September4,2012 3 Sign-Based Construction Grammar: An Informal Synopsis IVAN A. SAG 1 Introduction Thischapter1 isintendedasanintroductiontosomeofthecentralnotionsof Sign-Based Construction Grammar (SBCG).2 For a more general discussion of SBCG, including an informal, high-level summary of the framework, its historicaldevelopment,motivation,andrelationtootherapproachestogram- mar,thereaderisreferredtoSagetal.thisvolume. 1.1 Preliminaries SBCG isaframeworkblendingideasdevelopedoveraquartercenturyofre- search in Head-Driven Phrase Structure Grammar (HPSG3) with those pre- 1Aboveall,I’dliketothankPaulKayforhoursofdiscussionandusefulfeedbackatevery stageintheseeminglyendlessprocessofwritingthischapter.I’malsogratefultoChuckFillmore andLauraMichaelisformanyusefuldiscussionsofbothcontentandexpositionandforcom- mentsonanearlierdraft,whichwasalsobeenimprovedbythecommentsofRuiChaves,Adele Goldberg,BillCroft,AdamPrzepiórkowski,RussellLee-Goldman,andHansBoas.Inaddition, discussionswiththefollowingpeoplehavebeenquitehelpful:FarrellAckerman,EmilyBender, DanFlickinger,MarkGawron,AndreasKathol,BethLevin,BobLevine,StefanMüller,Frank VanEynde,TomWasow,GertWebelhuth,andSteveWechsler.Noneoftheviewsexpressed hereshouldbeattributedtoanyoftheseindividuals;norshouldtheybeheldresponsibleforany remainingerrors. 2SBCGowesaconsiderabledebttotheimplementationworkcarriedoutwithinCSLI’sLinGO LabandtheDELPHINconsortium,whosegrammardevelopmenteffortshaveproceededinparal- lelwiththeevolutionofSBCG.SeeCopestake2001,Flickinger2000,Uszkoreitetal.2000,and theonlineresourcesavailableathttp://lingo.stanford.edu/andhttp://www.delph-in.net/. 3SeePollardandSag1987PollardandSag1994,GinzburgandSag2000,andRichter2004, interalia. Sign-BasedConstructionGrammar. Editors:HansC.Boas&IvanA.Sag. Copyright(cid:2)c 2012,CSLIPublications. 69 September4,2012 70/IVANA.SAG sented within the tradition of Berkeley Construction Grammar (BCG4) over roughlythesameperiod.ItsgoalistoexpandtheempiricalcoverageofHPSG, whileatthesametimeputtingBCGonafirmertheoreticalfooting. To readers steeped in HPSG theory, SBCG will no doubt seem like a mi- norvariantofconstructionalHPSG(asdevelopedinSag1997,Ginzburgand Sag 2000, and elsewhere), with the principal innovation being the introduc- tion of the distinction between signs and constructs. There is a certain truth to this, but at the same time, it is my sincere hope that construction gram- marians of all stripes will find that SBCG is recognizable as a formalized version of BCG, with a few straightforward (and only minimal) notational adjustments.WhatisgainedfromtheincreasedanalyticprecisionofSBCGis plain:clearerempiricalpredictionandfalsifiability,enhanced comparability of analyses across languages, and a general theoretical clarity. Certain dif- ferencesbetweenSBCGandBCGorotherversionsofConstructionGrammar (CxG)willbenotedwhererelevant.(SeealsoSagetal.thisvolume.) Like Pollard and Sag (1987, 1994), Fillmore et al. (1988), Fillmore and Kay(1996),andKayandFillmore(1999),thisstudywilltakeEnglishasits focus. However, the goal of our enterprise, like that of other researchers in HPSG and BCG, is to provide a basis for the description of all human lan- guages.Construction-basedgrammarhasanadvantageinthisendeavor,asit concernsitselfdirectlywithwords,generalizationsaboutlexicalclasses,and thepatternsaccordingtowhichcomplexexpressionsareconstructed.Every human language has these components; hence there is no need to transform a language into a mold that ill suits it in order to provide a typologically realistictheoryofgrammar–onethatalsomeetstheobjectionstotheChom- skyan conception of Universal Grammar raised, for example, by Evans and Levinson(2009).Itisclearthatconstruction-based grammarhasdeeproots inStructuralLinguistics.Mygoalhereistoconvincethereaderthataprop- erlyformalizedtheoryofconstruction-basedgrammarcaninadditionsatisfy thedemandsofmodernlinguistictheory,e.g.thoseoutlinedbyLees(1957: 376): 1. freedomfromcontradiction, 2. maximalcohesionwithotherbranchesofknowledge, 3. maximalvalidityincoverageofknowndata,and 4. maximaleleganceofstatement. 1.2 Preview Let us take a language to be an infinite set of signs and assume that the job of a grammarian is to provide a systematic account of those signs and their 4See,forexample,Fillmoreetal.1988,Goldberg1995,FillmoreandKay1996,Kayand Fillmore1999,P.Kay2002a,andMichaelisandLambrecht1996. September4,2012 SIGN-BASEDCONSTRUCTIONGRAMMAR:ANINFORMALSYNOPSIS/71 properties,includinghowtheyfunctioninlanguageprocessingandlanguage use. The notion of ‘sign’ of course comes from Saussure (1916). However, whiletheSaussauriansignisanassociationofsound(signifiant)andmeaning (signifié),thesignsofSBCGembodymorecomponents.Theseincludeatleast phonologicalstructure,(morphological)form,syntacticcategory,semantics, andcontextualfactors,includinginformationstructure. Signs, like all linguistic entities in SBCG, are modeled as feature struc- .tures(FSs),whichareoftwobasickinds: . atoms5(e.g.accusative,+,finite,...), functions(asexplainedbelow). Afunctional FS mapseachfeatureinitsdomain(somepropersubsetofthe setoffeatures)toanappropriatevalue(atomorfunction).6 Ingeneralterms then,functionalFSsmapfeaturestofeaturestructures.Theparticularfeatures andvaluesutilizedinthischapteraregivenintheappendix. As in LFG (Lexical-Functional Grammar; see Bresnan 2001), HPSG, and BCG,SBCGmakesastrictdistinctionbetweenentitiesinthelanguagemodel (model objects for short) and descriptions of those objects.7 As in HPSG, the most important model objects are signs (the formal representations of actual words and phrases, including sentences). Each lexical sign or fixed phrasalexpressionislicensedbyalisteme(a‘listed’descriptionofawordor phrase).8Another,distinctkindofmodelobjectinSBCGistheconstruct.As 5Thesetofatomsincludes,foranalyticconvenience,aninfinitesetofindices. 6AfunctionF canbedefinedsimplyasasetoforderedpairs,wherethefirstmemberof eachpairisamemberofthesetthatisF’s‘domain’andthesecondisamemberofthesetthat isF’s‘range’.Theonlyfurtherconditionthathastobemetisuniqueness–theconditionthat foranydistinctbandcinF’srange,F cannotcontainboth(cid:2)x,b(cid:3)and(cid:2)x,c(cid:3)(foranyxinF’s domain). AtotalfunctionF isasetofsuchpairsthatcontainsapair(cid:2)a,...(cid:3) foreachmemberaofF’s domain.FinallynotethatallFSsmust‘bottomout’inatoms.Thatis,themostdeeplyembedded functionswithinaFSmustmaptoanatom.Otherwise,i.eifF mapstoanotherfunction,F is bydefinitionnotthemostdeeplyembeddedfunctioninthefeaturestructure. 7Thoughthedistinctionbetweenmodelandmodeldescriptionmayseemunfamiliartomany linguists, drawing such a distinction reflects the standard practice of research in most scien- tific disciplines. Representational models allow the theorist to abstract away from irrelevant propertiesofthephenomenaunderinvestigation,whilethedescriptionsthemselvesconstitute the theory of the modeling objects. By placing the modeling objects in correspondence with thereal-worldphenomena,thedescriptionsbecomeatheoryofthosephenomena–morepre- cisely,atheoryofthepropertiesofthoseobjectsthatarebeingmodeled.Inthepresentcase, forexample,wewanttomodeltheobservablephoneticandsemanticpropertiesofsigns.Other modelobjects,e.g.syntacticcategories,casevalues,orcontextualconditions,areintroduced becausewebelievetheyarenecessaryinordertomodelotherpropertiesofsignsdeemedof interest. 8Theterm‘listeme’isfirstproposedbyDiSciulloandWilliams1987asageneralizationof thenotion‘lexicalentry’toincludemultiwordexpressionsofvariouskinds.Multiwordexpres- sionsarediscussedinsection7.1below. September4,2012 72/IVANA.SAG in Generalized Phrase Structure Grammar (GPSG – see Gazdar et al. 1985), constructsarelocaltreesthatarelicensedbyaparticularkindofconstruction: a combinatoric construction. As we will see, a construct can be naturally accommodatedwithinaFS-systemasafunctionalFSthatspecifiesvaluesfor the MOTHER (MTR)feature and the DAUGHTERS (DTRS)feature. The value ofMTRisasignandthevalueofDTRSisanonemptylistofsigns. Signsandconstructs,asalreadynoted,are FSs–theyarepartofthelan- guagemodel.Listemesandconstructionsaredescriptionsthatlicenseclasses oflinguisticobjects(signsorconstructs)–theyarepartofthegrammar(the descriptionofthelanguagemodel;thetheoryofthelanguage). Thelinguisticobjectsin SBCG hereareclassifiedintermsofasystemof types,whichareorganizedintoalattice-likestructurethatreflectsalinguisti- callymotivatedclassification.Thus,polar-interrogative-clauseisamaximal type (a type without subtypes) that is instantiated by clausal constructs like (1):9 (1) {[Will][Sandy][bethere]?} But in order to be well-formed according to our theory, this construct must also satisfy the constraints the grammar imposes on all the supertypes of polar-interrogative-clause. These constraints take the form of further con- structions–thosewhichdefinetheparticularpropertiesofthesupertypesaux- iliary-initial-construct, headed-construct, interrogative-clause, core-clause, andclause. AnSBCGgrammaralsocontainsasignature.Justasthetimeandkeysig- natureofamusicalcompositionspecifyhowmusicaldescriptions(e.g.notes, rests, and measures) are to be interpreted, the grammar signature delineates the basic ontology of the grammar, and thus specifies how grammatical de- scriptionsaretobeinterpreted.Itisherethatthedetailsofthetypehierarchy are laid out, along with a characterization of the general properties of each typeoffeaturestructure.Thesignatureassociateseachtypeoffunctionalfea- ture structure with a domain (a set of features) and assigns an appropriate valuetypetoeachfeatureofthedomain,asillustratedin(2): ⎡ ⎤ FEATURE1 type1 ⎢ ⎥ (2) type0:⎣... ⎦ FEATUREn typen 9Theinformalnotationin(1),usedforabbreviatingconstructs,isduetoChuckFillmore.The entireconstructisenclosedincurlybraces,witheachdaughterinsquarebrackets.Annotations arealsopossible,asindicatedbytheFsin(i): (i) {F[FD1],...,[FDn]} SeeFillmoreetal.thisvolume. September4,2012 SIGN-BASEDCONSTRUCTIONGRAMMAR:ANINFORMALSYNOPSIS/73 Thisistobeinterpretedas:ThegrammarrecognizesFSsoftype0,whichare functionswhosedomainincludesFEATURE1 ...FEATUREn andwhichmap eachFEATUREitoaFSoftypetypei,asindicated.10 AgainstthisbackgroundofpossibleFSs,aparticularSBCGofagivenlan- guagespecifieswhichparticularfamilyofFSsexistsinthatlanguage–those thatarelicensedbyalistemeoraconstruction.11 Thegrammarsignatureas- sumedinthischapterissummarizedintheappendix. 2 FeatureStructures Asalreadynoted,grammaticalobjectsofallkinds(includingsigns,caseval- ues, parts of speech, and constructions) are modeled as FSs, either atoms or else functions from features to FSs.12 This is a simple, but powerful way of modeling linguistic objects, one that is already familiar from early work in generativephonology,wherespeechsegmentsareoftenanalyzedinthisway. Forexamplethefollowingfunctioncharacterizesthephone[t]inthefeature systemofChomskyandHalle(1968): ⎡ ⎤ − CONTINUANT ⎢ ⎥ ⎢ −⎥ ⎢VOICED ⎥ ⎢ +⎥ ⎢ANTERIOR ⎥ ⎢ ⎥ (3) ⎢CORONAL +⎥ ⎢ ⎥ ⎢SONORANT −⎥ ⎢ ⎥ ⎣ +⎦ CONSONANTAL − VOCALIC Similarly, a fundamental tenet of ‘X Theory’13 is that familiar atomic cate- gorieslikeNPorVParetobereanalyzedasfunctions,e.g.asin(4):14 ⎡ ⎤ + NOUN ⎢ ⎥ (4) ⎣VERB −⎦ BAR 2 10TherangeofagivenFStype,i.e.theunionofthepossiblevaluesofitsfeatures,doesnot generallyconstituteaparticularlyusefulorcoherentset.AFSofthissortcanberegardedasa particularkindoffunctioncalledan‘ad-hocpolymorphism’(Strachey1967). 11Thoughthedistinctionbetweenthesignature’stypedeclarationsandthegrammar’scon- structionsissometimesmoreoneofconvenience,therearecomputerimplementationsoftype systemswherethetypedeclarations(butnotthetypeconstraints)areusedina‘top-down’man- nerduringprocessing. 12Carpenter1992.SeealsoSagetal.2003. 13SeeHarris1946,Chomsky1970,Jackendoff1977,andKornaiandPullum1990. 14ThesearethedistinctivefeaturesofthecategoryNPintheanalysisproposedinGazdaret al.1985. September4,2012 74/IVANA.SAG Note that the functional nature of this kind of analysis can be obscured by linguists’tendencytowritethevalueofafeaturebeforethefeature’sname, e.g. [+CORONAL] or [+V] or to use other notations, e.g. X1 (Harris 1946) orX(Chomsky1970).Yetitisclearthattheanalyticintentofsuchnotions is accurately rendered by functions whose domain is a set of features and whose range is a set of feature values (e.g. the set {+,−} in the system of ChomskyandHalle1968orthatofChomsky1974).Theuseoffunctionsto model linguistic objects is thus nothing out of the ordinary, though lack of formalizationandidiosyncraticorabstrusenotation(especiallyinthecaseof generative-transformationalsyntactictheories)oftenobscuresthisfact. Building on the more explicit ideas pioneered by computational linguis- tic work of the late 1970s, e.g. Martin Kay’s Functional Unification Gram- mar (M. Kay 1979), and the extensive subsequent work in GPSG, LFG, and HPSG,15 every grammatical object used here is modeled by a function that mapseachmemberofasetoffeaturestooneofitspossiblevalues,asspeci- fiedin(2)above.Grammaticalcategories,forexample,areanalyzedascom- plexesofvariouspropertiesrepresentedasfeature-valuepairs:nounsinclude specificationsforthefeaturesCASE,NUMBER,andGENDER;verbsarespeci- fiedintermsofthefeatureVERB-FORM(VF)fortheirinflectionclass(as[VF finite],[VF present-participle],etc.)andwillhavea‘+’or‘−’valueforthe featureAUXILIARY(AUX).Thisapproachalsotakesadvantageofthepower offunctionstomodelcomplexlinguisticentities.Unlikethephonologicaland X illustrations given above, where the values of the features are all atomic, thevalueofanSBCGfeaturemaybeanatomoritmaybeafunction(another complexFS).Thisallowsforrecursiveembeddingoffeaturestructureswithin FSs,analogoustotheembeddingoffunctionsthatisnowstandardpracticein formalizedapproachestosemanticanalysis. Signsarenoexception.Saussure(1916)regardedasignasan‘associative bond’betweenasoundconceptandasemanticconcept.Addinginsyntactic information, we arrive at representations like the ones in Figure 1, rendered hereintheSaussureanstyle(CNstandsforcommonnoun;Nforpropernoun; Vforverb). As already noted, signs are modeled as functions that specify a phono- logicalandmorphologicalstructure,ameaning,contextualconnections,and relevant syntactic information (including traditional syntactic category and combinatoric potential). These functions are described intermsof attribute- valuematrices,i.e.diagramslikethefollowing:16 15SeeGazdaretal.1985,Bresnan1982,Dalrympleetal.1995,PollardandSag1987,1994, King1989,1994,andCarpenter1992. 16Theinformalsemanticsin(5)and(6)isatemporaryexpedient,andwillbereplacedbya morepreciserepresentationinsection3.4below. September4,2012 SIGN-BASEDCONSTRUCTIONGRAMMAR:ANINFORMALSYNOPSIS/75 /tejbl/ /kIm/ " CN N ‘beingatable’ ‘apersonnamedKim’ /læf-d/ V ‘aneventoflaughing (inthepast)’ FIGURE1 SaussureanRepresentation ⎡ ⎤ (5) a. PHONOLOGY /kIm/ ⎢ ⎥ ⎣SYNTAX NP ⎦ SEMANTICS ‘theintendedpersonnamedKim’ ⎡ ⎤ b. PHONOLOGY /læf-d/ ⎢ ⎥ ⎢⎢SYNTAX V[fin] ⎥⎥ ⎣SEMANTICS ‘alaughingeventsituated ⎦ priortothetimeofutterance’ And, following work in HPSG (Pollard and Sag 1987, 1994), the notion of ‘sign’ is extended to phrases, recognizing FSs like those in (6) for complex linguisticexpressions: ⎡ ⎤ (6) a. PHONOLOGY /Evri#lINwIst/ ⎢ ⎥ ⎣SYNTAX NP ⎦ SEMANTICS ‘thesetofpropertiesalllinguistsshare’ ⎡ ⎤ b. PHONOLOGY /pæt#læf-d/ ⎢ ⎥ ⎢SYNTAX S[fin] ⎥ ⎢ ⎥ ⎢ ‘thepropositionthattherewasalaughingevent⎥ ⎢ ⎥ ⎣ ⎦ SEMANTICS situatedpriortothetimeofutterancewherea certainpersonnamedPatdidthelaughing’ The non-atomic FSs used to model linguistic objects are total functions. That is, once an appropriate feature domain is established for a particular type of feature structure, every FS of that type assigns an appropriate, fully determinatevaluetoeveryfeatureinthatdomain.Thevalueassignedtoany September4,2012 76/IVANA.SAG featuremustalsobeafeaturestructure,i.e.eitheranatomorafunctionthat inturnassignsavaluetoeveryfeatureinitsappropriatedomain.AFSisthus always ‘complete’ in a simple, intuitive sense: every feature in a function’s domainisassignedavalueintheappropriaterange. ItisimportanttonotethatwhereasFSsaretotalfunctions,FSdescriptions areinpracticealmostalwayspartial.Forthisreason,mostofthediscussions below, indeed throughout the chapters of this book, will include partial de- scriptionsoflinguisticobjects.ListemesareformulatedaspartialFSdescrip- tions(typicallybeingtrueof(or‘satisfiedby’)alargeclassof FSs);combi- natoric constructions are also quite sparse, when compared with the feature structures that instantiate them. But underlying all our concerns will be the setofFSsthatislicensedbythegrammarwedevelop.Ifsomeaspectofour grammargoesawry,weshouldbeabletolearnwhybyisolatingcertain FSs thatshouldsatisfytheconstraintsofourtheory,butdonot.Alternatively,we shouldbeabletofindsomeFSthatincorrectlysatisfiesourtheory.Inpartic- ular, an SBCG grammar must not license signs that fail to model something in the target language (they must not license an unacceptable sentence, for example).Inaddition,itmustnotfailtolicenseasignthatisneededtoserve as a model of a sentence that the grammarian decides is part of the target language. FSs have one more property that is not part of the basic theory of func- tions(whichIwillassumeonlycursoryfamiliaritywith): FSsareorganized in terms of linguistic types. A type is a classification associated with a set of FSs that have certain stated properties in common. One benefit derived from assigning FSs to types is that we can then better organize the proper- tiesthatclassesofgrammaticalobjectshaveandsimplifytheirdescriptionin the process. Intuitively, it makes no sense (in English, anyway) to ask what caseaverbhasorwhetheranounisanauxiliary–certaingrammaticalfeatu- raldistinctionsareappropriateonlyforcertainkindsofgrammaticalobjects. Thisintuitionisgivenformalexpressionintermsofthetypesthatparticular FSsinstantiate.EachFSinstantiatesaparticularmaximaltypeandthefeature appropriatenessconditionstogetherdeterminewhichsubsetoffeaturesisap- propriatefor FSsofthattype,rulingoutverbalcategoriesthatspecify CASE values,nominalcategoriesthatspecifyVFvalues,andsoforth. Thespaceoftypesishierarchicallystructured.Infact,thetypesareinter- related in terms of a multiple inheritance hierarchy. If a type B is a sub- type of another type A, then FSs of type B must satisfy all constraints that thegrammarimposesonobjectsoftypeA,aswellasthegrammaticalcon- straints imposed on type B.17 In this case, the FSs of type B form a subset 17Iamassumingthatconstraintinheritanceismonotonic,i.e.thatthereisno‘overriding’of grammaticalconstraints.Butthisassumptionisdebatable,andthegeneralframeworkofSBCG September4,2012 SIGN-BASEDCONSTRUCTIONGRAMMAR:ANINFORMALSYNOPSIS/77 literarywork verse prose Asian European epic lyric Greek English Greek-epic English-epic English-lyric (e.g.OdetoaNightingale) (e.g.TheOdyssey) (e.g.Beowulf) FIGURE2 AMultiple-InheritanceHierarchy oftheFSsoftypeA.Thissituationisinformallycharacterized,followingthe terminologyofBCG,bysayingthat‘typeBinheritsfromtypeA’.18Inamul- tipleinheritancehierarchy,atypecaninheritfrommorethanoneimmediate supertype.Thatis,typehierarchiesbehavelikethenonlinguisticexamplein Figure2,whereaninstanceofeachmaximaltypeisgivenbelowit.Multiple- inheritance hierarchies are useful for analyzing cross-classifying properties ofasetofobjects,whethertheyareliteraryworks,words,orconstructs. InSBCG,themoregeneralnotionof‘typehierarchy’takesovertheinher- itance functions that constructional inheritance performed in some earlier traditionsof CxG.19 Forexample,Fillmore(1999)treatsthevariouskindsof auxiliary-initialclauseintermsofconstructionalinheritancefromthesuper- typehecalls‘subject-auxiliaryinversion’(SAI).SomeoftheseSAIconstruc- tionsareillustratedin(7): (7) a. {[Has][he][left?]} b. {[Am][I][tired!]} c. Never{[will][I][harmyou.]} d. What{[did][Merle][know?]} e. {[May][you][livelongandprosper!]} f. {[Had][he][beenontime]},hewouldn’thavegonehungry. couldbeoutfittedtoincludedefaultconstraintsandconstraintoverriding.Mostlinguistically interestingexamplesofdefaultconstraints,however,canbecastas‘nonpersistent’defaultsin thesenseofLascaridesandCopestake1999,andhencecanbestraightforwardlyreanalyzedin purelymonotonicterms,asIhaveattemptedtodohere. 18Thisterminologyisinfactsomewhatmisleading,asitisthefeaturestructuresinstantiating agiventypethatinheritthepropertiesassociatedwiththetype’ssupertypes. 19See,forexample,Fillmoreetal.1988,Goldberg1995,FillmoreandKay1996,Fillmore 1999,KayandFillmore1999,P.Kay2002a,2002b,MichaelisandLambrecht1996. September4,2012 78/IVANA.SAG IntheSBCGanalogofFillmore’sanalysis,eachofthebracketedsequencesin (7)instantiatesatypeofconstructthatisasubtypeofthemoregeneraltype auxiliary-initial-construct (aux-initial-cxt). The Aux-Initial Construction (a combinatoricconstruction)placesgeneralconstraintsoninstancesofthetype aux-initial-cxt,assketchedin(8): (8) Aux-InitialConstruction(PreliminarySketch) (cid:5) (cid:6) MTR S[...] Anaux-initial-cxtmustsatisfy: DTRS (cid:2)V[AUX+],...(cid:3) Amorespecificconstruction,i.e.aconstructionthatcharacterizes asub- typeofaux-initial-cxt,needstospecifyonlythepropertiesthatarespecificto thatsubtype.Inthecaseofaux-initialclauses,thesubconstructionsinques- tion specify primarily semantic information, but also syntactic constraints about independent clause status, modifier status, etc. For example, the In- vertedWishConstruction(ofwhich(7e)isaninstance)definesthecharacter- isticpropertiesoffeaturestructuresinstantiatingthemaximaltypeinverted- wish-construct (inv-wish-cxt). It specifies the appropriate semantics and the constraintthataFSofthistypemustbeanindependentclause.Invertedwish constructswillofcoursealsoexhibitthegeneralpropertiesof(obeythegen- eral constraints on) aux-initial constructs. This is accomplished simply by specifyingthatinv-wish-cxtisasubtypeofaux-initial-cxt. 3 Signs Thefollowingsectionsintroducethespecificfeatureswhosevaluesserveto distinguishthesignsofalanguagefromoneanother. 3.1 PHONOLOGYandFORM Little will be said here about morphology, and nothing at all about phonol- ogy,buttheintentionisthatphonologicalandmorphologicalentitiesbepart oflinguisticsigns.Iassumethatalargelyautonomoussetofconstraintschar- acterizetherelationbetweenthephonologicalandmorphologicalaspectsof signs.20Therearethustwodistinctsign-levelfeatures:PHONOLOGY(PHON) andFORM: (9) a. ThevalueofthefeaturePHONisaphonologicalstructurei.e.aFS oftypephonological-object. 20Anumberofresearchershavedevelopedrelatedapproachestotheanalysisofphonological andmorphologicalstructureintermsoftypedfeaturestructuresofthesortassumedhere.See BirdandKlein1994,Orgun1996,2000,Klein2000,AsudehandKlein2002,BonamiandBoyé 2002, 2006, Haji-Abdolhosseini 2003, Tseng 2003, 2008, and Bonami and Delais-Roussarie 2006,amongothers.
Description: