered to us by trans-
formational grammar regarding the universalgramnar that would be a part
of this rich innate schemaare valuable. However, it seems to me that this universal grammar is conceptually basedand thus must conform, toa large
extent, to the postulates that I have presentedhere. That is, we must begin to speak of describing what is goingon inside the child and not inside
the linguist. yr
References
Bach, Emmon. 1964. An Introduction To TransformationalGrammar. New York, Holt, Rinehart and Winston.
Bloch, Bernard. 1948. A Set of Postulates for PhonemicAnalysis. Lg. 24.3- )4-6.
Bloomfield, Leonard. 1926. A Set of Postulates for theScience of Language. Lg.2.153-64.
Chomsky, Noam. 1967. LinguisticContributions to the Study of Mind. Beckman Lectures, Berkeley.
Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, M.I.T. Press.
Fodor, J., and Garrett, M. 1966. Same Reflections on Competence andPerfor- mance. Psycholinguistics Papers, ed.J. Lyons, and R. Wales. Edinburgh, University Press.
Robinson, Jane. 1968. Dependency Structures andTransformational Rules. Specification and Utilization ofa Transformational Grammar 3, ed. by Warren J. Plath. YorktawnHeights, I.B.M.
Schank, Roger. 1969. A Conceptual DependencyRepresentation for a Computer- Oriented Semantics. Ph.D. Thesis,University of Texas, Austin. Stanford Artificial IntelligenceProject Memo 83, Stanford, Computer Science Department,Stanford University.
Schank, Roger,and Tesler,Lawerence. 1969. A Conceptual Parser for Natural Language presented to First InternationalJoint Conference on Artificial Intelligence,.Washington,D.C., May 7-9, 1969.