aronoff - in the beginning was the word

29
Access Provided by Univ.Lisboa - Fac.Letras e ICS at 09/20/11 2:45PM GMT

Upload: natalia-prado

Post on 18-Aug-2015

245 views

Category:

Documents


3 download

DESCRIPTION

livro sobre linguística

TRANSCRIPT

Access Provided by Univ.Lisboa - Fac.Letras e ICS at 09/20/112:45PM GMTINTHEBEGINNINGWASTHEWORDMARKARONOFFStonyBrookUniversityAlllinguistsassumethatthemeaningofacomplexsyntacticexpressionisdeterminedbyitsstructureandthemeanings of its constituents. Most believe that the meanings of words aresimilarly compositionally derived fromthe meanings of their constituent morphemes. The classicallexicalisthypothesisholdsinsteadthatthecentralbasicmeaningfulconstituentsoflanguagearenot morphemesbut lexemes. Thisarticlesupportsthat hypothesiswithevidencefromsyntax,lexical semantics, and morphology. Al-Sayyid Bedouin Sign Language (ABSL) is a new languagethat is compositional down to its smallest pieces, in which these pieces are lexemes. ABSL showsthat a language can emerge very quickly in which lexemes are basic. Next comes evidence againsttheclaimthatnewlyderivedwordsdivergefromcompositionalityonlybecausetheyarestoredinmemory. Finally,Hebrewverbrootsareshowntohaverobustmorphologicalpropertiesthatbear no relation to meaning and little to phonology, leaving room for morphemes within a lexeme-basedframework,butnotasthebasicmeaningfulatomsoflanguage.*Linguists have followed two ways in the study of words. One seeks to accommodatethe word, the other to obliterate it. I defend here an approach to language that respectstheautonomyofarbitraryindividualwordsorlexemesandprivilegestheinteractionbetweentheidiosyncrasiesoflexemesandahighlyregularlinguisticsystem.1Puttingthisapproachinthecontextofthehistoryofourfield,IshowthatChaquemotasonhistoire[Eachwordhasits[own]history](Gillieron; seeMalkiel 1967:137) inunsyste`meou` tout setient [asystemwhereeverythinghangstogether](Meillet 1903:407),butinsteadofphonology,whichwasGillieronsmainconcern,Iconcentrateonsyntax, semantics, andmorphology. I amnot claimingthat thelexemeis theonlyidiosyncraticelementinlanguage,butthatitisthecentralsuchelement.1. COMPOSITIONALITY. All linguists assume that natural languages are composi-tional, which we understand informally to mean that the meaning of a complex syntacticexpression is determined by its structure and the meanings of its constituents.2The mainargument for compositionality of natural language is rooted in syntactic productivity andwasstatedsuccinctlybyFrege:Thepossibilityofourunderstandingsentenceswhichwehaveneverheardbeforerestsevidentlyonthis,thatwecanconstructthesenseofasentenceoutofpartsthatcorrespondtowords.(Frege1980[1914]:79)* ThisarticleisarevisedversionofmyPresidentialaddress,deliveredattheannualmeetingoftheLSAinJanuary2006.MuchofthecontentpresentedhereistheresultofjointworkwithFrankAnshen,RobertHoberman, Irit Meir, Carol Padden, WendySandler, andXuZheng. I amgrateful for discussiontomydepartmental colleagues at Stony Brook: Christina Bethin, Ellen Broselow, Dan Finer, Alice Harris, HeejeongKo, RichardLarson, andLori Repetti;andtomyformercolleaguePeterLudlow. Thankstothemembersofmyimmediatefamilyforprovidingdataandeditorialassistance.Thisresearchwassupportedinpartbythe US-Israel Binational Science Foundation, the National Institutes of Health, and the Stony Brook ProvostsOffice.1Psycholinguists,forexampleLevelt(1989),usethetermLEMMAforthepurelysyntacticandsemanticaspectsoflexemesandthetermLEXEMEforwhatlinguistswouldcallphonologicalword. Iuselexemetorefertotheentirelexicalentry,includingsyntax,semantics,andforms.2Forageneral overviewofcompositionality(Fregesprinciple), seeSzabo2007. Iamnot suggestingthat all linguists agree on what the units are over which compositionality operates, only that compositionalityhasaroleinlanguage.803LANGUAGE,VOLUME83,NUMBER4(2007) 804Exactlywhatthesepartsthatcorrespondtowordsareinnaturallanguageshasbeendebatedsincethebeginningof linguistics(cf. theAncient Greekgrammatical termMEROS LOGOUpiece/part of speech). Saussurecalledthem(simple) signs, withoutbeingentirelyclearaboutwhatlinguisticunitshetooktobesimplesigns,thoughforthe most part his examples are what modern linguists would call lexemes (de Saussure1975, Carstairs-McCarthy 2005). A term likeANTIDECOMPOSITIONALIST opposed to theviewthatthemeaningsofwordscanbebrokendownintoparts,forexample,canbeanalyzedinto[[anti [[[de[compos-it]V]Vion]Nal]A]Aist]A.3Thestandardviewisthatthemeaningsofthiswordandotherslikeitarecompositionallyderivedfromthemeaningsoftheseconstituentmorphemes, makingmorphemes, notwords, thepartsthat correspondtowords innatural languages. Abit confusing, but linguists likeexoticizingtheordinary.Iamantidecompositionalist bytemperament, becauseIbelievethat what happensinsidelexemes(Matthews1972)isqualitativelydifferentfromwhathappensoutsidethem. Myadherencetothisclassicversionof theLEXICALISTHYPOTHESIS(Chomsky1970) is grounded in a simple love of words. As I wrote in my first publication (Aronoff1972),onthesemanticsoftheEnglishwordgrowthanditsrelevancetothelexicalisthypothesis, all a person had to do [in order to notice the peculiar differences in meaningbetweentheverbgrowandthenoungrowth]wastolookthewordsupinareputabledictionary,orthinkforalongtime:growgrowth(1972:161).Yetveryfewmodernlinguistseitherlookwordsupinareputabledictionaryorthinkaboutthemforalongtime. Philosophers do, but linguists dont often think much about the subtleties of wordmeanings,whichiswhymostlinguistsarenotantidecompositionalists.The locus classicus of antidecompositionalist lexicalismis Chomsky1970. Thisparticular work, however, can be read in a myriad of ways, much in the way that mostof Wittgensteins works can be read to support a great number of contradictory positions.What is beyond dispute is that Chomsky introduced the terms LEXICALIST and LEXICALISTHYPOTHESISindiscussingderivednominalslikerefusal,eagerness,andconfusion:Wemight extendthebaserulestoaccommodatethederivednominal directly(Iwill refertothisasthelexicalistposition). . . [T]hereisnoaprioriinsightintouniversalgrammar. . . thatbearsonthisquestion,whichisapurelyempiricalone.(Chomsky1970:88)Chomsky brings three types of empirical evidence to bear on the issue of whether suchderived nominals are, in fact, transformationally related to the associated propositions:restricted productivity, baselike syntax, and the fact that the semantic relations betweentheassociatedpropositionandthederivednominalarequitevariedandidiosyncratic(1970:188). Heconcludesthat theyarenot transformationallyrelatedandthat Thestrongest and most interesting conclusion that follows from the lexicalist hypothesis isthat derivednominalsshouldhavetheformof basesentences (212), bywhichhemeansthatderivednominalsareinsertedintolexicalstructuresinthesamewaythatsimplebasenounsare. Thisistheoriginal lexicalist hypothesis, theideathat somemembersofmajorlexicalcategories(lexemes)arenotderivedbythesameapparatusthat derives sentences but are inserted into lexical categories in the base just as simplelexemes are.4What I call theCLASSICAL LEXICALIST HYPOTHESIS (Jackendoff 1975, Aro-3Theformpos-itistheallomorphoftherootposethatappearsbeforethesuffix-ion.4Chomsky provides a sketch of an apparatus that might account for derived nominals nontransformation-ally,butthatapparatushasbeenverylittleexploredsince,exceptforX-bartheory,whichwasdesignedtoexpresstheparallelismbetweenthe internalstructuresofphrasalcategorieslikeNPand VPwithouthavingtoderiveonefromtheothertransformationally.INTHEBEGINNINGWASTHEWORD 805noff 1976) extendsthescopeof theoriginal hypothesistoall lexemesandlexeme-internalsignstructure(derivationalmorphologyandcompounding),thoughnottoin-flectional morphology(whichrelatesformsof asinglelexemetooneanother), andclaims that the level of the lexeme or lexical category is significant in the way that thelevel of the molecule is significant in chemistry: it has properties that are worth isolating.ANTIDECOMPOSITIONALISMisrelatedtothisclassicallexicalisthypothesis:it saysthatthescopeofsyntax-basedlogicalcompositionalityshouldnotbeextendedbelowthelexemebecauselexeme-internal structure(however it isdescribed) isdifferent fromlexeme-externalstructure.Animportant andoftenforgottenpoint about antidecompositionalismisthat, likeChomskys original lexicalist hypothesis, it is anempirical claim, not atheoreticalpostulate. Still, weantidecompositionalists dohave theoretical obligations that ouropponentsdonot, becausewemustaccount fortheinternalpropertiesoflexemesinadifferentwayfromhowweaccountforsyntax(includinginflection).5ThatiswhatIhavebeentryingtodoforthelastthirty-fiveyears.Alsoimportant is thefact that antidecompositionalists donot denythevalueofmorphemes, just aschemistsdonot denythevalueof atomsandparticles. Thereisplentyofevidence, linguisticandpsycholinguistic, formorphemesandrootsandformorphologicalrelatedness.Butnoneofthisevidence,paceStockall&Marantz2006,supportsapurelymorpheme-basedtheoryoveronethat recognizeslexemesbut alsorecognizes roots and morphemes as morphologically significant elements, albeit not asreliableSaussureansigns.6Therest of myarticleisdividedintothreeparts. First, I discussalanguagethatappears to be completely compositional down to its smallest pieces, Al-Sayyid BedouinSignLanguage(ABSL).IshowthatABSLisaninstantiationofSaussurespictureoflanguage, with little if any structure below the level of the lexeme. ABSL thus providesanunusual typeof evidenceforbothcompositionalityandthelexicalist hypothesis,showing that very quickly, a language can emerge on its own in which words are indeedthepartsthatcorrespondtowords.I thenturntowell-knownphenomenafromEnglishandpresent evidenceagainstonecommonmisunderstandingofthespecialnatureofwords,thattheydivergefromcompositionality only because they are stored in memory. I review old data and presentone new example to show that even productively coined new words can be noncomposi-tional.I closewithsometraditional linguisticanalysisandshowthat Semiticroots, theoriginal lexeme-internal linguistic units, have robust properties that bear no relation tomeaningandlittletophonology. Thereisthusroomforrootswithinalexeme-basedframework,butnotasthebasicmeaningfulatomsoflanguage.Theapproachtomorphologythat I takehere is lexeme-basedandrealizational,usingthetaxonomyofmorphologicaltheorieselaboratedinStump2001. Likemuchmorphological research of the last thirty years it is rooted in the work of P. H. Matthews(1972, 1974). It also has close affinities to Corbin 1984 and Anderson 1992. Carstairs-McCarthy2002, Booij 2005, andAronoff&Fudeman2005arerecent introductions5Decompositionalists havetheir ownburdens, most vexingamongwhichis the general problemofrecoverabilityofdeletion(Chomsky1970,n.11).6FordefinitionsandextendeddiscussionofthetermsLEXEMEandROOT,seeAronoff1994:Ch.1.IamusingMORPHEMEintheclassicsenseofHarris1951, extendedfollowingthediscussioninAronoff1976:Ch.2.LANGUAGE,VOLUME83,NUMBER4(2007) 806tomorphologywithinthistradition. Spencer2000providesagoodbriefoverview. Ido not think that it is unfair to say that the lexeme-based realizational approach has beenprevalent among practicing morphologists for the last quarter century. Nonpractitioners,however, seem to prefer morpheme-based approaches. This raises the question of howmuchoneshouldtrusttheexperts,butIdonotattempttoanswerthatquestionhere.2. AL-SAYYIDBEDOUINSIGNLANGUAGE. For several years, I have been privileged tobe a member of a team, along with Wendy Sandler, Irit Meir, and Carol Padden, workingonAl-SayyidBedouinSignLanguage(ABSL). ABSLhasariseninthelast seventyyearsinanisolatedendogamouscommunitywithahighincidenceofnonsyndromicgeneticallyrecessiveprofoundprelingual neurosensorydeafness(Scott et al. 1995).WhatdistinguishesABSLfromallotherwell-documentednewlanguagesarethecir-cumstances of its creation and use, which show neither discontinuity of social structurenortheinfluenceofotherlanguages.In the space of one generation from its inception, systematic word order has emergedin the language. This emergence cannot be attributed to influence from other languages,sincetheparticularwordordersthatappearinABSLdifferfromthosefoundbothintheambient spokenlanguagesinthecommunityandinother signlanguagesinthearea.TheAl-SayyidBedouingroupwasfoundedalmost twohundredyearsagointheNegevregionof present-dayIsrael. Thegroupisnowinitsseventhgenerationandcontainsabout3,500members, allresidingtogetherinasinglecommunityexclusiveof others. Consanguineousmarriagehasbeenthenorminthegroupsinceitsthirdgeneration. Suchmarriagepatternsarecommonintheareaandleadtoverystronggroup-internalbondsandgroup-externalexclusion.Withinthepastthreegenerations,over one hundred individuals with congenital deafness have been born into the commu-nity, all of them descendants of two of the founders five sons. Thus, the time at whichthe language originated and the number of generations through which it has passed canbepinpointed. All deaf individuals showprofoundprelingual neurosensoryhearingloss atall frequencies,have anotherwise normalphenotype, andare ofnormal intelli-gence. Scott andcolleagues(1995) identifythedeafnessas(recessive) DFNB1andshowthat it hasalocusonchromosome13q12similartothelocusofseveral otherformsofnonsyndromicdeafness.The deaf members of the community are fully integrated into its social structure andarenotshunnedorstigmatized(Kisch2004).Bothmaleandfemaledeafmembersofthe community marry, always to hearing individuals. The deaf members of the commu-nity and a significant fraction of its hearing members communicate by means of a signlanguage. Siblings and children of deaf individuals, and other members of a household(whichmayincludeseveralwivesandtheirchildren),oftenbecomefluentsigners.Membersofthecommunitygenerallyrecognizethesignlanguageasasecondlan-guage of the village. Hearing people in the village routinely assess their own proficiency,praisingthosewithgreaterfacilityinthelanguage. ThosewhohaveanyfamiliaritywithIsraeli SignLanguage, includingthosewhohaveattendedschoolsfor thedeafoutside the village, recognize that the two sign languages are distinct. Nor do Al-Sayyidsigners readily understand the Jordanian sign language used in simultaneous interpretingonJordaniantelevisionprogramsreceivedinthearea.Many of the signers in this community are hearing, a highly unusual linguistic situa-tion, but onethat is predictedtoariseasaconsequenceof recessivedeafnessinaclosedcommunity(Laneetal.2000).OneresultoftherecessivenessisthatthereareINTHEBEGINNINGWASTHEWORD 807aproportionatelylargenumberofdeafindividualsdistributedthroughoutthecommu-nity(over4percent).Thismeansthatmorehearingmembersofthecommunityhavedailycontact withdeaf members, andconsequentlysigningisnot restrictedtodeafpeople.Furthermore, eachnew generationofsigners isborn intoanativelike environ-ment with numerous adult models of the language available to them. ABSLthus presentsa unique opportunity to study a new language that has grown inside a stable communitywithout obviousexternal influence. Wehaveidentifiedthreegenerationsof signers.Thefirstgenerationinwhichdeafnessappearedinthecommunity(thefifthsincethefounding of the community) included fewer than ten deaf individuals, all of whom aredeceased. Information on their language is limited to reports that they did sign and oneveryshort videotape recordof one of these individuals. I report here onlyonthelanguageofthesecondgeneration.I discusstheorder of signswithinsentencesandwithinphrasesinthelanguage,based on a tally of all sentences in a database that consist of more than one sign (Sandleret al. 2005). Sentences consisting solely of a predicate and utterances consisting solelyofnounswereset aside. Wefirst talliedtheorderamongthemajorelementsoftheremainingclauses, thepredicate (henceforthV)and itsarguments. Figure1shows theorderofpredicatesrelativetoargumentswithintheclausesconsideredinourcount.Outof158clauses,136arepredicate-final.FIGURE1.ABSLispredicate-final.Figure2showsthefrequencyoftherelativeordersofS, O, IO, andVwithinthedata (following standard practice, we lump together O and IO unless a clause containsboth). One hundred and twenty-six clauses contain only one noun, of which fifty-eightcontain S and V, and sixty-eight O and V. Where two nouns are mentioned in a clause,SprecedesO(inallthirty-twocases). SneverfollowsV. ThusweconcludethattheprevalentwordorderforABSLsentencesisSOV.Abinomialtestyieldsstatisticallyreliable p-values of less than 0.00001 for this and all other word-order effects reportedhere.LANGUAGE,VOLUME83,NUMBER4(2007) 808FIGURE2.MajorconstituentordersinABSLdata.We next considered the order of modifier elements within phrases (adjectives, nega-tives,andnumerals)relativetotheirheadnounsandverbs.AsshowninFigure3,inallinstancesbutone,themodifierfollowsitshead.Overall, wefindstructural regularitiesintheorderofsignsinthelanguage: SOVorderwithinsentences, andHead-Modifierorderwithinphrases.7Thesewordorderscannot be attributed to the ambient spoken language. The basic word order in the spokenArabicdialect of thehearingmembersof thecommunity, aswell asinHebrew, isSVO. Thisgenerationofsignershadlittleornocontact withIsraeli SignLanguage,whose word order appears to vary more widely in any case. Nor can the Head-Modifierorderbeascribedtotheambient colloquial Arabicdialect ofthecommunity. Inthisdialect, and in Semitic languages generally, although adjectives do follownouns, numer-als precedenouns; andnegativemarkers onlyrarelyfollowtheir heads. Hencetherobust word-order pattern exhibited by the data is all the more striking, since it cannotbe attributed to the influence of other languages; rather, this pattern should be regardedasanindependentdevelopmentwithinthelanguage.This remarkablestructural clarity breaks downentirely when welook inside words,where we find very little structure, either morphological or phonological. The languagedoeshaveafairnumberofcompounds, butcompoundsarenotmorphologicalinthestrictsense,sincetheyarecomposedentirelyofwordsanddonotcontainanyboundelements.8Signlanguagesareknownfortheirrobust inflectional morphology, espe-7Thetwoordersarenotsymmetrical,butclaimsaboutsymmetryinthisdomainhavenotstoodupwellovertime(Croft2003).8BettercandidatesformorphologicalstructurearethesizeandshapeclassifiersthatarefoundinsomesignsinABSL. Thesearelimitedincomplexityandproductivity, but maybeprecursorsoftherichandproductiveclassifierconstructionsfoundinmoreestablishedsignlanguages(Emmorey2003).INTHEBEGINNINGWASTHEWORD 809FIGURE3.RelativeordersofamodifieranditsheadinABSL.ciallytheircomplexverbagreementsystems, whichariseveryquicklyinthehistoryof individual languages (Aronoff, Meir, &Sandler 2005). Our teamwas thereforesurprised to discover that ABSL has no agreement morphology, indeed no inflectionalmorphology at all (Aronoff, Meir, Padden, & Sandler 2005). Nor does it appear to haveaphonologicalsystemofthekindwearefamiliarwithinothersignlanguages.PhonologyisaninstantiationofwhatHockett(1960)callsdualityofpatterning,orwhatothershavecalleddoublearticulation(Martinet1957).Whenwesaythatasignlanguagehasphonology, wearesayingthatithasasetofdiscretemeaninglesscon-trastive elements(handshapes, locations,and movements) thatform theirown system,withconstraintsontheircombination, constitutingasecondarticulationindependentofthefirst articulationofmeaningful elements. ThissystemofcontrastiveelementsmakesupthesignifiantoftheSaussureansign.Itwastheexistenceofsuchasystemin American Sign Language that first led William Stokoe to proclaim its legitimacy asalanguage(Stokoe1960).ABSL appears not to have a system of discrete meaningless elements within words.Instead, each word has a conventionalized form, with tokens roughly organized aroundaprototype,butnointernalstructure,asthefollowingillustrationsoftheABSLsignsKETTLE(Figure4)andBANANA(Figure5)illustrate.KETTLEisacompound,ofwhichthefirstcomponentisCUPandthesecondeitherPOURorBOIL.Forlackofspace, it isimpossibletoshowmorethanafewexamplesofeach, but inboth, andespecially in BANANA,there is considerable variation among individualsigners. Thevariationfoundamongsignersinmanywordsinourdata(ofwhichBANANAisagoodexample)liespreciselyalongtheparametersthat createcontrastsinothersignlanguages. The difference in the amount of variance between KETTLE and BANANAmayalsobetakenasanindicationthat certainsigns, likeKETTLE, arebecomingmoreconventionalized, but conventionalizationdoesnot equal phonology. It isalsoLANGUAGE,VOLUME83,NUMBER4(2007) 810FIGURE4a.KETTLE(CUP POUR).FIGURE4b.KETTLE(CUP POUR).noteworthy that the signers in Figures 5ce, whose sign for BANANA is very similar,arebrothers.Wedofindmuchmoresimilarityinsignformwithinsinglefamilies.ABSL has been able to develop into a full-fledged linguistic system without benefitofphonologybecauseofthevisualmediumofsigning,whichhasmanymoredimen-sions than sound does and which allows for direct iconicity (Aronoff, Meir, & Sandler2005).AsHockettnotes(1960:95):Thereisexcellent reasontobelievethat dualityofpatterningwasthelast propertytobedeveloped,because one can find little if any reasonwhy a communicative system should have this property unlessitishighlycomplicated.IfaVOCAL-AUDITORYSYSTEM[emphasisaddedMA]comestohavealargerandlarger number of distinct meaningful elements, thoseelementsinevitablycometobemoreandmoresimilartooneanotherinsound.Thereisapracticallimit,foranyspeciesoranymachine,tothenumber of distinct stimuli that can be discriminated, especially when the discriminations typically havetobemadeinnoisyconditions.ItmaybethatABSLwilldevelopphonologyasitages,perhapssimplyasafunctionofthesizeofitsvocabulary,asHockettsuggests(seeNowak&Krakauer1999foraINTHEBEGINNINGWASTHEWORD 8115a. 5b.5c. 5d.5e.FIGURE5.BANANA.LANGUAGE,VOLUME83,NUMBER4(2007) 812mathematicalmodel). Indeed,our teamis currentlyinvestigating questionsof justthissort. Forthemoment, though, ABSLisalanguagewithacompositional syntax, butvery little word-internal structure of any kind. It is, from a certain perspective, a perfectlanguage(Chomsky1995,2000,Eco1995).3. THE MEANINGS OF MORPHOLOGICALLY COMPLEX WORDS. The tenet that every naturallanguageisaunitarycompositionalsystemofsentenceswhosebasicbuildingblocksaremorphemeshasanintellectualsisterinthenotionthatthemeaningandstructureof morphologicallycomplexwords cannot merelyberepresentedinterms of theirsentential paraphrases but should be formally reduced to these paraphrases, so that thewordsthemselvesare reducedtosentencesorpropositionsand morphologytosyntax.To proponents of this method, words and sentences really are the same kinds of things,becausewordsaresentencesinminiature, linguisticmunchkins.Thisideadatesbackat least to Panini and it has been popular among generativists almost from the beginning;not asurprise, sincethemost successful earlygenerativeanalyses, notablythoseofChomsky1957, makenodistinctionbetweensyntaxandmorphology, atleastinflec-tionalmorphology.9Precisely those morphologically complex word types that have seemed most akin tosentencesinthestructureof their meanings, however, whenstudiedinmoredetail,provideempirical supportforjust theoppositeconclusion: morphologicallycomplexwords are not sentences and their meanings are arrived at in an entirely different fashion.Almost all theevidencethat Ireviewisovertwenty-fiveyearsoldandmuchofthefieldhassuccessfullyignoredit, soI amverymuchintheminority. Nonetheless, Iwould like to take advantage of the occasion to make a new plea for what, to my mindatleast,havebeenveryimportantfindings.Thefoundinggenerativeworkintheenterpriseofreducingwordstosentences, asChomskynotes(1970:188), isLees1960.10ThedetailsofLeessanalysishavebeenforgotten,buthisideahasflourishedonandoffeversince,moreonthanoff,largelyunchanged except in notation. I offer two examples of Leess analysis. The first is whathecalls actionnominals. Herearetwosamples of suchnouns andtheir sententialsources,accordingtoLees.(1) ThecommitteeappointsJohn Johnsappointmentbythecommittee(2) ThecommitteeobjectstoJohn ThecommitteesobjectiontoJohnTheseandothers withdifferent morphologyareproducedbymeans of asingleaction nominal transformation. The second example is nominal compounds. Here Leesdiscusseseight types, whichincludesubject-predicate, subject-verb, verb-object, andothers, all basedonsyntacticrelations. Hisdiscussionof thesetypestakesupfiftypages,soIcantparaphraseithere,butafewsamplesaregivenbelow.(3) Theplantassemblesautos autoassemblyplant(4) Thecapiswhite whitecap(5) Theartisthasamodel artistsmodel(6) Thesheephasahorn.Thehornislikeaprong pronghorn9Synactic structuresitself has a prescientshort section (7.3) showingthat the adjective interestingis notderivedfromasentencecontainingtheverbinterest,forpreciselythesortsofreasonsthatIoutlinebelow.10Oneindicatorofitspopularityatthetimeisthefactthatitwasreprintedfourtimesby1966,thedateofissueofmyowncopy,nomeanfeatwhenonetakesintoaccountthesizeofthecommunityoflinguistsintheearly1960s.INTHEBEGINNINGWASTHEWORD 813Linguists are tempted to try to reduce the structures and meanings of complex wordsto those of sentences because they have a good set of tools for studying the syntax andsemantics of sentences, which they apply to words. But words have much less apparentinternalstructure.Thisdifferenceisbothablessingandacurse:wecantransferwitheasewhatweknowabouttheinternalstructureandmeaningsofsentencestowords,but we have no direct way of finding out the actual extent of the isomorphism betweenthetwodomains.Wemayverywellbedeludingourselves.Asecondproblemis that, unlikesentences, words arenot essentiallyephemeralobjects. AsPaul Bloomhasshown, thoughsyntacticdevelopment stopsfairlyearly,certainlybeforepuberty, individual speakers of alanguagecontinuetoaccumulatewords at an astonishing rate throughout their lives and retain them in memory (Bloom2000). They do not accumulate sentences. Because sentences are ephemeral, their mean-ings must be compositional sums of the meanings of their parts. This was Freges mainargument for the compositionality of sentences in natural languages. By the same token,becausewordsarenotsoephemeral,andmoreoftenthannotretainedinmemoryandsociety, complex words do not have to be compositional. Words that have been aroundforanytimeatalldevelopidiosyncrasiesthatarepassedontotheirnewlearners.Therehave beentwo linesof responseto thisobvious noncompositionalityof somecomplex words. The more common response is to say that complex words are composi-tionalatheartandthatthedeparturesfromregularityaccretepreciselybecausewordsarestoredandused. Thesemanticdifferencesbetweensentencesandcomplexwordsare thus purely accidental on this view, the result of nonlinguistic or noncomputationalfactors.One example in support of the position that words arent born with noncompositionalmeaning but have it thrust upon them is the trio of words cowboy, refugee, and skinner,whicharecuriously synonymousinonesense: oneofabandof loyalistguerillasandirregularcavalrythat operatedmostlyinWestchesterCounty, NewYork, duringtheAmerican Revolution (Websters third new international dictionary, henceforth WIII).Explainingwhythesethreewordscametobeusedtodenotethisparticulargroupisa philological adventure of precisely the sort that gives linguists a bad taste for etymol-ogy, but it is obvious that that story has nothing to do with the compositional meaningofanyofthem.Instead,thispeculiarmeaningaroseforallthreewordsundercertainveryparticularhistoricalcircumstances. Thefactthatthismeaningofthesewordsisnot compositional is, linguisticallyspeaking, anaccident. Themeaningitself is nomorethanacuriosity.Thisanalysisallowsustopreservetheintuitionthatwordsarecompositional at birth, just like sentences, but that they lose their compositional mean-ingsandacquireidiosyncraticmeaningslikethisonejustbecausetheyarestoredandused.I followedthis well-troddenroadat first firminthe faiththat the meanings ofproductivelyformedwordswouldturnouttobecompositionalifwecouldevercatchthewordsat theirmomentofbirth. But Ifoundthat roadtooeasyandsoItooktheone less traveled by, and that has made all the difference. This less traveled road followsaMiesianmethod, takingthesurfacestructureof complexwordsat facevalueandreplacing the analogy to sentences with a very sparse nonsentential syntax that mirrorsthis simple structure. The semantics of this sparse syntax for complex words is composi-tional, but only because it accounts for much less of the actual meaning of even newbornwordsthanother semanticsystemsattempt tocover, leavingtherest topragmatics(Aronoff 1980). Thisbifurcatingtreatment restsontheassumptionthat wordshavecomplexmeanings preciselybecauseneither words nor their meanings areentirelyLANGUAGE,VOLUME83,NUMBER4(2007) 814linguistic objects, but rather the bastard offspring of language and the real or imaginedworld; it is this union of sparse linguistic resources with the vastness of the nonlinguisticuniverse that makes all words so rich from birth. Noncompositionality, on this account,follows fromthenatureof words, not fromtheir nurture. Of course, newcomplexwordscannot beentirelyarbitraryinmeaning, because, except forHumptyDumpy,we use them expecting our interlocutors to understand us.11Some are more predictablethanothers,butnewcomplexwordsareneverentirelytheproductoflanguage.As James McClelland has kindly pointed out to me, on this viewthe meanings of wordsmay be compositional just incase the worldthat these words andtheir meanings intersectswith is itself a logical world. A good example is chemical terminology. It is well knownthat this terminological system is compositional, but this follows directly from the factthat thesystemisitselfentirelydeterminedbythescientificprinciplesofchemistry.Indeed, the name of any chemical compound is governed by the principles of the Inter-divisional CommitteeonTerminology, Nomenclature, andSymbols(ICTNS) of theInternational Union of Pure and Applied Chemistry. Not surprisingly, this terminologycanyieldverylongwords,thelongestoneattestedinprintinEnglishbeingthenamefor a strain of tobacco mosaic virus: acetylseryltyrosylserylisoleucylthreonylserylprolylserylglutaminylphenylalanylvalylphenylalanylleucylserylserylvalyltryptophylalanylaspartylprolylisoleucylglutamylleucylleucylasparaginylvalylcysteinylthreonylserylserylleucylglycylasparaginylglutaminylphenylalanylglutaminylthreonylglutaminylglutaminylalanylarginylthreonylthreonylglutaminylvalylglutaminylglutaminylphenylalanylserylglutaminylvalyltryptophyllysylprolylphenylalanylprolylglutaminylserylthreonylvalylarginylphenylalanylprolylglycylaspartylvalyltyrosyllysylvalyltyrosylarginyltyrosylasparaginylalanylvalylleucylaspartylprolylleucylisoleucylthreonylalanylleucylleucylglycylthreonylphenylalanylaspartylthreonylarginylasparaginylarginylisoleucylisoleucylglutamylvalylglutamylasparaginylglutaminylglutaminylserylprolylthreonylthreonylalanylglutamylthreonylleucylaspartylalanylthreonylarginylarginylvalylaspartylaspartylalanylthreonylvalylalanylisoleucylarginylserylalanylasparaginylisoleucylasparaginylleucylvalylasparaginylglutamylleucylvalylarginylglycylthreonylglycylleucyltyrosylasparaginylglutaminylasparaginylthreonylphenylalanylglutamylserylmethionylserylglycylleucylvalyltryptophylthreonylserylalanylprolylalanylserine.Mypointisthatchemicalterminologyiscompositional, notforlinguisticreasons,but becausetheterminologymirrorschemical theory, whichisalogical world. Butwhen the world that the words reflect is not entirely logical (and few are), then neitherarethemeaningsofthewords.To show what motivates this bifurcating treatment of the meanings of words, I returnto one of our trio of Revolutionary words, skinner, the one that appears to have traveledfurthest from its compositional roots to arrive at this peculiar meaning. I do not attempt11Theresgloryforyou!Idontknowwhatyoumeanbyglory, Alicesaid.HumptyDumptysmiledcontemptuously. Ofcourseyoudonttill Itell you. Imeant theresaniceknock-downargumentforyou! Butglorydoesntmeananiceknock-downargument, Aliceobjected.WhenIuseaword,HumptyDumptysaid,inratherascornfultone,itmeansjustwhatIchooseittomeanneithermorenorless.Thequestionis,saidAlice,whetheryoucanmakewordsmeansomanydifferentthings.Thequestionis,saidHumptyDumpty,whichistobemasterthatsall.LewisCarroll,ThroughthelookingglassINTHEBEGINNINGWASTHEWORD 815to elucidate the meaning under consideration, whose origin is entirely opaque, but lookonly at the meaning that seems to be most rooted in language and least tied to language-externalcontext. Itquicklybecomesapparentthateventhatsimpleattemptcomestogrief.A skinner is transparently one who skins and it is easy to derive the most commonmeaningofthenounskinnerfromthatoftheverbskinviajustaboutanysyntacticorsemantictheory. Sofar, sogood, but moveonestepfurtherbacktothemeaningofthe verb and the going gets tough. This verb skin is derived from the noun skin, whoseOldNorseorigintellsusithasbeenaroundforoveramillennium. Howarethetworelated semantically? Here are a few of the twenty-six senses of the verb listed in WIII.Ihaveomittedallthosethatareapparentextensionsofothers.(7) a. Tocover withor asif withskinfuselagesandwingswill beskinnedwithsteel. . .ortitaniumalloysWernervonBraunb. Tohealoverwithskinc. Tostrip,scrape,orrubofftheskin,peel,rind, orotheroutercoveringsof: to remove asurface layer from huge catfish areskinned and dressedbyhandd. To remove (skin or other outer covering) from an object: pull or strip offtoolatetoskinoutthehidethatnightCoreyForde. Tochip, cut, ordamagethesurfaceofskinnedhishandontheroughrockf. Tooutdistanceordefeatinaraceorcontestg. Toequalizethethicknessofadhesiveon(apastedorgluedsurface)byplacingasheetofwastepaperoveritandrapidlyrubbingorpressingh. Tobecomecoveredwithor asif byskintheseinkswont dryinthepress. . .norwilltheyskininthecanusuallyusedwithover.i. To climb or descendused with up or down skinned down inside laddersfromthebridgedeckK.M.Dodgsonj. To pass with scant room to spare: traverse a narrow openingused withthroughorbythebigshipbarelyskinnedthroughtheopendrawRestassuredIdidnotchoosethiswordbecauseofitsmanysenses. Truthfully,Ididnot chooseit at all, andthewholeexcursionthat ledtothecuriousincident of theWestchestersynonymsbeganquitebyaccident, duringthecourseofaconversationabout a group the country later chose to call Hurricane Katrina evacuees, on the groundsthat thetermfirst used, refugee, waspejorative. Suchrichvarietyof sensesissim-plytypicalofzero-deriveddenominalverbsinEnglish(henceforthzero-verbs).ClarkandClark(1979) provideacatalogof themajorcategoriesof sensetypesfor zero-verbs, based ontheir paraphrases, which numberin the dozens. A synopsisis given inTable1.Inthefaceofsuchawealthofsensetypes,thesyntacticallyinclinedexplainerhasnochoice buttodenyall thebestOckhamite tendenciesandresortto amultiplicityoftree-types, eachwiththenounorroot inadifferent configuration. Leeswouldhavedone so, as he did for compounds, and as have more recent scholars, notably Hale andKeyser(1993)andtheirfollowers.12Evenifweignoretheproblemofunrecoverable12Incidentally, anexamplelikeskindemonstrates onits facethewrongheadedness of approaches toconversionpairsbasedonrootsorunderspecification,inwhichthemeanings ofboththeverbandthenounare derived from some third category that isneither a verb nor a noun: the direction of the semantic relationLANGUAGE,VOLUME83,NUMBER4(2007) 816LOCATUMVERBS GOALVERBSa. on/not-on grease a. humanroles widowb. in/not-in spice b. groups gangupc. at,to poison c. masses heapd. around frame d. shapes loope. along hedge e. pieces quarterf. over bridge f. products nestg. through tunnel g. miscellaneous creamh. with trustee wordSOURCEVERBSLOCATIONANDDURATIONVERBS INSTRUMENTVERBSLOCATIONVERBS a. go bicyclea. on/not-on ground b. fasten nailb. in/notin lodge c. clean mopc. at,to dock d. hit hammersummer e. cut,stab harpoonDURATIONVERBSf. destroy grenadeg. catch traph. block dami. follow trackj. musicalinstruments fiddlek. kitchenutensils ladlel. places farmm. bodyparts eyeballn. simpletools wedgeo. complextools millp. miscellaneous ransomAGENTANDEXPERIENCERVERBS MISCELLANEOUSVERBSAGENTVERBS a. meals luncha. occupations butcher b. crops hayb. specialroles referee c. parts wingc. animals parrot d. elements snowwitness e. other house/s/EXPERIENCERVERBSTABLE1.Categoriesofzero-verbsfromClark&Clark1979.deletionthatplaguestheseanalyses, theproliferationofsyntacticandsemantictypesassembledunderasinglemorphologicalconstructionisembarrassing.Buthowcanoneuniteallthesetypesofsenses?Jespersenisprescient,asheoftenis. In the volume on morphology of A modern English grammar on historical principles(1943,vol.6),hesaysthefollowingaboutthemeaningsofcompounds:Compoundsexpressarelationbetweentwoobjectsornotions, but saynothingofthewayinwhichtherelationistobeunderstood. That must beinferredfromthecontext orotherwise. Theoretically,this leaves room fora large number of different interpretationsof one and the same compound. . . Theanalysisofthepossiblesense-relationscanneverbeexhaustive.(13738)Theonlymethodthathasbeenusedsuccessfullytosimultaneouslyavoidandexplainthischoiceamonganinfinityofsensesremovesboththecomplexityandthevariancefrom the linguistic to the pragmatic realm, as Jespersen suggests and as I am advocating.PamelaDowningpioneereditinher1977articleonEnglishnominalcompounds.Asisclearlyoneway,fromnountoverb.Thedifferentsensesoftheverbareallreadilytraceabletothesenseofthe noun,butthere isnoinverse relationforany senseofthe verb,sothat theverbmust bederivedfromthe noun.If we start fromsome third point,we cant accountfor the directionality. Theopposite derivation,fromverbtonoun,holdsforinstanceorresultnounslikehit.INTHEBEGINNINGWASTHEWORD 817Downingputsit, echoingJespersen:Indeed, becauseoftheimportantdifferencesinthe functions servedby compounds, as opposedto the sentence structureswhich moreor less accuratelyparaphrasethem, attempts tocharacterizecompounds as derivedfromalimitedsetofsuchstructurescanonlybeconsideredmisguided.Aparaphraserelationshipneednot implyaderivational one (1977:84041). Sheconcludes thattherearenoconstraintsontheN Ncompoundingprocessitself(1977:841).Theconstraintslieinsteadinhowpeoplecategorizeandrefer.English zero-verbs show a similarly rich and varied set of interpretations, as we haveseen in the case ofskin. Eve and Herb Clark come close to an account like Downingsfor these verbs, but they confine themselves to innovations, which they call contextuals.Inmy responseto theClarks (Aronoff1980), Isuggested thatit ispossible toaccountuniformlyfor all zero-verbs, not just theinnovations, byaconversion(rule) of thesimpleformN V, andthat themeaningoftheinnovativeverbalwayscompriseswhatIcallanEVALUATIVEDOMAINofthenounsdenotation(essentiallyadimensionalong which the denotation of the noun can be evaluated: a knife is good if it cuts well;amotherisgoodifshedoeswellwhatmothersdo;aclubisgoodforclubbing,etc.).For most nouns or other lexical items, there is no fixed evaluative domain, so that whatthe meaning of the novel zero-verb will be depends on the context of its use. Of course,anywordsmeaningwill becomefixedlexicallywithenoughuseandtime, but thatfixing should be of no interest to a linguist. This story holds most remarkably for verbslikeboycottandlynch,whicharederivedfrompropernounsandwhosemeaningsaretraceable to very specific incidents in which the named person, here Boycott or Lynch,playedanimportant role. Thereisnothingelsetosayabout thesemanticsof zero-verbs. Even the notion of an evaluative domain is superfluous, since Gricean principlesdictatethat theverbhavesomethingtodowiththenoun, nomoreandnolessthanwhatneedstobesaidtoaccountfortherangeofdata.Onthisanalysis,then,allthatthegrammarofEnglishcontainsisthenoun-to-verbconversion.Thebeautyof thiskindof account isthat it leavesthewordslargelyuntouched,freeingthemtovaryasmuchasspeakersneedorwantthemto. Criticshaverepliedthat thisvarianceisnobeautyandthat peoplelikemewhoadvocatesuchasparseaccount,fewthoughwemaybe,aresimplyirresponsibleandshouldbepurgedfromthe field. These critics, though, ignore the most important point about Downings con-clusionandmine,whichisthattheyarenottheoreticalbutempirical.Whenwelookattheactualmeaningsthatspeakersattributetonovelcompoundnounsunderexperi-mentalconditions, asDowningdid, wefindthat theyvaryinwaysthatareformlessandvoid. TheconstraintsonN NcompoundsinEnglishcannotbecharacterizedin terms of absolute limitations in the semantic or syntactic structures from which theyarederived (1977:840). Words haveidiosyncraticmeanings not just becausetheyarepreservedinmemoryandsocietybutbecausewordscategorizeandreferoutsidelanguage.IofferonemoreEnglishzero-verbtobuttressmypoint. Theverbisfriend, whichhasbeencitedonandoffforat least acentury, but hasnevercaught onmuchuntilveryrecently. TheonedefinitiongiveninWIIIistoact asthefriendof andWIIIcitesalinefromAShropshirelad(1896):andIwillfriendyou,ifImay,inthedarkandcloudydayA. E. Housman. Inthelast coupleof years, though, friendhasreemerged as a common term among the myriads who use the websites friendster.comandfacebook.com, bothof whichcruciallyinvolveindividual membersmaintaininglists of friends. Though the term seems to have originated with the earlier Friendster,I concentrate on Facebook, which began, by its own definition, as an online directoryLANGUAGE,VOLUME83,NUMBER4(2007) 818that connects people through social networks at school, because I have the best ethno-graphic sources of data for it (three children of the right age and their friends). Facebookwasinauguratedonlyinlate2003, butthecommunityofmembersisquitelargeandactiveandhasbeenthesubjectofarticlesinmajormediaoutlets.TofriendsomeonewithintheFacebookcommunityisnottoactasafriendtothat person, but toinvitesomeonetobeyour Facebookfriend, whichyoudobytryingtoaddthat persontoyour list of Facebook friends. Here is how that works, according to the Facebook helppage(fromlate2005).13Youcaninviteanyonethat youcanseeonthenetworktobeyourfriend. JustusetheSearchpagetofindpeopleyouknowandthenclickontheAddtoFriendsbuttonontherightsideofthescreen.Afriendrequestwillbesenttothatperson.Oncetheyconfirmthattheyactuallyarefriendswithyou,theywillshowupinyourfriendslist.Herearesomeusesoffriendinthissense.14(8) a. Eeew! This guy from my London seminar whos a total ASS just friendedme!b. Alltheserandompeoplefromhighschoolhavebeenfriendingmethisweek!c. ShouldIfriendthatdudeweweretalkingtoat1020lastnight?IdontwanthimtothinkImastalker.The invitation is crucial to the meaning of friendV, because the reciprocity of the friendrelationinFacebookisenforcedbythesystem:yourinviteeisnotaddedtoyourlistof friends unless he or she accepts the invitation. Oddly, if the invitee does not accept,the[inviter]willnotbenotified.Theyalsowillnotbeabletosendyou[theinvitee]anotherfriendrequestforsome amountoftime,sotothem,itwill justseemasifyouhavent confirmedtheir friendshipyet. ThecreatorsofFacebookmayseethisasapolitemethodofrejection. Ifindit odd, but Iamnot twentyyearsold, soIguessIjustdontunderstand.Thepoint ofallthisisthat themeaningoffriendxcannot justbeactasafriendofx, asit clearlyisintheHousmancitation(notetheuseofthemodalmaythere),but theinvitation, whichwemaythinkofasconativeaspect, must bebuilt intothemeaning. Tofriendxinthisworldistotrytobecomeafriendofxortoaskxtobeonesfriend, inthespecialsenseofthenounfriendthatappliesinthisworld.Ofcourse,weknowwhytheextrapredicateispartofthemeaningoftheverb:itisbuiltintotheprogramandthereforeintothesocialcommunitythatsurroundstheprogram!One could try to amend the Clarks list or its syntactic equivalent to allowfor multipredi-cateor aspectual meanings, but that wouldavoidtheunderlyingproblem, that themeaningdependsonextralinguisticfactorsfromtheveryfirstmomentofthecoiningoftheword,whichispreciselyDowningsempiricalpoint.154. LEXICAL ROOTS. The lexeme-centric view of word meaning, where lexical-seman-ticinformationisatleastpartlynonlinguisticanddoesnotresideinmorphemes, hasimplications for the meanings of roots. If I am right about how lexical semantics works,13TheverbfriendneverappearsinofficialFacebooktext,onlyamongusers.14TheseexamplesweresuppliedbymydaughterKaty.15I am not denying that noncausative individual verbs can have multipredicate meaning. The verb proposeto make an offer of marriage (WIII) is similar to friend in involving an invitation. The question is whetherthere is a systematic multipredicate syntactic relation between nouns and their derivative zero-verbs involvingconativeorsimilaraspect.INTHEBEGINNINGWASTHEWORD 819thentwowordscanhavethesamerootandnotsharemuchlexicalmeaning.Thereislittlerelationinmeaning,forexample,betweenthenameBorkandtheverbbork,yettheyareascloselyrelatedmorphologicallyasanytwowordscanbeandmust sharearoot,ifthetermROOTistohaveanycontent.Whatdoesonesaythenaboutlexicalroots, which are supposed to be the atomic meaningful units of language, if two instancesof the same root can share so little meaning? The simplest ploy is to deny the linguisticrealityofrootsentirely, what mycolleagueRobert Hobermancallstheantirootarianposition. But thereis amiddleground, wherewords havemorphological structureeven whenthey are not compositionallyderived, and whereroots are morphologicallyimportantentities,thoughnotparticularlycharacterizedbylexicalmeaning.4.1. LATINVERBROOTS. In the first piece of morphological analysis I ever did (pub-lishedinAronoff 1976), I showedthat certainLatin-derivedrootsinEnglishverbscanbeactivemorphologicallyevenwhentheyareobviouslymeaningless: eachrootconditionsacertainset ofaffixesandalternationsthat ispeculiartothat root alone.SomeexamplesaregiveninTable2.WORD-FINALROOT SAMPLEVERB ROOT ion ROOT ivesume resume resumption resumptivemit permit permission permissivepel repel repulsion repulsiveceive receive reception receptiveduce deduce deduction deductivescribe prescribe prescription prescriptivepete/peat compete competition competitivecur recur recursion recursiveTABLE2.SomeLatinaterootsinEnglishwiththeiralternations.These Latinate roots in English are the historical reflexes, through a complex borrowingprocess, of the corresponding verb roots in Latin. Even for Classical Latin, the alterna-tions exhibited by at least some of these verb roots were not phonologically motivated.ButinLatin,thesesamerootsweremorphologicallyactiveinotherwaystoothatareindependent of syntaxor semantics. Perhapsmost intriguingistheroleof rootsindeterminingwhetheragivenverbwasdeponent.Thetraditional Latingrammatical termDEPONENTis thepresent participleof theLatinverbdeponereset aside. Deponent verbs haveset asidetheir normal activeforms and instead use the corresponding passive forms (except for the present participle)in active syntactic contexts. The verb admetior measure out, for example, is transitiveand takes an accusative object, but is passive in form, because it is deponent. Similarlyfor obliviscor forget (oblitus sumomniaI haveforgotteneverything [Plautus]),scrutor examine, and several hundred other Latin verbs. There are even semideponentverbs,whicharedeponentonlyinformsbasedontheperfectstem.For some time, Xu Zheng, Frank Anshen, and I have been working on a comprehen-sivestudyof Latindeponent verbs, usingadatabaseofall themain-entrydeponentverbs intheOxfordLatindictionary, whichcovers theperiodfromthefirst LatinwritingsthroughtheendofthesecondcenturyAD(Xuetal. 2007). Ireporthereononlyonesmallpartofourresearch,thatinvolvingdeponentroots.WeshowinXuetal.2007,basedonanexhaustiveanalysisofallthesensesofalldeponentverbs, thatneithersyntaxnorsemanticsisthebestpredictorofdeponency,thoughbotharefactors. Ourdatabasecontains287deponent verbsnotderivedfromanother lexical category(about half thetotal number of deponent verbs), thegreatLANGUAGE,VOLUME83,NUMBER4(2007) 820majorityof themconsistingof either abareroot or aroot andasingleprefix. Thesenses of all deponent verbs are about evenly divided between transitive and intransitivevaluesandagivenrootmayappearinbothtransitiveandintransitiveverbs.Ofthese287verbsnotderivedfromanotherlexicalcategory,85percent(244)havedeponentroots, whichwedefineasaverbroot that occursonlyindeponent verbs. Therearefifty-twodeponent roots, twenty-twoof whichoccur infour or moredistinct verbs.Table3isalistoftheseroots.Thus, whetheragivenverbisdeponentisdeterminedtoagreat extent byitsroot, independent of themeaningof either. Inother words,besides thealternations that havesurvivedinEnglishandtheRomancelanguages,deponent Latinverbrootshaveat least oneother important morphological property(theirmeredeponency)thatisindependentofeithersyntaxorsemantics.LATINDEPONENT NO. OFVERBS EXAMPLEROOT WITHSUCHROOT VERB GLOSSgradi 22 gradior proceedla:b 16 la:bor glidesequ 15 sequor followmin 11 comminor threatenloqu 10 loquor talknasc 10 nascor bebornmo:li 9 mo:lior buildupmori 9 morior dieori 8 orior risetue 8 tueor lookatf(or) 7 affor addresshor(t) 7 hortor encourageluct 7 luctor wrestleme:ti 7 me:tior measurefate 5 fateor concedeprec 5 precor askforquer 5 queror regretapisc 4 apiscor graspfru 4 fruor enjoyfung 4 fungor performmedit 4 meditor contemplatespic 4 conspicor seeu:t 4 u:tor makeuseofTABLE3.Latindeponentverbroots.4.2. HEBREWROOTS. I now turn to the role of the root in the morphology of ModernHebrew verbs.16For more thana millennium, traditional grammarianshave character-ized Semitic languages as being morphologically grounded in meaningfulROOTS, eachroot consisting usually of three consonants, with some number of lexemes (verbs, nouns,andadjectives)builtoneachroot.ButnoteveryModernHebrewroothasaconstantmeaning.17Tables4and5eachcontainasingleroot withanapparentlyconstant meaning.18Themeanings of the16IsayMODERNHEBREWbecausethetranscriptioninthisarticleisofModernHebrew, but infact thepatternsIdiscussarelargelyidenticalintheearliestattestedformsofthelanguage.IuseBIBLICALHEBREWtorefer tothelanguageof theunpointedtext of theBible, datingbetweenabout 800and100BC. I useMASORETICHEBREWtorefertothepointedtextoftheBible,datingfromabout600AD.17Earlier stages of Hebrew or related Semitic languages do not show much more regularity in this regard.18Eventheseexamplesarealittleforced,sincesomeofthelexicalitemsareveryinfrequent, obsolete,orevenartificial.Inthe earlydaysofModernHebrew,linguisticzealotscoined numerouswordsthatneverINTHEBEGINNINGWASTHEWORD 821ROOT ZLP sprinkle,spray,dripNOUNSzelef sprinklingfluid(perfume)zalfan sprinklermazlef wateringcan,sprayerhazlafa sprinklingziluf sprinklingzlifa sprinklingVERBSzalaf topour,spray,sprinklezilef todriphizlif tosprinklezulaf tobesprinkledhuzlaf tobesprinkledADJECTIVEmezulaf sprinkledTABLE4.LexemesformedontheModernHebrewroot ZLP.ROOT SKR pay,rent,leaseNOUNSsaxar hire,wages,profitsoxer tenantsaxir hiredlaborer,employeesxirut rentsxira hiring,rentinghaskara leasehistakrut wages,profit,rentingmaskir landlordmaskoret salaryVERBSsaxar tohire,rentniskar tobehired,rewarded,paidhistaker toearnwages,makeprofithisker tolease,lethuskar tobeleasedADJECTIVESsaxur rentedmusxar let,leasedniskar hired,benefitedTABLE5.LexemesformedontheModernHebrewroot SKR.individual nouns, verbs, andadjectives inthesetables canbepredictedreasonablywellifeachmorphologicalpatternisassociatedwithitsowncompositionalsemanticfunction. This is what we expect from the traditional account. Table 6, though, quicklybrings such an enterprise to grief. What do pickles (kvu+im) have to do with highways(kvi+im)?Thestoryofhowthesetwowordsarerelatedisquitesimple. PavedroadsinearlymodernPalestineweremacadamized: madeof layersof compactedbrokenstone bound together with tar or asphalt. Modern roads are still built by pressing layers,butina moresophisticatedmanner.Traditionalpicklingalso involvespressing:what-everistobepickledisimmersedinbrineandpresseddownwithaweight, but thecaughton;manyofthesesurviveindictionaries, andonlyasensitivenativespeakercandistinguishthemfromwordsineverydayuse.LANGUAGE,VOLUME83,NUMBER4(2007) 822container should not be sealed. If it is, it may explode (as I know from personal experi-ence). But pressing alone without brine does not constitute pickling. Every cook worthhisorhersaltknowsthatbeefbrisketshouldbepressedafteritiscookedandbeforeit is sliced, regardless of whether it is pickled (corned) or not. Modern industrial picklingdoesawaywithpressinginvariouswaysandnoteverypavedroadinIsraeliscalledakvi+,only ahighwayis,so whileitisclear whypicklesandhighways bothoriginateinpressing, thereisnothingleft ofpressinginthemeaningsoftheseHebrewwordstoday. One who points out their related etymologies to a native speaker of the languageisrewarded withthetolerant smilereservedfor pedants,asI alsoknowfrom personalexperience.ROOT KB+ pressNOUNSkeve+ gangway,step,degree,pickledfruitkvi+ pavedroad,highwaykvi+a compressionkiv+on furnace,kilnmaxbe+ press,roadrollermixba+a picklingshopVERBSkava+ toconquer,subdue,press,pave,pickle,preserve,store,hidekibe+ toconquer,subdue,press,pave,pickle,preservehixbi+ tosubdue,subjugateADJECTIVESkavu+ subdued,conquered,preserved,pressed,pavedkvu+im conserves,preserves,picklesmexuba+ pressed,fullTABLE6.LexemesformedontheModernHebrewroot KB+.The reflex response to problems like this is to posit an underspecified core meaningfor the root, which is supplemented idiosyncratically in each lexical entry. It is logicallyimpossibletoshowthat underspecificationis wrong, but tryingtofindacommonmeaningsharedbypicklesandhighwaysbringsoneclosetoempiricalemptinessandthismethodological danger recursfrequentlyinanySemiticlanguage. Inanycase,there is no need to find a common meaning in order to relate the two words morphologi-cally, as I show. In fact, Hebrew verb roots can be identified on the basis of alternationclassesremarkablysimilarintypetothosethat operateonLatinaterootsinEnglish,withoutreferencetomeaningorregularphonology.Thisobservationisnothingnew;thesealternationclassesarethebaneofHebrewstudents.19First, onemustknowafewbasicsofHebrewverbmorphology. HebrewhasasetofwhataretraditionallycalledBINYANIM,seveninnumber, oftencalledconjugationsintherecenttheoreticalliterature.Eachbinyanrigidlyassignstoeachcelloftheverbparadigmastempatterncontainingaprosodicshapecompletewithvowels. Astempatternmayalsoincludeaprefix. Thehifil binyan, forexample, hasaprefixhi-orha-, andthepatternCCiC(e.g. higdilgrew). ThepielbinyanhasthestempatternCiCeCinthepast, meCaCeCinthepresent, andCaCeCinthefuture(e.g. megadelgrow). Traditionally,each binyan(like eachroot) issaid tohave aconstant meaning19Iconfinemydiscussiontoverbs, bothbecausetherearefewerverbpatternsthannounpatternsandbecauseit iseasiertobelievethat verbrootsarerelatedsemantically, soifwecantalkabout verbrootswithout recoursetomeaning, thenall themoresofornounsandadjectives, whosemeaningsshowmuchgreatervariety.INTHEBEGINNINGWASTHEWORD 823orsyntacticstructure.Somemodernscholarshavequestionedthatclaim,butwhetherbinyanimhavemeaningisorthogonaltothematterathand,whichisroots.Besides being determined by binyanim, verb forms depend on the alternation class oftheir root. Root alternation classes, except for the (default) regular class, are traditionallyidentified in terms of one of the consonants that occupy the three canonical root conso-nantpositions,whichIcallR1,R2,andR3.20ThetermFULLisusedforregularrootsbecausetheydonot alternateexcept formoregeneral phonological phenomenathatarenotroot-dependent,likespirantization(thealternationofstopsandfricatives).Forexample, theroot alternation classR3h is sonamed because thethird consonant oftherootish;inarootofalternationclassR3nthethirdconsonantisn.Thesealternationclasses,namedinthisway,figureprominentlyintraditionalHebrewgrammar.At somepoint intheearlyhistoryof thelanguage, thedifferencesinverbformsamongroot alternationclasseswereundoubtedlypredictablefromthephonologyofthe consonant that defined each type. R1n verbs, which I discuss in some detail below,patternthewaytheydobecauseof thephonologyof ninsomeveryearlystageofHebrew(longbeforetheoriginal consonantal text oftheBibleoranyotherexistingHebrewtext was written down). The passage of time, however, made these phonologicalconditionsopaque, sothat twoverbswithidentical consonantsinthesamepositioncouldshowdifferent alternationpatternseveninBiblical Hebrew. Thisdifferentialpatterning proves that the alternation classes were no longer predictable phonologically,even in Biblical Hebrew, but are more like verb classes of the sort found, for instance,inGermaniclanguages, wheretheclassesaredefinedbydistinctionsinablaut andverbsmustbemarkedformembershipinagivenSTRONGclass.21InHebrew, similarsorts of idiosyncratic alternations to those of Germanic are determined by the alternationclassofaverbsroot (whichalsoappearstobetrueinGermanic), afact that givesrootstheirlinguisticreality.TherootalternationclassesandbinyanimofSemiticconstitutetwodimensionsofa four-dimensional matrix, the others of which are person/number/gender combination(tenarepossible)andtense(inModernHebrew:past,present,future,andimperative,though the last is becoming increasingly rare). All the possible regularities of any givenindividual verb form are thus exhausted by four specifications: its root class, its binyan,itsperson/number/gender value, anditstense. Lest it bethought that thismatrixistrivial, areasonablycompletetraditional tableof verbs(Tarmon&Uval 1991)lists235 distinct combinations of root class and binyan alone, what they call verb types, ofwhichonlyninecontainjustasingleroot.Multiplying235bythetenperson/number/gendervaluesinall ofthetenses(atotal oftwenty-six)yieldsalittlemorethansixthousand distinct form types into which any verb form in the language must fall.22Notall of these six thousand types must be learned individually, only what alternation classaverbrootbelongsto(ifitisnotregular)andwhatbinyanimarootmayoccurin.In this article, I look only at the roots whose R1 is a coronal sonorant: R1n, R1y, andR1l. Mytaskistoidentifythemarkedalternationclassesthat suchrootsbelongto,thosealternationclasseswhosememberrootsmustbelisted(weedingout, alongthe20Therearerootswithmorethanthreeconsonants,butthesearenotgermanetothisconversation.21NoteventhemostabstractofphonologicalrepresentationscouldsavetheHebrewsystemfrombeingmorphologized.Inthecaseofn,forinstance,onewouldhavetopositn1andn2,withoutanyphoneticwaytodistinguishthembeyondthefactthattheypatterndifferently.22Finkel andStump(2007) adopt asimilar approachintheir computational approachtoHebrewverbforms.LANGUAGE,VOLUME83,NUMBER4(2007) 824way, thosecoronal sonorant-initial rootsthat donot belongtoanymarkedclassandare hence regular). Only the class of full roots is regular. It constitutes the default classfor Hebrewroots. Andjust likedefault classesinotherlanguages, it includesthoserootsthat exceptionallydonot patternwithamarkedclass, althoughtheymeet thecriteriaformembershipinthatclass(Aronoff1994).Forexample,therearetwomor-phologically distinct sets of R1n roots and they cannot be distinguished by their phono-logical makeup, which means that the R1n roots in one set must be marked as belongingtoamarkedalternationclass.Becauseonesetpatternsexactlylikeregularfullverbs,the other set constitutes the marked alternation class. Any root that belongs to a markedclassalternationmust beflaggedassuch, inpreciselythesamewaythat averbisflaggedforitsconjugationinaclassicalIndo-Europeanlanguage.To a great extent, root alternation classes resemble the conjugation classes of classicalIndo-Europeanlanguages morethanbinyanimdo, eventhoughtheoretical linguistsusually call the latter conjugations. One might say that there are two dimensions to theconjugationofaSemiticverb,therootclassandthebinyan.ThisiscertainlyclosetothetraditionalviewofHebrewgrammar.Todemonstratethevalidityofthisclaim,Ishow firstthattherearetwosetsofR1nrootsandthat onesetbehavesjust likeregularroots, makingtheotherset amarkedclassofroots.IthenshowthattherearethreesetsofR1yroots,oneofwhichpatternsidenticallytothemarkedR1nalternationclass, leadingtotheconclusionthattheso-called R1n alternation class contains verbs with both initial n and y (and one with initiall), notnewstotraditionalgrammarians. Inotherwords, thealternationclassofrootsthatincludesmarkedR1nrootsalsoincludesrootswhoseR1isnotn.Thisalternationclass has thus become morphological rather than phonological. One might wish to givethealternationclassformerlyknownasR1namoregenerallabel:theclassofmarkedR1coronalsonorantroots,buttheexistenceofadistinctclassofR1yrootsmakesthatlabel unenlightening. The main point is that the class has become morphological, evenbytheearliestattestationofBiblicalHebrew.The two morphologically distinct sets of R1n roots are those in which the n is deletedbetween the vowel of a prefix and R2, which are traditionally termedMISSING R1n roots,andthoseinwhichthenisnot deleted. Thelatter patternexactlyasdocompletelyregular roots, andthusareactuallyfull roots, not membersof amarkedalternationclass,exactlyaspredictedgenerallyforexceptionstomarkedclasses(Aronoff1994).Originally, the difference in patterning was phonologically motivated: an n went missingbecause it was assimilated to a following nonguttural R2 consonant, which was therebygeminatedand, if astop, consequentlyexempt fromspirantization(realizationasafricative).23The root npl, for example, when prefixed with the first-person plural imper-fect prefixni-, wouldloseitsinitial n, sothat theactual Masoreticformwasnippolwe(will)fall.24Inrootswithguttural (pharyngeal orlaryngeal)R2, thenwasnotassimilated, being somehow protected by the guttural. The root nhg, for example, whenprefixedwiththefirst-personplural imperfect prefixni-, wouldnotloseitsinitial n,so that the actual Masoretic and Modern form is ninhag we (will) drive. The assimila-tion of n has been unproductive in recorded history, however, so that, in a good numberof roots with a nonguttural R2, the R1n remains intact even in the assimilation context.WhetherR1nismissingisapropertyofindividual verbroots, not lexemes: npl isa23Therearenolongeranygeminateconsonantsinthelanguage, but formergeminatesarestill exemptfromspirantization.24InModernHebrew,thecorrespondingformisnipol.INTHEBEGINNINGWASTHEWORD 825missingR1nroot;ngdisnot.25Thisistheprincipalempiricalanalyticalevidenceforthe linguisticreality of Hebrew roots.We even findetymologically distinct homopho-nous roots, one of which is a missing R1n root, and the other not. Compare hibit lookwithhinbitsprout.26Importantly, individuallexemesbelongingtothesamerootdonot vary. Examples of missingR1nroots aregiveninTable7; full R1nroots areexemplifiedinTable8.Individualverbformsinwhichthenismissingareitalicized;thoseinwhichitisphonologicallyeligibletobemissingbutisnotareboldfaced.ROOT BINYAN PAST FUTURE INFINITIVE GLOSS REMARKSntp paal nataf yi-tof li-tof drip(intr.) ndropsbeforeR2(t)ntp hifil hi-tif ya-tif le-ha-tif drip(tr.)ncl nifal ni-cal*(cf. yi-nacel le-hi-nacel arrive *wouldbenincalifnwerenixtav) notassimilatednpl paal nafal yi-pol li-pol fallnpl hifil hi-pil ya-pil le-ha-pil causetofallnpl hufal hu-pal ju-pal NA bethrowndownnbt hifil hi-bit* ya-bit* le-ha-bit* look *bwouldbevifnwerenotassimilatedntn paal natan yi-ten la-tet* give *infinitiveisirregular,shouldbelitenntn nifal nitan*(cf. yi-naten le-hi-naten begiven *wouldbenintanifnwerenixtav) notassimilatedlkx paal lakax yi-kax la-kaxat take onlyexampleofinitiallthatbehaveslikemissingR1nnpc paal nafac yi-poc linpoc disperse,smash ndoesnotassimilateininfinitivenpk paal nafak yi-pok linpok go/comeout,result ndoesnotassimilateininfinitivengd hifil hi-gid ya-gid le-ha-gid tellTABLE7.MissingR1nroots.NOTE:individualverbformsinwhichthenismissingareitalicized;thoseinwhichitisphonologicallyeligibletobemissingbutisnotareboldfaced.ROOT BINYAN PAST FUTURE INFINITIVE GLOSS REMARKSnpk hifil hi-npik ya-npik le-ha-npik issuengd paal nagad yi-ngod li-ngod opposenbt hifil hi-nbit ya-nbit le-ha-nbit sproutngb paal nagav yi-ngov, li-ngov becomedry ndropsvariablyyigov inthefuturengb piel nigev ye-nagev le-nagev dryngb hifil hi-ngiv ya-ngiv le-ha-ngiv gosouthncx hifil hi-nts ax ya-nts ax le-ha-nts ax memorialize,eternalizeTABLE8.FullR1nroots.NOTE:individualverbformsinwhichthenismissingareitalicized;thoseinwhichitisphonologicallyeligibletobemissingbutisnotareboldfaced.TherearethreesetsofR1yroots,distinguishedbytheirhifilbinyanforms.(i) In the most numerous set, the vowel of the hifil prefix is o instead of i throughoutthehifilparadigm.Thisprefixvoweloccursonlywiththisrootalternationclassand25Theroot ngdisancient, but itsuseinverbsisnot, whichshowsthat whatexemptsaverbroot fromassimilationofnisnottheageoftheroot,butratherthepointatwhichitwasfirstusedasaverb.26Thefailureofbtospirantizetovinthefirstexampleshowsthatthereisamissingnbeforethebintherootofhibit.LANGUAGE,VOLUME83,NUMBER4(2007) 826onlyinthehifilbinyan). Also, theyismissing. Becauseitismostnumerous, IcallthisthemissingR1yalternationclass.(ii) Inthreeroots, thehifil paradigmisidentical tothemissingR1nroot classparadigm throughout.27These are clearly members of the missing R1n root class, eventhoughtheirR1isyandnotn.(iii) Infourroots,theyisneverdeletedandthehifilprefixvowelise.Table9comparesthethreevarietiesofR1yrootswithR1nroots.SAMPLEROOT ROOTVARIETY PAST3M.SG FUTURE3M.SG GLOSSnmx fullR1n hi-nmix ya-nmix lower,humiliatenzq missingR1n hi-ziq ya-ziq damageyld missingR1ydefault, ho-lid yo-lid begetprefixvowel oycg(only3roots) missingR1y,like hi-cig ya-cig show,exhibit,presentmissingR1nynq(only4roots) fullR1y,prefix he-yniq ye-yniq suckle,water,saturatevowel eTABLE9.R1nrootsandR1yrootsinhifil.ThefourR1yrootsinwhichtheyneverdeletesinhifilarebestviewedasfullroots,parallel to full R1n roots (with the prefix vowel additionally becoming e for phonologicalreasons). Oneoftheseroots(ynq)andafewothersappearinthepaal binyanwiththeyintactandshowthenormalivocalismoffullrootsinpaal(e.g.yiynaq,yiyra,yiy+an),lendingfurthersupporttotheirclassificationasfullroots.28The missing R1y roots that follow the local default o vocalism in hifil and also occurinthepaalbinyanshowtwotraitsintheirpaalthatfurthercharacterizethismissingR1y alternation class: unlike all other root alternation classes, they have an e: vocalismthroughoutthefutureandtheydropsinboththefutureandtheimperative.To summarize, I have shown that there are two marked alternation classes of coronal-sonorant-initial verbrootsinHebrewandthat, thoughthesearetraditionallylabeledR1n and R1y, that characterization is not entirely correct: the R1n class contains severalR1yrootsandoneR1lroot;theR1yclassdoesnotcontainallmarkedR1yroots;andtherearebothR1nandR1yrootsthatpatternlikeregularroots.Therearemanyotherrootalternationclasses, somemorecomplexthanthese, but,ratherthanconfusenon-Semitistreadersevenfurther,Isimplyaskthemtoacceptontrust my assurance that the analysis I have given here for coronal sonorant roots extendsto the remaining marked alternation classes as well. Every root in every marked alterna-tion class must be flagged or marked individually for membership in its particular class(andnoroot canbelongtomorethanoneclass), muchinthewaythat membersofverbconjugationclassesareflaggedinotherlanguages(Aronoff1994). Mostimpor-tantly, thefact that thesealternationclassesinHebrewoperateoverindividual rootsand not lexemes or stems demonstrates the linguistic reality of Hebrew verb roots, even27All threeR1yrootsinwhichthehifil formsfollowthemissingR1npatternarealsoR2c. TwootherR1yR2crootsdonot occur inhifil, but their paal formsfollowthemissingR1npattern(ycqandycr),supportingtheobservationthat R1yR2crootsfollowthemissingR1npatterngenerally. However, theoneother root of this form, yc, falls into the local default R1y class in hifil (hoci, yoci). This sort of distribution,inwhichacircumscribedexceptiontoanexceptionfollowsalocaldefaultratherthanageneraldefault,isnot uncommon(Fraser &Corbett 1997), andsupportstheassignment of thesefiveR1yR2croots tothemissingR1nclassthroughaverylocalphonologicalgeneralization.28Theserootsalsosharethelesscommonrootvocalismainthefutureandimperativestem.INTHEBEGINNINGWASTHEWORD 827thoughtheserootshavenoconstantmeaning.Intheend,HebrewrootsaremuchlikeLatin roots. Their patterns are much more complex but they are also much more perva-sive: no Hebrew verb form can exist outside the matrix determined by the interplay ofroot alternationclassandbinyan. Eventheformsof averbderivedfromaforeignborrowingmustfallintooneofthecellsdeterminedbythematrix, resultinginsuchcelebratedformsastilfenhetelephoned or hizdangef hestrolleddownDizengoffStreet. Theformof aModernHebrewverbformiscompletelypredictable, thoughlargelyunmotivatedbyeithersyntaxorsemantics.There is one more kind of pattern that distinguishes among the root alternation classesunder discussion, their distributionamongthebinyanim. Althoughthere aresevenbinyanim, veryfewrootsoccurinall seven. Table10isbasedonatabulationofallthetypesofverbrootsdiscussedsofar,exceptfortheR1nrootsthatpatternlikefullroots.Itincludesallrootsofthesealternationclassesthatoccurineitherthehifilornifal binyanandcomparestheir distributioninother binyanim. What isespeciallyinteresting is the distribution of hifil/hufal or hifil/nifal pairs. Both hufal and nifalcanbe passiveinmeaning,and agivenhifil verb,whichisalways activeandusuallytransitive, isusuallypairedwithoneortheother, thoughhufalisthemorecommonpairing, sinceit isthestructurallyappropriatepassiveformforhifil. It isthereforeremarkablethatthetwolargestalternationclassesunderdiscussionhere,missingR1nrootsandmissingR1yroots,differmarkedlyfromoneanother:themissingR1nclassfavors hufal overwhelmingly, as expected of all roots, but the missing R1y class favorsneitherhufalnornifal,whichisveryunusual.OftheremainingtwovarietiesofR1yroots, the one that I have grouped together with missing R1n roots has the same distribu-tionasthatclass,whiletheonethatIhaveclassifiedasfullshowsnohufalornifalformsatall,againverycurious. Theoveralldistributionclearlysupportsthegroupingof the root varieties into traditional alternation classes that I have suggested. Addition-ally, it suggests another intriguing path to explore, which I do not follow here, for lackofspace: thepossibilitythat root alternationclassesdiffergenerallyinthewaytheydistributethemselvesamongthebinyanim.R1yROOTSLIKEMISSINGOCCURRINGBINYAN MISSINGR1n R1yDEFAULT R1n R1yFULLCOMBINATIONS ROOTS ROOTS ROOTS ROOTShifil/hufal/nifal 4 2 0 0hifil/hufal 19 7 3 0hifil/nifal 1 8 0 0hufal/nifal 1 0 0 0hifilonly 5 7 0 4nifalonly 9 5 0 0TOTAL 39 29 3 4TABLE10.Roottypeandbinyandistribution(R1n/y).There are no other missing R1 sonorant roots in the language besides the ones I havediscussed. IhavegroupedtogetherinasinglealternationclassthemissingR1nrootsandthemissingR1yrootsthatfollowthesamepatternasmissingR1nroots.ThereisalsoonemissingR1lroot(lqx).Iconcludeasfollows:(i) There is an alternation class of missing R1coronal sonorant roots whose R1rangesoverallcoronalsonorants, [n, l, y], allofwhosemembersmustbelexicallymarked,whichincludesthetraditionalmissingR1nalternationclass.(ii) AllremainingR1nroots,thosethatdonotdropthen,havenolexicalmarks.LANGUAGE,VOLUME83,NUMBER4(2007) 828WeareleftwithtwovarietiesofR1yroots, thosethatwemaynowcalltruemissingR1yroots, andthosethat behavemuchlikefull roots. Ifweagreethat nofull rootsshouldbelexicallymarked, then(true)missingR1yrootsmust bemarkedlexically.We now have two marked alternation classes of missing sonorant-initial roots: missingR1yandmissingR1coronalsonorant,eachcharacterizedbyasetofformsdistributedacrosstheparadigm.NothingofwhatIhaveshowndependsontherootsthatbelongto these alternation classes having any constant meaning. Indeed, the important generali-zationsarenotaboutindividualrootsbutratheraboutrootalternationclasses. Theseform a system of conjugation alternation classes and, like all good conjugation systems,have no discernible motivation of any sort but are purely arbitrary grammatical categor-ies.Exceptthatthemarkedrootalternationclasseseachcontainasubsetoftherootsthat beginincoronal sonorants, nopart ofthissystemcanbereducedtophonologyeither.Instead,rootalternationclassesareidentifiedbytheirparadigms.29Thisclaimshouldalsobeamenabletoexperimentalverification.ModernHebrewverbrootsandtheir alternation classes, like those of English and Latin, thus furnish yet another exam-ple ofmorphology by itself,leaving lexical meaningto reside whereit belongs, notinroots, which are purely grammatical objects, but in lexemes, where language meets theworld.5. CONCLUSION. Mygoalinthisarticlehasbeentoshowthat,thoughitisdifficulttodolinguisticswhilerespectingtheintegrityofwords, imposingsucharestrictionon ones analyses can lead to interesting results that are otherwise unavailable or unex-pected, not just inmorphologybut alsoinsyntaxandlexical semantics. Fromthatgeneral vantage point,we expect that alanguage newly arisen with littleor no outsideinfluencewill havearobust syntaxandnosublexicalstructureat all;weexpect thatnewlycoinedwordswillhaveidiosyncraticmeaningsthatareunderstandableonlyinthe extralinguistic context of their creation and whose internal semantic structure is notreadilyorentirelyamenabletosyntacticallyderivedanalysis;andweexpecttofindalanguagewithaverycomplexmatrixofinflectedword-formswhosemorphologicalstructureisunmotivatedinanystrictSaussureansign-basedsense.Ihopethatatleastsome small fraction of my readers nowbelieve that the kind of work that I have reviewedhere is not only possible but also necessary if we are to make progress in understandinglanguage.REFERENCESACKERMAN, FARRELL, and ROBERT MALOUF. 2007. Parts and wholes: Patterns of relatednessincomplexmorphological systemsandwhytheymatter. Analogyandgrammar, ed.byJamesBlevinsandJulietteBlevins.Oxford:OxfordUniversityPress,toappear.ALBRIGHT, ADAM. 2002. The identification of bases in morphological paradigms. Los Ange-les:UniversityofCalifornia,LosAngelesdissertation.ANDERSON, STEPHEN R. 1992. A-morphous morphology. Cambridge: Cambridge UniversityPress.ARONOFF, MARK. 1972. The semantics of growth. Quarterly Progress Report of the ResearchLaboratoryofElectronics, MIT113.15962.ARONOFF, MARK. 1976. Wordformationingenerativegrammar. Cambridge, MA: MITPress.29Therehasbeenagooddeal of recent workinwhichmorphologyisviewedprimarilyasacomplexsystem of word-forms (e.g. Albright 2002, Harris 2004, Blevins 2006, Gurevich 2006, Ackerman & Malouf2007,Finkel&Stump2007).Thisotherworkdoesnotdealspecificallywiththemorphologyofrootsandthe theoretical perspectives are by no means uniform, but none of them is morpheme-based in the traditionalsense.INTHEBEGINNINGWASTHEWORD 829ARONOFF,MARK.1980.Contextuals. Language56.74458.ARONOFF, MARK. 1994. Morphologybyitself:Stemsandinflectional classes. Cambridge,MA:MITPress.ARONOFF,MARK,andKIRSTENFUDEMAN.2005. Whatismorphology?Oxford:Blackwell.ARONOFF, MARK; IRITMEIR; CAROLPADDEN;andWENDYSANDLER. 2005. Morphologicaluniversalsandthesignlanguagetype. YearbookofMorphology2004.1939.ARONOFF, MARK; IRITMEIR; andWENDYSANDLER. 2005. Theparadoxof signlanguagemorphology. Language81.30144.BLEVINS,JAMESP.2006.Word-basedmorphology. JournalofLinguistics42.53173.BLOOM, PAUL. 2000. Howchildrenlearnthemeaningsof words. Cambridge, MA: MITPress.BOOIJ,GEERT.2005.Anintroductiontolinguisticmorphology.Oxford:OxfordUniversityPress.CARSTAIRS-MCCARTHY, ANDREW. 2002. An introduction to English morphology. Edinburgh:EdinburghUniversityPress.CARSTAIRS-MCCARTHY,ANDREW.2005.Basicterminology. Handbookofword-formation,ed.byPavolStekauerandRochelleLieber,523.Dordrecht:Springer.CHOMSKY,NOAM.1957. Syntacticstructures.(Janualinguarum4.)TheHague:Mouton.CHOMSKY,NOAM.1970. Remarksonnominalization. Readingsintransformationalgram-mar, ed. byRoderickA. JacobsandPeter S. Rosenbaum, 184221. Waltham, MA:GinnandCo.CHOMSKY,NOAM.1995. Theminimalistprogram.Cambridge,MA:MITPress.CHOMSKY, NOAM. 2000. Newhorizons inthestudyof languageandmind. Cambridge:CambridgeUniversityPress.CLARK, EVEV., andHERBERTH. CLARK. 1979. Whennounssurfaceasverbs. Language55.767811.CORBIN, DANIELLE. 1984. La formation des mots: Structures et interpretations. Lille: PressesUniversitairesdeLille.CROFT, WILLIAM. 2003. Typology and universals. Cambridge: Cambridge University Press.DESAUSSURE,FERDINAND.1975. Coursdelinguistiquegenerale.Publie parCharlesBallyetAlbertSechehaye,aveclacollaborationdeAlbertRiedlinger;editioncritiquepre-pareeparTulliodeMauro.Paris:Payot.DOWNING,PAMELA.1977.OnthecreationanduseofEnglishcompoundnouns.Language53.81042.ECO,UMBERTO.1995. Thesearchfortheperfectlanguage.Oxford:Blackwell.EMMOREY,KAREN.2003.Perspectivesonclassifierconstructionsinsignlanguages.Mah-wah,NJ:LawrenceErlbaum.FINKEL, RAPHAEL, and GREGORY STUMP. 2007. Adefault inheritance hierarchy for computingHebrewverbmorphology. LiteraryandLinguisticComputing22.2.11736.FRASER,NORMAN,andGREVILLECORBETT.1997.DefaultsinArapesh.Lingua103.2557.FREGE,GOTTLOB.1980[1914].LettertoJourdain.Philosophicalandmathematicalcorre-spondence, ed. byGottfriedGabriel, Hans Hermes, FriedrichKambartel, ChristianThiel,andAlbertVeraart,7880.Chicago:ChicagoUniversityPress.GUREVICH, OLGA. 2006. Construction morphology: The Georgian version. Berkeley: Univer-sityofCaliforniadissertation.HALE, KENNETH, and SAMUEL JAY KEYSER. 1993. On argument structure and lexical expres-sion of syntactic relations. The view from Building 20, ed. by Kenneth Hale and SamuelJayKeyser,53109.Cambridge,MA:MITPress.HARRIS, ALICE. 2004. History in support of synchrony. Berkeley Linguistics Society13.14259.HARRIS, ZELLIG. 1951. Methodsinstructural linguistics. Chicago: UniversityofChicagoPress.HOCKETT,CHARLESF.1960.Theoriginofspeech. ScientificAmerican203/3.8896.JACKENDOFF, RAY S. 1975. Morphological and semantic regularities in the lexicon. Language51.63971.JESPERSEN,OTTO.1943.AmodernEnglishgrammaronhistoricalprinciples,part6:Mor-phology.London:GeorgeAllenandUnwin.KISCH, SHIFRA. 2004.NegotiatingdeafnessinaBedouincommunity. Genetics, disability,anddeafness, ed. byJohnV. vanCleve, 14873. Washington: Gallaudet UniversityPress.LANGUAGE,VOLUME83,NUMBER4(2007) 830LANE,HARLAN;RICHARDPILLER;andMARY FRENCH.2000.OriginsoftheAmericanDeaf-world:Assimilatinganddifferentiatingsocietiesandtheirrelationtogeneticpattern.Thesignsof languagerevisited, ed. byKarenEmmoreyandHarlanLane, 77100.Mahwah,NJ:LawrenceErlbaum.LEES,ROBERTB.1960.ThegrammarofEnglishnominalizations.(IndianaUniversityRe-search Center in Anthropology, Folklore, and Linguistics publication 12.) Bloomington:IndianaUniversity.LEVELT, WILLEM. 1989. Speaking: Fromintentiontoarticulation. Cambridge, MA: MITPress.MALKIEL,YAKOV.1967.Eachwordhasahistoryofitsown. Glossa1.13749.MARTINET,ANDRE.1957.Arbitrairelinguistiqueetdoublearticulation.CahiersFerdinanddeSaussure15.10516.[ReprintedinReadingsinlinguistics2,ed.byEricP.Hamp,Fred W. Householder, and Robert Austerlitz, 37178. Chicago: University of ChicagoPress,1966.]MATTHEWS, P.H. 1972. Inflectionalmorphology:AtheoreticalstudybasedonaspectsofLatinverbconjugation.Cambridge:CambridgeUniversityPress.MATTHEWS,P.H.1974. Morphology.Cambridge:CambridgeUniversityPress.MEILLET, ANTOINE. 1903. Introduction aletude comparative des langues indo-europeennes.Paris:Hachette.NOWAK,MARTINA.,andDAVIDC.KRAKAUER.1999.Theevolutionoflanguage.Proceed-ingsoftheNationalAcademyofSciences96.802833.SANDLER,WENDY;IRITMEIR;CAROLPADDEN;andMARKARONOFF.2005.Theemergenceof grammar: Systematic structure in a new sign language. Proceedings of the NationalAcademyofSciences102.266165.SCOTT, D.; R. CARMI; K. ELDEBOUR; G. DUYK; E. STONE; and V. SHEFFIELD. 1995. Nonsyn-dromicautosomal recessivedeafnessislinkedtotheDFNB1locusinalargeinbredBedouinfamilyfromIsrael. AmericanJournalofHumanGenetics57.96568.SPENCER,ANDREW.2000.Morphology. Thehandbookoflinguistics,ed.byMarkAronoffandJanieReesMiller,21337.Oxford:Blackwell.STOCKALL,LINNAEA,andALECMARANTZ.2006.Asingleroute,fulldecompositionmodelofmorphologicalcomplexity:MEGevidence. TheMentalLexicon1.85123.STOKOE,WILLIAM.1960.Signlanguagestructure:Anoutlineofthevisualcommunicationsystemsof theAmericanDeaf. (Studiesinlinguisticsoccasional papers8.) Buffalo:DepartmentofAnthropologyandLinguistics,UniversityofBuffalo.STUMP, GREGORY. 2001. Inflectional morphology:Atheoryof paradigmstructure. Cam-bridge:CambridgeUniversityPress.SZABO, ZOLTAN. 2007. Compositionality. Stanfordencyclopedia of philosophy. Online:http://plato.stanford.edu/entries/compositionality/.TARMON,ASHER,andEZRI UVAL.1991. Hebrewverbtables.Jerusalem:Tamir.Webstersthirdnewinternational dictionaryof theEnglishlanguageunabridged. 1986.Ed. by Philip Babcock Gove and the Merriam-Webster editorial staff. Springfield, MA:Merriam-Webster.XU,ZHENG;MARKARONOFF;andFRANKANSHEN.2007.DeponencyinLatin.ProceedingsoftheBritishAcademy145.12743.DepartmentofLinguistics [Received14November2006;StonyBrookUniversity accepted10May2007]StonyBrook,NY11794-4376[[email protected]]