0% found this document useful (0 votes)
287 views22 pages

Three Factors in Language Design

Noam chomsky: language faculty as an "organ of the body" with other cognitive systems. He says biolinguisticperspective sees language faculty as organ of the body. Biolinguistics began in 1974 at a MIT meeting on the topic of "biolinguistics" he says many of the leading questions remain alive today.

Uploaded by

mychief
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
287 views22 pages

Three Factors in Language Design

Noam chomsky: language faculty as an "organ of the body" with other cognitive systems. He says biolinguisticperspective sees language faculty as organ of the body. Biolinguistics began in 1974 at a MIT meeting on the topic of "biolinguistics" he says many of the leading questions remain alive today.

Uploaded by

mychief
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 22

Three Factorsin Language Design

Noam Chomsky
The biolinguisticperspectiveregardsthe languagefacultyas an "organ of the body," along with other cognitive systems. Adopting it, we expect to find three factors that interactto determine(I-) languages attained: geneticendowment(the topic of UniversalGrammar), experience, and principlesthatare language-or even organism-independent. Researchhas naturallyfocused on I-languagesand UG, the problems of descriptiveand explanatoryadequacy.The Principles-and-Parameters approachopened the possibility for serious investigationof the third factor, and the attemptto account for propertiesof language in terms of generalconsiderationsof computational efficiency, eliminating some of the technology postulated as specific to language and providingmore principledexplanationof linguistic phenomena. Keywords: minimalism, principled explanation, Extended Standard Theory, Principles-and-Parameters, internal/external Merge, singlecycle derivation,phase

Thirty years ago, in 1974, an internationalmeeting took place at MIT, in cooperationwith the RoyaumontInstitutein Paris,on the topic of "biolinguistics," a term suggestedby the organizer, Massimo Piattelli-Palmarini, and the title of a recentbook surveyingthe field and proposingnew directionsby Lyle Jenkins (2002).1 This was only one of many such interactionsin those years, including interdisciplinary seminarsand international conferences. The biolinguisticperspectivebegan to take shape over 20 years before in discussions among a few graduatestudentswho were much influencedby developmentsin biology and mathematics in the early postwaryears, including work in ethology that was just coming to be known in the United States.One of themwas EricLenneberg,whose seminal 1967 studyBiological Foundations of Language remains a basic documentof the field. Many of the leading questions discussed at the 1974 conference, and in the years leading up to it, remainvery much alive today. One of these questions,repeatedlybroughtup in the conferenceas "one of the basic questions to be asked from the biological point of view," is the extent to which apparentprinciples of language,includingsome thathad only recentlycome to light, are uniqueto this cognitive system or whether similar "formal arrangements"are found in other cognitive domains in humans or

This article is expandedfrom a talk presentedat the annualmeeting of the Linguistic Society of America,9 January 2004. Thanksto Cedric Boeckx, Samuel David Epstein, Robert Freidin,Lyle Jenkins,HowardLasnik, and Luigi Rizzi, among others, for comments on an earlierdraft. ' The conference, titled "A Debate on Bio-Linguistics," was held at Endicott House, Dedham, Massachusetts, 20-21 May 1974, and organizedby the CentreRoyaumontpour une science de l'homme, Paris.
Linguistic Inquiry,Volume 36, Number 1, Winter 2005 1-22 ?) 2005 by the MassachusettsInstituteof Technology

NOAM

CHOMSKY

other organisms.An even more basic questionfrom the biological point of view is how much of language can be given a principled explanation,whether or not homologous elements can be found in other domains or organisms. The effort to sharpenthese questions and to investigate them for language has come to be called the "Minimalist Program" in recent years, but the of theoreticalpersuasion,in linguisquestionsarise for any biological system and are independent the tics and elsewhere. Answers to these questions are fundamentalnot only to understanding natureand functioningof organismsand their subsystems,but also to investigatingtheir growth and evolution. For any biological system, languageincluded,the only generalquestionthat arises about the programis whetherit can be productivelypursuedor is premature. In these remarks,I will try to identify what seem to me some of the significant themes in the past half-century of inquiryinto problemsof biolinguisticsand to considertheircurrentstatus. Several preliminaryqualificationsshould be obvious. One is that the picture is personal;others would no doubt make differentchoices. A second is that things often seem clearerin retrospect in this account,but not I think too much. A third than at the time, so there is some anachronism is that I cannot even begin to mention the contributionsof a great many people to the collective as the relatedfields have expandedenormouslyin the years since the 1974 enterprise,particularly conference. The biolinguistic perspective views a person's language as a state of some component of the mind, understanding"mind" in the sense of eighteenth-century scientists who recognized that after Newton's demolitionof the only coherentconcept of body, we can only regardaspects of the world "termedmental" as the result of "such an organicalstructure as that of the brain" (JosephPriestley).Among the vast arrayof phenomenathatone might loosely considerlanguagerelated,the biolinguisticapproachfocuses attentionon a componentof humanbiology thatenters into the use and acquisitionof language,however one interprets the term "language." Call it the "faculty of language," adaptinga traditionalterm to a new usage. This component is more or less on a par with the systems of mammalianvision, insect navigation, and others. In many of these cases, the best availableexplanatory theoriesattribute to the organismcomputational systems and whatis called "rule-following" in informalusage-for example, when a recenttext on vision presents the so-called rigidity principle as it was formulated50 years ago: "if possible, and other rules permit,interpretimage motions as projectionsof rigid motions in three dimensions" (Hoffman 1998:169).In this case, laterworkprovidedsubstantial insightinto the mentalcomputations that seem to be involved when the visual system follows these rules, but even for very simple organisms, that is typically no slight task, and relating mental computationsto analysis at the cellular level is commonly a distantgoal. Adopting this conception, a language is a state of the faculty of language, an I-language,in technical usage. The decision to study language as part of the world in this sense was regardedas highly controversialat the time, and still is. A more careful look will show, I think, that the arguments advancedagainstthe legitimacy of the approachhave little force (a weak thesis) and thatits basic assumptionsare tacitly adoptedeven by those who strenuouslyreject them, and indeed must be, even for coherence (a much stronger thesis). I will not enter into this interesting chapter of

THREE

FACTORS

IN LANGUAGE

DESIGN

contemporaryintellectualhistory here, but will simply assume that crucial aspects of language can be studied as part of the naturalworld, adoptingthe biolinguistic approachthat took shape half a century ago and that has been intensively pursuedsince, along differentpaths. The languagefacultyis one componentof whatthe cofounderof modem evolutionarytheory, Alfred Russel Wallace, called "man's intellectualand moral nature":the humancapacities for and recordcreativeimagination,languageand symbolism generally,mathematics,interpretation ing of naturalphenomena,intricatesocial practices, and the like, a complex of capacities that seem to have crystallized fairly recently, perhapsa little over 50,000 years ago, among a small breedinggroupof which we are all descendants-a complex thatsets humansapartrathersharply fromotheranimals,includingotherhominids,judgingby tracesthey have left in the archaeological record.The natureof the "humancapacity," as some researchers now call it, remainsa considerable mystery.It was one element of a famous disagreement between the two foundersof the theory of evolution, with Wallace holding, contraryto Darwin, that evolution of these faculties cannot be accounted for in terms of variation and naturalselection alone, but requires "some other influence, law, or agency," some principle of naturealongside gravitation,cohesion, and other forces withoutwhich the materialuniversecould not exist. Althoughthe issues are frameddifferently today within the core biological sciences, they have not disappeared(see Wallace 1889: chap. 15, Marshack1985). It is commonly assumed that whatever the human intellectual capacity is, the faculty of language is essential to it. Many scientists agree with paleoanthropologist Ian Tattersall,who writes that he is "almost sure that it was the invention of language" that was the "sudden and emergent" event that was the "releasing stimulus" for the appearance of the humancapacity in the evolutionaryrecord-the "greatleap forward"as JaredDiamondcalled it, the resultof some genetic event that rewired the brain, allowing for the origin of modem language with the rich syntax that provides a multitude of modes of expression of thought, a prerequisitefor social development and the sharpchanges of behavior that are revealed in the archaeologicalrecord, also generally assumedto be the triggerfor the rapidtrek from Africa, where otherwisemodem humanshad apparently been presentfor hundredsof thousandsof years (Tattersall1998:24-25; see also Wells 2002). Tattersall takes language to be "virtually synonymous with symbolic thought." Elaborating,one of the initiatorsof the Royaumont-MITsymposia, Francois Jacob, observedthat "the role of languageas a communicationsystem between individualswould have come about only secondarily,as many linguists believe" (1982:59), perhapsreferringto discussions at the symposia, where the issue repeatedlyarose, among biologists as well. In the 1974 conference, his fellow Nobel laureateSalvadorLuriawas the most forceful advocateof the view that communicativeneeds would not have provided "any great selective pressureto produce a system such as language," with its crucial relation to "development of abstractor productive thinking" (Luria 1974:195). "The quality of language that makes it unique does not seem to be so much its role in communicatingdirectives for action" or other common features of animal communication,Jacob continued,but rather "its role in symbolizing, in evoking cognitive images," in "molding" our notion of reality and yielding our capacity for thought and planning, of allowing "infinitecombinationsof symbols" andtherefore"mental throughits uniqueproperty

NOAM

CHOMSKY

creationof possible worlds," ideas thattraceback to the seventeenth-century cognitive revolution that answersto questionsaboutevolu(1982:59). Jacob also stressedthe common understanding tion "in most instances ... can hardly be more than more or less reasonableguesses" (1982: 31). We can add anotherinsight of seventeenth-and eighteenth-century philosophy: that even the most elementaryconcepts of human language do not relate to mind-independent objects by means of some reference-likerelationbetween symbols and identifiablephysical featuresof the external world, as seems to be universal in animal communicationsystems. Rather, they are creations of the "cognoscitive powers" that provide us with rich means to refer to the outside world from certainperspectives,but are individuatedby mentaloperationsthatcannotbe reduced to a "peculiarnaturebelonging" to the thingwe aretalkingabout,as Humesummarized a century of inquiry.Those are critical observationsabout the elementarysemantics of naturallanguage, suggesting that its most primitive elements are related to the mind-independent world much as the internalelementsof phonologyare,not by a reference-likerelationbutas partof a considerably more intricatespecies of conceptionand action. It is for reasons such as these, thoughnot clearly graspedat the time, thatthe early work in the 1950s adopteda kind of "use theoryof meaning," pretty much in the sense of John Austin and the later Wittgenstein:language was conceived as an instrumentput to use for various humanpurposes,generatingexpressions including arrangements of the fundamental elements of the language, with no grammatical-ungrammatical divide, each basically a complex of instructionsfor use (see Chomsky 1955, hereafterLSLT).2 If this much is generally on the right track,then at least two basic problemsarise when we considerthe origins of the faculty of languageand its role in the suddenemergenceof the human intellectual capacity: first, the core semantics of minimal meaning-bearingelements, including the simplest of them; and second, the principles that allow infinite combinationsof symbols, hierarchically organized,which providethe meansfor use of languagein its manyaspects.Accordingly, the core theory of language-Universal Grammar (UG)-must provide, first, a structured inventoryof possible lexical items that are relatedto or perhapsidentical with the concepts that are the elements of the "cognoscitive powers," sometimes now regardedas a "language of thought" along lines developed by Jerry Fodor (1975); and second, means to construct from these lexical items the infinite varietyof internalstructures thatenter into thought,interpretation, planning, and other human mental acts, and that are sometimes put to use in action, including the externalizationthat is a secondary process if the speculationsjust reviewed turn out to be correct. On the first problem, the apparentlyhuman-specificconceptual-lexicalapparatus, there is importantwork on relationalnotions linked to syntactic structuresand on the partiallymindinternalobjects that appearto play a critical role (events, propositions,etc.).3 But there is little beyond descriptiveremarkson the core referentialapparatus that is used to talk aboutthe world. The second problemhas been centralto linguistic researchfor half a century,with a long history before in differentterms.
2 For later discussion, see among others Chomsky 1966, 2001b, McGilvray 1999, Antony and Hornstein2003. 3For insightful review and original analysis, see Borer 2004a,b.

THREE

FACTORS

IN LANGUAGE

DESIGN

The biolinguistic approachadopted from the outset the point of view that C. R. Gallistel (1997) calls "the norm these days in neuroscience" (p. 86), the "modularview of learning": the conclusion that in all animals, learning is based on specialized mechanisms, "instincts to learn" (p. 82) in specific ways. We can think of these mechanismsas "organs within the brain" (p. 86), achieving states in which they performspecific kinds of computation.Apartfrom "extremelyhostile environments"(p. 88), they change states underthe triggeringand shapingeffect of externalfactors, more or less reflexively, and in accordancewith internaldesign. That is the "process of learning" (Gallistel 1997, 1999), though "growth" might be a more appropriate term, avoiding misleading connotationsof the term "learning." The modularview of learning of course does not entail that the componentelements of the module are unique to it: at some level, everyone assumes that they are not-the cellular level, for example-and the question of the level of organizationat which uniquepropertiesemerge remainsa basic one from a biological point of view, as it was at the 1974 conference. into evolutionaryand Gallistel's observationsrecallthe conceptof "canalization"introduced over 60 yearsago, referring to processes "adjustedso developmentalbiology by C. H. Waddington as to bring about one definite end result regardlessof minor variationsin conditions duringthe courseof the reaction,"thusensuring"the production of the normal,thatis optimaltype in the face of the unavoidable hazardsof existence" (Waddington1942). Thatseems to be a fairdescriptionof the growth of language in the individual.A core problemof the study of the faculty of language is to discover the mechanismsthat limit outcomes to "optimal types." It has been recognized since the origins of modernbiology that such constraintsenter not only into the growth of organismsbut also into their evolution, with roots in the earliertradition that StuartKauffmancalls "rationalmorphology" (1993:3-5).4 In a classic contemporary paper, John Maynard Smith and associates trace the post-Darwinianreformulationback to Thomas Huxley, who was struckby the fact thatthereappearto be "predetermined lines of modification" that lead naturalselection to "producevarietiesof a limited numberand kind" for every species (MaynardSmith et al. 1985:266). They review a varietyof such constraintsin the organicworld and describehow "limitationson phenotypicvariability"are "caused by the structure, character, or composition, dynamics of the developmentalsystem," pointing out also that such "developmentalconstraints. .. undoubtedlyplay a significantrole in evolution" thoughthereis yet "little agreementon theirimportanceas comparedwith selection, drift,and othersuch factorsin shaping evolutionaryhistory" (p. 265). At about the same time, Jacob wrote that "the rules controlling embryonic development," almost entirely unknown, interactwith other constraintsimposed by general body plan, mechanicalpropertiesof building materials,and other factors in "restricting possible changes of structuresand functions" in evolutionarydevelopment(1982:21), providing "architectural constraints"that "limit adaptivescope and channelevolutionarypatterns"(Erwin 2003:1683). The best-knownof the figures who devoted much of their work to these topics are
4 For comment in a linguistic context, see Boeckx and Hornstein2003. For more general discussion, see Jenkins

2000.
5

For review of some of these topics, see Stewart 1998.

NOAM

CHOMSKY

D'Arcy Thompson and Alan Turing, who took a very strong view on the central role of such factors in biology. In recent years, such considerationshave been adducedfor a wide range of problemsof developmentand evolution, from cell division in bacteriato optimizationof structure and functionof corticalnetworks,even to proposalsthatorganismshave "the best of all possible Chemiak(1995:522).6The probneuroscientistChristopher brains," as arguedby computational lems are at the borderof inquiry,but their significance is not controversial. Assuming thatthe faculty of languagehas the generalpropertiesof otherbiological systems, we should, therefore,be seeking three factors that enter into the growth of language in the individual: l. Genetic endowment,apparentlynearly uniformfor the species, which interpretspart of the environment as linguistic experience, a nontrivial task that the infant carries out reflexively, and which determinesthe generalcourse of the developmentof the language faculty. Among the genetic elements, some may impose computationallimitationsthat disappearin a regularway throughgenetically timed maturation.KennethWexler and his associates have provided compelling evidence of their existence in the growth of language, thus providing empiricalevidence for what Wexler (to appear)calls "Lenneberg's dream." 2. Experience,which leads to variation,within a fairly narrowrange, as in the case of other subsystems of the humancapacity and the organismgenerally. 3. Principlesnot specific to the faculty of language. The third factor falls into several subtypes:(a) principlesof data analysis that might be used in languageacquisitionandotherdomains;(b) principlesof structural architecture anddevelopmental constraintsthat enter into canalization,organic form, and action over a wide range, including principlesof efficient computation,which would be expected to be of particular significance for computationalsystems such as language. It is the second of these subcategoriesthat should be of particularsignificance in determiningthe natureof attainablelanguages. Those exploring these questions 50 years ago assumedthat the primitivestep of analysis of linguistic experience would be feature-basedphonetic analysis, along lines describedby Roman Jakobson and his associates (see Jakobson,Fant, and Halle 1953). We also tried to show that basic prosodicproperties reflect syntacticstructure thatis determined by otherprinciples,including crucially a principleof cyclic computationthat was extended much more generallyin lateryears (see Chomsky,Halle, and Lukoff 1956). The primitiveprinciplesmust also providewhat George Miller called "chunking," identificationof phonologicalwords in the stringof phoneticunits. In LSLT(p. 165), I adoptedZellig Harris's(1955) proposal,in a differentframework, for identifying morphemesin terms of transitionalprobabilities,though morphemesdo not have the required beads-on-a-string property.The basic problem,as noted in LSLT,is to show that such statistical

6 See also Laughlinand Sejnowski 2003, Chemiak et al. 2004, and Physics News Update 2001 reportingHoward, Rutenberg,and de Vet 2001.

THREE

FACTORS

IN LANGUAGE

DESIGN

methods of chunkingcan work with a realistic corpus. That hope turnsout to be illusory, as has recently been shown by Thomas Gambell and CharlesYang (2003), who go on to point out that the methods do, however, give reasonableresults if applied to materialthat is preanalyzedin terms of the apparentlylanguage-specificprinciplethat each word has a single primarystress. If so, then the early steps of compiling linguistic experience might be accounted for in terms of general principles of data analysis applied to representations preanalyzedin terms of principles specific to the languagefaculty, the kind of interactionone should expect amongthe threefactors. In LSLT,it was assumedthatthe next step would be assignmentof chunkeditems to syntactic categories,again by generalprinciplesof data analysis. A proposalwith an information-theoretic flavor was tried by hand calculations in that precomputerage, with suggestive results, but the matterhas never been pursued,to my knowledge. Surely what are called "semantic properties" are also involved, but these involve nontrivialproblemsat the most elementarylevel, as mentioned earlier.The assumptionof LSLTwas that higher levels of linguistic description,including morphemes, are determinedby a general format for rule systems provided by UG, with selection among them in terms of a computational procedurethat seeks the optimalinstantiation,a notion defined in terms of UG principles of significant generalization.Specific proposals were made then and in the years that followed. In principle,they provided a possible answer to what came to be called the "logical problemof languageacquisition,"butthey involved astronomical calculation and thereforedid not seriously addressthe issues. The main concerns in those years were quite different,as they still are. It may be hard to believe today, but it was commonly assumed50 years ago that the basic technology of linguistic descriptionwas availableand thatlanguagevariationwas so free that nothingof much generality was likely to be discovered. As soon as efforts were made to provide fairly explicit accounts of the propertiesof languages, however, it became obvious how little was known, in any domain. Every specific proposalyielded a treasuretrove of counterevidence, requiringcomplex and varied rule-systems even to achieve a very limited approximationto descriptive adequacy. That was highly stimulatingfor inquiry into language, but it also left a serious quandary,since the most elementary considerationsled to the conclusion that UG must impose narrow constraintson possible outcomes-sometimes called "poverty of stimulus" problemsin the study of language, thoughthe termis misleadingbecausethis is just a special case of basic issues thatariseuniversally for organic growth. A numberof paths were pursuedto try to resolve the tension. The most successful turned out to be efforts to formulategeneral principles, attributedto UG-that is, the genetic endowment-leaving a somewhat reduced residue of phenomena that would result, somehow, from experience. Early proposalswere the A-over-A Principle,conditions on wh-extractionfrom whphrases (relatives and interrogatives),simplificationof T-markersto base recursion (following observationsby Charles Fillmore) and cyclicity (an intricatematter, as shown in an important paperof RobertFreidin's(1978) and insightfullyreviewed in a currentpaperof HowardLasnik's (to appear) which shows that many central questions remain unanswered),later John Robert Ross's (1967) classic study of taxonomy of islands that still remains a rich store of ideas and observationsto explore, then attemptsto reduceislands to such propertiesas locality and structure

NOAM

CHOMSKY

preservation,and so on. These approacheshad some success, but the basic tensions remained unresolved at the time of the 1974 conference. Within a few years, the landscape had changed considerably.In part this was because of greatprogressin areasthat had hithertobeen explored only in limited ways, includingtruth-and In partit was the result of a vast arrayof new model-theoreticsemanticsand prosodic structures. materialsfrom studies of much greaterdepththan previously undertaken, and into a much wider variety of languages, much of it traceableto RichardKayne's work and his lectures in Europe, which inspiredfar-reaching inquiryinto Romanceand Germaniclanguages,laterotherlanguages, also leading to many fruitfulideas about the principlesof UG. About 25 years ago, much of this work crystallizedin a radically differentapproachto UG, the Principles-and-Parameters (P&P) framework, which for the firsttime offeredthe hope of overcomingthe tensionbetweendescriptive and explanatoryadequacy.This approachsought to eliminatethe formatframeworkentirely,and with it, the traditional conceptionof rules and constructionsthathad been prettymuch takenover into generativegrammar.That much is familiar, as is the fact that the new P&P frameworkled to an explosion of inquiry into languages of the most varied typology, yielding new problems previously not envisioned, sometimes answers, and the reinvigoration of neighboringdisciplines concernedwith acquisitionandprocessing,theirguidingquestionsreframedin termsof parameter setting within a fixed system of principlesof UG with at least visible contours.Alternativepaths, variouslyinterrelated, were leading in much the same direction,includingMichael Brody's highly illuminatingwork (1995, 2003). No one familiar with the field has any illusion today that the horizons of inquiryare even visible, let alone at hand, in any domain. Abandonmentof the format frameworkalso had a significant impact on the biolinguistic program. If, as hadbeen assumed,acquisitionis a matterof selectionamongoptionsmadeavailable by the format provided by UG, then the format must be rich and highly articulated,allowing relatively few options; otherwise, explanatoryadequacyis out of reach. The best theory of language must be a very unsatisfactoryone from other points of view, with a complex arrayof conditions specific to human language, restrictingpossible instantiations.The only plausible theorieshad to impose intricateconstraintson the permissiblerelationsbetween soundand meaning, all apparently specific to the faculty of language.The fundamental biological issue of principled explanationcould barely be contemplated,and correspondingly,the prospects for serious inquiryinto evolutionof languagewere dim;evidently,the morevariedandintricatethe conditions specific to language, the less hope there is for a reasonableaccount of the evolutionaryorigins of UG. These are among the questionsthat were raised at the 1974 symposiumand others of the period, but they were left as apparentlyirresolubleproblems. The P&P frameworkoffered prospects for resolution of these tensions as well. Insofar as this frameworkproves valid, acquisitionis a matterof parameter settingand is thereforedivorced entirely from the remaining format for grammar:the principles of UG. There is no longer a conceptualbarrierto the hope that the UG might be reduced to a much simpler form, and that the basic propertiesof the computational systems of languagemight have a principledexplanation insteadof being stipulatedin termsof a highly restrictivelanguage-specificformatfor grammars. Within a P&P framework,what had previously been the worst theory-anything goes-might

THREE

FACTORS

IN LANGUAGE

DESIGN

be within the realm of conceivability at least, since it is no longer refuted instantlyby the fact that it rendersacquisitionimpossible. Returningto the three factorsof languagedesign, adoption to shiftingthe burdenof explanation of a P&P frameworkovercomesa difficult conceptualbarrier from the first factor, the genetic endowment,to the thirdfactor,language-independent principles and computational of data processing, structural architecture, efficiency, therebyprovidingsome answers to the fundamentalquestions of biology of language, its nature and use, and perhaps even its evolution. As noted by Luigi Rizzi (pers. comm.), whose comments I paraphrase here, abandonment of the construction-based view of earlier(and traditional)work had furtherconsequencesfor the biolinguistic program.In the earlier framework,not only rules but also UG principleswere expressed in terms of grammaticalconstructions(islands, specified-subjectand other constraints on operations,Emonds's structure-preserving hypothesis, filters, etc.), all inherentlyspecific to in other biological systems. Within the P&P framelanguage, without even remote counterparts work,the basic computational ingredientsareconsiderablymoreabstract (locality,minimalsearch, basic recursion,etc.), and it becomes quite reasonableto seek principledexplanationin terms that may apply well beyond language, as well as relatedpropertiesin other systems. As noted, the third factor subsumes two kinds of principles: (a) data processing, and (b) constraints.Consider the first category. Within the architectural/computational-developmental general P&P framework,language acquisition is interpretedin terms of parametersetting; an acquisition theory would seek to set forth the mechanisms used to fix parametervalues. That of what the parametersare, and how they are organized,perhaps requires some understanding into a hierarchicstructurewith each choice of value setting limits on subsequentchoices. The most far-reachingapproachof this sort that I know of is Mark Baker's (2001). A somewhat differentapproach,also within the P&P framework,is suggestedby CharlesYang (2002). Developing suggestions of Thomas Roeper's, Yang proposes that UG provides the neonate with the full arrayof possible languages, with all parameters valued, and that incoming experience shifts the probabilitydistributionover languagesin accordwith a learningfunction that could be quite general.At every stage, all languagesarein principleaccessible,butonly for a few areprobabilities high enough so that they can actually be used. He provides interesting empirical evidence in supportof this approach,which carriesvery early proposalsaboutgeneraldata-processing principles to a new plane of sophistication. Considernext language-independent principlesof structural architecture and computational efficiency, the subcategoryof the third factor in the design of any biological system that might be expected to be particularly informativeaboutits nature.With the conceptualbarriersimposed by the format frameworkovercome, we need no longer assume that the means of generating structured expressions are highly articulated and specific to language.We can seriously entertain the possibility that they might be reducible to language-independent principles, whether or not there are homologous elements in otherdomains and organisms.We can, in short,try to sharpen the question of what constitutesa principledexplanationfor propertiesof language, and turnto one of the most fundamentalquestions of the biology of language:to what extent does language approximatean optimal solution to conditions that it must satisfy to be usable at all, given

10

NOAM

CHOMSKY

extralinguistic structural architecture? These conditionstakeus backto the traditional characterizaa links sound tion of language, since Aristotle at least, as system that and meaning.In our terms, the expressions generatedby a languagemust satisfy two interfaceconditions:those imposed by the sensorimotorsystem SM and by the conceptual-intentional system C-I that enters into the human intellectualcapacity and the variety of speech acts. Let us put aside the possibility that the linkage to sound is a secondary one, for reasons of language.We can mentionedand in the light of recent discoveries on modality-independence regard an explanation of propertiesof language as principled insofar as it can be reduced to propertiesof the interface systems and general considerationsof computationalefficiency and the like. Needless to say, these "external" conditions are only partiallyunderstood:we have to learn about the conditions that set the problemin the course of trying to solve it. The research task is interactive:to clarify the natureof the interfaces and optimal computationalprinciples throughinvestigationof how language partiallysatisfies the conditions they impose, not an unfamiliarfeatureof rationalinquiry.Independently,the interfacesystems can be studied on their own, including comparativestudy that has been productivelyunderway.And the same is true of principles of efficient computation,applied to language in recent work by many investigators with importantresults (see, e.g., Collins 1997, Epstein 1999, Epstein et al. 1998, Framptonand Gutmann1999). In a variety of ways, then, it is possible to both clarify and addresssome of the basic problems of the biology of language. It is perhapsworth recalling that the intuitionthat language is "well designed" relative to interfaceconditionshas been a valuableheuristicfor many years. The searchfor some reasonable notion of "simplicity of grammars"goes back to the earliest explorationof generativegrammar in the late 1940s. It has also provenvery useful over the yearsto ask whetherapparent redundancy of principlesis real, or indicates some errorof analysis. A well-known example is passive forms of exceptional Case-markingconstructions,which seemed at one time to be generatedby both raising and passive transformations. Dissatisfactionwith such overlappingconditions led to the realizationthat the transformations did not exist: rather,just a general rule of movement, which we can now see to be a virtualconceptualnecessity. More recent work suggests that such apparently language-specificconditionsas Jean-RogerVergnaud' s Case Filterand its descendantsmay be reducibleto virtualconceptualnecessity as well, in this case, interpretability of constructions at the interface,topics reviewedandelaborated recentlyby FreidinandVergnaud(2001). Throughout, the guiding intuitionthatredundancyin computational is a hint of errorhas proven structure to be productiveand often verified, suggesting that somethingdeeperis involved, a topic opened to more direct investigationas conceptualbarrierswere removed within the P&P framework. To proceed further,for concreteness, consider the Extended StandardTheory (EST), now often called the "Y-model," which was modifiedandimprovedextensively as it was reconstructed within the P&P framework-though I should perhapsstress again that the minimalistquestions to which I am now turningarise within any theoreticalorientation.In the early 1990s, Howard Lasnik and I publisheda sketch of our own best understanding of UG in these terms (Chomsky and Lasnik 1993). Taking that as a point of departure, let us ask to what extent its assumptions

THREE

FACTORS

IN LANGUAGE

DESIGN

11

can be revised or eliminated in favor of principledexplanationin terms of interfaceconditions and general principles. The EST/Y-modelpostulatesthreeinternallinguisticlevels in additionto the interfacelevels: and LF. A linguistic level is postulatedto host specific operations.Thus, D-Structure, S-Structure, D-Structureis the locus of lexical insertionand projectionto X-bar constructions;S-Structure is the point of transferof the computationto the sound interface,and the transitionfrom overt to covert operations;LF is the outputof overt and covert operations,and the input to the mapping to the meaninginterface.Three internallevels requirefive operations,each assumedto be cyclic: the operationsformingD-Structures by the cyclic operationsof X-bartheoryfrom selected items from the lexicon; the overt syntacticcycle from D- to S-Structure, the phonological/morphological cycle mappingS-Structure to the sound interface,the covert syntacticcycle mappingS-Structure to LF, and formal semantic operationsmapping LF compositionally to the meaning interface. The similarityof the overt and covert cycles was recognized to be problematicas soon as the frameworkwas developed 35 years ago, but the problemis in fact much broader:there are five cycles operatingin rathersimilar fashion. One crucial question, then, is whetherinternallevels, not forced by interface conditions, can be eliminated, and the five cycles reduced to one. If possible, that would be a substantialstep forward,with many consequences. One naturalpropertyof efficient computation,with a claim to extralinguisticgenerality,is that operationsforming complex expressions should consist of no more than a rearrangement of the objects to which they apply, not modifying them internallyby deletion or insertionof new elements. If tenable, that sharply reduces computationalload: what has once been constructed can be "forgotten" in later computations,in that it will no longer be changed.That is one of the basic intuitionsbehindthe notion of cyclic computation.The EST/Y-modeland otherapproaches violate this conditionextensively, resortingto barlevels, traces,indices, and otherdevices, which both modify given objects and add new elements. A second question,then, is whetherall of this technology is eliminable, and the empiricalfacts susceptibleto principledexplanationin accord with the "no-tampering"condition of efficient computation. Other questions arise about the variety of operations-phrase structure,transformations, reconstruction, and so on; about the crucial reliance on such notions as governmentthat seem to have no principledmotivation;and aboutmany principlesthat are hardeven to formulateexcept in terms specific to the language faculty. The general question is, How far can we progress in showing that all such language-specifictechnology is reducible to principledexplanation,thus isolatingthe core properties thatareessentialto the languagefaculty,a basic problemof biolinguistics? An elementaryfact about the languagefaculty is that it is a system of discrete infinity. Any such system is based on a primitiveoperationthat takes n objects alreadyconstructed,and constructsfrom them a new object:in the simplestcase, the set of these n objects. Call thatoperation Merge. Either Merge or some equivalent is a minimal requirement.With Merge available, we instantlyhave an unboundedsystem of hierarchically structured expressions.The simplestaccount of the "Great Leap Forward" in the evolution of humanswould be that the brain was rewired,

12

NOAM

CHOMSKY

perhapsby some slight mutation,to provide the operationMerge, at once laying a core part of the basis for what is found at that dramatic"moment" of humanevolution, at least in principle; to connect the dots is no trivial problem.There are speculationsabout the evolution of language that postulatea far more complex process: first some mutationthat permitstwo-unit expressions (yielding selectional advantagein overcoming memory restrictionson lexical explosion), then mutationspermittinglarger expressions, and finally the Great Leap that yields Merge. Perhaps the earlier steps really took place, but a more parsimoniousspeculationis that they did not, and that the Great Leap was effectively instantaneous,in a single individual,who was instantlyendowed with intellectual capacities far superiorto those of others, transmittedto offspring and coming to predominate, perhapslinkedas a secondaryprocessto the SM systemfor externalization and interaction,includingcommunicationas a special case. At best a reasonableguess, as are all speculationsaboutsuch matters,but aboutthe simplest one imaginable,and not inconsistentwith anythingknown or plausibly surmised.In fact, it is hardto see what accountof humanevolution would not assume at least this much, in one or anotherform.7 Similarquestionsarise about growthof languagein the individual.It is commonly assumed thatthereis a two-wordstage, a three-wordstage, and so on, with an ultimateGreatLeapForward to unboundedgeneration.Thatis observedin performance, but it is also observedthatat the early stage the child understandsmuch more complex expressions, and that randommodification of longer ones-even such simple changes as placementof functionwords in a mannerinconsistent with UG or the adult language-leads to confusion and misinterpretation. It could be that unboundedMerge, and whateverelse is involved in UG, is presentat once, but only manifestedin limited ways for extraneousreasons (memory and attentionlimitations and the like)-matters discussed at the 1974 symposium and now possible to formulatemore explicitly, for example, in the frameworksdeveloped by Wexler and Yang, alreadymentioned.8 Suppose,then, thatwe adoptthe simplestassumption: the GreatLeap Forwardyields Merge. The fundamentalquestion of biology of language mentionedearlierthen becomes, What else is specific to the faculty of language? Unless some stipulationis added, there are two subcases of the operationMerge. Given A, we can merge B to it from outside A or from within A; these are external and internalMerge, the latterthe operationcalled "Move," which thereforealso "comes free," yielding the familiar displacementpropertyof language. That propertyhad long been regarded,by me in particular, as an "imperfection" of language that has to be somehow explained, but in fact it is a virtual conceptualnecessity;9some version of transformational grammar seems to be the null hypothesis,

7For discussion in a broaderframework,see Hauser,Chomsky, and Fitch 2002; and for intriguingextensions, see Piattelli-Palmarini and Uriagereka,to appear. 8Rizzi (to appear)argues that performancerestrictionsinteractwith UG principles:inconsistency with the target language in early productionis grammaticallybased, with parametersfixed and principles observed, but performancedriven until the productionsystem matures.The logic is similar to that of Gambell and Yang (2003) and Wexler (to appear). 9 Options that "come free" may, of course, not be used, for example, in invented symbolic systems that keep to externalMerge-possibly becausethey do not have to make use of the edge properties thataretypicallydiscourse-related.

THREE

FACTORS

IN LANGUAGE

DESIGN

13

and any other mechanisms,beyond internalMerge, carry a burdenof proof. Assuming the notamperingcondition that minimizes computationalload, both kinds of Merge to A will leave A intact. That entails merging to the edge, the "extension condition," which can be understoodin different ways, including the "tucking-in" theory of Norvin Richards(2001), which is natural to accommowithin the probe-goalframeworkof recentwork, and which can also be interpreted date head adjunction. The no-tampering conditionalso entailsthe so-called copy theoryof movement,which leaves unmodifiedthe objects to which it applies, forming an extendedobject. As movementoperations began to emerge with improvementsof the more intricateLSLTframework,it was assumedthat of the system. they must compoundmovement and deletion, because of the general architecture In the EST/Y-model architecture,that was unnecessary,but since there did appearto be clear effects of the position left by Move, it was assumedthatthe operationleaves a trace,a new object coindexed with the moved element; thus, the objects subject to the operationare modified, and new entities are introduced,violating elementaryconditions of optimal computation.That was an error-mine in this case. Neither indexing nor the notion of traces is requiredif we adoptthe the copy theory.An important additionalgain is thatrules of reconstruction simplest assumptions: can be eliminated,and the phenomenacan be accountedfor more effectively, as has been shown in a great deal of recent work. Move leaves a copy, perhaps several, all transmittedto the phonological component. If language is optimized for communicativeefficiency, we would expect all to be spelled out: that would overcome many of the filler-gap problems faced by processing models. If language is optimized for satisfactionof interfaceconditions, with minimal computation,then only one will be spelled out, sharplyreducing phonological computation.The latter is the case, one of many reasons to suspect that speculations of the kind mentioned earlier about language design and evolution are on the right track. Which one is spelled out? There are several ideas about that. The most convincing seems to me Jon Nissenbaum's (2000), in phase-theoreticterms to which I will return.At the phase level, two basic operationsapply:transferto the interface,and Merge, either externalor internal.If internalMerge precedes transfer,movement is overt; otherwise, it is covert. If movement is covert, transferhas already spelled out the lower copy; if overt, the choice is delayed to the next phase. Zeljko Boskovic (2001) has pointed out that there should be an exception to such minimal computationif the phonetic interfacerequirespartialspell-out of a copy, and he has producedinterestingexamples. To take a different one discovered by Idan Landau(to appear),VP-frontingin Hebrew sometimes leaves the V behind, namely, when it has to satisfy what Lasnik called the "StrandedAffix Filter," a phonetic condition on affixes.10If these are the only exceptions, then syntacticand phonologicalcomputationare optimallyefficient in this regard,at least. The two kinds of Merge appearto correlateclosely with interfaceproperties,capturingmuch of the dual aspect of semanticsthatwas becoming clearerin the 1960s in work of Ray Jackendoff

10Landau'sformulationis somewhat different.

14

NOAM

CHOMSKY

(1969), Peter Culicover (1970), Lyle Jenkins (1972), and others, sometimes expressed in terms of deep and surface interpretation. ExternalMerge correlateswith argumentstructure,internal Merge with edge properties,scopal or discourse-related(new and old information,topic, etc.). The correlationis not perfect, at least on the surface, but close enough to suggest that apparent violations should fall under broaderprinciples. For external Merge, these go beyond argument structure associatedwith substantive includefunctionalcategories categories,andthey presumably of the kind coming to light in a varietyof inquiries,notablythe very fruitfulcartography projects (Cinque 2002, Rizzi 2004, Belletti 2004). Possibly such apparentexceptions as morphological identificationof typicaledge propertiesinvolve covertMove, also mattersstudiedin current work. External Merge yields the "base structures"of the EST/Y-model and earlier proposals. Since structuresare constructedin parallel, there is no locus of lexical insertion. Accordingly, the notion of D-Structureis not only superfluous,but unformulable,a more far-reachingand desirableresult. If base structures requirenothing more than Merge, bar levels can also be eliminated, extending ideas of Naoki Fukui (1986) and MargaretSpeas (1986) to full "bare phrase structure."Proceedingfurther,along lines to which I will briefly return,S-Structure and LF seem to be unformulableas well, leaving us with just the interfacelevels, the optimal outcome. The specifier-complement distinctionloses independentsignificance,except in thatthe complementof a head H shouldbe the only domainaccessible to operationsdrivenby H, by conditions of minimal search,the core propertyof c-command,but barringm-commandand specifier-head relations (unless the specifier is a head, hence a possible probe), a strong and controversial empiricalconsequenceof minimal search. "Complement"and "specifier" arejust notationsfor first-Mergeand later-Merge.Banning of multiple specifiers requiresstipulation,hence empirical justification. The empirical evidence seems to me to leave the null hypothesis viable, maybe supported,though the issue is very much alive. Each syntacticobject generatedcontains informationrelevantto furthercomputation.Optimally, that will be capturedentirely in a single designatedelement, which should furthermore be identifiablewith minimalsearch:its label, the element takento be "projected"in X-bar-theoretic systems. The label, which will invariablybe a lexical item introducedby externalMerge, should be the sole probe for operationsinternalto the syntacticobject, and the only element visible for furthercomputations.We would hope that labels are determinedby a naturalalgorithm,perhaps even by methodsas far-reachingas those that ChrisCollins (2002) has proposedin his label-free analysis. Again, these optimal possibilities are explored extensively in recent work by many researchers,along differentthough related paths. Note that labels, or some counterpart, are the minimum of what is required,on the very weak assumptionthat at least some informationabout a syntactic object is needed for further computation,both for search within it and for its external role. Labeling underlies a variety of asymmetries:for example, in a head-XP construction,the label will always be the head, and the XP a "dependency"; for substantiveheads, an argumentwith a semantic role that it assigns. Therefore,such asymmetriesneed not be specified in the syntacticobject itself, and must not be, because they are redundant. A head-complement is a set, not an orderedpair.There are structure many similar cases.

THREE

FACTORS

IN LANGUAGE

DESIGN

15

One asymmetryimposed by the phonetic interfaceis that the syntactic object derived must be linearized. Optimally, linearizationshould be restrictedto the mapping of the object to the SM interface,where it is requiredfor language-external reasons.If so, then no orderis introduced in the rest of the syntacticcomputation: the narrowsyntax and the mappingto the C-I interface. That has been a research objective for the past 25 years at least, ever since Tanya Reinhart's (1979) workon c-command,and it seems at least a plausibleconclusion,thoughthe consequences are quite restrictiveand inconsistentwith some of the most interestingwork: papersby Danny Fox and Jon Nissenbaumon quantifierraising, antecedent-contained deletion, and late Merge of adjuncts,to mention some currentexamples (Fox 2002, Fox and Nissenbaum 1999). But I think there is reason to suspect that the simplest possibility can be maintained.How syntactic objects are linearized is also a lively topic. An early proposal is the head parameter,perhapsgeneral, A different approachis based on Kayne's Linear Correspondence perhapscategory-dependent. Axiom, which has inspireda greatdeal of productivework, includingKayne's (to appear)discovery of surprisinggaps in language typology, and accounts for them. Other ideas are also being is thatthe conclusion introducefew assumptions,optimally explored.A desideratum, throughout, none, that cannot be reduced to the interfacerequirementand conditions of computationalefficiency. If linear orderis restrictedto the mappingto the phonetic interface,then it gives no reason to requirethe basic operationMergeto departfromthe simplestform.As noted,manyasymmetries can be derivedfrom configurations,from propertiesof lexical items, and from the independently needed principles of labeling, so need not-hence should not be specified by the elementary operationthat forms the configuration.If that is universal, then the basic operationwill be the simplest possible one: unstructured Merge, forming a set. Theremay be-I thinkis some reasonto introducean asymmetrical operation,for adjuncts, the basically serving function of composition of predicates.The most elementarychoice is an orderedpair, which, at some point, must be simplified to a set for the purposesof linearization. Simplificationoptimallyshouldbe at the point of Spell-Out,thatis, transferto the two interfaces. Adopting this assumption,we can derive at least some, maybe all, of the intricatepropertiesof reconstruction of adjunctsexploredby RobertFreidin,Henk van Riemsdijkand Edwin Williams, David Lebeaux, and others since (Chomsky 2004). It might be useful to allay some misgivings that appear in the technical literature.With minimal technical apparatus,a syntactic object X with the label A can be taken to be the set {A, X}, where X itself is a set {Z, W}. If A happensto be one of Z, W, then the set {A, X} has the formal propertiesof an orderedpair, as NorbertWiener pointed out 90 years ago; that is, it will have formalpropertiesof the pairsconstructed for adjuncts. Thathas been wronglyassumed to be problematic.It isn't. The possibility of giving a set-theoreticinterpretation of orderedpairs, along Wiener's lines, has no bearingon whetherorderedpairs are formal objects with their own

1 The standardWiener-Kuratowski " definitionof orderedpairis slightly different,HowardLasnikpoints out, taking Ka, b) = {{a}, {a, b}}. Wiener's original proposal was more complex.

16

NOAM

CHOMSKY

thatare distinctfrom sets, includingthose properties,"primitiveobjects" for mentalcomputation indicatinglabel. Without further stipulations, external Merge yields n-ary constituents. Particularlysince Kayne's (1981) work on unambiguouspaths, it has been generally assumed that these units are mostly, maybe always, binary.If so, we would like to find a principledexplanation.For internal Merge, it follows from elementary conditions of efficient computation,within the probe-goal framework, thoughpreciselyhow this worksraisesinterestingquestions,with considerableempirical bite. For external Merge, one source could be conditions of argumentstructureimposed at the C-I interface.Anotherpossibility has been suggested by Luigi Rizzi (pers. comm.), in terms of minimizationof search in working memory. The most restrictivecase of Merge appliesto a single object,forminga singletonset. Restricnumbers tion to this case yields the successorfunction,from which the rest of the theoryof natural can be developed in familiar ways. That suggests a possible answer to a problem that troubled Wallace over a century ago: in his words, that the "gigantic development of the mathematical capacity is wholly unexplained by the theory of natural selection, and must be due to some altogetherdistinctcause" (1889:467), if only because it remainedunused. One possibility is that the naturalnumbersresult from a simple constrainton the language faculty, hence are not given by God, in accord with Kronecker'sfamous aphorism,though the rest is createdby man, as he continued. Speculations about the origin of the mathematicalcapacity as an abstractionfrom dissociaproblems,includingapparent linguisticoperationsarenot unfamiliar.Thereare apparent phenomena is unclear tion with lesions and diversity of localization,but the significance of such for many reasons (including the issue of possession vs. use of the capacity). There may be somethingto these speculations,perhapsalong the lines just indicated,which have the appropriate minimalistflavor. On very weak assumptionsaboutefficient computation,there should be no linguistic levels beyond those that are imposed by interfaceconditions:that is, the interfacelevels themselves. In fact, it is not even obvious that these exist. Thus, one can imagine, say, a computationalprocess that sends parts of generatedobjects to the SM systems in the course of derivation(say, certain but not all phonetic properties),not just at its final stage; and the same on the meaning side. Though such proposals have never been explored, to my knowledge, they could turn out to be correct.Beyond this, however,we would like to determinewhetherall internallevels aredispensable, with the five cycles of the EST/Y-model reduced to one-better still unformulable,as DStructureis. That will follow if computationrelies solely on Merge, perhapspair- as well as setto the two Merge, yielding syntactic objects that, at some point in the derivation,are transferred interfaces:transferto the sound interface is often called "Spell-Out." Let us call the syntactic it shouldbe mapped "phases." Optimally,once a phase is transferred, objects that are transferred directly to the interfaceand then "forgotten"; later operationsshould not have to refer back to what has alreadybeen mappedto the interface-again, a basic intuitionbehindcyclic operations. We thereforehope to be able to establisha "Phase Impenetrability Condition," which guarantees that mappingsto the two interfacescan forget about what they have alreadydone, a substantial

THREE

FACTORS

IN LANGUAGE

DESIGN

17

saving in memory. Whetherthat is feasible is a question only recently formulated,and barely explored. It raises many serious issues, but so far at least, no problemsthat seem insuperable. If these general ideas are on the right track,then all internallevels are unformulable, hence dispensablein the strongestsense. We are left with only the interfacelevels, and the five cycles of the EST/Y-model are reducedto a single one, based on Merge. The cyclic propertiesof the mappingsto the interfacefollow without comment. Pure cyclic computationis also requiredfor features. Their values are redundant,determinedby an the simplest account of uninterpretable features.They should thereforebe unvaluedin the lexicon, and Agree relationwith interpretable when assigneda value, mustbe treatedas whatJuanUriagereka(1998) calls a "virus": eliminated to the phonological as soon as possible, hence at the phase level, where they are transferred componentfor possible phoneticrealizationbut eliminatedfrom the syntacticderivationand from transferto the semantic interface to avoid crash. That is again an immediate consequence of phase-basedcomputation.'2 The topic of uninterpretable featureshas moved to center stage since Vergnaud'sproposals in the past few years. Vergnaud'soriginal concerningCase theory 25 years ago, and particularly thesis that such elements as structural Case may be presenteven when not spelled out has received considerableempiricalsupport.A very strongversion of the thesis holds that inflectionalfeatures are in fact universal, ideas developed particularlyby Halldor Sigurosson (2003) as part of an extensive analysis of the role of these features,including Case, in speech events, and their finegrained syntactic analysis. Since the copy of a moved element remains,it is in principleaccessible for higher syntactic computations (muchas it can be accessedmoreglobally at the meaninginterface,underreconstruction). How deeply is such access allowed? Evidence so far is meager, but as far as I know, it never goes beyond one phase. We find examples of this, for example, with IcelandicquirkyCase: a phase-internal nominativeobjectcan agreewith an externalinflection.The exampleis interesting and meritsfurtherinquiry,because it appearsto underminea possible phase-theoretic explanation for some of Vergnaud's account of subjects that cannot appearwhere Case cannot be assigned, namely, thatthey are "invisible" because they are in the interiorof a phase thathas alreadybeen passed. In the Icelandiccase, the object is not invisible, thoughthe Phase Impenetrability Condition, which preventsany tamperingwith an earlierphase, blocks extraction.A differentaccount, based on a richervarietyof linguisticmaterialsand some differentassumptions, has been proposed by Andrew Nevins (2004). What objects constitutephases?They should be as small as possible, to minimize computational load. For reasons mentioned earlier, they should at least include the domains in which featuresare valued; agreementfeaturesapartfrom nouns and structural uninterpretable Case for nouns. There is evidence, which I will not review here, thatthese domains are CP and vP (where v is "small v" thatintroducesfull argumentstructure, transitive,or experiencerverbalstructures) and that these are the only phases in the core clausal structures. Perhapsphases also include DP,
12

See Chomsky 2001a. A consequence is the EarlinessPrincipleof David Pesetsky and EstherTorrego (2001).

18

NOAM

CHOMSKY

as is worked out by Peter Svenonius (2003), developing furtherthe parallels between DP and CP. Keeping to CP and vP, the objects derived have a skeletal structurein which C and v are the labels driving internaloperationsand relevant to externalMerge, and are also the points of feature valuation (as is a virtual necessity) and transfer.If so, a number of propertieshave a unified account. A possible speculation is that the more elaborate structuresrevealed by the cartographicinquiries are based on linearizationof features in these labels, and possibly labels closely linked to them (as in the C-T connection). The phases have correlatesat the interfaces:argumentstructure or full propositionalstructure at the meaning side, relative independence at the sound side. But the correlationis less than perfect, which raises questions. This computationalarchitecture,if sustainable,seems to be about as good as possible. SStructure and LF are no longer formulableas levels, hence disappearalong with D-Structure, and computationsare reduced to a single cycle. It should be that all operations are driven by the designatedelement of the phase, the phase head C, v. That should include internalMerge. But a problem arises, discovered by Julie Legate (2003): there are reconstructionsites at smaller categories, though movement cannot stop there but must move on to the phase head that drives the operation.A solution to the conundrumis developed by Cedric Boeckx (2003) in a slightly differentframework,adaptingideas of Daiko Takahashi' s (1994). In brief, the phase head drives the Move operation,but it proceeds category by category, in a manneroccasionally suggested for other reasons, until it reaches a point where it must stop. That is usually the phase head, though there can be an exception, namely, if the phase head assigns its uninterpretable features to the head it selects. Thus, the phase head C may be the locus of agreement,selecting T and assigning it (unvalued) +-features, so that when raising of DP driven by C-agreementreaches the TP level, its uninterpretable features are valued and it is frozen, unable to proceed further. Thatyields many of the propertiesof the ExtendedProjectionPrinciple(EPP) and, if generalized, also yields interesting empirical consequences that would carry us too far afield. But partial movementis impossiblephase-internally in othercases; in particular, we can overcomethe annoying lack of structuresof the form *There seems a man to be in the room, without the apparatus that has been proposed to account for the gap, which generalizes to the specifier of participial phrasesand other such constructions.In a somewhatdifferentframework,Jan Koster (2003) has developed the interestingidea that V-movementto T is "partialmovement" to C, the true locus of Tense, analogous to the partialwh-movementin Germandiscussed by Henk van Riemsdijk (1983) and Dana McDaniel (1989). InternalMerge seems to be driven in part at least by uninterpretable features of the phase head, as a reflex of probe-goal matching. Insofar as that is true, it carries us part of the way toward a principledexplanationfor why languages should have uninterpretable features at all. But it is not clear how far. For example, there is considerableevidence for successive-cyclic movement throughphase edges, but for A-movement it is unclearwhetherthere is independent motivationfor the featuresthat would be requiredto formulatethe operationswithin this framework. It may be that phase heads have an edge feature,sometimes called an "EPP-feature"(by extension of the notion EPP) or an "occurrencefeature" OCC (because the object formed is an

THREE

FACTORS

IN LANGUAGE

DESIGN

19

occurrenceof the moved element in the technical sense). This edge featurepermitsraisingto the phase edge without feature matching. Similar questions arise for A-movement, particularlyin languages where EPP is satisfied by a nonnominativeelement, as in locative inversion, which may agree with T, as in Bantu,thoughnot in English, thoughthereis reasonto doubtthatlocative inversion involves A-movement to subject in English. Mark Baker (2003) suggests a parameter in this respect,with interestingempiricalconsequences, distinguishingBantufrom Indo-European to T by a D-featurein Slavic and it is possible that his conclusions extend to apparentattraction andJapanese-type languages,along lines developedby JamesE. LavineandRobertFreidin(2002) and by Shigeru Miyagawa (2004). It seems possible that the variety of cases might fall undera somewhat along the lines that Baker suggests, but probe-goal frameworkwith T parameterized thatthe goal be active, keeping to +-featureprobe-goalmatchingwith or withoutthe requirement that is, with unvalued Case; or, along lines that Miyagawa develops extensively, with Focus playing the role of agreementin some class of languages.If the phase heads have both agreement features(or Focus features)andedge features,then we would expect both to applysimultaneously. That assumptiondirectlyyields some curiousphenomenaaboutinterventioneffects discussed by (2003), and also an Ken Hiraiwa (2002) and by Anders Holmberg and ThorbjorgHroarsdottir account of some apparentexceptions to the Subject Island Condition.We move here well into the terrainof open problems-a vast terrain. I have barely touched on the rich range of inquiries with these general motivationsin the summary,with many new ideas and new past few years, and now in progress.One broad-ranging empiricalmaterials,is Cedric Boeckx's recentbook Islands and Chains (2003). It seems reasonably clear thattherehas been considerableprogressin moving towardprincipledexplanationthat addressesfundamental questionsof the biology of language.It is even moreclearthatthese efforts for a sensible researchprogram:stimulatinginquirythat has have met one primaryrequirement been able to overcome some old problems while even more rapidlybringing to light new ones, previously unrecognizedand scarcely even formulable,and enrichinggreatlythe empiricalchallenges of descriptiveand explanatoryadequacythat have to be faced. References
Antony, Louise M., and NorbertHornstein,eds. 2003. Chomskyand his critics. Malden, Mass.: Blackwell. Baker, Mark. 2001. The atoms of language. New York: Basic Books. Baker,Mark.2003. Agreement,dislocation,andpartialconfigurationality. In Formalapproachestofunction in grammar,ed. by AndrewCamie, Heidi Harley,and MaryAnnWillie, 107-132. Amsterdam: John Benjamins. Belletti, Adriana,ed. 2004. The cartographyof svntacticstructures.Vol. 3, Structuresand beyond.Oxford: Oxford University Press. Boeckx, Cedric. 2003. Islands and chains: Resumptionas stranding.Amsterdam:John Benjamins. Boeckx, Cedric, and NorbertHomstein. 2003. The varying aims of linguistic theory. Ms., University of Maryland,College Park. Borer, Hagit. 2004a. Structuringsense. Vol. 1, In name only. Oxford:Oxford University Press. Borer, Hagit. 2004b. Structuringsense. Vol. 2, The normal course of events. Oxford: Oxford University Press.

20

NOAM

CHOMSKY

Boskovi6, Zeljko. 2001. On the nature of the syntax-phonologyinterface. Amsterdam:Elsevier. Brody, Michael. 1995. Lexico-LogicalForm: A radically minimalisttheory.Cambridge,Mass.: MIT Press. Brody, Michael. 2003. Towardsan elegant syntax. London: Routledge. Chemiak, Christopher.1995. Neural componentplacement. Trendsin Neuroscience 18:522-527. Raul Rodriguez-Esteban, and Kelly Changizi. 2004. Global Chemiak, Christopher,Zekeria Mokhtarzada, optimizationof cerebralcortex layout. Proceedings of the National Academyof Science online, 13 January2004 (printversion 27 January2004, 101(4):1081-1086). Chomsky, Noam. 1955. Logical structureof linguistic theory (LSLT).Ms. Partsof 1956 revision published in New York: Plenum, 1975, and Chicago: University of Chicago Press, 1985. Chomsky, Noam. 1966. Cartesian linguistics. New York: Harper& Row. Chomsky,Noam. 2001a. Derivationby phase. In Ken Hale: A life in language, ed. by Michael Kenstowicz, 1-52. Cambridge,Mass.: MIT Press. Chomsky,Noam. 2001b. New horizonsin the studyof language and mind.Cambridge: Cambridge University Press. Chomsky, Noam. 2004. Beyond explanatoryadequacy.In The cartographyof syntactic structures.Vol. 3, Structuresand beyond, ed. by AdrianaBelletti. Oxford:Oxford University Press. Chomsky, Noam, MorrisHalle, and Fred Lukoff. 1956. On accent andjuncturein English. In For Roman Jakobson, compiled by MorrisHalle and others, 65-80. The Hague: Mouton. In Syntax:An internaChomsky,Noam, and HowardLasnik. 1993. The theoryof principlesandparameters. tional handbookof contemporaryresearch, ed. by JoachimJacobs, Arnim von Stechow, Wolfgang Sternefeld,and Theo Vennemann,506-569. Berlin: de Gruyter.Reprintedin Noam Chomsky, The MinimalistProgram, Cambridge,Mass.: MIT Press, 1995. Cinque, Guglielmo, ed. 2002. The cartographyof syntactic structures.Vol. 1, Functional structurein DP and IP. Oxford:Oxford University Press. Collins, Chris. 1997. Local economy. Cambridge,Mass.: MIT Press. Collins, Chris. 2002. Eliminatinglabels. In Derivation and explanationin the MinimalistProgram, ed. by Samuel David Epstein and T. Daniel Seely, 42-64. Malden, Mass.: Blackwell. Culicover, Peter. 1970. Syntacticand semanticinvestigation.Doctoraldissertation,MIT, Cambridge,Mass. Epstein, Samuel David. 1999. Un-principledsyntax:The derivationof syntacticrelations.In Workingminimalism, ed. by Samuel David Epstein and Norbert Hornstein, 317-345. Cambridge,Mass.: MIT Press. Epstein, Samuel David, Erich M. Groat,RurikoKawashima,and HisatsuguKitahara.1998. A derivational approach to syntactic relations. Oxford:Oxford University Press. Erwin, Douglas. 2003. The Goldilocks hypothesis. Science 302 (5 December 2003):1682-1683. Fodor, Jerry. 1975. Language of thought.New York: Crowell. Fox, Danny. 2002. Antecedent-contained deletion and the copy theory of movement.LinguisticInquiry33: 63-96. Fox, Danny, and Jon Nissenbaum. 1999. Extrapositionand scope: A case for overt QR. In Proceedings of the 18th West Coast Conferenceon Formal Linguistics,ed. by Sonya Bird, Andrew Carnie,Jason D. Haugen, and Peter Norquest, 132-144. Somerville, Mass.: CascadillaPress. Frampton, John,andSam Gutmann.1999. Cyclic computation, a computationally efficient minimalistsyntax. Syntax2:1-27. Freidin,Robert. 1978. Cyclicity and the theory of grammar.LinguisticInquiry9:519-549. Freidin,Robert,and Jean-RogerVergnaud.2001. Exquisiteconnections:Some remarkson the evolution of linguistic theory. Lingua 111:639-666. Fukui, Naoki. 1986. A theory of category projectionand its applications.Doctoraldissertation,MIT, Cambridge, Mass. Revised version publishedas Theoryof projection in syntax. Cambridge:Cambridge University Press, and Stanford,Calif.: CSLI Publications, 1995.

THREE

FACTORS

IN LANGUAGE

DESIGN

21

Gallistel, C. R. 1997. Neuronsand memory.In Conversationsin the cognitiveneurosciences,ed. by Michael S. Gazzaniga.Cambridge,Mass.: MIT Press. Gallistel, C. R. 1999. The replacement of general-purposelearning models with adaptively specialized learningmodules. In The cognitive neurosciences,ed. by Michael S. Gazzaniga.2nd ed. Cambridge, Mass.: MIT Press. Gambell, Thomas, and CharlesYang. 2003. Scope and limits of statisticallearningin word segmentation. Ms., Yale University, New Haven, Conn. Harris,Zellig S. 1955. From phoneme to morpheme.Language 31:190-222. Hauser, Marc D., Noam Chomsky, and W. Tecumseh Fitch. 2002. The faculty of language. Science 198: 1569-1579. Hiraiwa, Ken. 2002. Multiple Agree. Paperpresentedat the 25th GLOW Colloquium,Utrecht. Hoffman, Donald. 1998. Visual intelligence. New York: Norton. 2003. Agreementand movement in Icelandic raising conHolmberg, Anders, and ThorbjorgHroarsdottir. structions.Lingua 113:997-1019. in English. Doctoral dissertation,MIT, CamJackendoff,Ray. 1969. Some rules of semanticinterpretation bridge, Mass. Jacob, Francois. 1982. Thepossible and the actual. New York: Pantheon. Jakobson, Roman, GunnarFant, and Morris Halle. 1953. Preliminaries to speech analysis. Cambridge, Mass.: MIT Press. Jenkins,Lyle. 1972. Modality in English syntax. Doctoral dissertation,MIT, Cambridge,Mass. Jenkins, Lyle. 2000. Biolinguistics. Cambridge:CambridgeUniversity Press. Kauffman,Stuart. 1993. The origins of order. Oxford:Oxford University Press. Kayne, Richard. 1981. Unambiguouspaths. In Levels of syntactic representation,ed. by RobertMay and Jan Koster, 143-183. Dordrecht:Reidel. Kayne, Richard.To appear.Anti-symmetryin Japanese.In Variationand universals in biolinguistics, ed. by Lyle Jenkins. Koster, Jan. 2003. All languages are tense second. Ms., University of Groningen. Landau,Idan. To appear.Chain resolutionin Hebrew V(P) fronting.Syntax. on the move, ed. by Lisa Lai-Shen Lasnik, Howard.To appear.Conceptionsof the cycle. In Wh-movement Cheng and NorbertCorver.Cambridge,Mass.: MIT Press. in neuronalnetworks.Science 301 (26 Laughlin,Simon, and TerrenceJ. Sejnowski. 2003. Communication September2003): 1870-1874. Lavine, James E., and RobertFreidin. 2002. The subject of defective T(ense) in Slavic. Journal of Slavic Linguistics 10:251-287. Legate, Julie Anne. 2003. Some interfacepropertiesof the phase. LinguisticInquiry34:506-516. Lenneberg,Eric. 1967. Biologicalfoundations of language. New York: Wiley. of remarksat "A Debate on Bio-Linguistics," a conferenceorganizedby Luria,Salvador. 1974. Transcript the CentreRoyaumontpourune science de l'homme, Paris,held at EndicottHouse, Dedham,Mass., 20-21 May 1974. Marshack,Alexander. 1985. Hierarchical evolution of the humancapacity. New York: AmericanMuseum of NaturalHistory. MaynardSmith, J., R. Burian, S. Kauffman,P. Alberch, J. Campbell, B. Goodwin, R. Lande, D. Raup, and L. Wolpert. 1985. Developmental constraintsand evolution. QuarterlyReview of Biology 60: 265-287. McDaniel, Dana. 1989. Partial and multiple wh-movement.Natural Language & Linguistic Theory 7: 565-604. McGilvray,James. 1999. Chomsky:Language, mind, and politics. Cambridge:Polity. ed. by Martha McGinnis Miyagawa,Shigeru.2004. On the EPP.In Proceedingsof the EPPiPhase Workshop,

22

NOAM

CHOMSKY

and Norvin Richards.MIT WorkingPapersin Linguistics. Cambridge,Mass.: MIT, Departmentof Linguistics and Philosophy, MITWPL. Nevins, Andrew.2004. Derivationswithoutthe Activity Condition.In Proceedings of the EPPiPhase Workshop, ed. by MarthaMcGinnisandNorvinRichards.MITWorkingPapersin Linguistics.Cambridge, Mass.: MIT, Departmentof Linguistics and Philosophy, MITWPL. Nissenbaum,Jon. 2000. Investigationsof covert phrasemovement.Doctoral dissertation,MIT, Cambridge, Mass. Pesetsky, David, and EstherTorrego. 2001. T-to-C movement:Causes and consequences. In Ken Hale: A life in language, ed. by Michael Kenstowicz, 355-426. Cambridge,Mass.: MIT Press. Physics News Update.2001. AmericanInstituteof Physics Bulletinof Physics, News Number570, December 21, 2001, reportingstudy by M. Howard,A. D. Rutenberg,and S. de Vet, Physical Review Letters, 31 December 2001. Piattelli-Palmarini, Massimo, and JuanUriagereka.To appear.In Variationand universalsin biolinguistics, ed. by Lyle Jenkins. Reinhart,Tanya. 1979. Syntactic domains for semantic rules. In Formal semantics and pragmaticsfor natural language, ed. by Franz Guenthnerand Siegfried J. Schmidt, 107-130. Dordrecht:Reidel. Richards,Norvin. 2001. Movementin language. Oxford:Oxford University Press. Riemsdijk,Henk van. 1983. Correspondence effects and the EmptyCategoryPrinciple.In Studiesin generative grammarand language acquisition, ed. by Y. Otsu et al., 5-16. Tokyo: International Christian University. Rizzi, Luigi, ed. 2004. The cartographyof syntacticstructures.Vol. 2, The structureof CP and IP. Oxford: Oxford University Press. Rizzi, Luigi. To appear.On the grammatical basis of language development.In Handbookof comparative syntax, ed. by Guglielmo Cinque and RichardKayne. Oxford:Oxford University Press. Ross, John Robert. 1967. Constraintson variablesin syntax. Doctoraldissertation,MIT, Cambridge,Mass. Publishedas Infinitesyntax!. Norwood, N.J.: Ablex, 1986. Sigur6sson, Halld6r.2003. Meaningful silence, meaningless sounds. Ms., Lund University. Speas, Margaret.1986. Adjunctionsandprojectionsin syntax.Doctoraldissertation,MIT, Cambridge, Mass. Stewart,Ian. 1998. Life's other secret. New York: Wiley. Svenonius, Peter. 2003. On the edge. Ms., University of Troms0. Takahashi,Daiko. 1994. Minimalityof movement.Doctoraldissertation,Universityof Connecticut,Storrs. Tattersall,Ian. 1998. The origin of the humancapacity. New York: AmericanMuseum of NaturalHistory. Uriagereka,Juan. 1998. Rhymeand reason. Cambridge,Mass.: MIT Press. Waddington,C. H. 1942. Canalizationof development and the inheritanceof acquiredcharacters.Nature 150:563-565. Wallace, Alfred Russel. 1889. Darwinism.London:Macmillan. Wells, Spencer. 2002. Thejourney of man: A genetic odyssey. Princeton,N.J.: PrincetonUniversity Press. Wexler, Kenneth. To appear. Lenneberg's dream: Learning, normal language development and specific language impairment.In Variationand universals in biolinguistics, ed. by Lyle Jenkins. Yang, Charles.2002. Knowledgeand learning in natural language. Oxford:Oxford University Press. Departmentof Linguisticsand Philosophy 32-D840 MIT Cambridge,Massachusetts02139 @mit.edu chomsky

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy