Preface
In one of the most infamous episodes of twentieth -century intellectual history , the linguist -anthropologist Benjamin Lee Whorf argued (i ) that language shapes thought and reality , (ii ) that the tense system of a language can tell us about the metaphysics of time entailed by that language, and (iii ) that for the Hopi , among other cultures, the tense system (if it can be called that) is so radically different from ours that those cultures may not have a concept of time at all . " I find it " gratuitous , writes Whorf ( 1956, p . 57), to assumethat a Hopi who knowsonly the Hopi languageandonly thecultural ideasof his own societyhasthe samenotions, often supposedto be intuitions, of time andspace that we have, and that are generallyassumedto be universal. In particular, he has no generalnotion or intuition of TIME as a smoothflowing continuum in which everything in theuniverseproceedsat anequalrate, out of future, througha present,into a past; or, in which, to reversethe picture, the observeris being carriedin the streamof duration continuouslyawayfrom a pastand into a future. . . . The Hopi languageis seento containno words, grammaticalforms, constructionsor expressions that refer directly to what we call " time," or to past, present, or future or lasting, or to motion askinematic ratherthandynamic(i.e. asa continuoustranslationin spaceandtime ratherthanasan exhibitionof dynamiceffort in a certainprocess ), or thatevenreferto spacein sucha way as to excludethat elementof extensionor existencethat we call "time ," and so by implication leavea residuethat could be referredto as " time." Hence, the Hopi language containsno referenceto " time," eitherexplicit or implicit . I think that Whorf was more right than wrong in the abovepassage. Oh, I don ' t mean that he was right in thinking that the Hopi are so different from us. The ' past few decades work in generative grammar has shown us that the differences among human languagesare superficial at best. Moreover , it is now pretty clear that evenon the surfacethe Hopi have a temporal systemnot unlike our own. ( See Malotki 1983 for a discussion.) Still , I think Whorf was on target on a number of points. I think he was correct in thinking that one can argue from the sttucture of human language to the nature of reality , and I think he was most likely correct in seeing a close connection between language and thought .
xiv
Preface
But there is anotherpoint on which I think Whorf was right in the above . I think thaton a certainlevelof deepanalysishis descriptionof theHopi passage tensesystemwasbasicallycorrect- notjust for theHopi, but for all of us. That is, I think that a closestudyof Englishdoesnot supportthe thesisthat thereis sucha thing astense- at leastnot the sortof tensesystemthatis compatiblewith currentlyfavoredphilosophicaltheoriesof time. More to the point, I doubtthat weactuallyhavea " generalnotionor intuition of TIME asa smoothflowing continuum in which everythingin the universeproceedsat an equalrate, out of future " " , througha present,into a past. I am alsoquite surethatwehave no words, grammaticalforms, constructionsor expressionsthat refer directly to what we " call "time ," or to past, present,or future. Theseare pretty provocativeclaims. At leastWhorf had the good senseto restrict his claims to (in our eyes) exotic peoples, thus allowing us room to concedethat exoticpeoplesmay haveexotic realities. But who would suppose thatwedon' t havea tensesystemthatallowsusreferenceto pastandfuture, etc.? Isn' t it obviousthat we do? Evenif it couldbe shownthat our tensesystemlacksreferenceto future and pastevents, why shouldwe draw conclusionsabouttime itself aboutthe nature ' of reality?Can t we simply saythatthereis tense, andthereis time, but they don' t havemuchto do with eachother? ' What I will try to showis that Whorf s unarguedintuition- that thereis a closeconnectionbetweenlanguageandreality- is basicallycorrect. His error . Oncethat error wasin exaggeratingthe differencesbetweennaturallanguages is corrected, thereis no reasonfor us to be driven to the kind of cultural relativism ' thatfollowedfrom Whorf s originalthesis.Thestructureof languagedoes havemetaphysicalconsequences , but the structureof languagedoesnot differ in relevantwaysbetweenEnglish andHopi, betweenFrenchandFarsi, or between ChineseandUrdu. It followsthathumansall sharethesamemetaphysicsthe samereality. Thus, oneof the centralgoalsof this book is to illustratehow onecanstudy . The specific metaphysicalquestionsfrom a linguistic/semanticalperspective issuethatI havechosento investigateis thewell-entrencheddisputebetweenAtheorists andB-theoristsaboutthe natureof time. Accordingto B-theorists, thereis no genuinechange; rather, thereis a permanent sequenceof unchangingevents, ordered(lined up, if you will ) by an earlier-than/later-thanrelation. Forexample,WorldWarI (andall the sub-events containedin it ) is just asreal asthe eventof your readingthis preface, which in turn is just asreal asthe eventof the deathof the sun. Whenwe saythatWorld War I is past, we meanthat it is earlier than the event of our utterancethat
Preface
WorldWar1is past. Whenwe saythatthedeathof thesunis future, we meanthat thedeathof the sunis laterthanour utterancethat the deathof the sunis future. " - events are In this sense , B-theoristsconsiderreality to be " untensed not intrinsically , or future; rather,theysimplyexist(outthere, somewhere past, present ), and 'past' and ' future' aremerelywaysof talking aboutwherethoseeventslie relativeto the utteranceeventsin which we speakaboutthem. This view may seemcounterintuitive , but it is mostlikely thereceivedview in bothphysicsand philosophy,havingbeenadvocated(or saidto havebeenadvocated ) by figures rangingfrom Albert Einsteinto BertrandRussell. Accordingto A -theorists, on the otherhand, time is not a frozensequences of unchangingevents. The picturegivenby A -theoristsvariesfrom theoristto theorist, but 1 will be defendingan alternativedueto A. N. Prior (andperhaps to SaintAugustinebeforehim) in which, strictly speaking,thereis no futureand no past" out there" or anywhere.Wecansaythatit will betruethata certainstate of affairs (say, the deathof the sun) will hold andthat it wastrue that a certain stateof affairs (say, World War I ) held, but that this doesnot involveour referring to future and past eventsor to therebeing sucheventsfor us to refer to. Accordingto this idea(often calledpresentism ), what makessomethingfuture or pastis how the world standsright now. The allegedproblemswith eachof thesetwo positionsare now fairly well mappedout. The chief problemwith the B-theoryis that it fails to accountfor the indexicalnatureof our temporaldiscourse.As an illustration, supposethat 1know 1haveanimportantappointmentat 20 ' clock, but thatbecausemy watch hasstopped1do not know that it is now 2 0' clock. 1blissfully think out loud: " I havean appointmentat 2 o' clock." Suddenly, the radio announcesthat it is 2 0' clock. 1nowthink out loud: " Oh no, 1haveanappointmentnow! IIThealleged problemfor the B-theoristis that thereis no way to distinguishthe contentof thesethoughts/utteranceswith B-theory semanticalrelations. As far as the B' ' , now just meansthe sameas ' the time of this utterance', theoryis concerned ' ' which is just to say 2 o clock' . Somethinghasgonewrong. The responsestrategiesof the B-theoristsarelimited to two generalroutes. doesn't haveanythingto do with metaphysics First, onecansaythatsemantics , so we canallow indexicalityin the semanticswithout it infectingour metaphysics . Second,onecansaythatthetwo sentences in my exampleof the2 0' clock meeting , despiteappearances , actuallyhavethe samesemantics , or at leastthe same semantic" content." Theextra" meaning" suppliedin thesentence with theindexical is not really semantical ; however,it may be psychological and , psychology doesnot havethe samemetaphysicalcommitmentsthat semanticsdoes. In this book 1 will arguethat both of thesestrategiesfail - that semanticscannotbe
Preface
. divorcedfrom metaphysics , andindexicalitycannotbedivorcedfrom semantics defective . is will be that conclusion aBtheory metaphysics inherently My On the otherside, therearetwo centralobjectionsto theA -theory: first, that it allegedlyfalls victim to a logical paradoxobservedby J. ME . McTaggart, ' and, second , thatit can t beintegratedwithin a semanticsof tensebecauseit cannot ' ' accountfor temporalanaphora(expressions , suchas then , that apparently ' referto timesin thepastandfuture) andthereforeit alsocan t accountfor complex tenses(e.g., thepastperfect). I will arguethattheobjectionsrelatedto temporal ' anaphoraand McTaggarts objectionsturn out to be related, and that a theoryof temporalanaphoracanbe incorporatedinto a semanticsof tensethat doesnot havefuture and pastevents. The idea will be to developa theory of " " Etype temporalanaphora in which temporalanaphorsare not referring expressions but ratherstandproxy for temporalconjunctions- e.g., when-clauses which can thenbe treatedin a nonreferentialway). My conclusionwill be that ( thereareno compellingsemanticalobjectionsto theA -theory. My casefor theA -theorywill not endthere. If thereis a fact of the matteras to what semanticaltheorya languageuseris exploiting (andhow the language useris representingthat knowledge), andif languageusersactuallyhave(tacit) knowledgeof their semanticaltheory (and of how they are representingit ), then theremay be psycholinguisticprobesthat canhelp us detenninewhether . As we will the speakeris usinganA -theorysemanticsor aBtheory semantics see, evidencefrom languageacquisitionandfrom acquiredlinguistic deficits supportsthe ideathat the structureof our semanticalknowledgeis consistent only with theA -theorypicture. The main thesesof this book are, then, the following: . We can gain insight into the metaphysicsof time by studyingthe semantics of natura1language , wherethis constitutes(in part) our knowledgeof language world relationsandhow we representthat knowledge. . TheB-theoristcannotaccountfor thesemanticsof temporalindexicals;hence, the possibility of aBtheory metaphysicsis undennined. . TheA -theoristcananswersemanticalobjectionsabouttemporalanaphoraand metaphysicalobjectionsaboutthe McTaggartparadox. . Psycholinguisticevidenceaboutthe semanticaltheory that humansactually . andhencefavorsA -theorymetaphysics employalsofavorstheA -theorysemantics Defendingsomeof thesetheseswill obviouslyrequirethatwe do sometechnical work, both on the philosophicalendandon the linguistic end. This need to go into technicalmattershaspresentedmewith a dilemma. Philosopherswill puzzleovertheneedfor formal fragmentsandwill find theempiricaldiscussion
Preface
xvii
of grammaticaltenseandanaphorato be mind-numbingly detailed. Linguists will puzzleover all the attentiongiven to the McTaggartargumentand truthvalue links , and will find the empirical discussionof tensephenomenaand . Thereis no helping this, I ' m afraid. I have anaphorasurprisinglycondensed tried to give the minimumlevel of detailnecessaryto securemy argument,and I havereliedheavilyon notesto point theway to furtherdiscussionof manycrucial issues. Readersinterestedin digging deeperwill haveto pursuethe references . My goalsherearesimply to layout theform of the argumentandto give enoughdetail to showhow the pieceshangtogether. As I havealreadyhinted, my centralconcernin this project is not with the A -theoryandtheB-theory, or evenwith thephilosophyof time. Rather,my goal in this exerciseis to illustratean approachto metaphysicsin which semantical theory andthe philosophyof languageare central. Fifty yearsagoperhapsno one would havebotheredmaking this final point, sinceanalyticphilosophers thentook it for grantedthat language(andthe philosophyof language ) would standat thecenterof anyphilosophicalendeavor . In subsequent decadesthephilosophyof languagewasremovedfrom this centralposition, sometimesto be replacedby the philosophyof mind and sometimesto be replacedby nothing at all (leaving the various sub-branches of analytic philosophy to spin off in numerousunrelateddirections). It is not all badthat the philosophyof languagelost its placeat the centerof analyticphilosophy.The view of languagethat guidedmostof the mid-century research was, in my view, fundamentallymistaken.Further,manyof thesupposed whichwereto flow from thephilosophyof languageweretenuous consequences at best. Today, if the philosophyof languagehassomeclaim to makein metaphysicsor elsewhere , it mustbe carefullyargued,andthatis all to the good. Yet, asthecenturydrawsto a closethe philosophyof languagehasreturned in a newform. It hasbeensuccess fully naturalized,in my view, andintegrated into thesemanticsof naturallanguageandlinguistic theory. ( Fora surveyof relevant literatureand a glosson this naturalizationproject, seeLudlow 1997a.) Returningin this form, the philosophyof languagedoeshavepowerful claims to makeaboutour variousphilosophicalendeavors , notjust in the metaphysics of time, but alsoin the theoryof causality, in the theoryof action, andin value theory. I hopethat this book will serveto illustratejust how centralthe philosophy of languageis, and how, executedcorrectly, the philosophyof language deservesto reclaimits placeat the very heartof analyticphilosophy.
Acknowledgements
WhenI conceivedof this project, in 1991, the ideawasto put togethera short manuscriptthatwoulddeveloptwo differenttheoriesof tensewithin anabsolute semanticsfor naturallanguageandwould showthe connectionbetweenthose theoriesand two leading theoriesof time. In 1992I circulated a short draft manuscriptthat containedthe core of the idea(althoughnot workedout in detail ). That draft receiveda numberof helpful commentsfrom Barry Schein, Ed Zalta, andRichardLarson, which led to somesignificantrevisions. In the fall of 1993the revisedmaterialwaspresentedat a seriesof lectures at the University of Padua, and it benefitedgreatly from the criticisms and commentsof someof theparticipants(includingPaoloLeonard i , Mario Mignuce, andErnestoNapoli). Theselectures(andthe discussionperiods) led to the first 1994draft, which I againcirculatednarrowly. Again I was fortunateto have importantfeedbackfrom RichardLarsonandBarry Schein, and in particular from Ernie Lepore. Incorporatingthosecomments, I revisedthe manuscript again. Late in 1994I submittedit to The MIT Press.Although I receivedhelpful commentsfrom two anonymousreviewers, and althoughthe project was " " greenlighted, it wasclearto me that a lot of technicalproblemsremainedto be workedout. Subsequentrevisionsof this material, particularly the developmentof the technicaldetails, werepresentedin a courseon the semanticsof tensewhich I taughtat in the Diploma Programin Philosophyof LanguageandLinguistics at the University of Venicein the summerof 1995, and at a seminaron tense whichI taughtin theDeparbnentof Linguisticsat StonyBrook in 1996. Theless technicalportionsof this materialwerepresentedin mini-coursesthat I taught at the University of Novi Sad, Jugoslaviain 1997and at the Fourth Central EuropeanSummerSchoolin GenerativeLinguistics, in Olomouc, the Czech Republic, 1997.
Acknowledgements
Portions of chapters5- 10 were presentedin talks at the Departmentof of Linguistics, PhilosophyattheUniversityof Maryland( 1995), attheDepartment HumboldtUniversityin Berlin ( 1995), attheInter-UniversityCenterConference on Truth in Bled, Slovenia ( 1996), at the Conferenceon the Question of Temporalityat theUniversityof Venice( 1996), andat the Moral SciencesClub at CambridgeUniversity( 1997). Thepenultimatedraft, completedin late 1997, wasagaincirculatedfor comments . Onceagain, invaluableassistance andencouragement camefrom Richard Larsonand Ernie Lepore, aswell asfrom Lori Repetti, GabeSegal, andJason Stanley.Crucially, LarsonandRepettiboth convincedme that, while I hadperhaps ironedout thetheoryto my satisfaction , theexpositionwasutterly opaque. I accordinglyspentthe nextsix monthstrying to streamlinethe manuscript , remove technicaldetails, andin generalmakethebookmoreuserfriendly. To this end, in the springof 1998I usedseveralvenuesto experimentwith waysof presenting the material: the Universityof Milan, the ScuolaNormaleSuperlorein Pisa, anda workshopon tensesponsored by UCLA' s Departmentof Linguistics. The last yearanda half of work on this book wasdonein someidyllic settings. In the spring of 1997I was a visiting scholarat CambridgeUniversity, whereapartfrom the MSC talk, I benefitedgreatly from discussingtheseissues with HughMellor, JeremyButterfield , IsaacLevi (who wasalsovisiting), andMichaelPotter. (I might addthattheoffice I wasprovidedat Cambridgewas furnishedwith portraitsof J. ME . McTaggart,C. D. Broad, andG. E. Moore quite inspirational for someoneworking on the reality of time.) In the fall , work continuedat the Universityof Venice, whereI held the FulbrightChair in / Linguistics.I amindebtedto AnnaCardinaletti,Michal Philosophyof Language Starke, and Guglielmo Cinquefor logistical and intellectual supportwhile in Venice. I am also indebtedto my departmentchair, Ed Casey, and to the administration at SUNY Stony Brook for supportingmy leavesof absenceduring this period. Certainindividualsdeservethanksfor contributionsthat aresomewhatmore existentialin nature. Lori Repetti and our daughterChiaradeservesomesort of awardfor toleratingme while I workedon this. Fortunatelyfor Chiara, she wasnot aroundduring the first four yearsof this effort. I shouldadd that Lori providednot only existentialsupportbut also somecrucial additionsandcorrections - particularly in the sectionson the natureof I -languageand in the sectionon eliminatinggrammaticaltense.A differentsort of existentialcontribution camefrom Noam ChomskyandRichardLarson, who (apartfrom providing me with commentson portions of the manuscript) taught me to be courageousin my thinking, and more importantly, by their examples, taught
Acknowledgements
mehowto havethekind of mentalstrengthnecessary to grind out thearguments of a required minority intellectualposition. I nowrealize, however,thatthefoundation of what I learnedfrom them I had alreadylearnedfrom my father. It washe who taughtme abouthavingat leastone new ideaeveryday, andhow to staywith eachnew idea, sometimesfor years, reworkingit until it cameto fnrition. He alsotaughtmethat, evenafterall this work, onehasto expecta high rateof failure amongradicalideas. As this projectis way out on severalphilosophicallimbs andcould, eventomorrow, crashin a most spectacularway, the only personI canimaginededicatingthis book to is my father. H the projectin this booksucceeds , no oneelseis moredeservingof havingit dedicatedto them. If the projectfails . . . well, asmy fatherwould say, " it ' s all in a day' s work."
Introduction
The A -Series vs. the B - Series
Metaphysicsis, in part, the studyof what is real. The laypersonoften supposes thatwhenphilosophersworry aboutwhatis realtheymustworry aboutwhether " tables,chairs, anddinnerplatesarereal. Somephilosophersdo worry aboutthese questions, but there are a numberof other metaphysicalissuesthat philosophers worry abouttoo. Amongthe areasof interestis a classof metaphysicalquestionssurrounding thenatureof time. Forexample,is timereal?If so, is thefutureasrealasthepast? Can we changethe future? If yes, why? If not, why not? If time is real, then whatexactlyis it? Is time, assomehavesuggested , reallyjust physicalchange? If so, then how do we makesenseof this changeapartfrom its occurring in time? But if changetakesplacein time, thenhow cantime be change? As with othermetaphysicalquestions , questionsaboutthenatureof time arenotoriously difficult. Certainproblemsposedby the pre-Socraticphilosophersare still debated , andthe numberof metaphysicalpuzzlessurroundingthe natureof time continuesto multiply. This work will focusonjust oneof themanyissuesin thephilosophyof time. Theissue, which in someform wasdiscussedasfar backasthethird century(by the NeoplatonistIamblichus'), hasbeenat the centerof the twentieth-century discussionof the philosophyof time. Briefly , the problemis asfollows: lWo esto the philosophyof time canbe distinguished.Accordingto broadapproach oneapproach, adoptedby Russell, Einstein, Reichenbach , and others, time is events.Futureevents,pastevents, of unchangingandtenseless simplya sequence and presenteventsare all equally real. McTaggart( 1908) called this the Bseriesconceptionof time; others, including Mellor ( 1981), havecalled it the untensedconception.Accordingto the alternativeapproach,it is fundamental
Introduction to the notion of time that events, or perhapspropositions, have genuine temporal status. So, for example, there is a fundamental metaphysical distinction between events that are future and those that are present or past. This fundamental difference is supposedly deeper than a simple ordering of events by the earlier than/ later -than relation . McTaggart referred to this as the A -series conception of time ; others have characterized it as the tensed conception . Following Gale ( 1967), we might find it useful to distinguish the A - seriesand the B - series according to the following criteria . A - series
B - Series
The B-seriesis reducibleto theA series. Temporalbecomingis intrinsic to all events. Thereareimportantontological differencesbetweenpastandfuture. Changeis analyzablesolelyin terms of A -seriesrelations(past, present, future).
TheA -seriesis reducibleto the B series. Temporalbecomingis psychological. The B-seriesis objective. All events areequallyreal. Changeis analyzablesolely in terms of B-seriesrelations(earlier-than, later-than).
As we will see, thesecriteria do not provide necessaryand sufficient conditions for identifying theA -seriesandthe B-series; indeed, someof the criteria will haveto be relaxedif logical conundrumsareto be averted.For now, however , theycanprovideuswith a usefulway of thinking aboutthedistinction. But what exactlyis at stakein this distinction? Questionslike thesehavebeenpursuedthroughoutthehistoryof philosophy, notjust becauseof their intrinsic interestbut alsobecausemetaphysicshasoften beenseenasa pointof departurefor otherphilosophicalinvestigations . Questions aboutthe metaphysicsof time havebeenthoughtto haveconsequences for the of the of mind the of , , philosophy language philosophy philosophy religion, the philosophyof science,epistemology and other branch esof philosophy. , As we will seein chapter10, the decisionbetweenthe A -theory andthe B. But how doesonedecidebetweenalternatives theoryis rich in consequences like these? One answerwould be that questionsabout the natureof time are bestaddressed by physicists. For example, Einsteinheld aBseries conception of time} He presumablyhadgoodphysicalreasonsfor doing so. Shouldn't we thereforedefer to Einstein, or to whatevercurrent physical theory dictates? Putnam( 1967, p. 247) appearsto adoptsucha view:
Introduction I concludethat the problemof reality and the determinateness of future eventsis now solved. Moreover, it is solvedby physicsand not by philosophy. . . . Indeed, I do not believethat thereare any longeranyphilosophicalproblemsaboutTime; thereis only thephysicalproblemof detenniningtheexactphysicalgeometryof thefour- dimensional continuumthat we inhabit.
On the other hand, Sklar ( 1981, p. 2493), specificallyaddressingthis passage from Putnam, notesthat sucha position reflectssomeignoranceof the nature of the scientificenterprise: I thinkthatsucha naiveviewis aswrongasit canbe. Justasacomputer is onlyasgood ' asitsprogrammer in, garbage out' ) onecanextractonlysomuchmetaphysics ( Garbage fromaphysical , beconsistent theoryasoneputsin.Whileourtotalworld-viewmust,of course withourbestavailable scientifictheories , it is a greatmistaketo readoff a metaphysics fromthetheory's overtappearance , andanevengravermistaketo superficially havegoneintotheformulation of the neglectthefactthatmetaphysical presuppositions , asit is usuallyframed , in thefirstplace. theory If Sklar is correct, thereis still a philosophicalinvestigationto be undertaken , evenif it is to be a part of the scientific programof physics.4Othershaveargued that subjectmattercalled " time" in physicsreally hasnothingto do with the metaphysicalconceptionof time but is just an appropriationof the tenDto discusscertainaspectsof light relationswithin the theorys Perhapsthemosttelling problemis that it is not obviousthat currentphysics as ( opposedto physicists) actuallyfavorsonepositionover the other. As Sklar " ( 1974, p. 275) hasnoted, the scientifictheorycan changethe philosophyand , put the dispute[ betweenthe A -theory andthe B-theory] in a new perspective butit cannotresolvethedisputein anyultimatesense ." Furthennore Sklar 1981 , ( ), Stein( 1968), Dieks( 1988), andShimony( 1993) havearguedthatnon-B-theory 6 interpretationsof the specialtheoryof relativity arepossible. In sum, evenif philosopherswish to passthe burdenof metaphysicalinquiry ontothephysicist, thephysicistsimplycannotshouldertheburden. There are still metaphysicalquestionsto be answered,and it is not up to physicists aloneto answerthem.' But what exactlycanphilosophybring to the table? Philosophers , after all, havebeendebatingmetaphysical for more than 2000 without much puzzles years success . Dummett . 12 the 1991 situation as follows: , p ) puts ( apparent Themovesandcountermovesarealreadyfamiliar, havingbeenmaderepeatedly by thephilosophers . Thearguments of onesideevokea response in throughthecenturies certainof thespectators of thecontest , thoseof theothersideswayothersof them;but wehavenocriterionto decidethevictors.No knock-outblowhasbeendelivered . We mustawardthedecisiononpoints;andwedonotknowhowto awardpoints.
Inb"oduction
Or, asVoltaireput it (lesspugilistically) 200yearsearlier, metaphysics is a dance of elegantstepsin which we endup backwherewe started. Is thereanythinga philosophercanbring to this dancebesidespurespeculationandelegantsteps? One strategyin philosophyhasbeento reject the ideaof metaphysicsas a startingplaceandto arguefor movingfrom the bottomup - from the theoryof . thoughtto metaphysics The generalideais broadlyKantianin character . Wecanneverknow things " astheyare" in themselves , sincethemind is activelyinvolvedin organizingour . Thebestwe cando is elucidatethecategoriesor structureof reason. experience For example,accordingto Kant, time wasnot itself a propertyof thingsin themselves ; rather, it was imposedupon our experienceby the mind. Of course, from this perspectiveit would be futile to beginan investigationinto the nature of time apartfrom a considerationof the natureof thoughtor reason.And indeed , after an investigationinto the natureof time asa categoryof humanreason hadtakenplace, therewould be little left to do in the way of metaphysics , saveperhapsto dotthei ' s andcrossthet ' s. Or, to usea metaphordueto Dumrnett, the philosopher's task is like that of the optometrist, who cannottell us what we will seewhenwe look about, but who, by providing us with adequatecorrective lenses,cannonetheless help us to seemoreclearly. In thetwentiethcentury, a numberof analyticphilosophershaveconcededto Kant thegeneralview in which aninvestigationinto metaphysicscannotbeconducted apartfrom an investigationinto the natureof thought, but haverejected Kant' sconceptionof thought, with its attendantcategoriesof reason.In its place, theyhaveproposedthatthoughtis inherentlylinguisticin nature.Thus, theyhave proposedthat the properstartingplacefor investigationshouldbe the language in which we think, andthis hasgenerallybeentakento be naturallanguage.8 esto thephilosophyof time , theB-theoryandA -theoryapproach Interestingly esto the semanticsof tensein naturallanguage. parallel two distinct approach On the one hand, there are approaches to the semanticsof tense(see, e.g., Reichenbach1947) that appealto referenceeven~s in accountingfor complex tensesandtemporalanaphora . On theotherhand, thereareapproach esto tense that are morein the spirit of Prior ( 1967, 1968) - approach es in which ' past' , ' ' ' ' , andin which thereareno pastand present, and future areprimitive operators 9 future eventsper se. If onesupposesthat thereis an interestingconnectionbetweenmetaphysics andthe semanticsof naturallanguage,and if one supposesthat the semantics of naturallanguagecanhelp illuminate our metaphysics , thenonemight hope that the semanticsof tensecanhelp illuminatethe metaphysicsof time. For example ' , onemight supposethatthechoicebetweenReichenbachs theoryof tense
Introduction
andPrior' s theoryof tensemighthaveprofoundmetaphysicalconsequences (favoring eitherthe B-seriesor theA -seriesconceptionof time). HereI amnot advocatinganapproach , like thatof Dumrnett( 1991), in which we aresupposed to reasonbottom-up from thetheoryof meaningto metaphysics . It seemsto me that the constructionof a theoryof meaningwithout someprior senseof ontology would haveus climbing blind. That is, without somesense of theconstituentstructureof the world we would haveno ideaof how the theory of meaningis to link up our languagewith the world. Nor am I advocating the oppositeposition in which we are to sort out our ontology before we undertake the constructionof a theory of meaning. It is only throughthe theory of meaningthatwe areableto differentiatetheelementsof our ontology. For example , evenif a priori metaphysicsis ableto tell us that somethingin the world hasan abstractpropertyfoo, whatin our ontologytells us thatfoo is a temporal property- that it hassomethingto do with the natureof time? Accordingly, I assumethat we havepartial knowledgeof the natureof reality andpartialknowledgeof thetheoryof meaning, andthatour taskis to solve a kind of complexequationinvolving information from semanticson the one sideandmetaphysics on theother. Whatwe knowaboutthenatureof reality will help shapeour semantictheories, but it is also the casethat semantictheory will help to shedlight on the natureof reality. Of coursemanyphilosopherswill hold that eithermetaphysicsor the theory of meaningmustbe morefundamentalthanthe other, but to me this hasall the " " makingsof a chickenor egg argument.Theremay be somedeeptruth about whetherchickensor eggsaremorefundamental , but no seriousbiologistwould in such a debate nor I , ( hope) would any seriousphilosopherbe exercised engage by the question.Likewise, in my opinion, philosophersshouldworry less aboutwhethermetaphysicsor the theoryof meaningis morefundamentaland shouldworry moreaboutthe relationsthat musthold betweenthemin view of whatwe alreadyknow abouteach. Roughly,whenI saythatthereis aninterestingrelationholdingbetweenmetaphysics andsemanticsI meanthat concretequestionsaboutthe natureof reality canbeilluminatedby whatweknowaboutsemantictheory, andthatimportant questionsin semantictheorymay be adjudicatedby certainof our metaphysical intuitionsabouttheconstitutionof reality. Clearlymoreneedsto be said, and chapter4 will takeup theissuein detail. Of course,this bookis intendedto serve asan illustrationof this generalpoint. Indeed, the goal of this book is to providea semanticalargumentin support of the A -theory conceptionof time. Or, better, the goal is to arguesimultaneously for theA -theoryconceptionof time andfor a theoryof tensethat I will
Introduction
call theA -theoryof tense.As will be seen, if the connectionbetweenlanguage andthe world holdsup, thenthesetwo doctrineswill be mutuallyreinforcing. My main argumentwill be as follows: First, there are certain semantical weaknesses inherentin the B-theory semanticalposition. Specifically, the Btheorist cannotadequatelyaccountfor the indexical natureof temporal discourse . Since the B-theory of time cannot be detachedfrom the B-theory . semantics , this effectivelyundenninesthe B-theory metaphysics On theotherside, it hasbeenheldthattheA theorysemanticshasweaknesses of its own. Accordingly, I arguethat thoseweaknesses , to the extentthey exist, areeasily repaired. The A -theoryconceptionof time thus remainsa plausible andundamagedalternativeto the B-theoryconception. But I will arguefurther that independentpsycholinguistic evidencesupports the thesisthat theA -theorysemanticsis in fact the semanticaltheorythat usersof naturallanguageinternalizeand" know" (in a senseto be spelledout in chapter2). To get a betterideaof how this argumentwill unfold, let us first briefly review the semanticalchallengesfacingboth theA -theoryandthe B-theory.
:for theB-Theorist: Temporal TheSemanticalChallenge My wedding anniversaryis March 12. Supposethat I have memorizedthis date. Maybe I had it inscribedin my weddingring. So I know the following : My weddinganniversaryis March 12. Now supposeI am in my office late in the afternoononeday next March. I " may say to myself: My fifth anniversaryis March 12. I should think about buyingmy wife an anniversarypresent." I might thenwonderhow muchtime I have. I takeout a calendarto find today' s dateanddiscoverto my horror that it is March 12! I shout" My fifth anniversaryis today!" : In this little episodeit is clearthat I hadtwo distinct utterances ( 1) My fifth anniversaryis March 12. (2) My fifth anniversaryis today. It is alsoclearthat whenI uttered(2) I hadknowledgethat I did not havewhen I uttered( 1), andthis extraknowledgeappearsto be reflectedin the difference between( 1) and(2). Thus, it is arguablethat ( 1) and(2) havedifferent semantical contents.As intuitive asthis mayseem,therearesomepowerfularguments
Introduction
designedto showotherwise- to showthat the semanticcontentsof ( 1) and(2) arethe same. Indeed, the standardphilosophicaltreatmentof indexicals- for example, by Perry( 1969, 1977) andKaplan( 1977, 1979, 1990) - hasbeento arguefor a distinctionbetweenthe contentof a demonstrativeexpressionand its character or role. Thecontentwouldbetheindividualor objectrefelTedto by thedemonstrative /role would be the additionalcognitive significance , andthe character " (sometimestheexpressionlinguistic meaning" is usedto characterizetheextra element) suppliedby the indexicalin caseslike (2). Although the literatureis sometimesunclearon this point, it appearsthattheseauthorsareadvocatingthat character /role shouldnot be part of the semanticcontent(or literal truth conditions ) of the utterance. Whatdoesthat mean? If the semanticsof naturallanguagetakesthe form of a T-theory, andhencethe semanticsof a sentenceis givenby theoremslike (3), thentheright-handsideof thetheorem- theportionfollowing ' if andonly if ' statesthe literal truth conditionsof the sentenceon the left-handside.
(3) 'Snowis white' is trueif and onlyif snowis white. In thiscase arethatsnowis white. If weassume , thetruthconditions a framework of thiskind(I will explainandarguefor it in chapter2), thenonewayof -Perrythesisis assayingthatcharacter /roledoesnotmakeit takingtheKaplan intotheright handsideof aT-theorytheorem .loForexample , thetruthconditions ofa sentence like 'I amhungry'wouldbenotasin (4) butakinto (5). (4) 'I am ' hungrynow is trueif andonlyif I amhungrynow. (5) An utterance u, attimet, by speakers , of 'I amhungry now ' is ttue if and only if s is hungryatt. Herethe only things that makeit into the truth conditionsare the individuals andthetime t. Theextraindexicalelementfoundin ' I ' and' now' mustlie somewhere outsidethe semanticsproper. As we will seein chapter3, thereare many argumentsfor keepingthe semantics free of character /role. For starters, havingit in the semanticsleadsto headachesin modal constructions. But, as we will also see, theseheadaches - therearewaysto retain canbe ameliorated indexicalityin the semanticsand alsocopewith technicalproblemsaboutmodality.
Introduction
Accordingly, 1am going to arguethat the receivedtreatmentof indexicality . Obviously, - that indexicalityshouldnotbedivorcedfrom semantics is mistaken this is not an argumentthat canbe madecarelessly , anda greatdealof groundwork concerningthe natureof languagewill haveto be laid. For example, leavingindexicalityin the semanticsleadsto certainanalyses that appearshockinglynaive. A casein point would be (4) above. This is alleg' ' edly naive, sincean utteranceof I am hungry by me to a hearerH canhardly be interpretedby H using(4) - that would forceH to concludethat 1am saying that H is hungry. The advicewe aregivenby PerryandKaplan, therefore, is to sweepawaythis naiveview of indexicalsand retain only the contentswithin the truth conditions(asin (5 . The problem with this brief chain of reasoningis that it restson assumptions aboutthe nature of languagethat 1 considerto be fundamentally mistaken . If thefunctionof languageis communication , thentheobjectionhassome merit. But why shouldwe supposethat languageis for communicationasopposed to, say, representingour thoughts? Indexicalswill be discussedat length in chapter3, but it is alreadyevidentthat certainassumptionsaboutthe nature of languagewill haveto be laid our first. This will be donein chapter1, which will lay the foundationsfor the discussionof the natureof semanticsin chapter 2 andfor the treatmentof indexicalsin chapter3. Later, in chapter6, we will seepreciselywhy this problemis insurmountablefor the B-theorist.
for theA-Theorist TheChallenges Therearetwo centralchallengesfor theA -theorist. The first is a philosophical challengeinvolving an allegedparadoxoriginally discussedby McTaggart. Thesecondchallenge- moresemanticalin nature - is thattheA -theoristhasno . Thesetwo problemsturn out to be related way of accountingfor temporalanaphora . , but we canbeginby treatingthemseparately The McTaggart Paradox : Is the A -Theory Contradictory ? One of the earliest and most influential critiques of the A -theory is found in ' ' McTaggart s ( 1908, 1927) argument for the unreality of time . McTaggart s argument begins with the observation that certain pairs of properties are such that it would be inconsistent for one object to have both properties. Forexampleal though a table can be both round and red, it cannot be both round and square, for roundness and squarenessare inconsistent properties . Likewise , according to McTaggart, it would be inconsistent for certain events(e.g ., the death of Queen
Introduction
Anne) to be both pastandfuture. Thus, in suchcases , if we affinn (6), we have statedsomethingthat is inconsistentif not contradictory . (6) future(X ) & past(X ) But accordingto McTaggartthis is exactlywhattheA -theoryentails, for a given eventE will at somepoint be past, at somepoint be present, andat some point be future.II Thus, we havethe following conjunction: (7) future( E) & past( E) & present(E) Theinitial reactionto this partof theargumentis oftenthatit is absurd , for surely one is not sayingthat E is alwaysfuture and alwayspastand always present, but ratheroneis asserting(for example) thatE is futureat a certaintime segment t, presentat sometime t * , andpastat anothertime segmenttIeBut accordingto McTaggartthis moveis a cheat; it amountsto smugglingin B-theoryresources ; theA -theoristcan' t appealto a sequenceof eventsor times. The B-seriestime line cannotbe introducedhereto savetheA -theorist. Let ussetthisparadoxon thebackburnerfor themomentandturnto the problem of temporalanaphora . As we will see, thetwo problemsarelinked, andonly by solvingtheproblemof temporalanaphorafor theA -theoristcanwe cometo grips with the McTaggartargument. The Problem of Temporal Anaphora Considerthe following example(partee1973, 1984): (8) 1turnedoff the stove. Clearly (8) doesnot merelymeanthat at sometime in the past1turnedoff the stove.Withouta doubttherehavebeenmanysuchepisodesin my past. According to Partee( 1973), (8) is informativebecausethereis animplicit referenceto some time or somereferenceevent. 1 might equally well haveuttered ' I turned off the stovethen' (with ' then' serving as a temporalanaphorreferring to some segmentof time or eventin thepast). This problemseemsto lie at the heartof anotherobjectionto Prioreantheories : that they arenot ableto accountfor complextenses.The objectionis that, for example, [ PAST[ PAST[ S ll ] simply collapsesinto the simplepast. To see this, first considerthecasewheretime is discrete. Let us call the minimumunit
Introduction
of time a " chronon." Then, at best, [ PAST[ S ]] is true iff S wastrueat leastone chronon ago. But then [ PAST[ PAST[ S ]]] is true iff S was true more than onechrononago. But this doesn't seemto capturewhat we intendedto sayby ' ' a pastperfectsentencelike I hadleft . Onemight try to get aroundthis difficulty by talking aboutdegreesof pastness ' ' , buteventhis moveis boundto fall short. I hadleft mightbeaboutanevent at anyarbitrarydistancein the past. Plus, thereis the strongintuition that there ' ' really is a referenceeventhere- thatonecouldvery well continue I hadleft . . . ' with ' whenSmith arrived . How is that to be cashedout on a Prioreantheoryif thereis no way to avail ourselvesof temporalreference? The Solution The solutionthat I will proposeto the problemof temporalanaphorais to develop - essentially a theory of temporal a notion of E-type temporalanaphora or events. to times reference not involve that does anaphora In the caseof ordinary E-type pronominal anaphora, as in (9), the idea is ' thatthepronoun' He doesnot referto somesalientindividual, but ratherstands proxy for a definitedescription,sothattheanalysisof (9) is somethingalongthe linesof (9'). (9) A mancamein the room. He trippedover the chair. ' (9 ) A mancamein theroom. The manwho camein theroomtrippedoverthechair. ' ' Crucially, The manwho cameinto the room is a Russelliandescription, not a ' ; hence, the secondsentencein (9 ) is not aboutsomeparticular referringexpression individualbut makesa generalclaim abouttheworld- i .e., thattheworld containsexactlyonemanwho cameinto theroom, andhetrippedoverthechair. Wecansaythatthe secondsentencein (91) is thereforeageneralpropositionand not a singular or object-dependentproposition. This differencemay not seem like a big deal; however,aswe will see, it is very importantin certaincontextsfor example, within the scopeof modalsandpropositional-attitudeverbs. For example, consider( 10). ( 10) I believethat a unicornis in the gardenandthat it is eatingmy roses. If the pronoun' it ' is a referring expressionand I havesuccess fully utteredan to the exisare committed that we then it appears object dependentproposition,
Introduction
tenceof unicorns. However,if the pronounstandsproxy for a description,asin ' ( 10), thenif we treatdescriptionsa la Russell( 1905) we arenot forcedto admit theexistenceof unicorns. ' ( 10) I believethat a unicornis in the gardenandthat the unicornin the gardenis eatingmy roses. A similarstrategycanbeexecutedfor temporalanaphora . Theoperativeidea is thattemporalanaphorslike 'then' do not referto timesbut ratherstand proxy for temporalconjunctionslike when-clauses.So, for examplein ( 11), the pronoun ' then' doesnot refer to a time, but is a placeholderfor a when-clausethat be extracted from the text. Thus, ( 11) might havethe glossgivenin ( 12) . might
( II ) Samaddressed Bill. Bill didn't respond then. ( 12) Samaddressed Bill . Bill didn' t respondwhenSamaddressed him. Crucially, on this proposal, the when-clausedoesnot refer to a time, but will expressa generalproposition(at leastgeneralin the sensethat the proposition is not dependent uponparticulartimesor eventsdescribedtherein). Furthermore , the generalnatureof thesepropositionswill be crucial when they areembedded in intensionalenvironmentslike thosecreatedby modalsandby propositional -attitudeverbs, andalso, I shall argue, in the scopeof temporaloperators ' like past' and'future' . In brief, whengeneralpropositionsareembeddedin such environmentsthe resultsare innocentclaimsto the effect that statesof affairs matchingcertaindescriptionsdid hold or will hold. Nothingfollows aboutthere beingpastor future eventsor times. I realize that so far this is a big promissory note. It certainly sounds incredible that a when-clause need not refer to a time . But , as we will see, the idea can be cashed out handlly using only off -the- shelf philosophical resources. These resources(including the distinction between general and singular propositions ) will be introduced and incorporatedinto a full theory of E- type temporal anaphora in chapter 8. Of course, in most casesthe temporal anaphor is implicit ( as in ' Bill didn ' t ' respond ) and the number and range of temporally anaphoric constructions is vast. I will not be able to chart the entire territory , but I will survey enough of them in chapter 8 to suggesthow the theory of E- type temporal anaphora might be developed. In addition , one of the central hypotheses of chapter 8 will be
Introduction
that every sentencehas a when-clause, or a temporal adjunct of someform ' ' ' ' (e.g. before. . . , after. . . ), or a temporalanaphorthat standsin for a whenclause. Accordingly, every sentencein naturallanguagehaseither an explicit or an implicit temporaladjunctclause. This insight turns out to lie at the root of the connectionbetweentemporal ' ' anaphoraandMcTaggarts paradox. The idea, very simply, is that one s initial intuition aboutthe McTaggartargumentwasbasicallyon track. One wantsto saythat it is possibleto avoidthe contradictionin (6) by addingthatX is future at a certaintime andpastat another. (6) future(X ) & past( X) That moveis certainlyblockedfor theA -theoristif timesareconstruedin aBseries way, but that doesnot meanthat thereis no analogousA -theorymaneuver . The idea, in short, is thattheA -theoristwill wantto invokeE-typetemporal . anaphora of the fonn given Hereis the gambit: ( 13) is shorthandfor a representation ' in ( 13). ( 13) X is future. ' ( 13) X is FUT when [ . . .] . ' But thenthe semanticsdeliversthe following truth conditionsfor ( 13): ( 13* ) . ' X is FUT when . . . ' is true iff X will be true when . . . . [ ] [ ] Similar considerationsapply to ( 14). ( 14) X is PAST. It will havethe following truth conditions: ( 14* ) ' X is PASTwhen . . . ' is true iff X wastrue when . . . . [ ] [ ] The trouble is that we canneverget to the point wherewe havea conjunction of two conflicting A -theory tensedclaims. Clearly ( 13* ) and ( 14* ) are not incompatible , sincethe when-clauseswill havedifferentcontents. In short, once we introduceE- typetemporalanaphora , theMcTaggartargumentwill fail to get off the ground.
Introduction
ThePlanof theBook Clearly, the lines of argumentI will be taking on temporalindexicalsandtemporal anaphoralead us directly to the semanticsof indexicals and anaphors . Thereis no question generally,andultimatelyto thevery foundationsof semantics that the argumentsgiven by B-theoristslike Mellor haverestedupon a very respectablesemanticalfoundation: the work of David Kaplan and John Perry on indexicals. for example. However, for all its respectability, I believe that it is a foundationof sand, not rock, andthat it is fundamentallyunstable. . Accordingly, I will beginwith a detaileddiscussionof the natureof semantics NothingI sayaboutthe foundationsof semanticsis goingto be original; for the mostpart, it will involveelucidatinga numberof ideasaboutindexicalsand anaphorsthathavebeenadvancedby GarethEvans, andit will alsoincorporate a generalsemanticalpicture that hasbeenproposedby JamesHigginbotham anddevelopedby a numberof my peersin thephilosophyof language . This work remainsa minority position, however,andI will needto developanddefendthe leadingideasin somedetailhere. Underlyingthis alternativeapproachto the semanticsof indexicals, for example , thereis anevendeeperphilosophicalissueaboutthe natureof language. Is it a socialobjectconstructedfor purposesof communication ? Or, asChomsky hasproposed, is languagea naturalobject which is part of our biologicalendowmentandnot necessarily for communicationat all?My answerto this question will haveconsequences for the semanticsof indexicals, so my first steps will haveto be thoughthis terrain. Accordingly, the generalplan will be to movegraduallyfrom a discussion of the natureof languageto a generaldiscussionof the semanticsof natural language. After adding someneededtechnical resourcesto the semantics, I will takeup thegeneralissueof theconnectionbetweenlanguageandtheworld, thendeveloptheA -theory andB-theory semanticaltheories, andthenproceed to a discussionof the challengesfacedby theA -theoryandthe B-theory. All thesepreliminariesmay seemlike a very " long march" just to arriveat a discussionof theproblemposedby indexicalityfor the B-theoristandtheproblem posedby temporalanaphorafor the A -theorist, but in my view the march is unavoidable . The issuesherearevery subtleandvery deep, andany attempt to skirt foundationalquestionsis goingto resultin anexchangeofuncompelling claimsandcounterclaims . readers Impatient may want to begin with the challengeto the B-theory in 6 and thedefenseof theA -theoryin chapters7 and8 andthenwork backward chapter ' . That s fine by me. Theorderof expositionherereflectsmy understanding
Introduction
of the logical precedenceof the issues, not necessarilythe friendliest or most enticingpathfor the reader. Here, then, is the completeorganizationof thebook. In chapter1, I first take up generalconsiderationsaboutthe natureof language " " , opting for a conceptionof language(called I -language by Chomsky) / representational as an innatecomputational systemthat is part of our biological endowmentbut is not necessarilyevolvedfor purposesof communication. I thendiscussthe possibility that I -languagemight the languageof thought. In chapter2, I beginto developthesemantictheoryat a very simplelevel, beginning with a discussionof why a referentialsemanticsshouldtaketheform of . Amongtheissuesthatwill bepivotalin anabsolutetruth-conditionalsemantics later chaptersis the questionof whetherT-theoriescandisplay senses(and if so, how), the distinctionbetweenmodestandrobustT-theories, andthe possibility of ontologically parsimonioustheoriesof predication- theories thatdon' t requirereferenceto properties. In chapter3, I devotesometime to showinghowresourcesforpropositionalattitudeenvironmentsandfor the theoryof indexicalscanbe incorporatedinto . ( Bothsetsof resourceswill be necessaryto developmy an absolutesemantics semanticsof tense.) Crucially, I will arguefor ananalysisof indexicalsin which . FollowingGarethEvans, I will alsoarguethat theyaretreateddisquotationally we canrely on certaintrackingabilitiesto underwriteour knowledgethat ' I am ' esthe samepropositionas ' I washungryyesterday'saida hungrynow express day later. Of coursethe real challengefor any alternativetheory of indexicals ' is whetherit can handlemodal environmentsas well as Kaplan s theory can. Hence,I alsodiscusspossiblelinesof investigationthatwouldallow anEvansian themodalenvironmentsthathavebeen semanticsof indexicalsto accommodate broughtto attentionby Kaplanandothers. In chapter4, I returnto the questionof the metaphysicalcommitmentof semantic commitmentsarisewhen theory, showingexactlywherethemetaphysical and an absolute semantical one is utilizing theory exploring certainexample cases. In chapter5, I extendthe generalsemanticalapparatusto includea semantics of tensein the spirit of Reichenbach( 1947). As we will see, this semantical - in particular, it is committed theoryhasrobustmetaphysicalconsequences to aBtheory metaphysicsof time in the senseof McTaggart( 1908). In chapter6, I developthe criticism of the B-theorysemanticsby constructing the problemof indexicality. I arguethat thereareprofounddifficulties surrounding the analysisof temporal indexicals- difficulties that may well be . insurmountable
Introduction
In chapter7, I constructa semantics of tensein thespiritof Prior( 1967, 1968), some of the of the view (inparticular againexploring metaphysicalconsequences , showingthat it is committedto a versionof theA -theoryof time that Prior himself advocated ). I thenarguethat the Prioreansemanticaltheorycaneasily avoidthe difficulties with temporalindexicalsthatplagueReichenbachian theories . I thencanvassa numberof standardobjectionsto Prioreansemanticsof tenseandarguethat thoseobjectionscanbe overcome. In chapter8, I takeup thetreatmentof temporalanaphorain theA -theoryand " " . The idea is that, ratherthan developa theory of E-type temporalanaphora to time or to and future events, temporalanaphorsareimplicit referring points past when-clauses(or, moregenerally, implicit temporalconjunctions). I then extendthis ideato a numberof casesof complextemporalanaphora . Finally, I ' apply the theoryto the refutationof McTaggarts paradox. In chapter9, I expandthe investigationof therelativemeritsof theA -theory andthe B-theory, drawingin particularon datafrom researchinto the acquisition of tense. I arguethat the availableevidenceclearly leansin favor of the Prioreansemanticsandits attendantA -theoryconceptionof time. Finally, I argue that, if onewantsto takethe phenomenology of time seriously, the revisedversion of theA -theory presentedin chapters7 and 8 comportsquite nicely with at leastsomeof the leadingphenomenologies of time. In chapter1O,I takeup someof thephilosophicalandlinguisticconsequences of the theory, first pursuingthe philosophicalconsequences for the natureof logic , epistemology, and memory and then pursuing somelinguistic consequences - in particulartheprospectsfor eliminatingthelinguisticnotionof tense altogether. It is admittedlya long andtortuousjourney to get to the issuesof indexicality andtemporalanaphora , not to mentionthe relativemeritsof the A -theoryand the B-theory, but onceagaincarefulgroundworkand stagesettingwill be crucial to the argumentspresentedin the end. As I havealreadynoted, somewill want to jump aheadandseewhat the destinationlooks like. For the rest of us, the long marchbeginsdirectly.
Chapterl -The Nature of Language
1.1 I -Languagevs. E- Languagel Chomsky( 1986) distinguishes betweentwo conceptionsof the natureof language : I -languageandE-language.An I -languageis not a spokenor writtencorpus of sentences , but is rathera stateof an internalsystemwhich is part of our .2Thus, I -languagerepresentations biologicalendowment arenot to beconfused with spokenor written natura11anguage sentences . They are, rather, datastructures in a kind of internal computationalsystemwith which humansare born andwhich theyhaveco- optedfor communicationandfor other .3 purposes From the E- languageperspective , on the otherhand, a natura11anguage is a kind of social object the structureof which is purportedto be established by ' is to be 4 convention(however'convention understood ), andpersonsmayacquire in their knowledgeanduseof that socialobject. varyingdegreesof competence I gatherthat, on Chomsky's view, suchsocialobjectsdo not exist andwould ' be of little scientific interest if they did existisTo seewhy Chomsky s basic ideais right, considerthe problemof trying to individuatesuchsocial objects. For example, simply considerthe linguistic situationin Italy. We speakof "the Italian language," and we say it is distinct from Spanish, but why ? In large measure , CastillanSpanishandStandardItalian aremutually intelligible when reador spokenslowly. Why don' t we saythat they areregionalvariantsof the samelanguage ? In variousregionsof Italy, variousdialectsare spoken. For example, a different dialectis spokenin Parmathanin Venice. For all practicalpurposes , these dialectsareno moremutually intelligible thanCastillanSpanishandStandard Italian. Why do we saythat both aredialectsof Italian? Why not saythat they are separatelanguageson equalfooting with StandardItalian (and Castillan " " ), not merely dialects? Spanish
1 Chapter In Italy, somedialects(e.g., Friulian andSardinian) are recognizedasoffi ciallanguagesby the government.What makesthosedialects special? What ' makesthemlanguagesin a way that the Venetiandialectisn t? The answer, of course,is thatwhatcountsasa distinctlanguageis a political decision. (As Max " Weinreichsaid, " a languageis a dialectwith anarmyanda navy. ) Politicalidentity precedesE-languageidentity. But this isn' t just an observationabout the distinction betweenlanguage anddialect; it also appliesto the distinction betweendialect andidiolect. My wife , who teacheswhereI teachand lives whereI live, speaksslightly differently ' ' ' ' ' ' thanI do. For example, shepronouncesMary , marry , and merry differently . I pronouncethem all the same. Do we speakthe samedialect? How muchdivergenceis necessarybeforewe can saythat we do? This isn' t a caseof fuzzyboundaries , suchasthatbetweena couchanda chair. The problem is that sufficient divergencewill dependupon factorsthat have nothingto do with whatwe areactuallysayingandeverythingto do with whether we feel disposedto identify with eachother. Nor is the problem resolvedsimply by throwing out talk of E-languages andE-dialectsandrevertingto talk aboutE-idiolects. Whatportionof thenoises I makecountasbelongingto my idiolect? If I coughin the middleof a particular utterance , doesthe extranoisebecomepart of my idiolect? How do we distinguish partsof my idiolect from simpleerrors? But supposewe had an error theory. If we wrote down everythingI ever (correctly) uttered, would that exhaust my idiolect? Would everythingI everhavesaidor everwill sayexhaust Don' t we haveto considerwhat I might say or could say? But idiolect? my ? how do we makesenseof that from the E-languageperspective . Or supposethat I speakdifferently at home and at school Doesit follow that I employtwo idiolects? Sincemy speechvariesslightly with everyperson I talk to, why not saythat I havea different idiolect for eachof my conversation partners? And sinceI speakto them differently at different times, . . . . It shouldbe clear wherethis is going. Why not saythat I employ a different idi olecteverytime I openmy mouth?At thatpoint, theconceptof idiolect(andlanguage 6 ) haseffectivelycollapsed. Clearly thereis no way to sort out anyof thesequestionsif we arethinking or inscriptions. of languageor idiolectin termsof somesetof externalutterances Wecanmakechoicesaboutwhatcountsasan error or a possibleutterance,but on the E-languageperspectivethesechoicesaregoing to be stipulativeor prescriptive . The difficulty is that thereis no way we can stipulateeverythingthat we needto. Considerthe following two examplesfrom Chomsky1986:
The Natureof Language
( 1) Johnfiled everyletter without readingit. (2) What letterdid Johnfile without readingit? Somehow , speakersof English- even thoseof us who neverrememberto say ' whom' ratherthan' who' in dativecase- know thatif we deletethepronoun' it ' in thesetwo sentences the effectson meaningareratherdifferent: ' ( 1) Johnfiled everyletter without reading. ' (2 ) What letterdid Johnfile without reading? Sentence(2') is ambiguousin a way that sentence( 1') is not. Both ( 1') and(2') havethe meaningin which the filing wasdonewithout some( arbitrary) reading ' taking place, but (2 ) alsopreservesthe meaningof (2) - it can still be understood asaskingwhatletterJohnfiled withoutreadingit (thefiled letter). Who us taught that? Factslike theseareubiquitousin naturallanguage , andit would be impossible for anyinstitutionto prescribeor stipulateevena smallsubsetof them. The AcademieFran~aiseis supposedto dictatethat a certainset of propertiesthat cover the Frenchlexicon and the pronunciationof the words in it . However, the Academiedoesn't dictateas much as it thinks it does. At bestit dictatesa small rangeof superficialrulesaboutFrench. All the while, it relies on shared tacit knowledgeof I -language , which providesthe substrateuponwhich those rules are prescribed parasitic. What theAcademiehasis a collectionof proclamations about what the structureof Frenchshould be but which is not even sufficientto constitutea respectablecandidateE-language. In short, the only way out of theseimbrogliosis to give up talk of language andidiolect asa naturalobjector to think aboutthemechanismthatexplainsthe linguistic competencethat eachof us obviouslyhas! The only sensibleway to do that is to supposesomethingalong the lines suggestedby Chomsky: that our linguisticcompetence is atbibutableto aninternalcomputationalsystemthat is part of our biological endowment.That systemand the representations encoded in it areobjectsthat we canstudyandaboutwhich we canhavetheories. Anything elseis just vaporware.
1 Chapter 1.2 What Is I -LanguageFor? " for the The question" What is I -languagefor ? hasprofound consequences in advanced attitudes of the and for indexicals of theory propositional theory viewthe standard On ordinarily presupposing subsequentchaptersof this book. anE-languagepictureof language- it is assumedthatthepurposeof language is communication, and henceit is assumedthat a semanticaltheory (including the theory of indexicalsand propositionalattitudes) must respect . the communicativefunction of language ' ' In the caseof indexicals, for example, an utteranceof I am hungrynow is takento havea componentto its meaningthatholdsat all timesandfor all speakers ). (i.e., that the speakerof the sentenceis hungryat the time of its utterance " of the indexicalcharacter " nowness . the e. more ( g, Allegedly, insertinganything of the utterance) into the meaningwould throw off your communicative partnerswhen you left messageson their answeringmachines. The communicative function of languagethusrequiresthat propositionalcontentnot vary from individual to individual or from time to time. Thereis probablyroomto arguethatthecommunicativefunctionof language (if it exists) doesnot put such seriousconstraintson propositional content. However,it seemsto methatthebasicpremiseitself needsto becalledinto question . Specifically, oncewe supposethatthe I -languageperspectiveis correct, it immediatelyraisesquestionsaboutwhy I languagesshouldbefor communication - or , for that matter, why they shouldbefor anythingat alIiS The issuecanbe put in a generalway. If onetakesthe E-languageperspective that languagesare establishedby humanbeingsby convention, it is entirely sensibleto supposethat languagescanhavea certainend. In this casethe endis establishedby humanintentions. Personssimply agree(perhapstacitly) to follow certainlinguistic conventionsin orderto further the endof communicating with oneanother. But if we reject the E-languagepicture and adoptan I -languagepicture in , humanintentions which the languagefaculty is part of our biological endowment ' t enterinto the -languageis to have I individual . If an don equation simply ' somesort of end, it can t be establishedby humanconvention; it must be established by biology. Herewe areon very tricky terrain. In the first place, teleologicalexplanationin biology is a controversialdoctrine . Thereis a strongtemptationto appealto evolutionarytheoryto establish thepurposeof a biologicalsystem(for example,by arguingfor selectionalvalue of a faculty to which we want to attribute somepurpose) . But this requires somestrongassumptionsaboutevolutionarytheory- in particular, it requires
The Natureof Language
a gradualistapproachto evolution. In the caseof attributinga communicative endto language , it alsoassumesthat we havea faculty dedicatedto communication (asopposedto just an ad hocbundleof abilitiesthat we employin order to communicate ). Theissueaboutgradualismis, of course,familiar from recentwriting on evolutionary theory. On oneside, StephenJayGould et al. hold that humanevolution hasbeenfar from gradualandthat in manycasesorgansthat evolvedfor certainpurposes weresubsequently co- optedfor quitedifferentpurposeS .9Humans are not uniquein this respect; evolution is typically of this character.Hence, for example, membranes thatevolvedfor thermalregulationmight be co-opted for the purposeof flight . In the contextof the languagefaculty, the ideais that we cannotreasonbackwardfrom thecurrentwaysin which we usethelanguage faculty to the conclusionthat the languagefaculty evolvedfor thosepurposes . And evenif we could, thereis no reasonto supposethat the languagefaculty is . Biology, asChomskyandLasnik therebyanoptimalsystemfor thosepurposes "' ' ( 1993) haveobserved , is typically messy, intricate, the resultof evolutionary ' ' andby physicalconditions tinkering, andshapedby accidentalcircumstances thatholdof complexsystemswith variedfunctionsandelements ." Perhaps , then, I -languageis a very imperfectsystemfor purposesof communication . More to thepoint, its handlingof indexicalexpressions maynotcoincidewith philosophical intuitionsabouthow anoptimal systemof communicationoughtto be built. Ontheothersideof thisissue,PinkerandBloom( 1990) havearguedthatevolution is in fact a muchmoregradualprocessandthatit makesgoodsenseto think of the languagefaculty ashavingevolvedin responseto the selectionaladvantage of communication . But evenif we accepttheirarguments for gradualismand thethesisthattheI -languagefacultyevolvedin response to selectionalpressures , it doesnot follow that I -languageevolvedfor purposesof communication . For , it is still possiblethatit evolvedin orderto serveasa mediumof thought example (seesection1.3 andappendixPI ). Or, in view of the rich metricaltheoryfound in variousI -IanguagesiOit could evenhaveevolvedfor purposesof producing poetic soundpatterns. Clearly thereare advantagesto having communicative abilities; however,thereareadvantages to havingotherabilities, andif oneis to backward from the existence of I argue languageto someselectionaladvantage onemustarguecarefullythatthe main advantageis its communicativefunction andnot someotherfunction. In summary:Thelanguagefaculty (I-language ) is not a productof humanintentions but rathera productof humanbiology, so any end that we attributeto I -languagemustbebiologicallybased.This meansthatanargumentfor thecommunicative function of languagewould requirean evolutionarystory aboutthe
Chapter1
selectionaladvantageof having a languagefaculty for purposesof communication . This strategyassumesa controversialview aboutthe natureof evolutionary theory; further, it assumesthat communicationis the only (or the chiet) . Finally, evenif we wereto assume thatI -languageevolved functionof I -language for the expresspurposeof humancommunication , it is a very long distanceto the conclusionthat the propositionalcontentof an utterancemustcontainonly infonnation that is stableamonglanguageusers. Evolutionarytinkering may well havefound an alternativestrategy.
1.3 Is I -LanguagetheLanguageof Thought?II in thisbook, it turnsout Althoughit is notcrucialto thelineof argumentation of thoughtwouldhelp andthelanguage betweenI-language thatequivalence . usto cutsomecomers Justto beclear, we needto distinguishthought(or, better, thehavingof thatall human . No onewouldsuppose ) from cognitionin general thoughts . Thequestion that is not what is at issue in I but takes , cognition place language abouttheworld, arethosethoughtssimply is this: Whenwehavethoughts in our tokens I -language , or aretheytokensto befoundelsewhere interpreted ? cognitivearchitecture thiscouldeasilyslideintoa terminological Because disputeoverwhatgets of argument tobecalled"thought," I amperfectlywillingtoadmit(forpurposes ) interest is in . thatmy definitionis stipulativeMy primary cognitivestatesthat I amcalling of thisinvestigation purportto beabouttheworld, andfor purposes " However " of defining on the side there is a fair bit of . suchstates , history thoughts ' sclaimthatintention with Franz Brentano in this way, beginning thoughts a broad I am not mental . But the mark of the , advocating ) is again ality(aboutness between anequivalence andcognition;I amsuggesting of I -language equivalence states and and those certainI-language thoughts cognitive representations thatareabouttheworld. ? Prima of thought bethelanguage Withthisclarification , couldI-language not, on thegroundthatevenif thereis a closeconnection facie, onemightsuppose of thought andrealitytheremustbesomelanguage between language this?Indeed andtheworld. Butwhysuppose I-language between thatmediates , me as redundant strikes medium of an of thepostulation intermediary thought atbest. andunmotivated : If thereis a for thisreason of a thirdmediumis redundant Thepostulation of thoughtLar) , then ontothelanguage thatmapsfromI-language mechanism eithertheLOTmustbeisomorphicto I -language (in whichcaseit cando no
TheNatureof Language
23
morethanI -languagecan) or the propertiesof the LOT arerecoverablefrom 1via somealgorithm. In eithercase, I -languagerepresentations languagerepresentations all for servingasthelanguage alreadycarry theinformationnecessary of thought. Oncewe movefrom theE-languageconceptionof languageto theI -language conception, most of the stock argumentsagainstnatural languagebeing the languageof thoughtcollapse.Forexample,Jackendoff( 1993) andPinker( 1994) -as-LOT hypothesis raise severalargumentsthat seemto leavethe I -Ianguage ' untouched . I ll considerthesetechnicalargumentsin appendixPI , but someless formal argumentsdue to Pinker (ibid., p. 68ff.) are worth exploring here. For " example, Pinker notesthat babiescannotthink in words becausethey have " " not yet learnedany, that monkeys. . . cannotthink in wordsbecausethey are " incapableof learningthem, andthat manyhumanadultsclaim to do their best thinking without words. Clearly theseargumentshaveconsequences for any thesisthat I -languageis the languageof thought. If humanbabiesandchimps are thinking thoughtsbut haveno I -language, that certainly underminesany identificationbetweenI -languageand thought; likewise if humanadultshave thoughtswhich arenonlinguistic. But aresuchthingspossible? Oneproblemwith babiesis knowingpreciselywhentheybeginhavinggenuine thoughts(in the sensespecifiedabove) . Thereis no questionthat cognition is takingplacein the womb, but whendo babieshavethoughtsaboutstates of affairsin theworld? I know of no illuminating developmentalstudieson this question. It is onething to find evidenceof a conceptof number, or of the permanence of objects, or of the ability to identify one' s mother, but by what line of argumentcanit be shownthat thesearegenuinerepresentational thoughts? And insofaraswe areinclinedto call thesethoughts(andthe moresophisticated andworld-relatedthe cognitionthe moreapt we areto do so) it becomes lessclear that babiesdon' t havethe relevantI -languagerepresentations . For example, linguistic comprehensionprecedesproductionby a goodbit (as any parentknows). On thetheoryof I -languageadoptedhere(from ChomskyI 995b), sentences areconstruedasorderedpairsof representations , {PF, LF} , in which -articulatory" component the PF representations interfacewith the " perceptual -intensional" system. andthe LF representations interfacewith the " conceptual It is entirely possiblethat the I -languagesystemis intact in babiesbut that the interfaceof PF representations with the perceptual-articulatorycomponentis not yet developed . aboutintelligentadults By theway, thispossibilityalsoundenninesarguments with no apparentlinguistic abilities. Suchindividualsmaywell havefull -blown linguistic abilities, which may well be servicing the conceptual-intensional
1 Chapter . The opposite system, yet be incapableof speechproductionandcomprehension true: an individual be and also be may producing may respondingto wellbut if the LF arenot being formednaturallanguagesentences , representations utilized asthoughts(thatis, if theyarecausallyinert in plansandactions) he or shewill appearto havelanguagebut no thoughts. In effect, suchan individual would havecausallyinert thoughtforms. The caseof animalsraisesmoresubtleissues. On the onehand, we want to " " say Fido thinks his dish is empty, andpresumablyto saythis is to ascribea thoughtto Fido. It is certainly a cognitive stateaboutthe world. On the other to beuniqueto humans.SohowcanFido' s thought hand, I -languageis supposed ?CouldFido (or at leastchimps) have beaninterpretedI -languagerepresentation ' rudimentaryforms of I language? Here, I tendto be in Pinker s camp. The evidence thatnonhumananimalslackanythingremotelylike our language suggests comethey all think? . So how faculty thatnatureprovidesonly oneway Thequestionis vexingonly if onesupposes to solvea problem. For example,we know that variousmarineanimalsemploy variousmethodsof propelling themselves(contrastthe squid with the manta ray), yet it is perfectly correctfor us to saythat they all swim. We also know thatnatureprovidesdifferentwaysfor animalsto be sexedmaleandfemale(our X -and-Y-chromosomestrategybeinga minority strategy), yet we arenot blundering " whenwe say"That is a malechicken" or "That is a femalespider. The ' ' ' ' point is that the terms swim and male covera broadclassof phenomenaas ' ' we movefrom speciesto species.What I wantto suggestis that thinks is a bit like this. We areperfectlywithin our rights to saythat Fido is thinking, just as we arewithin our rights to saythat Samthe squidswimsandthat Charlottethe ' spideris a female; wejust haveto keepin mind that Samdoesnt swim like we do, that Charlotte is sexedby different biological mechanismsthan we are, and, crucially, that Fido thinksby differentmechanismsthanwe do. Fido maywell havea languageof thought. His languageof thoughtmayhave somerudimentarysyntax, and he may (as the Greekphilosopherssupposed ) . But his languageof thoughtisn' t human becapableof someinferentialreasoning I -language. It ' s that simple. It is entirely reasonableto supposethat Fido and otheranimalsco- opt othercognitivesystems(with relevantsyntacticproperties) to do their thinking. They may evenhavefacultiesthat evolvedfor that express purpose.But humansappearto be different. It is entirely possiblethat we have co-optedI -languagefor the purposeof thinking (or perhapsit evolvedfor that ). In any case, the rudimentary purposeandhasbeenco-optedfor communication our useof I -languagefor not tell the do Fido and of against chimps thoughts the kind of robustthinking of which we arecapable.
The Natureof Language
In regardto adulthumanswho purportto havenonlinguisticthoughts,Pinker givesseveralanecdotalstoriesaboutfamouscreativeindividualswhoclaimsome role for imageryin their thinking. Of course, for everysuchanecdotethereis an anecdotethat deliversjust the oppositeconclusion. One famousexample, reportedin Monk 1996, is the caseof BertrandRussell, who claimedto be incapable of thinking imagistically.12Clearly there are individual differences concerningour useof imagesin cognitionandcreativeproblemsolving, but this doesn't entail individual differencesin the languageof thought. Evenif weconcedefor thesakeof argumentthattherearementalimages,and thattheseimagesarefundamentallypictorial andnot text-based,13we still have to beon guardagainsta fallaciousstepin reasoning . It is surelytruethatimages canbe usefulin our thinking. Doesit follow thatour thoughtsareimagistic? For example, mapsare very useful in planninginvasionsand family vacations , but whenwe aredeliberatingwhetherto go north or southon Thesdayit doesnot follow that our thoughtsarein any way composedof maps. Likewise, in working out an argumentI may find it usefulto doodlepictureson a pieceof paper, but no one would supposethat I am having" doodlistic" thoughts. Why shouldwe chooseto identify thoughtsandimageswhenthe imageshappento be mental? I may dreamof a snakebiting its tail , andthis may inspire a thoughtabout the structureof carbon-basedmolecules, but it doesnot follow that the dream andits attendingimageconstituteda thought. The thoughtcamelater. This fact is actuallyevincedin Pinker' s accountof how RogerShepardcameup with his famousimage-rotationexperiments : betweensleepandawakening Earlyonemorning,suspended in a stateof lucidconsciousness a " spontaneous , Shepard kineticimageof three-dimensional images experience ." Withinmoments andbeforefully awakening majestically turningin space , Shepard hada clearideafor thedesignof anexperiment . (Pinker1994,p. 71; emphasis added ) I am perfectlywilling to concedethat the kinetic imagehelpedShepardcome to havethe thoughtthat he did. Somethingmustinspireus to havethe thoughts that we do. Weblunder, however,whenwe supposethat theseinspirationalimages mustbe thoughtsthemselves . In any case, we haveknown for nearly400 yearsthat imagesare very poor candidates for thoughts.Descartes observedthatalthoughwehavethoughtsabout -sidedfigures) we do nothaveimagesof them. Even chiliagons(thousand among theBritish Empiricists, theequationof imageswith thoughtswasunravelingby the time of Berkeley. It washe who observedthat vaguethoughts(say, about the conceptof a triangle) are often associatedwith a very clear image(say, a
Chapter1
red equilateraltriangle), andthat thoughtsaboutparticularindividuals(say, a with a very vagueimage. long-lost friend) canbe associated -grained Moreover, asFodor( 1975) observed , imagesappearto be too coarse to serveasthoughts. I may havethe thoughtthat SantaClausis bearded, and this may well be accompanied by an imageof Santa; however,that imageis the sameimagethat accompaniesmy thoughtthat Santais red-cheeked.Herewe haveone imageandtwo thoughts. Conclusion: the thoughtscannotbe identified with images. If imagesdon' t work, could somethingelsebesideslanguageserveas the mediumof thought? One idea, advancedby Johnson-Laird ( 1983), is that we " " work. Mental mightintroducenonlinguistic mentalmodels to do thenecessary modelscanbe asfine grainedaswe please.The only problemwith this family of strategiesis that suchmodels, while often presentedin pictorial form, necessarilyhavea syntacticstructurethat is crucial to our interpretationof them (Rips 1986). Moreover, they aresuccessfulonly insofarastheymimic the syntactic form we would useto expressthe thoughtsthey are intendedto model. That is, our expressionof ~ thoughtis conceptuallyprior to the constructionof a modelfor the thought. It is certainlyanempiricalquestionasto whatthe languageof thoughtmust look like, butthereis a heavypresumptionin favorof theI -languagealternative. Thenonlinguisticalternatives , suchasimagesandmentalmodels, fail to getout of the blocks, andoncewe areworking with an I -languageconceptionof language any intermediarylanguageof thoughtis simply redundant. Obviously there is a lot more to be said about the identification of I -language with the languageof thought. I don' t presumeto havegivenknock-down evidencefor thethesishere(or evenin appendixPI ). Rather, my goal hasbeen to suggestits plausibility. For thosewho find the thesisplausible, someof the argumentsin this book will unfold muchmorerapidly.
Chapter 2 The Fonn of the Semantic Theory
2.1
The Nature of Semantic Knowledge
Oneof the key assumptionsof this work is the ideais that part of thejob of semantics is to characterizethe semanticknowledgethat an agenthas. " Semantic " knowledge is not intendedto suggesta thesislimited to how an agentrepresents meaningto himselfor herself. Rather, it is assumedthat agentsstandardly haveknowledgeaboutthingsin the world (tables, chairs, etc.), andthat a great dealof semanticknowledgeis a speciesof this sort of knowledge. So, for example , a speakerof Englishmight knowthattheword ' snow' refersto snow(the stuff in the world), andthis bit of knowledgewould be an exampleof genuine semanticknowledge. Talk of " semanticknowledge" hasa tendencyto setoff alarmbells among , no doubtbecausethesephilosopherstake ' knowledge' to be referring philosophers to a Cartesianconceptionof knowledgeI- knowledgethattheagentcould havecompletelyindependentlyof anyenvironmentalrelations. Thatis not what is intendedhere. The ideais that onecanperfectlywell haveknowledgeof the externalworld (asnotedabove) andthat onecanknow thingsthat link expressions of the languagewith thosethings in the externalworld; semanticrules beingcasesin point! It will alsobe assumedthat muchof an agent's semanticknowledgeis "tacit " knowledge- knowledgewhich the agentmay not assentto having, but knowledge which nonetheless underwritesabilities which the agenthas. This sort of knowledgehasbeenproposedin syntactictheory, whereit is arguedthat tacit .3 knowledgeof certainlinguistic principlesexplainsour linguistic competence Analogously, it will be assumedherethat tacit knowledgeof a numberof semantic rulesunderliesour semanticabilities. Varioustheoriesof tacit knowledgearepossible(e.g., tacit knowledgeasa dispositionto act in a certainway), but hereI shall be adoptingthe view that
2 Chapter genuine tacit knowledge is in some sense represented by the knower in the
is asemi-stablesyntactic fonDof a datastructure- thatis, thattherepresentation ' statein the agents rnind/ brain. Again , the claim is not that the object of semantic knowledge is the representation . Rather, it is that the object of semantic knowledge is related to the representation ' ' in an interesting way . So, for example, my knowledge that snow refers to snow does not consist solely in my representation of snow, since the representationitself is a syntactic object. Rather, the object of my semanticknowledge is snow itself , since that is what my representation determines.
? 2.2 Why anAbsoluteSemantics So far , I have discussedsemantic theory in very general terms and have left open the question of what form such a theory should take. In this section I will argue " " that semantic theory should take the form of an absolute truth theory, in the form of a truth -conditional semantics. This general approach is controversial , but for the most part the controversy has been due to misunderstandings. The controversy has been fostered in part by a failure to articulate the difference between the goals of absolute truth conditional semantics and the goals of alternative semantic theories. I will therefore begin by trying to get clear on how an absolute semantics differs from the other candidates, and why an absolute semantics is best suited for the kinds of concerns that arise in the semantics of natural language. Let me begin by identifying three broad classesof semantic theories: . structural semantics . model -theoretic semantics . absolute truth -conditional semantics. As indicated above, I will be defending the third type of semantic theory . To see why , we need to take a closer look at the other two alternatives.
Structural Semantics Structuralsemantictheorieshavebeenadvocated by Katz( 1972), KatzandFodor ( 1963), Katz andPostal( 1964), andJackendoff( 1972, 1987). The key ideabehind all theseproposalshasbeenthe notion that providing a semanticsfor a natural languageexpressionrequiresthat one provide a mappingof that expression ontosomerepresentational language(possiblythelanguageof thought). For example, in Katz 1972, Katz and Fodor 1963, and Katz and Postal 1964 " " theideais to mapa givennaturallanguageexpressionontoa semanticmarker,
The Fonn of the SemanticTheory
which is in turn a symbolin a particularrepresentational language.(Following " . In turn Lewis 1972, we cancall this language" SemanticMarkerese ) , Semantic Markereseis designedsothatambiguousexpressions of a givennaturallanguage aredisambiguated in Markerese . In addition, entailmentrelationsandsynonymy relationsbetweennaturallanguageexpressions aresupposedto follow by virtue of theforms of their SemanticMarkeresecounterparts . Lewis ( 1972) raisedanimportantandwidely acceptedobjectionto sttuctural semantics . Accordingto Lewis, if we providea mappingof Englishexpressions ontotheexpressions of Markerese of Markeresewill still stand , theexpressions in needof interpretation . Whatwe haveat bestis a mappingfrom onesystemof notation onto another, but what we wantedwas someindication of what the systemof notationis about. Lewis' s point canalsobecharacterizedasa point aboutthe semanticknowledge that we have. We all cometo know the referenceof variousEnglish expressions , andit is naturalto expecta semantictheoryto characterizewhatit is that we havecometo know. But noticewhat kind of knowledgeis attributedto the speakerwhenall one hasaremappingsinto Markerese.As Lepore( 1983) notes, at bestonegetstheoremslike ( 1). ( 1) ' ' Fragola in Italian translatesinto the languageof SemanticMarkereseasS. We might evenfollow currentfashionandrenderthe expressionsof Semantic Markereseascapitalizedversionsof their naturallanguagecounterparts , asin ' ( I ). ( 1') ' ' Fragola in Italian translatesinto the languageof SemanticMarkereseas FRAGOLA. But clearly neitherof thesetheoremscharacterizes what a speakerof Italian is ' ' sayingwhenhe or shesays fragola . That is, thesetheoremsdo not tell us that ' ' fragola meansstrawberry. This generalpoint is not particularlycontentious.Evenadvocatesof sttuctural semanticshaveessentiallygrantedthatthekind of semantictheorytheyare offering will not yield a connectionbetweenlanguageandthe world. What is contentiousto the advocatesof sttucturalsemanticsis the questionof whether -world thereis anyreasonfor a semantictheoryto characterizesucha language connection.Katz ( 1972, p. 183) hastakenpreciselythis tack, arguingthat it is not the businessof semanticsto " study the relationsbetweenobjectsof one sortor anotherandthe expressionsof a languagethat speakaboutthem."
Chapter2 There is no point in debating the question of which enterprise gets the honorary title semantics. There may well be an important place for the type of enterprise advocated by Katz , Jackendoff, and others. My main concern is to note that such an enterprise will not addressthe sorts of concerns raised earlier in this chapter and hence cannot exhaust our interest in language. In particular , such an account cannot help us to characterize the very basic knowledge that underlies our ability to know what the expressions of our language are about. Some readers may find this discussion old hat , but for all its familiarity it seemsto me that the moral has been lost on many semanticists. As we shall seein the next section, precisely the sameconcerns can be raised against model theoretic semantics.
Model-Theoretic Semantics Manyproponentsof model-theoreticsemanticshavearguedthatit canovercome thelimitationsof the structuralsemantics- that model-theoreticsemanticscan success fully characterizethe relationbetweenlanguageandthe world. Thequestion,aswe shallsee, is whethera model-theoreticsemanticsdoesin -world connections . Specifically, it hasbeenobserved fact provideus with language othersthatmodel-theoretic and 1990 1983 ), ( ), ( Higginbotham by Lepore semanticsfalls victim to the sameconsiderationsthat havebeenraisedagainst theories. Let' s seewhy. structuredrepresentation A model-theoreticsemanticsmight characterizethecontentof anexpression asin (2). (2) ' ' ' The contentof the Italian word ' fragola is the extensionof fragola in every possibleworld. ' ' Alternatively, onemight find fragola characterizedasin (3). (3) ' ' ' The contentof the Italian word ' fragola is that which fragola is true of in everymodel. , a nativespeakerof Englishwhoknows But, asHigginbotham( 1990) hasstressed and the Italian a bit of meaningsof virtually no Italian words, could syntax only while know both (2) and(3) knowing essentiallynothing aboutthe contentof ' . Thekind of ' knowledgesucha speakerwouldnothave,butwhichwould fragola ' be crucial to understanding' fragola, is expressedby (4). (4) ' ' . fragola refersto strawberries
The Fonn of the Semantic Theory
The point to be stressedhereis that model-theoreticsemanticsfails in exactly the sameway that structuralsemanticsfails. What we expecta semantictheory to deliver is a characterizationof the speaker's knowledgeaboutthe connection betweenlanguageand the world. Structural semanticsdoesnot deliver this becauseit only gives us a mappingfrom one languageonto someother . Model-theoreticsemanticslikewise fails, becauseit systemof representation nevermakestheconnectionbetweenlanguageandthe world; at bestit givesus a mappingfrom expressionsof a given languageonto certainmodel-theoretic objects. It might be objectedthat we canalwaysspecifyan intendedmodel, andthat theactualworld shouldbetheintendedmodel.4But Lepore( 1983) hasobserved thatthis movewill not work. If thereis to be anintendedmodel, thenwe should like to know how to identify it. But, on the faceof it , this is no easytask(ibid., p. 184): Howmuchabouta worlddo we needto knowbeforewecandistinguishit from all otherworlds? Presumably a lot. Therepresumably is a classof worldsin whichthe number of treesin Canada isevenandonein whichthenumber is odd. Sofararewefrom beingableto singleouttheactualworldfromall othersthatwedonotevenknowwhich classit fallsin. Butdoweneedto distinguish theactualworldfromall othersto understand ourlanguage ? However , I thinkthereis a wayof specifyinganintendedmodel.Whatis needed is a recursivetheorythat will tell us that ' snow' is intendedto refer to snowin the model, ' white' is to be true of white thingsin the model, andsoon. Suchan intendedmodel must also specify the intendedinterpretationof complexexpressions in the model- indeedan infmite numberof them. Therearetheories that cando this; after all, that is preciselywhatT-theoriesdo! In effect, talk of intendedmodelsamountsto a kind of handwavingat preciselythe point where T-theoriesarerequired. Noneof the aboveis to suggestthat model-theoreticsemanticsis uselessor flawedin anysense . Whatis suggested is thatclaimsmadeon its behalfareoften excessive . In particular, onemust be wary of claims that a model-theoreticsemantics -world connections canprovidea theoryof language .s
AbsoluteTruth-ConditionalSemantics - thatof Davidson( 1967a OnthetheoryI will bepursuing theory ) - a semantic is a systemof rulesthatyieldstheorems of thefollowingform: (T) s is trueiff p.
Chapter2
Heres is an expressionin the object language(i.e., the languageaboutwhich the investigatoris theorizing), andp is an expressionin the metalanguage(the might languagethe investigatoris usingto statethe theory). The metalanguage be English, or it might be someother language.Of course, sincethe theory is hasto be a languagethat the theoristalreadyunderstands statedin it , the metalanguage . The schemaabove(sometimescalled a T-schema ) saysthat the sentences ' ' is true if andonly if it is the casethat p. The phrase is true if andonly if is actually a very weakconstraint. What it meansis that eithers andp areboth true or they areboth false. For example, the systemof rulesmight yield theoremslike the following:
(i) ' Snowis white' is trueiff snowis white. (ii ) ' Snowis white' is trueiff . grassis green Systemsof rules ( T-theories) that yield only theorems like (i ) are said to be interpretive , becausethe expressionsin the metalanguagesuccessfully give the contents of the expressions in the object language. T-theories that yield theorems like ( ii ) are said to be non- interpretive T-theories. Here I will be interested only in interpretive T-theories, i .e., those that deliver the contents of the object- language expressions. I assumethat those theories are the ones that speakersexploit in the generation and comprehension of language. As it turns out , this assumption has some strong consequencesfor the origin and nature of our semantic knowledge . For example, it follows that we must have knowledge of a certain class of constraints on T theories. If we did not , we would be unable to construct interpretive T- theories. Of course we must know more than just the rules of an interpretive T-theory. We must know that our theory is 6 interpretive and that it can be exploited in the comprehensionof language. Larson " " and Segal ( 1995) argue that it follows that we must be hard wired to construct interpretive T- theories and to exploit them in language comprehension. For our current interests, the key idea in aT -theory is that one can have a theory that will provide the language- world connections for any expression of the language (or part of language) under study. Typically , this is accomplished via an axiomatic theory in which certain axioms give the referents of the tenninal symbols of a sentential phrasemarker and other rules show how the semantic values of the tenninal symbols contribute to the semantic value of the nontenninal nodes of the phrase marker. It will be helpful to consider a concrete example of a T-
The Form of the SemanticTheory
will suitourpurposes theory.Thefollowingexamplefrom LarsonandSega11995 in this . nicely regard Sincewe will beconsideringa relativelysimplefragmentof naturallanguage , the syntaxwill be describedby a context-free phrase-structuregrammar,using rewriting rulesof the fonn A ~ BC Roughly,thisrulestatesthata nodeof categoryA mayhavethedaughters B andC:
A ~ / / / / """" B C
In this and subsequent fragments , we will proceed by specifying the syntax and then stating the semantic (interpretive) rules for the fragment .
SyntaxS -+ SI andS2 S -+ SI or S2 S --+it is notthecasethatS1 S --+NPVP NP--+~
,~
VP--+~
, ~.@ !b
Semantics Herewe introducethe predicateVal(A , B), which is to be readas "A is the semanticvalue of B." .The theory will include two kinds of axioms: axiomsthat assignsemanticvaluesto the lexical items (i.e., words) and axioms that showhow we candeterminethe semanticvalueof a mothernodeA , giventhe semanticvaluesof its daughternodesB andC. Examplesof axioms for the lexical itemsaregivenin (5).7 (5) a. Val(x, mg ) iff x = Dick Val(x, ~ ) iff x = Sally b. Val(x, ~ ) iff x leaps Val(x , ~ ) iff x walks
2 Chapter
34
nodesaregivenin (6) and(7). of axiomsfor thenontenninal Examples (6) a. Val( T, [sNPVP]) iff for somex, Val(x, NP) andVal(x, VP) b. Val(x, afi]) iffVal(x, fi) (wherea rangesovercategories , andfi rangesover items and lexical ) categories (7) a. Val( T, [s81~ 82]) iff it is boththecasethatVal(T, 81) andVal(T, 82) b. Val( T, [s81QI82]) iff eitherVal( T, 81) orVal( T, 82) c. Val( T, [sit is notthecasethat81]) iff it is notthecasethatVal(T, 81) valueof the8 nodein thestructure Forexample , rule(6a) showshowthesemantic s ' / NP
" " " VP
is detennined by the semantic values of the NP and the VP. The rule says, in effect , that the semantic value of S will be true just in case there is something that is both the semanticvalue of the NP and the semanticvalue of the VP. Similar considerations apply to the rules in (7) . For example , (7b ) tells us that in the structure
5 " / 1 " " " or 52 51 the semantic value of S will be the true just in caseeither,theTrueis the~emantic value of S 1 or the True is the semantic value of S2.8
The ideais that we canusetheseaxiomsto derivethe truth conditionsof a sentenceof the objectlanguage.In orderto carry out thesederivations,we will needto introduceproductionruleslike thoseshownin (8) and(9) .
TheFoInl of theSemantic Theory
35
(8) ProductionRule (SoE) ...a ... aiffp therefore . . . 13.. . ( Thebasicideaof this rule is that we can substitutep for a at any stepin a derivation , just in casea iff p holds.)
(9) Production Rule(Soi)
x, x = (Xand. . . x . iff forsome
. . .
therefore . iff .. .a ... It is important to keepin mindthatthesearenotlogicalrules,andthatthesteps in eachderivationaremuchmoretightlyconstrained thantheywouldbeif the full resources of logicwereatourdisposal . Forexample , wearenotallowedto substitute unlesstheyhavebeengenerated in a logicallyequivalent expressions .9Thesearesimplyrulesfor derivingT-theorems previous stepin thederivations . " " Givensimpleruleslike (8) and(9), andtheaxiomsabove one can , prove T-theorems of theaxiomsto thestructural by recursive application description of theobject-language sentence . To seehowthiswill work, considera simple 'Dickwalks : thederivation forthestructural of thesentence example descriptions ' or it is notthecasethatSallyleaps: Val(T, [S[S.[NP Dick][vpwalks]] or [S2it is notthecasethat[S3 [Np Sally][vp ]]])) iff . . . leaps ( 1) eitherVaI (T, [s.[NPDick ][vpwalks]]) orVaI( T, [S2it is notthecasethat[S3 [NP )])) Sally][vpleaps of (7b)] [instance (2) eitherVaI(T, [s.[NpDick walks]]) or it is notthecasethatVaI( T, [S3 ] Vp [NP ])) Sally][ypleaps [from( 1) and(7c), by SoB] (3) eitherfor somex, Va1 (x, [NPDick ]) andVa1 (x, [ypwalks]) or it is notthecase thatfor somex, Va1 and Va1 x (x, [Np ( ,[ypleaps Sally]) ]) of SoB] [from(2) and(6a), by applications
2 Chapter
36 (4) eitherfor somex, Val(x, I } ig ) andVal(x, ~ ~ somex, Val(x, Smly) andVal(x, ~ ) [from (3) and(6b), by applicationsof SoE]
or it is not the casethat for
(5) eitherfor somex, x = Dick andx walksr it is not the casethat for somex, x = Sally andx leaps [from (4), (Sa), and(5b), by applicationsof SoE]
(6) eitherDickwalksorit isnotthecasethatSallyleaps [from (5) by Soi] This fragment, obviously extremely simple, is introducedhereto give some flavor of how theoremsarederivedin aT-theory. In thetechnicalappendixeswe will examineT-theoriesof greatercomplexity. Of course, it is crucial to the project that we establishthat aT -theory can . Crucially, the theoremsof a truth-conditional deliverlanguage-world connections . That is, theyexpressrelationsbetween semanticsarealsodisquotational expressionsof the object languageand the world. For example, ( 10) tells us that the Englishexpressionspelleds-now - #-i -s- #-w-h-it -e is true iff snowis white. ( 10) ' snowis white' is true iff snowis white. like ( 10) straddlelanguageandthe world, describingthe relation Thus, statements betweenthe two. As LeporeandLoewer( 1981, 1983) note, sinceT-theoriesaredisquotational , . semantic that knowledge intuitively accompanies they support reasoning Considerhow a disquotationalT-theoremlike ( 10') allowsinferencesfrom a fact aboutlanguageto a fact aboutthe world, andfrom a fact aboutthe world to a fact aboutlanguage. ' ( 10) ' der Schneeist wei ' is true iff snowis white (i ' ' der Schneeist weip is true iff snowis white ' ' der Schneeist weip is true therefore: snowis white ' ' der Schneeist weip is true iff snowis white snowis white therefore: ' der Schneeist weip' is true
TheFormof theSemantic Theory
37
Not all theorieshavethis feature, however.We sawabovethat structuralsemantics andmodel-theoreticsemanticsdo not havethis property. We caneasily imagineothersemantictheoriesthatwouldnotbedisquotational , e.g., theories which issuetheoremslike ( 11). ( 11) ' der Schnee ist weip' is true iff ' snowis white' is true. ' By itself, ( 11) doesnot tell usenoughabouttheexpression'derSchneeist weiJ3 , for one could know ( 11) but not know the conditions under which it is true. Likewisewith ( 12). ( 12) ' der Schnee ist wei~' is true iff the Englishtranslationof ' der Schneeist ' wei~ is true. Also, it will not do to simply translatethe object-languagesentenceinto some . So, for example, ( 13) would not be disquotational. uninterpretedlanguage ( 13) ' snowis white' is true iff SNOWIS WHITE is true. Exampleslike this simply fail to deliver the semanticvaluesof expressions , becausethey arenot disquotational . As obviousasthis featureofT -theoriesmayappear,therearethosewho have -world connections thatT-theorieslack the ability to makelanguage suggested . For example,ChierchiaandMcConnell-Ginet ( 199O , p. 82) appearto hold that suchconnectionscould not be delivered: Whata Tarski-styledefinitionof truthdoesis to associate sentences with a description of theconditions underwhichtheyaretruein a certainmetalanguage . It therebyseems to shifttheissueof meaning fromtheobjectlanguage to themetalanguage withoutreally is. tellinguswhatmeaning " object However,it is not accurateto saythataT-theoryconsistsin " associating -languageexpressionswith . Rather, the idea is metalanguageexpressions that thetruth conditionsarestatedin the metalanguage . The contrastis equivalent to that between( 14) and( 15). ( 14) ' snowis white' is true iff snowis white. ( 15) ' snowis white' is ' ' associated with themetalanguage expressionsnowis white . Clearly the latter sort of claim is lessinformative than the former. Crucially, it fails to be disquotational, and thereforefails to provide a language-world
Chapter2 connection. But, just ascrucially, aT-theorydeliverstheoremslike ( 14) rather than ( 15) . -world connections In additionto providinglanguage , T-theoriesarecompositional . As both ChomskyandDavidsonhavestressed , speakerstypically have . How canthis knowledgeof the meaningsof an infinite numberof expressions ' be, giventhat we havefinite memories?Wepresumablycan t representaninfinite numberof rules. The ideabehinda compositionalsemantictheory is that it mustbepossibleto statetheagent's theoryof meaningin termsof a finite number of axiomsplusa finite numberof rulesfor determiningthemeaningsof longer from the meaningsof their parts. Youknow the meaningsof a finite expressions numberof words, andyou know somerules for combiningwords into mean. Theresultwouldbeknowledgeof aninfinite numberof sentences ingful expressions from a finite base. Put anotherway, insteadof attributingknowledgeof an infinite numberof T-sentencesof fonD shownin ( 16), we can attributeknowledgeof axiomsfor eachof the lexical entries, and then combinatorialaxiomsthat showhow the canbe derivedfrom the contentsof the parts. contentsof the sentences ( 16) a. ' the cat died' is true iff the cat died. b. ' the cat that atethe rat died' is true iff the cat that atethe rat died. c. ' the cat that atethe rat that lived in the housedied' is true iff the cat that ate the rat that lived in the housedied.
2.3 Modestvs. Robust1ruth Theories Dummett ( 1975; 1991, chapter5) hasarguedthat the theory I havesketched thus far would constitutea modest, rather than a robustsemantictheory. The , thoughit might tell us somethingaboutthe semantic theoryis modestbecause has that an agent , it falls far shortof telling us everything.For example knowledge ' ' , thekind of semantictheoryenvisionedabovemay tell us that snow refers to snow, but it doesnot tell us aboutthe knowledgewe useto identify snow. According to Dummett, it thereforefails to explain the abilities that underlie . our semanticcompetence
The Fonn of the SemanticTheory
Onemight reply that theabilities that underliesemanticcompetencearenot part of semanticsproperbut ratherpart of someotherandperhapsdeeperpsychological theory. However,this reply seemsunhelpful, sinceour concernis not with drawingpretheoreticalboundariesaroundthevariousdisciplinesbutrather with investigatingtheknowledge(semanticor otherwise)thatunderliessemantic . competence Ultimately, however,thereis no reasonto takeDummett's point asan objection to T-theories.To think of it asan addendumto the overallprojectwould be moreuseful. Sooneror later, semanticists will beinterestedin thepsychological abilities that underwritethe knowledgethat, for example, Val(x , ~ ~ iff x laughs. We might then regardT-theory axiomsto be descriptionsof semantic . For certainpurposes knowledgeat a certainlevelof abstraction , therewill beno harmin our operatingat this level of abstraction . For otherpurposes , important cluesto the natureof semanticknowledgemay be obscuredat higherlevelsof abstraction. Whatexactlywoulda robustaxiomlook like?Higginbotham( 1989) hasmade someproposalsin this regard, suggestingthat suchaxiomswould include the kind of informationthat recentresearchhasproposedto be part of our lexical knowledge. For example, Higginbotham, following Hale and Keyser( 1987), " " proposesthe following elucidation of the meaningof the term ' cut' : aV thatappliestrulyto situations e, involvingapatientyandanagentx who, bymeans of someinstrument z, effectsin e a linearseparation in thematerialintegrityof y. Thus, a robustT-theoryaxiom would incorporatethe informationthat hasbeen the subjectof muchrecentwork on the natureof the lexicon, both in linguistics (Hale andKeyser1987, 1993; Grimshaw1990; Pustejovsky1995) and in natural languageprocessing(NirenbergandRaskin1987; PustejovskyandBergier 1991; BoguraevandBriscoe1989).10 The properperspective , then, is not to view robustandmodestsemantictheories ascompetitors, but rather to seethat they are theoriesof the samephenomenon atdifferentlevelsof abstractionandidealization.Oncetheyareviewed in this way, the insightsgainedfrom the investigationof the lexicon becomea potentially powerful tool in the evaluationof competingT-theories, and ultimately in detenniningmorepreciselywhattheontologicalcommitmentsof such theoriesmight be. 2.4 PsychologicalEvidencefor the Nature of SemanticKnowledge If we pushour investigationno furtherthantheconstructionof truth-conditional semantic theories, we may find that it soon stalls. Supposethat we have
2 Chapter conflicting metaphysicalintuitions; perhapsyou haveB-theoryintuitions andI haveintuitions that areconsistentwith theA -theory. With adequateingenuity, it maybepossibleto constructseveralpossiblesemantictheoriesfor a givennatural , theorieswith differentmetaphysical (byhypothesis languagephenomenon them? Hwe view semantics between to some ). Is there way pick consequences asthe studyof theknowledgethat an agenthas, it providesus a fair bit of leverage in adjudicatingbetweencandidateT-theories. The questionbecomesthis: Which of theT-theoriesdoesthe agentactuallyknow? Evans( I985b) hasgiven an exampleof what sortsof considerationsmight . becomerelevant. For the time being, let us think only aboutfinite languages ' In fact, let s imaginean agentwhoselinguistic competenceis limited to twenty . sentences . The first, TI , introduces Wecanimagineat leasttwo T-theoriesfor this language twentyrules, beginningwith thesesix andcontinuingin obviousfashion: ( 1) [5Dick walks] is true iff Dick walks (2) [5Dick leaps] is true iff Dick leaps (3) [5Dick sings] is true iff Dick sings (4) [5Dick runs] is true iff Dick runs (5) [5Dick laughs] is true iff Dick laughs (6) [5Janewalks] is true iff Janewalks In 1' 2, on the other hand, there are only eleven rules. There is one rule for each of the names, and one for each of the properties, and one rule for the nontenni nal node:
(1) iff x = Dick VaI(x, ~ (2) VaI(x,~ ) iff x = Jane (3) VaI(x, ~ ) iff x = Sally
TheFormof theSemantic Theory (4) Va1 (x, Sm1 ) iff x = Spot (5) Va1 (x, ~ ) iff x walks (6) Va1 (x, ~ ) iff x leaps (7) Va1 (x, ~~ iff x sings (8) Va1 (x, mna ) iff x runs (9) Va1 (x, ~ gb ) iff x laughs ( 10) Val( T, [8NP VP]) iff , for somex , Val(X, NP) andVal(x, VP) ( 11) Val(x, [a (:i]) iffVal (x, (:i) (wherea rangesovercategories , and(:i rangesover categoriesandlexical items) Both TI andT2 success fully yield interpretiveT-theoriesfor this speaker's . Is thereanyreasonto preferoneoverthe other? That is, is thereany language reasonto supposethat the speakerknowseitherTI or T2?Accordingto Evans, theremay well be empiricalconsiderationsfavoringone of the.proposals.For example,supposethatwe hada recordof theorderin whichthespeakeracquired thelanguage . Suppose , for example,at somepoint thespeakersuddenlyshowed the ability to understandthe following four new sentences : Dick laughs. Janelaughs. Sally laughs. Spotlaughs. On theoryTI , this wouldhaveto beexplainedby thehypothesisthatthe speaker had suddenlylearnedfour newandunrelatedrules. On theoryT2, the explanation wouldsimplybethatthespeakerhadacquiredthesemanticrule for thepredicate ' laughs'.'1 Likewise, considerlossesof semanticcompetence . Supposethat a speaker suddenlylosesthe ability to understandthe four expressionsabove. On TI , the
Chapter2
explanationwouldhaveto bethatthespeakerhadsuddenlylost thetacit knowledge of four different and unrelatedrules. On T2, the explanationwould be that knowledgeof a singlerule hadbeenlost. LarsonandSegalnotethatin theorieslike T2, which admittwo kindsof rules andrulesfor predicateexpressions for referringexpressions rules ), it is entirely ( possiblethat thereshouldbe acquiredlanguagedeficits that affectoneclassof rules and not the other. Indeed, not only do suchcasesseemto be possible; as Larson and Segalnote, there in fact seemto be suchcaseson record. They point to a casestudy(SemenzaandZettin 1989) in which a subjectsuffereda " severeheadinjury and subsequentlylost the ability to dealwith purely referential " nondescriptivesemanticrelations. In effect, the subjectlost the ability to know rulesof the form of ( 1)-(4) in T2. Of course, asthe linguistic theoriesunderconsiderationbecomemorecomplex , the possiblepsychologicalprobesalsobecomemore interesting.A number of psycholinguisticprobeshavebeenproposedfor elementsrangingfrom clausalboundariesto themorphologicalconstituentsof words. Whetherthespecific probesofferedthusfar turn out to be fruitful , the importantconsideration for our purposesis the ideathat we can, in principle, appealto suchconsiderations in groundinga particularversionof semantictheory. to bearon the In sum, it appearsthatwe canbring a numberof considerations we realizes . a what T of may find it useful Specifically, theory speaker question sciences to enlist datafrom other branchesof the cognitive , andfollowing the hypotheticalsituationsenvisionedabove, we may find relevantevidencein the interactionof semantictheorywith the theoryof languageacquisition, and with the theoryof acquiredlanguagedeficits. 2.5
Do T- Theories Display Senses?
, Much of the interestin absoluteT-theorieshasstemmedfrom the suggestion 12 truth conditions than the more due to severalauthors, that T theoriescandeliver of expressions , but that they candeliverthe truth conditionsin a way that " " shows" or " . displays the senseof expressions ' ' ' ' It hasbeenarguedsinceFregethattwo expressionslike Cicero and Thlly havethe samereferentbut havedistinct senses , for they havedifferent modes of presentation that is, they presentthe referentof Cicero/ fully in different ' ' ' ' , thereis supposedlyno ways. BecauseCicero and Thlly havedifferentsenses both who believes someone the beliefs of in ( 17a) (a belief about irrationality aboutthe same a belief and 17b Ciceroundera certainmodeof presentation ) ( )( individual undera differentmodeof presentation ).
The Formof the SemanticTheory
( 17) a. Cicerois bald. b. Thlly is not bald. With respectto T-theories,thequestionnaturallyarisesasto whetherthetheorems of sucha theorydeliveronly thereferentsof expressions or thesensestoo. Thatis, doestheright-handsideof aT-theorytheoremsimplygivethereference of theexpressionon theleft, or doesit alsogivethereferencein a waythatshows or displaysits sense? This questionturnsout to haveimportantconsequences . Indeed, someauthorshavesuggestedthat the ability of T-theoriesto display sensesis centralto whethera T-theory can serveasa meaningtheory. For example , LeporeandLoewer( 1987) claim that therearecertainadequacyconditions whicha theoryof truthmustmeetin orderto functionasa theoryof meaning suitablefor interpretinga languageL. On their view, if aT-theoryis to serveas a meaningtheory, it mustbe the sort of T-theorythat candisplaysenses . But canT-theoriesdisplaysenses ?Accordingto LeporeandLoewer( 1987, p. 104), an adequateT-theoryfor an agent(call herArabella) will containtheorems ( 18) and( 19) without containingtheorems(20) and(21).
(18) 'Ciceroisbald'istrueiff Ciceroisbald. (19) ' ' Tullyisbaldistrueiff Tullyisbald. (20) 'Ciceroisbald'istrueiff Tullyisbald. (21) ' ' Tullyisbaldistrueiff Ciceroisbald. Thereasoningis thatArabellamaybelievethatCicerois baldbut not thatThlly is bald, soaT-theorythatgenerates (20) asa theoremwill not correctlycharacterize Arabella' s semanticcompetence . The dangerof introducing(20) into a semantic for Arabella to theory appears be that it leadsto a falsebelief attributionin the following fashion: (a) Arabellautters' Cicerois bald' .
44
2 Chapter
(b) Arabellabelievesthat ' Cicerois bald' is true (from (a) andcertainassumptions aboutArabella' s truthfulness, etc.). (c) Arabellabelievesthat ' Cicerois bald' is true iff Tully is bald (by hypothesis ). (d) aboutclosure). Arabellabelievesthat' n1l1yis bald(from (b), (c), andassumptions Notice that step(c) relies on the assumptionthat Arabella actually believesa theoremof the T-theory.13This assumptionis interesting, sinceit suggeststhat Arabella(presumablynot a semanticist ) musthavebeliefsabouttheoremsof a it kind of T . Since is theory implausibleto supposethatArabella(unless particular sheis trainedin semantics ) hasconsciousknowledgeof suchmatters, it appears to entailthatArabellamusthavesomesortof "tacit " knowledgeof (20} - a possibilitythat is entirelyconsistentwith thetheorylaid out this chapter. Now we needto avoida confusion- indeed, oneto which I havebeenparty.14 . Clearly, The ideais not thattheT-theorywill statethe senseof the expressions IS aT theorycannotdo that. If we go by what is literally stated, (20) attributes no moresemanticknowledgeto an agentthan( 18) does. Both areliterally talking aboutthe sameindividual. But whatis statedis not the sameaswhat is displayed. Dummett( 1973) and Evans( 1981) havearguedthat this distinctionwasmadeasfar back asFrege, ' andDummett( 1973, p. 227) hasinvokedWittgensteins distinctionbetweensaying andshowingto elucidatethe distinction: of a wordor symbol to givethesense . . . evenwhenFregeis purporting , whatheactually therelation whohasnotclearlygrasped statesis whatthereference [is] : and, for anyone betweensenseandreference , this fact makeshis holdon thenotionof sense : in of thereferent is themodeof presentation . Thesenseof anexpression precarious a particularwayof sayingthis. . .. In a is, wehaveto choose sayingwhatthereference of theexpression to convey casein whichweareconcerned , we , thesense , or stipulate thesense : wemight iswhichdisplays of statingwhatthereferent thatmeans shallchoose hereborrowa famouspairof termsfromtheTractatus , wesay , andsaythat, for Frege is. whatthereferentof a wordis, andtherebyshowwhatits sense Of course,Wittgenstein's distinctionbetweensayingandshowingis notoriously difficult , so perhaps, ratherthan worry aboutWittgensteinexegesis , I should try to elucidatea similar sort of distinctionwithin my framework. In the beginningof this chapter, I madea distinctionbetweenthe semantic knowledgethat an agenthasandthe way in which the agentrepresentedthat
The Fonn of the SemanticTheory
' ' knowledge.For example, an agentmight know that snow refersto snow, and thatknowledgemightberepresented by a kind of datastructurein themind/ brain. Let' s call the first sort of knowledge(e.g., that ' snow' refersto snow) firstorderknowledgeof meaning. Canan agentalsohavesecond-orderknowledge of meaning(tacit or otherwise)? That is, can the agentnot only know certain axiomsof aT-theorybut also(i ) know that he knowsthe axiomsand(ii ) know that he is representingthe knowledgein a way that displaysa certainsense ? It seemsto me that the answerhasto be yes. We know that the way we packageour claims canradically affect the way ' ' ' ' peoplebehave. For example, whetherI choosethe name Cicero or Tully for my utterancemay well affectthe behaviorof my interlocutor. Similar considerations apply to the way we tacitly representsemanticinformation. How we representour semanticalknowledgemay well haveconsequences for our behavior. It seemsto follow that a correctT-theorynot only mustcorrectlycharacterize our semanticknowledgebut mustdo it in a way that correctlycharacterizes our second-orderknowledgeof how the semanticalknowledgeis represented . -orderknowledgeand Thisraisesa numberof interestingissuesabouttacitsecond aboutthe transparencyof the way our T-theory axiomsarerepresented . What -orderknowledge seemsbeyonddisputeis thatthereis suchsecond . Whatremains to be investigated(anotherday) is the natureandreliability of this knowledge. 2.6
A Word on Predication
In chapter4 wewill considersomealternativeaxiomatizations . Fornow, it is worth noting that the axioms for the predicatesgiven above- the VP axioms- are not whatmanypeopleareaccustomedto. That is, peopleworking in semantics areaccustomedto thinking of predicatesasreferring to either propertiesor to setsof individuals. In the lattercase, we would expectto find axiomslike (22) . (22) Val(x, ~ ~
iff x = { x: x walks}
That is not whatis offeredup here, andit is not the pathtakenelsewherein this book. Nor needit be. Instead,I am supposingthatin a sentencelike 'Janewalks' the semanticvalueof ' walks' is Jane. Thus, the semanticvalueof the predicate will dependon the sententialcontextin which it appears . Wecanthink of thisapproachasincorporatinganideadiscussed by Carruthers ' ( 1989),16who attributesit to Wittgensteins Tractatus. The core of this idea is
Chapter2
thatpredicatesdo not refer to their extensions ; rather, their sensesspecifyrules of classification which are applied to the referent of the subjectexpression. Therule appliesto thatreferentin virtueof some" property-token" possessed by the referentof the subjectexpression : . . . thesenseof a predicate is not. . . a modeof thinkingabouta referent . It is rathera ruleof classification , applyingto thereferentof thenamein virtueof someproperty 'F' istoknow tokenwhichthatthingpossess es. Onsuchaview,tounderstand apredicate thedifference between thingswhichareF andthingswhicharenotF, wherethisknowledge in thegraspof theruleof classification consists whichconstitutes thesense of 'F' . . Carruthers 1989 170 ,p ( ) What is the advantageof this approach? The difficulty with directly referring to extensionsis thatonly by virtueof our understanding thepredicatearewe able to identify the extension.And, indeed, no onereally knowsthe extensionof ' x walks' ; theclosestwe cancometo specifyingthatextensionis by sayingthat" it is the setof all thosethingsthat walk" - in effect, that we musthavesomerule of classificationwhich is prior to the setof thingsthat walk. It is worth noting, asCarruthersdoes, that everyoneneedsto appealto some " " suchnonreferringpredicates , otherwisewe run into the third man problemdiscussed in Plato' s Parmenidesdialogueandemphasizedearly in the twentieth centuryby Wittgensteinandby F. H. Bradley. Considera standardsemantical treatmentof predicatesas extensions. On suchaccounts, the referent of the subjectandthe referentof the predicate(the extension) muststill be relatedby -of' . But thenwe wantto knowif thispredicate therelationalpredicate' is-an-element too mustrefer. If it does, thenwe haveembarkedon an infinite regress , for how is this newreferentto be relatedto thetwo that areaIreadyon the table? Finally, I shouldmakeit clearthattheintroductionof nonreferringpredicates . doesnot amountto abandoningreferentialsemanticsfor proceduralsemantics I am not identifyingthe semanticvalueof the predicatewith a procedureor a rule of classification. That rule is in the background(after all , Carruthersis talking aboutthe senseof the predicate, not its referent). I am still after a theory -to-world connections . The point hereis that properties that deliverslanguage andextensions(setsof objects) don' t haveto be part of that picture.
Chapter Attitudes
3 and Indexicals
Section2.6 told a bit aboutwhatthe pictureneednot include. However, a number of resources will haveto beincludedif we areto tacklethesemantics of tense. I will now beginaddingthoseresources . Sofar, my expositionof truth-conditionalsemanticshasfocusedon very simple . However, it is crucial to theprogramin this book that something constructions be said aboutindexicals(at a minimum, abouttemporalindexicals) . in particular, it is requiredthat indexicalsbe handledin a nonstandardway; temporal indexicalscan' t be treatedasreferringexpressions , andfurthermoretheir indexicalcharactermustbe reflectedin the semantics . (More on this in a bit.) Somethingsmustalsobe saidaboutpropositional-attitudeconstructions . In laterchaptersI am goingto arguethattensemorphemes areindexicalpredicates thattakeinterpretedlogicalforms(D..Fs) astheir arguments . Accordingly, I need to saysomethingabouttheconstructionof I LFs andabouttheir usein the treatment of propositional-attitudereports. As will be seen,the ILF theoryof the attitudes leadsto a certainshift in perspectivein our understanding of what must be stableamonglanguageusersfor us to saythattheyaresayingor thinking the samething. This shift in perspectivewill serveuswell whenwe turn to theproblem of indexicals.Accordingly, my plan is to beginwith propositionalattitudes andthenmoveon to the generalquestionof theproperhandlingof indexicals.
3.
PropositionalAttitudes
Oneof the centralproblemsin the semanticsof naturallanguagehasbeenthe " " " " problemof accountingfor the semanticsof opaque or hyperintensionalenvironments . To illustratetheproblem, considerthecontrastbetween( 1) and(2). ( 1) a. Max met [NPJudyGarland].
Chapter3
b. Max met [NPFrancesGumrn] . (2) a. Max believed[sthat JudyGarlandwasa fine actress ]. b. #Max believedfsthat FrancesGumrnwasa fine actress ]. If it is true that JudyGarlandis FrancesGumrn, then ( la ) entails( 1b) . Indeed, asa generalrule, co-referringtenDScanbe substitutedfor eachotherwithout affecting thetruth valueof the sentencein which they occur. The examplesin (2) constitutean exceptionto this generalrule. (2a) maybe true, but it needn't follow that(2b) is true. Let ussaythatopaqueenvironmentsarepreciselythoseenvironments wheresubstitutionof co-referringtenDSfails. Partof our taskin constructingaT-theoryis to determinethe " semanticvalues " of the constituentsof natural . Ordinarily, semanticists languagesentences would like to saythat the semanticvalueof a referringexpressionis the thing that it refersto. But noticethat the semanticistapparentlycannotsaythis in the faceof (2). ' JudyGarland' and ' FrancesGumrn' both refer to the sameindividual ; hence,thetenDShavethe samesemanticvalue; hence,we wouldexpectthat thetenDSshouldbe intersubstitutablewithout semanticeffect. However, asthe examplesin (2) show, that expectationis wrong. The semanticvalueof referringexpressionsis not the only difficulty facing . It is alsounclearwhatsemanticvaluesshouldbeassignedto the thesemanticist embeddedclausesof belief reports. Consider(3), for example. (3) [s Galileobelievedthat [s the Earthmoves]] Intuitively, (3) expresses a relation betweenGalileo and somesort of object, esa relationbetween but what sortof object? Onepossibleproposalis that it express Ga1ileoandthe clauseitself- in otherwords, betweenGa1ileoand(4). (4) [s the Earthmoves] But this doesnot seemplausible, since(4) doesnot carry asmuchinfonnation asthebelief reportintuitively does. For example,pointing at the moon, I might utter (5). (5) [s Ga1ileobelievedthat sthat moves]]
AttitudesandIndexicals
ButI mightalsoutter(5) whenpointingatthesun.Thetroubleis this: if thecontent of thebeliefattributionin bothcasesis simplythesyntacticobjectin (6), thenthesetwobeliefattributions arenotdistinguished in thesemantics . (6) [sthatmoves ] Therehavebeenmanyresponses to thisproblemin theliteratureof semantics overthelastthreedecades . Oneresponse hasbeento arguethatpropositional -attitude like(3) express relations between andsetsofpossible reports agents worlds(for example , thesetof all worldsin whichtheEarthmoves ).1Another has been to that an attitude response esarelationbetween argue reportlike(3) express Galileoanda Russellian propositionconsistingof theEarthitself and thepropertyof moving! Thereare, of course onthesetwoproposals , variations . In recentliterature , however , a numberof authors , includingHigginbotham , 1991 ( 1986a ), Segal( 1989), LarsonandLudlow( 1993), andLarsonandSegal analternative ( 1995 ), havesuggested in whichthesemantic proposal problems -attitudeverbsmightberesolved arisingwithpropositional bytakingsuchverbs to express relationsbetween andinterpreted markers agents logicalforms(phrase whosenodesarepairedwith semantic values ). " wouldbethe Forexample , theILF for "The Earthmoves syntacticphrase markerin (7) (abstracting fromdetailhere), with semantic valuesassigned to eachnodeof thetree.Wemighttherefore the ILF as in 8 where each represent ( ), nodeis pairedwithits semantic value. (7) S - - - - - -- -----, , - ----,,- - ---. NP VP "The Earth"
"Moves"
(8) (S, True ) - - - -.. -----, - - ------~ --(NP, TheEarth) (VP, TheEarth) .
" " ( The Earth, TheEarth)
" " ( Moves, TheEarth)
Chapter3
Suchobjectsin effect representa semanticvalue togetherwith its linguistic " " modeof presentation. Unlike the naive syntactic theory discussedabove, ILF theories, by including the semanticvalues, candistinguishboth of the beliefs " " reportedas Galileo saidthat that moves, becausethe two different acts of pointing pick out different semanticvalues(the moon in one caseand the sunin the other). Thus, in the casewhereI point at the moon, we get the ILF in (9). In the casewhereI point at the sun, we get the ILF in ( 10).
5 True , ) ( ~ / " " . . . Moon P Moon P ,"The ) (V,The ("N ) "~ ",The Moon M oves Moon ,The ) ( )(S (that True , ) . . . . " , . . -.,The ..Sun Sun P ) ,"The )P "'-(V ("N Sun oves Sun ) ,The ) (M",The (that
(9)
(10)
ILF theories thus provide an alternative to the treatments of propositional - attitude constructions that traditionally make use of possible - world semantics , Russellian propositions, or Fregeansenses. Becausethey eschewtheseresources, ILF theories are more austere- they attempt to cover as much of the data (or more ) with fewer theoretical resources. Furthermore , because of their austerity , they can be embedded in a truth - conditional semantics for natural language of the kind advocated in chapter 2. My A -theory of tense will regard tenses as being predicates of propositionlike objects. Strictly speaking, however, propositions will not do - propositions , ' as ordinarily construed, are eternal; they don t change over time . Instead we will be looking for proposition -like objects that are not eternal- objects whose truth values can shift over time . Before the A -theory can be incorporated into a truth conditional semantics, then, it will be necessaryto introduce technical resources that will do the work that propositions are routinely designed to do yet will be
AttitudesandIndexicals
non-eternal. Accordingly, I will be proposingthat ll..Fs can serveasthe arguments of tensein addition to being the objectsof the attitudes. Clearly I L Fs will playa centralrole in my semantics , soit is critical that I layout the details of ll..F formation carefully. (HereafterI will indicate ll..Fs with structuraldescriptions enclosedin closedbrackets( ' [] . . . [] ' ); e.g. the ll..F in (8) would be abbreviatedas ' [] [sThe Earthmoves] [] ' ). LarsonandLudlow ( 1993) proposethatthell..F theoryconsistof threeparts: -attitudeverbssuchas' believes', ' thinks' , and'claims' axiomsfor propositional ; an axiom for V Ps containinga complementclause; and a recursivedefinition of I L Fs. The axiomsfor propositional-attitude verbsin ( 11) exploit the idea that thesepredicatesarerelational. Intuitively, x will be an agentandy will be an ll ..F. ( 11) a. Val x,y), believes, 0' ) iff x believesy. b. Val x,y), .thi~ , 0' ) iff x thinks y. c. Val x,y), claims, 0' ) iff x claimsy. Axiom ( 12) introducesI LFs in the interpretationof V Ps of the generalform [vpVS ]- that is, VPs containinga clausalcomplement. ( 12) Val(x , [vpV S], 0' ) iff for somey, Val x,y), V, 0' ) andy = [] S[] w.rit. 0' . Finally, thereis a generalinductivedefinition of the interpretedlogical form of a with respectto a sequence0' .3ll..Fs are constructed" on the fly " by having theT-theoryrun throughthecomplementclause,detenniningthesemanticvalue of eachnode, and then constructingthe correspondingILF. AT -theoremfor ' Smithbelieves Joneswalks' would look like ( 13). ( 13) Val(T, [S[NPSmith] [vp[vbelieves][s[NpJones][vp[vS~ [] [S[NpJones] VpV~ ~ ]] [] w.rit. 0'
]]], 0' ) iff Smith believes
Constructionof thell ..F([][s[NPJones] VpV~ 1U] ]] [)) proceedsasfollows: The truth definition must be appliedto the embeddedclauseto determinethe semantic valuesof eachnodeof the embeddedp-marker. Recallthat the embeddedp-markerin this caseis just asshownin ( 14).
~
( 14)
-
Chapter3
~ We can get the semantic value of each node from a derivation utilizing axioms like those employed in chapter 2. Given the calculation of these semantic values, the ILF shown in ( 15) can be constructed. ( 15)
- ..--., Jones ( NP ) I } ~ .s., Jones
-
(S, True ---- ) - - - -' - - - "
- - - ) (VP, Jones I ( V, Jones ) I ) (l : alb, Jones Finally,then, wehavederivedthetheoremshownin ( 16). ( 16) Smithbelieves Joneswalksis trueiff Smithbelieves ) (S, True ---- - ---""" ' "' -' - - .--.) ( VP,Jones ) (NP, Jones I I ) , Jones ( V, Jones ) (Jones I ) (walks,Jones thatexhibitinattitudesarenottheonlyenvironments Overtpropositional effects , andit is quitenaturalto askhowILF theoriesareto beextended tensionality acomplete . Of course totheseotherenvironments storyis abookin itself, buthereI canat leastshowhowtheframeworkmighthandleso-calledintensional . transitiveconstructions
AttitudesandIndexicals
Theseconstructionsinvolve intensionalverbs, suchas ' seeks', that do not takeclausalcomplements showintensionalityeffects. Consider yet nevertheless 17 . 19 ( )-( ) ( 17) Johnwantsa donkey. ( 18) Johnseeksa donkey. ( 19) Johnlooks for a donkey. Of course, everythingturnson whetherthe clausein thesecasesis really missing , andeversincethe Middle Ageslinguistsandphilosophershavetoyedwith the ideathat theremight actuallybe an implicit clausein thesesentences .4 Oneway to makesenseof the hiddenclauseanalysiswithin the framework outlinedhereis to supposethat the LF representations of thesesentences actually have a hidden clause in them- perhapssomethingalong the lines of ' ' ( 17)-( 19). ( 17') Johnwants[PRO(to have) a donkey] . ' ( 18) Johnseeksa donkey[ pRO(to find) a donkey] . ' ( 19) Johnlooks for [PRO(to find) a donkey] . The evidencefor the hiddenclauseincludesan adverb-attachmentambiguity , which is mostevidentin caseslike ( 17). Consider(20). (20) Johnwantsa donkeytomorrow. Intuitively it is not the wanting which is to take placetomorrow, but the having . This is naturallyaccountedfor if we assumean implicit clause, asin ( 17'), andallow an additionaladjunctionsite for the adverb:
' (20) Johnwants[[PRO(to have)a donkey ] tomorrow ]. Of courseananalysisof this form makesa very strongclaim uponour syntactic theory, andthis claim hasto be arguedthroughon standardsyntacticgrounds.s Butif thesyntacticclaimcanbesustained , thenit is possibleto unburdentheseman tics of the needfor heavymachineryto handletheseintensionalenvironments .
Chapter3
In short, they canbe assimilatedto propositionalattitudesand subjectedto an ll...F analysis. I do not introduce this analysisof intensional transitive verbs merely to showhow onecanachievesemanticalausterityandstill treatattitudeconstructions . In chapter8, I will explorea similar sort of strategyfor the semanticsof tense,introducingimplicit clausesin supportof a particulartheoryof tense.Once againthe implicit clauseswill haveto obeythe constraintsof currentsyntactic theory, but providedthey can, I hopeI haveshownherethat the moveis quite . legitimate, if not yet entirelycommonplace Questionsabout ~ Fs The ll..F theory sketchedin this sectionhasled to a numberof misunderstandings . One of thesehasto do with the relationbetweenan ll..F andthe agentto whom an attitudeis ascribed. As wasnotedin chapter1, we routinelyascribebeliefsto dogsandothercreatures thatdo not havelanguage- certainly not anI -languagesimilar to whathumans have. A numberof individualshavesupposedthat an ll..F is intendedto mirror the structureof someobject in the mind/ brainof the ascribee; as Segal ( 1989) andLarsonandLudlow ( 1993) argue, this suppositionis mistaken.I may " " say Fido believeshis dishis empty, employinganll..F with Englishwords, but I do not therebysupposethatthereareEnglishlexical itemslocatedsomewhere in Fido' s languageof thought. I do not evensupposethat what is happeningin .6 Fido' s languageof thoughtmustbe" similar" to thell..F thatI haveconstructed For all I know, Fido may havea languageof thoughtthat is radically different from my own. WhenI say" Fido believeshis dish is empty" I am sayingthat Fido believes a particular ll..F with English expressionsin it . If we were interestedonly in modestT-theoriesthis would be the endof the discussion,but we might want to know more aboutthe psychologicalabilities which underlieour useof the term ' believes'. Exactly what is going on if ' believe' isn' t supposedto express a relationbetweenan ILF andsomethingthat is going on in Fido' s head? How do we go aboutelucidatingtherobustbackgroundtheorythatunderliesourunderstanding of ' believes'? As we sawearlier, the elucidationof a verb like ' cuts' is nontrivial. As we will see, ' believes' is evenmorecomplex. Sincewe canfaithfullyascribeattitudesto creatureswithout I -language,it is crucial that ll..Fs not be understoodasdescribingepisodesin the creature's languageof thoughtbutratherasprovidinginformationfor thebenefitof a hearer who wantsto constructa theoryof the creature's menta1life. Put anotherway, the emphasisshouldbe on the relation betweenthe ascriptionand the hearer
Attitudes;andIndexicals
ratherthanon therelationbetweenthe ascriptionandthe agentto whomthe attitude is ascribed. LarsonandLudlow ( 1993) makea preliminarysketchof how that could be possible, suggestingthat a systemgoverningbelief ascriptions(andour ability to usethe phrase' x believesy ' ) will involve at leastthe following threecomponents : . the tacit theoryof belief . thetacit theoryof the goalsof belief ascription . the tacit theoryof belief-ascription logistics. Thepreliminarysketchproceededasfollows: Considera speakerS,who wishes to ascribean attitudeto an agentA for the benefitof a hearerH. In orderfor S to succeedSandH mustsharea theoryof what beliefs are, for it is the shared ontology of beliefs that will guide the way in which beliefs are ascribed. S mustalsohavetacit know,ledgeof the goalsof belief ascription. That is, S must havea theory that allows him or her to determinewhat featuresof A' s belief will assistH in the relevantway. Finally, S musthavea tacit theorythat allows him or her to deliverthe kind of ascriptionthat will be helpful to H. Thecommon-sensetheoryof belief will statethepropertiesthatspeakers tacitly ascribeto beliefs, including their relationsto othercomponentsof thought andto action. The theoryof the goalsof belief ascriptionwill statethe theorya ' speakerdeploysin detennininga hearers interest. In almosteverycase,thegoal of a speakerSis to causea hearerH to form a certaintheory aboutthe belief structureof an agentA. Finally, thetheoryof belief-ascriptionlogisticswill statewhich expressions mustbe usedin a givencontextto achievespecificbelief-ascriptiongoals. This theory interactswith I L Fs directly and incorporatesthe knowledgerequired for determiningwhich U shouldbe usedin reportinga givenattitude. For example ' , dependingupon H s interests, it is sometimesthe referential component of an ILF and sometimesthe syntacticcomponentthat will be important to the goalsof ascription. We canidentify severalrules of thumb in characterizingwhich component of an U will be relevantin a givenattitudereport! For example, if H is interested in informationthatA hasaboutthe world (for example,the distancefrom Earthto Venus), thentheobjectualcomponentof theU will be of primaryimportance to the goalsof belief ascription. H will thereforebe indifferent to the choicebetween"the MorningStar" and"the EveningStar" in anattitudeascription concerningVenus. By contrast, if H is interestedin explainingor predictingA' s behavior(for example, whetherA will assentto an utteranceof "The Morning
Chapter3
Staris theEveningStar," or whetherA will actin a waycompatiblewith theknowledge that the Morning Staris the EveningStar), thenH may well be interested in the syntacticexpressions A' s belief. that S usesto characterize In casesof the latter kind, whereprediction or explanationof behavioris the goal, Larson and Ludlow envisionthe speaker's choice of syntacticconstituents in an ILF to involvea two- stageprocesswith respectto the hearerH. In the first stage, S detenninesthe way in which H modelsA' s belief structure " with H the . Then S " negotiates expressionsto be usedin speakingof the of that model . Both es. For example components stepsinvolve complexsubprocess ' s modelof A' s belief structureS would in H , inferring appearto draw at leaston all of the following: . S' s knowledgeof H' s interests . generalprinciplesof common-sensepsychologythatS supposes thatH believes . knowledgewhich S knowsH to haveaboutA . Supposethat S knowsH to be interestedin the behaviorof A - for example, in whetherA will train her telescopeon a particularregionof the dawnsky. Then, H to share, , whichS supposes by generalprinciplesof common-sensepsychology S may infer that H will deploy a fine-grainedmodel of A' s psychology- one that distinguishes Morning Starbeliefs from EveningStarbeliefsS may also rely uponinformationsupplieddirectly by H or someothersource. ForexampleS may learn that H knows that A is unawarethat the Morning Star is the EveningStar. In the secondstageof selectingan ILF, S andH must agreeon expressions of H' s modelof A' s belief structure.Expressions usedto speakof thecomponents " usedin attitudeascriptionswill betacitly " negotiated by participantsin thediscourse , following quitegeneralprinciplesholdingof discoursesof all kinds. The generalprocessby which discourseparticipantsnegotiatea way to speakof " " ) is currently a subject objects(sometimescalled entrainment by psychologists Brennan and Clarke 1992for a discussion . (See of researchin psycholinguistics .) Ultimately, this work must be extendedto the studyof how statesof mindcometo bedescribedandof why subtledifferencesin expressionwill have for the truth of an attitudeascription. greatconsequences All this is preliminary, thoughit may be a serviceableenoughtreatmentof what lies beneaththe surfaceof our useof the verb ' believes' (and of related propositional-attitudeverbs). But notice what a powerful resourcethis collection of abilities canbe: theseabilities allow us to keeptrack of a singlebelief in manyformal guises.For example,LarsonandLudlow ( 1993) examinedcases in whichI LFswith differentstructures(indeeddifferentobjectualcontents ) might be usedto characterize"the same" belief or report- for example, whenI hear
Attitnde.~:andIndexicals
A andB uttersentences of differentstructure andI saythatA saidwhatB said,8 ' triestoelucidate Therobusttheoryof 'believes theabilitytokeeptrackof asingle beliefundertheguisesof differentthat-clauses , A similarsortof ability will haveto beemployed in thetheoryof indexicals , 3.2
Indexicals
One of the most vexing issuesfor semanticaltheory hasbeenthe questionof howto treatindexicalexpressions like ' I ' , ' you' , ' this' , ' that' , ' here' , and' now' . All theseexpressionsappearto be referring expressions,but they refer in a ' ' very context-sensitiveway. I as utteredby me picks out a ratherdifferent individual than ' I ' asutteredby you. ' ' " Clearly, thereis somestablemeaningto an indexical like I (e.g., the utterer " of thesentence) that allowsusto determinewho is beingspokenof in different contexts.This part of the meaningof ' I ' might be calledthe " character " of the indexical(Kaplan 1977) or the " role" ( perry 1977). Whetherwe choose to call it thecharacter , therole, or somethingelse, thebig questionis whetherthis extraelementis to be consideredpart of the semanticsof indexicalexpressions . The receivedview hasbeenthat character /role is not part of the semantics proper- thatit is notpartof thepropositionalcontentof anutterance.This view is associatednotoriouslywith KaplanandPerry, but alsowith theoristsworking within absolutetruth-conditionalsemantics . Forexample,LarsonandSegal, that T theoriescanotherwisedisplaysenses despiteholding , hold thatthe character /role of a demonstrativedoesnot makeit into the theoremsof aT-theory. " " Specifically, LarsonandSegalproposethat T-theoriesbe conditionalized so that the extrameaningwe associatewith the characterof an indexicalmust befixed outsideof theT-theorybiconditionalproper.As anillustration, consider the casewhereKatherineHepburnutters(21) andCary Grant, gesturingat Ms. Hepburn, utters(22). (21) I needa haircut. (22) Sheneedsa haircut. On the proposalof Larson and Segal( 1995, p. 220), the conditionalized Ttheoremsfor thesetwo utteranceswill be (21') and(22'), respectively . ' (21 ) Ifu is an utteranceof [S[NPI ] [ypneeda haircut]] andx is the uttererofu , then u is true iff x needsa haircut.
Chapter3
' (22 ) Ifu is an utteranceof [S[NPShe] [ypneedsa haircut]] andx is the object demonstrated by the uttererof u, thenu is true iff x needsa haircut. Notice that in (21') and (22') the truth conditions themselvesare identical. Consistentwith the basicKaplan-Perry line, Larsonand Segal(p. 220) argue asfollows: Whataretheactualtruthconditions of theparticular utterances ([21]) and([22])1In both casestheRHSsof theconditionalsis "x needsa haircut." Whois xi it is Katherine , theuttererof ([21]) andtheobjectdemonstrated Hepburn by theuttererof ([22]). As farasourtheoryis concerned . No , it is thisobjectx thatentersintothetruthconditions that of is no content enters into the , particularway specifying object given descriptive . truthconditions to keepcharacter ? Partof the /role out of the semantics Why this eagerness ' ' answerto thisquestionliesin Perrys ( 1977) critiqueof Fregestheoryof demonstratives . The objectionis that for a Fregeansemanticaltheory(and, by extension for , anysemantictheorywith similargoals) thereis simplynothingthatcan describethesenseof anindexicallike ' I ' . Evenif we arepreparedto believethat there are distinct sensesthat can explain the difference in cognitive significance between' Cicero' and 'Tully ' , thereis nothingthat is going to do all the work that we expectthe character /role of an indexicalto do. from Frege( 1956, p. 296): Therootproblemis illustratedby a famouspassage ' ' If someone wantsto saythesametodayasheexpressed yesterday usingtheword today, ' ' . Althoughthethoughtis thesame hemustreplacethiswordby yesterday , theverbal mustbedifferentsothatthesense beaffectedby , whichwouldotherwise expression .Thecaseis thesamewithwordslike 'here' thedifferingtimesof utterance , isreadjusted ' ' and there. In all suchcases themerewording,asit is givenin writing, is notthecomplete conditions of thethought of certainaccompanying , buttheknowledge expression thethought for its correct of utterance of expressing , areneeded , whichareusedasmeans . Thepointingof fingers , handmovements , glances maybelonghere apprehension '' differentthoughtsin the too. Thesameutterance containingtheword I will express mouthsof differentmen,of whichsomemaybetrue, othersfalse. with a thought As Perrynotes, if Fregeis goingto identify thesenseof a sentence he is in trouble. Roughly, if the senseof the indexicalis to be identified either with its character /role or with somesortof descriptivecontent, thena sentence ' ' ' ' containing today anda sentencecontaining yesterdaywill haveto havedifferent sensesand henceexpressdifferent thoughts. No descriptivecontentis ' ' ' ' to going be able to reflect the distinct sensesof yesterday and today while ' ' preservingthe consequencethat yesterdaywas fine expresses the thought that ' todayis fine' expressedthe day before.
AttitudesandIndexicals
Somethinghasto give. EitherFregemustgive up theidentificationof senses with thoughts(in which casethe semanticalmachineryseemsincapableof giving thesemanticsof our thoughts) or Fregemustgiveup the ideathatthetwo utterancesexpressthe samethought. But then we run into trouble with our ability to communicatewhat we thought. Canit really be that we can' t express todaywhat we thoughtyesterday? Anotherproblemwith trying to keepcharacter /role in the semanticshasto do with the embeddingof demonstratives in modals. As Kaplan( 1977) hasargued , if the characterof a demonstrativemakesit into the truth conditions, what arewe to sayaboutexampleslike (23)? (23) You arethe personI ' m addressingwith this utterance. We surelydon' t want it to be a consequence that (24) is true, for (24) appears to be necessarilytrue whereas(23) is not. (24) The personI ' m addressingwith this utteranceis the personI ' m addressing with this utterance. Clearly therearecounterfactualenvironmentswhereanotherpersonis my addressee . Someoneelsemight havebeenstandingbeforeme. But (24) doesnot to appear allow this possibility. Still, despitethesepowerfulobjections,the conclusionseemsunsatisfactory . Recallthis familiar story: In a houseof mirrors, someonemight point at a man who is aboutto be attackedby a dog, saying ' He is being attackedby a dog' , andI may assentto thisjudgement, not realizingthat I am aboutto be attacked. ' ' Intuitively, someonewho says You areaboutto be attackedby a dog is saying somethingmore than the first speaker.To saythat this extra information lies outsidedIeprovinceof semantics seemsto beswrendering all tooquickly. Speaking 's to the of directly import Kaplan modality argument,Higginbotham( 1995, p. 248) arguesthat "the quirksof modalityshouldnot. . . be allowedto undermine the thesisthat what we sayandthink is literally androbustlyexpressedby the wordsthat we use." Havewe surrendered ? I tendto think that we have. Let mebegin prematurely widt themodalityargument . Onesolutionwouldbeto arguethatthis talk of evaluating thesentence in a counterfactual environmentreallyamountsto a shorthand way of sayingthatwe evaluatea modalizedpropositionthat assertsthepossibility dtatI am not addressingyou:
Chapter3
(25) It couldhavebeenthatyou arenotthepersonI ' m addressing with this utterance . It is arguablethat this is all it meansto evaluatea propositionin a counterfactual situation. Wecan' t, afterall , travelto otherpossibleworldsto evaluate(23) . (I will return to this issuein chapters4 and 6; for now I merely want to mark this asa possibleway out.) But why does(25) help? Don' t we get the samedifficulties when we unpack ' ' you asbefore? Consider(26). (26) It could havebeenthat the personI ' m addressingwith this utteranceis not the ' personI m addressingwith this utterance. We must movewith cautionhere. We havea definite description( ' the person I ' m addressingwith this utterance') andwe havea modal ( ' it could havebeen that' ), and we needto be on guardfor the possibility that the descriptionhas takenwide scopeoverthe modal. No conundrumsneedarisein (26) if we take the descriptionto havescope over the modal, asin (26'). ' (26 ) ' [the x: personI m addressingwith this utterancex] It could havebeenthat x is not the personI ' m addressingwith this utterance. In sum, the strategywould be to showthat " counterfactualevaluation" is in fact parasiticon our ability to evaluatecorrespondingmodal sentences(in the actualworld, not someother world) . That is, to evaluate' I am speaking' in a counterfactual situationwouldbenothingmorethanevaluating' I mightnot have ' beenspeakingin theactual situation. Our graspof otherpossibleworldswould . If thatis right, thenbeposteriorto our graspof therelevantmodalizedsentences admit of scope thenwe would merelyneedto showthat suchmodalsentences ambiguities, andto showthat thereis no problemif the descriptionscanhave wide scopewith respectto thesemodals(asin (26' . Perhapstherearewaysto answerthe modalargumentsof Kaplan, but what ? Despite /role from the semantics of Perry' s argumentsfor banishingcharacter ' the eleganceof Perry s critique of Frege, Evans( 1981) hasarguedthat it falls short: . . . there is no headlongcollision betweenFrege's suggestionthat graspingthe same thoughton differentdaysmay requiredifferent thingsof us, andthe fundamentalcriterion of differenceof thoughtswhich restsupon the principle that it is not possiblecoherently to takedifferentattitudestowardsthe samethought. For thatprinciple, properly
AttitudesandIndexicals stated, precludesthe possibility of coherently taking different : attitudes towardsthe samethoughtat the sametime.
On Evans's interpretationof Frege, we needn't think of thoughtsasholding at slices of time; we can think of them as enduring. Thus, our ability to grasp themaswe moveaboutin theworld requiresthatwe haveanability to keeptrack of themfrom differentvantagepoints: ' statemayrequiredifferentthingsof usatdifferent Freges ideais thatthesameepistemic times;thechanging circumstances forceustochange in orderto keepholdof a constant - wemustruntokeepstill. Fromthispointof view, reference andaconstant thought theacceptance ond2of ' Yesterday wasfine' , givenanacceptance ondl of 'Todayis fine' canmanifestthepersistence of a beliefin just thewayin whichacceptance of difference 'Thesunsetsin theWest'can. utterances of thesentence This idea, whetheror not it is properly attributedto Frege,9is appealingfor a numberof reasons . Thoughts,afterall, aresupposed to bepersistent , andour ability to graspthemmustthereforebe quite dynamic. The ideacanbe refraInedas follows: Thereis the objectualcontentof a thought- what the truth conditions . literally state.Thenthereis theway in whichthetruthconditionsarerepresented In chapter2, I spokeof suchrepresentations asdisplayingthe senseof the utterance or thought. It would be anerrorto supposethatthe senseis to be identified with a singlerepresentation at a particulartime. On thecontrary,a singlethought a sense must be in differentwaysfrom differentspatiohaving single represented . If this ideais appliedto T-theorems temporalperspectives , the way a theorem is represented candisplaya sense ; however,it cannotbeidentifiedwith thesense , for that sensemustbe displayedin differentwaysat differenttimes.10 What would underliethe ability to keeptrack of senses ? Herewe might opt for anextensionof theunderlyingtheoryof 'believes'thatwasofferedin Larson andLudlow 1993anddiscussedin theprevioussection. On the Larson-Ludlow proposal, the idea was to explain our ability to know when two that-clauses attribute the samebelief or count as sayingthe samething. In that case, the puzzlewasto showhow usingdifferentwordsin an ILF presentedto different hearerscould countashavingattributedthe sameattitudeto an agent. The answer wasfound in a rich theoryunderlyingthe useof ' believes'- a theorythat took accountof the hearer's interests, the speaker's goals, sharedknowledge aboutthe structureof the agent's beliefs, andso on. The relation betweenthat analysisof propositional attitudesand Evans' s , sinceattitudereportsroutinely proposalfor indexicalsshouldnowbeobvious. Indeed include indexicals, what is called for is someamplification of the ILF theory that can explain how we keeptrack of when two indexical utterances (beliefs) say(mean) the samething.
Chapter3
Clearly, muchof theLarson-Ludlow machinerywill beusefulhere. Whatwe will needto do is expandon thetacit theoryof belief-ascriptionlogistics, which held that our ability to ascribebeliefsinvolved, amongotherthings, . S' s knowledgeof H' s interests, . generalprinciplesof common-sensepsychologythatS supposes H to believe, and . knowledgethat S knowsH to haveaboutA . It seemsthat we neednot only a theoryof the hearer's interestsbut alsoa theory of the hearer's spatiotemporalperspective . Whenreportingwhat the agent saidor thought, I needto know wherethe heareris relativeto the agent. If on ' ' Wednesdayan agentsays It is fine today , when I report this to a heareron butrelativeto thehearer'stemThursdayI will wantto reportthesamestatement ' ' poral position. Thus we get A saidthat it wasfine yesterday. Roughly, then, we will wantthe theoryto keeptrack of the following: . S' s knowledgeof H' s spatiotemporalposition . S' s knowledgeof H ' s andA' s relativepositions . H ' s knowledgethat S knowsA' s spatiotemporalposition (with standardassumptions aboutcommonknowledge). Whatgoesfor attitudereportsandindirectdiscoursegoesfor determiningwhen two indexical I -languagetokeningsexpressthe samethought. Thesetokenings , after all, are alsoreportsof a certainkind, sincethey aretokeningsof an I -languagesentenceto oneself. The twist is that in suchcasesthe agent, the " " " " , andthe hearer are all the sameindividual. The puzzlesthat arise speaker heredo so becausethe tokeningstakeplaceat different times and at different places.Accordingly, we mustadjustto thenewtemporalposition, just asthough the original tokeninghadbeenmadeby anotherperson. " Clearlya lot of work mustbedoneto elucidatetheexactnatureof thesetracking " abilities. Evans 1982 hasmadea start but , ( ) realisticallythis picturemust befleshedout by a full scalepsychologicalresearchprogram- at leastif we are interestedin therobustunderlyingtheory. In themeantime , whatwouldthemodest semanticaltheorylook like? It seemsto me that therearetwo possibilities. The first is that we simply introduce thedescriptiveinformationinto theT-theorems , thusobtainingtheorems " like (21 ) and(22") for (21) and(22). " (21 ) If u is an utteranceof [S[NPI ] [vpneeda haircut] ] thenu is true iff the uttererof u needsa haircut.
Attitudes and lndexicals
" (22 ) If u is an utteranceof [S[NPShe] [vpneedsa haircut] ] thenu is tl1leiff andthe objectdemonstrated by the uttererof u needsa haircut. As we will seein chapter6, there are seriouslimitations to this sort of approach . For now I just want to put the strategyon the table. Alternatively, we cansimply disquotationallyenterthe indexicalexpression into theright-handsideof a biconditional. For example, we might haveaxioms andtheoremslike (27) and (28). (27) Val(x, ' I ' ) iff x = I (28) Val(T, ' I walk' ) iff I walk ' Admittedly, this ideaappearsnaive. Wouldn t we get absurdresultswhen interpreting theutterancesof others? For example, if Smithsays" I walk," it is no goodfor meto haveaT-theorythatinterpretsthatutteranceassayingthatI walk. But that is preciselywherethe abovesort of axiom seemsto lead. Likewise, if someonesendsme an email on Tuesdaysaying" Smith went to the dentisttoday," if I only readthe email today(severaldayslater) it is no goodfor meto interpretthis assayingthatSmithwentto thedentisttoday. Thus, axiomslike (29) for temporalindexicalspresumablyarealsoout of court. (29) Val(x, ' today' ) iff x is tl1letoday On the other hand, perhapsthis objection is given far more weight than it is due. In the first place, if we aretalking aboutI -language , andif (asI suggested in section1.3) theprimaryuseof I -languageis not communicationbut thought, thenaxiomslike (27) and(29) aresuitedto a broadrangeof I -languagetokenings . Suchaxiomsarealsoentirely suitablefor the interpretationof any speech we produce. The only drawbackfor suchaxiomsappearsto be associatedwith usingthemin interpretingthe utterancesof others. But evenasappliedto theinterpretationof othersit is far from clearthatthese sortsof axiomsareinadequate . For example, onestrandof thinking would be thatwhenwe interpretotherswe try to " simulate" their thoughts.II That is, perhaps when we interpretanother's utterancewe try to " project ourselvesinto" thatperson's egocentricspace.If that is how we interpretothers, thenof course we wanttheaxiomsto reflecttheegocentricperspective of thespeaker . Anything lesswould get in the way of our ability to simulate, andhenceunderstand , our interlocutor.
Chapter3
A lessradical solutionis alsoavailable. Perhaps , when we interpretthe remarks of another,we amendthe axiomsof our T-theoryto accountfor the position . Accordingly, we mighthaveconditionalizedaxiomslike (30), of thespeaker where' Val( X, Y, Z)' saysthatX is the semanticvalueof Y asutteredby Z. (30) if s is my interlocutor, thenVal(x, " here" , s) iff x is there(nears) This sort of paraphrase might allow us to track the indexicalutterancesof others from different spatialandtemporalvantagepoints, just aswe did in determining whentwo tokeningsexpressedthe samething.12 I shall return to thesestrategieswhen we take up the issueof temporalindexicalsin chapters6 and7. For now, I merelywantto suggestthe plausibility of thesestrategiesandto proposethattheymight helpusto avoidtheprimafacie . /role from the semantics unattractivesolutionof banishingso-calledcharacter traditionin thephilosophyof languagethatwould Thereis nowanentrenched haveusfaceproblemsin semantics by bracketingthemoff andtIying to castthem . (Suchexorcismsusuallyareaccompanied out of semanticsandinto psychology " directedat thosewho seecloseconnectionsbetween " of by charges sloppythinking thesedisciplines.) Wettstein1986andKaplan 1990arecanonicalexamples of this phenomenon , but I too havebeencaughtup in the game(seeLudlow 1989). This sortof divide-and-wall-off strategyis almostneversuccessfulin the naturalsciences ,I) andit is hardto seewhy it shouldbe successfulhere(where we areat leastkneedeepin naturalscience). Artificial disciplinaryboundaries simply do not solveproblems. In sum, it seemsto me unwiseto try anddivorcesemanticsfrom the theory of meaning(in therobustsenseof a theorythat alsoincorporatescognitivesignificance ) . As will be seenin the next chapter, I think it is equally unwise (if . not impossible) to try anddivorcesemanticsfrom metaphysics
Chapter 4 Drawing Metaphysical Consequences from a T- Theory
The ideathat the studyof languagecanprovideinsightsinto metaphysicshasa long pedigreein thehistoryof philosophy.It seemsto havebeenoperativeasfar back asthe pre-SocraticphilosopherParmenides ,I and it hasbeencommonin twentiethcenturyphilosophy. A numberof linguists havealso flirted with the idea that the study of natural for metaphysics . Perhapsthemost languagemighthavestrongconsequences notoriousexample, notedin the prefaceto this volume, is the caseof Whorf ( 1956), who attemptedto drawconclusionsabouttheHopis' metaphysics of time from their temporallanguage.Therearealsoa numberof morerecent, if more tentative,claimsaboutthelink betweensemanticsandmetaphysics . Bach( 1981, 1986), for example,hasspokenof therelationbetweensemanticsandthe" metaphysics of English," andhasspecificallyaddressed the questionof what metaphysical are entailed the semantics of tense.But althoughBach consequences by offerssloganslike " No semanticswithout metaphysics !" ( 1986, p, 575), helater retreatswith remarkslike the following: I 'venowsaida little (butperhaps morethanenough ) aboutsomeof thekindsof things weseemto needin ourontologyfor Englishanda littlebit (notnearenough ) abouthow wemightgetthemintoa semantics for English . It wouldbeimmoralof measa linguist ' aphrase fromMontague (I mstealing ) tomakeclaimsonewayortheotheraboutwhether or notthesesortsof thingscoI Tespond to realthingsin therealworld, perceptual or conceptual thatareindependent of language categories , or nothingatall. (ibid., p. 592) It is, of course,a centralassumptionof thepresentwork that it is far from immoral to draw metaphysicalconsequences from a semantictheory (so long as it is an " absolute" theory), whetherone is a linguist or a philosopher.Yet this , if it is to be helpful, mustbe clarified. Although philosophersand assumption have linguists longpressedfor a connectionbetweenlanguageandmetaphysics , have not they alwaysbeenclear about what sort of connectionthere is supposed to be, muchlesswhy thereshouldbe sucha connection.
4 Chapter Therefore, we needto get clear on how languageand metaphysicsare supposed to hangtogether.Therearetwo questionsto consider: . Justwhat kinds of metaphysicalconclusionscanbe drawnfrom the kind of semantictheoryoutlinedin chapters2 and3, andhow canwe identify the specific metaphysical commitments of the theory ? . Can such a semantic theory be revised to avoid metaphysical commibnents
4.1 The Nature of the Metaphysical Commitment Onestandardprocedurein semantictheoryhasbeento definethedomainof relevant objectsin advanceof statingthe theory. However, it is arguablethat such a proceduresimply papersover the most interestingand contentiousissues.A numberof questionscanoften be raisedaboutwhat a semantictheoryof such andsuchform actuallycommitsus to. To simply stipulatea domainin advance } requiresthat the mostinterestingquestionshavealreadybeenanswered be drawn from semantic can thatmetaphysical I havesuggested consequences theory, but I havesaid little about what featuresof the semantictheory give . The shortanswerto the questionis that we rise to metaphysicalcommitments will be committedto whateverobjectsserveas a semanticvaluesin a correct . T-theoryfor naturallanguage To be moreprecise, let us suppose , on the basisof considerationslaid out in chapter2, that thereis a correctT theorythat describes(at a certainlevel of abstraction ) the semanticknowledgethat an agenthas. The axiomsandtheorems of thatT-theorywill introducecertainsemanticvalues- implicitly quantifying ' over thosesemanticvalues. For example, ( 1) is shorthandfor ( 1 ), wherethe metalinguisticquantificationover semanticvaluesis madeexplicit. ( 1) Val (x, ~ ) iff x = snow ' (I ) For all x , Val (x , ~ ) iff x = snow ' Becausethequantificationin ( 11 ) is not vacuous,( 1) commitsusto theexistence I of snow. So far, nothing saysthat a true semantictheory must employ ( 1 ); however, any theory that doesemploy it will havea clear metaphysicalcommitment . A fairly straightforwardclaim canbe madeaboutthe ontological commitments of semantictheoriesof the form discussedin this book. It is, with apologies to Quine( 1953), that to be is to be a semanticvalue.
from a T-Theory DrawingMetaphysicalConsequences
4.2 SampleCases Considerhow two distinct semantictheoriesmight introducedifferentkinds of objectsas semanticvalues, and might hencehavedifferent ontological commitments . The Commitment to Properties Consider, for example, the adjective' red' and the two ways of giving the semantics for the adjectiveshownin (2) and(3). (2) Val(x, ~
iff x = thepropertyof beingred
(3) Val(x, ~
iff x is red
In thefirst casethereis referenceto thepropertyredness- a kind of abstractobject . In thesecondcase(discussedin section2.6) it appearsthatreferenceto redness hasbeencircumvented . Evenif axiom (3) is employedin giving the truth conditionsfor a true sentenceof English, thereis merelya commitmentto one or more objectsthat happento be red! The differencebetween(2) and (3) reflects theintuitive differencebetweenaxiomsthatintroducereferenceandthose that introducesomethingmoreakin to predication. (2) refers, in the sensethat it identifiesa specificsemanticvaluefor ' red' . (3), on the otherhand, doesnot refer to a particular property; rather, it predicatessomethingof thosevalues that satisfy ' red' . Forexample,thetechnicalfragmentsintroducedin theappendixesavoidreferenceto propertiesandsets. Froma technicalpoint of view, therewould be no barrierto introducingthe ax.ioms shownin (4) instead.4 (4) a. Val(x, dQ&, 0 ) iff x = propertyof beinga dog Val(x, ~ , 0 ) iff x = propertyof beinga cat Val(x, ~ , 0 ) iff x = propertyof beinga man Val(x , woman, 0 ) iff x = propertyof beinga woman b. Val(x, hm , 0 ) iff x = the propertyof barking Val(x , ~ ~ , 0 ) iff x = the propertyof walking c. Val(x, ~ , 0 ) iff x = the seeingrelation Val(x, ~ , 0 ) iff x = the likes relation
68
4 Chapter
The nontenninalaxiomswould thenhaveto be modified accordingly: (5) a. Val(T, [5NP VP], 0 ) iff for somex, p, Val(x, NP, 0 ) andVal(p, VP, 0 ) andx hasp b. Val(p, [vpV NP], 0 ) iff for somer, Z, Val(r, V, 0 ) andVal(z, NP, 0 ), andp is thepropertyof bearingr to z Thesubstitutionof roleslike theseinto thefragmentwouldyield theoremslike (6). (6) [5 [NPSmith] VpVsees][NPJones]] is true iff Smithhasthe propertyof bearing the seeingrelationto Jones This alternativeaxiomatizationwould havethe sameempiricalcoverageasthe axiomatizationintroducedin chapter2, but it would haveprofoundlydifferent . The chief differencewould, of course, be its commetaphysicalcommitments itions. alternativeaxiomatizationsis arguedon a priori metaphysicalgrounds.Of course, it is alsopossibleto arguefor onepositionor the otheron linguistic grounds.The questionis, just how far canwe get with a priori metaphysicsalone? On the onehand, thereareindependentreasonsfor supposingthat properties oughtto exist. Wright ( 1983) haspointedto casesof apparentpronominal referenceto propertiesin exampleslike (7). (7) A : At leastJohnis honest. B: Yes, he certainlyis that. And Chierchia( 1984) hasnotedthat we appearto routinelyquantifyoverproperties in exampleslike (8) andin reasoninglike (9). (8) Johnis everythinghis motherwantedhim to be. (9) Johnis everythingthat Mary is. Mary is intelligent. Johnis intelligent. But noneof theseargumentsis particularly compelling. In the first place, the argumentfrom anaphorais weak. Pronominalreferenceusuallyallowsusto use plural anaphorsto pick up referenceto severalobjects, asin ( 10).
from a T-Theory DrawingMetaphysicalConsequences
( 10) A: Baccala is deliciousandsois polenta . B: OK, I ' ll try them. But similar plural anaphors are not available for alleged property references Consider ( 11) .
( 11) A : Johnis tall andhonest. B: #Yes, he certainlyis them/those. The allegedquantificationover propertiesin (8) is also suspect , asonecannot continuethe utterancein the usualway by listing the thingswe aresupposedly quantifyingover. Compare( 12) with ( 13). ( 12) I tried everythingmy motheraskedme to try : the chicken, the beef, the tuna surprise,. . . . ( 13) #Johnis everythinghis motherwantedhim to be: kindness,thriftiness, the propertyof beinga Republican,. . . . It is simplyno goodto list theproperties,butthosearethethingsthatwe aresupposed to be quantifyingover. Thereis somethingvery wrongwith the ideathat we arequantifyingoverpropertieshere. Similarconsiderations underminetheargumentfrom inferenceslike thosein 9 . We ( ) may well havesecond-orderquantificationin suchexamples , but there is no apparentquantificationoverproperties. If theusualargumentsfor andagainstpropertiesarenot persuasive , perhaps we cangainsomeleverageby seeingwhatconstraintsthesemantictheorymight placeon our choice. LarsonandSegalhavegivenan exampleof a linguistic argumentthat might decidebetweenan analysislike that in chapter2 and the property analysis. Accordingto them, the propertyanalysisencountersdifficulties whenone attempts " e. . ' Smith to developanaccountof " coordinationconstructions walked ( g, ' andswam). For example, onecanextendthe fragmentLl to handleconstructions suchasthosein ( 14). ( 14) Val(x, [sVPl . andVP2], 0' ) iff Val(x, VPl , 0' ) andVal(x, VP2, 0' ) On theotherhand, if thesemanticvaluesof sentences areto berenderedasproperties the account becomesmuch morecomplex perhaps,following Larson , andSegal, alongthe lines of ( 15) .
Chapter4
( 15) Val(p, [sV Pl ~
VP2], 0' ) iffVal (pl , V Pl , 0' ) andVal(p2, VP2, 0' )
As Larson and Segalnote, this will require the introduction of somekind of ? Intuitively, whatis . But whatsortof mechanism property-combiningmechanism them into newproperties wantedhereis a rule that will takepropertiesandcombine d1atareexemplifiedonly by d1oseobjectsthathaddie originaltwo properties . But how is this mechanismto work, andhow is it to be integratedinto die axiomsof die T-d1eoryin an intelligible way? The answeris far from clear. Wemaynot evenneedto considercaseslike conjunction; thepropertyanalysis may not get off the ground, in view of the kinds of constraintswe would " like to put on T-theories.We wantto saythat an agent" knows the axiomsof a T-d1eory , but what doesit meanto know an axiom like ( 16)? ( 16) Val(x, ' red' ) iff x = redness Redness , if diereis sucha thing, is anabstractobjectwell outsidespaceandtime. Our knowledgeof suchthings (if suchknowledgeis possible) is, of necessity , mediatedby other sortsof knowledged1atwe have- for example, our knowledge ' of the meaningof ' red , and, by extension,our havinga rule of classification that allows us to identify red things. Strictly speaking, knowing that x is red is morefundamentalthanknowingwhatrednessis. But if knowing that x is red is more fundamentalthan knowing what redness ' is, onehasto wonderwhatwork die propertyis doing here. It isn t needed to accountfor our semanticalknowledge, that is for sure; hence, therecan be no purely semanticalmotivationfor positingsuchproperties.Perhapsthereare other motivations(say, in physicaltheory), but it seemsto me that properties arevery poor candidatesfor admissioninto our ontology.6 Keepin mind that our goal hereis not to argueagainstpropertytheory, but ratherto showhow the questionis refraInedon this approachto metaphysicsit becomesa questionof simultaneouslysolving the constraintsof semantical theory andconstraintsimposedby our metaphysicalintuitions. Neither setof constraintsis a priori privileged; the resultingtheorymust satisfyboth. A similar situationholds whenwe movefrom the investigationof predicationto the role of nameswithin the semanticaltheory. Names I saidearlier in this chapterthat to be is to be a semanticvalue. Although the claim is straightforward, in this formulation it is also a claim with somevery
from a T-Theory DrawingMetaphysicalConsequences
. Justhow strongthethesisis becomesclearwhenwe consider strongconsequences an axiom like ( 17). ( 17) Val (x, Peiasus) iff x = Pegasus If our true semantictheory includesan axiom like ( 17), we are committedto an ontologythat includesPegasus ! Presumably(Meinong 1904), Pegasusis a kind of nonexistentobject. We may well recoil from this sort of commitment, but we mustobservethat inconsistenciesin the theory of Meinongian objects havebeenexorcisedin recentyears- for example, in Parsons1980, in Routley 1980, in Zalta 1983, andin Zalta 1988. Thus, in evaluatingsuchproposals , what we haveto go on are our metaphysicalintuitions and whateverpurchasesemantical theorycangive us. As far asthesemanticaltheoryis concerned , nothingsaysthatT-theoriesthat ' correctly characterizean agent s semanticknowledgemust employ axioms ' like ( 17). Indeed, undera numberof plausiblecircumstances , the agents semantic knowledgemay divergefrom ( 17) significantly. In particular, if we attend to robustT-theorieswhentrying to ascertainour ontologicalcommitments , thena carefulstudyof theagent's knowledgemaydeflectconcernsaboutnonexistent objects. Onepossibility is that closerinvestigationwill showthat the lexical entry ' is much for ' Pegasus morecomplexthan is reflectedby ( 17). This would not be an unusualstateof affairs. Recentwork on the lexicon suggeststhat words arein fact highly structuredobjects. For example,if thelexicalentrywereto encode the rich structureof a Russelliandescription, the metaphysicalcommitments would doubtlessbe much more austere. For example, the entry for the word we pronounce" peg~s~s" would be the following: on a naiveRussellian ' picture N: [the x: x is white, x is winged, . . .J' . For a richly structuredlexical entry like this, there would not evenbe a single semanticaxiom. Rather, the axiomsfor ' the' , ' white' , ' winged' , etc., would comeinto play. The problemsinherentin descriptivetheoriesof namesarewell knownisof course, but perhapsnot decisive. For example, muchof Kripke' s ( 1980) argument turnson our metaphysicalintuitions aboutrigidity . But we havealready seenin section3.5 that there may be ways to accommodaterigidity without giving up on the cognitivecontentof a term.9 For example, following a suggestionmadeby Dummett( 1973), we might let the descriptiontake wide scopeover any modaloperatorsin the sentence . ' DespiteKripke s protestationsin the prefaceto Namingand Necessity( 1980),
4 Chapter " " oneprobably can go somedistancetowards capturing the relevantrigidity ' intuitions in this way. Clearly, Kripke s characterizationof sucha stipulation " " as unaccountable! (ibid., p. 13) overstatesthe case. At worst, the stipulation is unaccountedfor, not unaccountable(just as many other scopeand movement restrictionsin generativegrammararenot yet accountedfor). That something is not accountedfor by currenttheoryis the noon in empiricalinquiry. It is somethingto investigate , not somethingto despair.over. ' Nor doesonenecessarilyhaveto bow to Kripke s argumentthat rigidity effects are apparentin ( 18) (which containsno modals) andthat henceapparent rigidity cannotbe accountedfor by scope. ( 18) Aristotle wasfond of dogs. Accordingto Kripke, ( 18) canbe consideredundervariouscounterfactualsituations , asit surelycan. But herethe unarguedassumptionis that whenwe consider ( 18) undervariouscounterfactualsituationsthe descriptivecontentof the namemustremainconstant.Weneedto considerwhatit meansto evaluatea sentence underother counterfactualsituations. Thereis no a priori reasonwhy a properevaluationcannotinclude a correlativeshift in the descriptivecontent of the name.10 Of course, this gambit againstKripke requiresa particularperspectivetowards rigidity andtransworld identity. The ideawould be that we do not simply rigidly designatean individual, but ratherthat transworld individuals are " " constructedfrom descriptive manifestationsin variousworlds. Theindividual . Theoriesof transworld is synthesized , asit were, from thesemanifestations identityin thisveinhavebeenproposedby Hintikka ( 1969b, 1972). Smith('1983) hasdrawnan interestingparallelbetweentechnicalefforts like Hintikka s and " " Kantian/ Husserliantheoriesof how individualsare constituted. This discussionis not intendedto be anythinglike a seriousdefenseof the descriptivetheory of names.That would takeanotherwork asleastaslengthy anddetailedasthis one. Rather, the point is to illustratethe shift in perspective , andmetaphysicsto be that follows if we areallowing epistemology , semantics of connectedin an interestingway. The precisenature our commitmentwill not turn solely on a priori argumentationaboutthe plausibility of suchobjects or our intuitions aboutrigidity ; it will alsoturn on a carefulelucidationof the ' agent s lexicon and of the correspondingsemanticknowledge. The point is that within this framework the questionof namesis not closed; indeed, it is very muchopen. Thereis still muchinterestingwork to be done.
from a T-Theory DrawingMetaphysicalConsequences
Events It is well known that Davidson( 1967b) arguedfor the existenceof eventson thebasisof thelogical fonn of actionsentences . Roughly, thereasoningwentas follows. Onecaneasilymakethe inferencefrom ( 19) to (20).
( 19) Johnatethechipsgracefully . (20) Johnatethe chips. Davidsonproposedthat this inferencewasa logical inference, andthat the inference could be madeformally oncethe underlyinglogical fonn of 'Johnate the chips gracefully' was revealed. Davidsonsuggestedthat its logical form shouldbe asin (21).
(21) (3e)[ate(John,thechips,e) & graceful (e)] assayingthattherewasanevente whichwas , (21) canbeunderstood Roughly aneatingof the chipsby John, ande wasgraceful. The conclusion(20) follows by simpleconjunctionreduction. Davidsontook this to beonepieceof evidence for the existenceof events. But noticethat it is not evennecessarythat therebe explicit eventquantification in the object language.What mattersis whetherthereis quantification overeventsin themetalanguage . For example,in thefragmentintroducedin appendix T2 we havesemanticruleslike (22). (22) a. Val(T, [sNPVP], 0' ) iff , for somee, Val(e, VP, 0' ), andfor somex , x is the agentof e andVal(x, NP, 0' ) b. Val(e, VpVNP], 0' ) iffVal (e, V, 0' ) andfor somey, y is the themeofe and Val(y, NP, 0' ) c. Val(e, VpV ADV ], 0' ) iff Val(e, V, 0' ) andVal(e, ADV, 0' ) The crucial quantificationfor our purposesis the metalinguisticquantification over the e position. H theseaxiomsarecorrect, and if at leastsomeof the action sentences we utteraretrue, thenwe arecommittedto theexistenceof events. It mightbearguedthatwe areindeedcommittedto somethingwhenwe quantify into the e position, but that is a far cry from sayingthat we arecommitted
Chapter4
to theexistenceof events.It is certainlytruethatwe arenot committedto events ' simply becausethe variableshappenedto be labeledwith little e s. What we do know, however,evenfrom simpleaxiomslike the above, is that we arequantifying over things (whateverwe may chooseto call them) that haveagents(cf. ' ' (22a and themes(cf. (22b , and of which adverbslike intentionally and ' aretrue. These ' 11 certainlysoundlike candidatesfor eventsto me. slowly Notice too that, if this approachis right, much of what we know aboutthe structureof events(or whateverwe chooseto call them) will notbededucedfrom a priori intuitions aboutthe natureof events(whetherthey haveparts, whether theycanhavemultipleagents,andsoon); rather, it will flow from theconstraints imposedby the demandsof constructingthe T-theory. The demandsof semantic theorywill help to elucidatethe natureof ontology. 4.3
? Can aT . Theory Avoid Having Metaphysical Consequences
that ontological commitmentmight be tied In the precedingsectionI suggested to the semanticvaluesintroducedby the semantictheory. In supportof this suggestion , it canbenotedthatthereis, afterall, implicit quantificationoversemantic valuesin theaxiomsof thesemantictheory. If thereis quantificationoversemantic values, thentheremustbe an ontological commitmentto thosevalues. However, it hasbeenarguedthat one can neutralizethe ontological commitments entailedby quantificationby treatingthe quantifiersassubstitutional quantifiers. The basicidea is that a sentencelike (23) will be truejust in case ' (23 ) (where is a concatenationsymbol) is true. (23) A dog barked. ' (23 ) ' For someterm t, t ' is a dog' is true andt ' barked is true. ' with ' is a dog' and Sincethereis a term (say, ' Lassie) that whenconcatenated ' barked' resultsin a true sentence 23' mustbe true. Noticethat this is accomplished ,( ) without everquantifyingover actualdogs. It hasbeenproposedthatthis devicemight beusefulin a numberof contexts, including the philosophyof mathematics(Gottlieb 1980), the semanticsof intentional ' constructionslike ' Johnseeksa unicorn (Ioup 1977), and the treatment of classesandattributes(Sellars1963). Thequestion,then, is whetherutilizing substitutionalquantificationcaneliminate . In this section, theontological commitmentin naturallanguagesemantics we shall seethat it cannot.
DrawingMetaphysicalConsequencesfrom aT -Theory A number of objections have been raised against substitutional quantification in connection with its compatibility with truth - conditional semantics and the Davidsonian program .12I will not be addressing this concern; rather, I will be addressing the independent issue of whether substitutional quantification can free us from ontological commitment when doing semantics. The most notorious statement of the view that substitutional quantification doesnot embody a genuineconceptof existenceis found on p . 106 of Quine 1969: . . . substitutionalquantificationgives no acceptableversionof existenceproperly socalled , notif objectualquantificationdoes. Substitutional quantificationmakesgoodsense , explicablein tenDS of truth andsubstitution,no matterwhat substitutionclasswe take eventhat whosesolememberis the left-handparenthesis . To concludethat entitiesare beingassumedthat trivially, andthat far out, is simply to drop ontological questions. Against Quine, Parsons( 1971a, p . 232) has argued that substitutional quantification in fact " has a genuine claim to expressa concept of existence." Parsons (ibid ., p . 233) responds to the above argument from Quine by arguing that there are two fonnal featuresof the category of singular tenD Sthat make substitutional quantification with respect to singular tenns far less trivial than substitutional quantification with respect to the left parenthesis: First [substitutionalquantification] admitsidentity with the property of substitutivity salva veritate. Second, it hasinfinitely manymembersthat are distinguishableby the that ' (\7'x)Fx' is strongerthananyconjunction identityrelation. This hastheconsequence that canbe formedof sentences of the form 'Ft ' , while ' ( 3x)Fx' is weakerthanany disjunction of suchsentences . After considering the specific case of introducing a predicative theory of classes, Parsons(ibid ., pp . 234- 235) concludes that the substitutional quantifier will in fact have certain ontological commitments : . . . in the casewherethe termsinvolvedhavea nontrivial equivalencerelationwith infinitely classes , substitutionalquantificationgivesriseto a genuine" doctrine manyequivalence of being" to besetalongsideQuine's andothers.It parallelscertainidealistictheories of theexistenceof physicalthings, suchasthe accountof perceptionin Husserl's ldeen. What does Parsonshave in mind here? Parallels to Husserl are generally problematic , since there is wide variation in the interpretation of his work .13By some accounts Husserl is an idealist ; by other accounts he is a realist ; by still others ( e.g . Hall 1982) he is neither. On each account, however, the theory of object perception is highly complex . There is , however, a section in chapter 1 of Ideas ( 1972) in which Husserl suggestsa deep connection between syntactical forms and Stoffen ( sometimes translated as ' matter ' , sometimes as ' elements' ) . One possible interpretation of
Chapter4
Parsons's suggestionmight be that certain syntacticpositions of a sentence (for example, the subject or noun phraseposition) might show a privileged connectionto Stoffen. The word ' connection' is usedloosely here, since it suggestssomesort of chasmbetweenthe syntacticobject andStoffenwhenin fact theseforms are better thought of as playing an active role in organizing the world. " Theseare, as Dummett is fond of saying, " deepwaters, and somemight addthat the watersalsoaremurky. Nevertheless , thereareclearly manycandidate notionsof existencein the idealisttradition andelsewhere , and in at least somepartsof the tradition, existenceor beingis closelytied to certainkinds of . It standsto reasonthatnot all syntacticfonDS privilegedsyntacticrepresentations aresoprivileged, andthatthepositionsinto which we quantify substitutionally (nounphraseor subjectpositions) may well be privileged. A reconstructionof certainidealistconceptionsof beingcould easilyinvolveresourceslike substitutional quantification. In short, the moveto substitutionalquantificationdoes not necessarilyeliminateontological commitmentin the sensethat we areinterested in.14 , in view of the kind of deflationarymetaphysicalinvestigation Fundamentally beingproposedhere, it shouldnot appearparticularlybold or surprisingthat ourmetaphysical commitmentsaretiedto ouruseof quantificationoversemantic . A moveto substitutionalquantificationwouldhardly valuesin themetalanguage undercutthis project. In fact, onemight askwhetherit is not preciselythe best way of thinking aboutmetalinguisticquantificationwithin this program. " Thediscussionin this chapterhasbeenin theserviceof seeinghowwe can"read off a T-theory (and seeingthat if we haveagenuine metaphysicalconsequences T-theory we cannotavoid theseconsequences ) . But so far the discussion hasconcernedthe metaphysicalcommitmentsimposedon us by our use of . In thenextfew chaptersI will extendthis generalstrategy NPsandbasicpredicates es to two morecomplexT-theoriesthathaveincorporateddifferentapproach have different of tenseandwhich, accordingly to thesemantics , very metaphysical commitments .
ChapterS The B - Theory Semantics
Having sketchedthe basicsemanticframeworkand shownhow metaphysical commitmentscanarisewithin that framework , I will now turn to the semantics of tenseand begin to explorein a generalway a classof semanticaltheories that areconsistentwith the B-theorymetaphysics . I will beginwith what is essentially the receivedview: the Reichenbachiantheory of tense, which I call theB-theorysemantics . 5.1 The BasicTheory HereI will utilize theresourcesof eventquantificationto developa specificsemantic . theoryfor temporaldiscoursethatis consistentwith B-seriesmetaphysics Recallthe criteria that Gale( 1967) gavefor that metaphysicalpictureof time: . TheA -seriesis reducibleto the B-series. . Temporalbecomingis psychological. . The B-seriesis objective. All eventsareequallyreal. . Changeis analyzablesolelyin tenDS of B-seriesrelations(earlier-than, laterthan). A first attemptat a semanticaltheory consistentwith this picture would be to " " . Thatis, we wanttherighthand give tenselesstruthconditions for tensedsentences sidesof the T-theoremsto utilize only B-theory resources(e.g., the B seriestime line andthe before/afterrelation), andwe wantthe right-handsides to befreeof A -seriespredicates(including'past' and 'future' aswell astemporal indexicals). This ideacould be executedsomewhatasin ( 1), wherewe takethe tenseoperatorsto be quantifyingoverarbitraryfuture andpasttimes.. ( 1) a. An utteranceof ' Fredis hungry' at time t, is true iff Fredis hungryat time t
78
ChapterS
b. An utterance of 'Fredwashungry'attimet, is trueiff Fredis hungryattimet' earlierthant An utteranceof ' Fredwill be hungry' at time t , is true iff Fredis hungryat time t' later thant We might carry that generalidea over into our Val notation as in ( 1I), where Val(A ,B ,C) saysthat A is the semanticvalue of B relative to the utterance time C. ' (I ) a. Va1(e, PAST, t) iffe is temporallybeforet b. Va1(e, PRES, t) iffe temporallyoverlapst c. Va1(e, FUT, t) iffe is temporallyaftert Ironically , the simple semanticsjust sketchedis often the one that is sometimes imputedto Prior. Whateverthe meritsof the analysis, it pretty clearly is not onethat Prior could countenance ; it is repletewith quantificationoverpast andfuture times. The main problemfor a semanticsof tenselike the oneI havejust sketched is thatit is suitableonly for simpletenses , andfuture. When , suchaspast, present we getto morecomplextenses , suchasthefutureperfect, mattersbecomemuch like (2). moreinvolved. For example, we will needa semanticsfor sentences (2) Smithwill haveswumin the lake. ' Clearly, the event of Smith s swimming is not past, nor is it merely future. ' What is suggestedis that at somefuture point the eventof Smith s swimming will be past. 1\ vo other complextensesthat will haveto be dealt with are the presentperfectandthe pastperfect. Is thereanythingwe can do to accommodate these? OnepromisingB-theorystrategyis to borrowfrom thework of Reichenbach ( 1947) andfrom its usein grammaticaltheory by a legion of linguistsin various traditions,includingAqvist ( 1976), Guenthner( 1979), Hinrichs( 1986, 1988), Hornstein ( 1990), and Giorgi and Pianesi ( 1997) . The central idea in 's Reichenbach proposalwasthataccountingfor complextensesrequiresthatwe
TheB-Theory Semantics makereferenceto three distinct temporalpoints (or, on somefonnulations, events): S, R, andE, whereS is the speechtime, E is the time of the event, and R is a (possiblyindependent ) referencetime. As an illustration of the relationship these consider the caseof the future perfect in (2) again. among points, 2 that the timeof Smith' s swimming( E) is laterthanthespeech Intuitively, ( ) says time (S) but earlierthan someestablishedreferencetime ( R) . Thinking of a schematictime line (H ) on which pointsto the left areearlier thanpointsto theright, we havethe following standardReichenbachian analysis of the tenses2 : present: +--FJR/S- + past: +--FJR- S- + future: +--S- FJR- + pluperfect: +--E- R- S- + futureperfect: +--SE - R- + future in future: +--S- RE - + future in past: +--R- SE - + or +--R- ES - + 's We can introduceReichenbach proposalinto a truth-conditional semantics 3 in the following fashion. We beginby augmentingthe Val predicateto be a six-placepredicateVa1(A , BS , R, E, 0 ) assertingthatA is the semanticvalue of B at time of utteranceS, referencetime R, eventtime E, andassignmento. Axioms wouldincludethenonterminalaxiomsshownin (3) to coverthecaseof with intransitiveverbsandwould includethe axiomsshownin simplesentences . (4) for the tensemorphemes (3) a. Va1(T, [IPNP 1'], S, R, E) iff for somex , e, Val(e, 1', S, R, E) andat(e, E) and for somex, x is the agentof e andVal(x, NP, S, R, E) b. Val(e, [.. 1VP], S, R, E) iffVal (e, I , S, R, E) andVa1(eVP S, R, E) , c. Val(e, [vpV ], S, R, E) iffVal (e, V, S, R, E) (4) a. Va1(e, PAST, S, R, E, 0 ) iff RIB is earlierthan S
80
5 Chapter
b. Va1 , S, R, E, 0) iff S, R, andEtemporally (e, PRES overlap c. Va1 (e, FUT,S, R, E, 0) iff SisearlierthanR/ E d. Va1 PERFECTS , RE , 0) iffE isearlierthanSIR (e, PRES e. Va1 PERFECI thanRandRisearlier thanS , S, R, E, 0) iff Eisearlier (e, PAST f. Va1 , RE , 0) iff SisearlierthanE andE isearlierthanR (e, FUTPERFECTS g. Val(e, FUT IN PAST, S, R, E, 0 ) iff R is earlierthanS andR is earlierthanE Clearlyit wouldbedesirableto axiomatizethe semantictheoryin sucha way that the complextensescould be derivedfrom the simplerones. For example, one might introducean axiom for ' had' that when usedin conjunctionwith a past-tensemorphemewould yield the pastperfect, asin (5). (5) a. Val(e, h,@ ,d, S, R, E, 0 ) iff E is earlierthanR, andAt (e,E) b. Val(e, .:will, S, R, E, 0 ) iff S is earlierthanE, andAt (e,E) Consideragainthe caseof ' Smithwill haveswumin the lake' . In combination, the axiomsgivenin (5) statethat S is earlierthanE andE is earlierthanR- in otherwords, whencombinedthey yield the future perfect.4 This Reichenbachian analysisof tensealsoleadsto severalnaturalanalyses of temporaladverbssuchas ' yesterday', ' today' , and ' tomorrow' . For example , theycanbeconstruedasfixing thetime of thereferenceevent. Wemight introduce the auxiliary axiomsshownin (6) . (6) a. Val(e, ~esterday , S, R, E, 0 ) iff R is the day beforeS b. Val(e, ~ , S, R, E, 0 ) iffR is the sameday asS c. Val(e, tomorrowS , R, E, 0 ) iffR is the day after S of numerousauthorsin this framework, we canlikewise Followingthe suggestions ' ' ' ' 5 incorporatetemporalconnectivessuchas before and after asin (7).
TheB-Theory Semantics (7)
a. VaI(T,[1P IPI .b~ IP2], S, R, E, 0) iffVaI( T, IPI, S, RI, EI, 0) andVaI(T, IP2, S, R2, E2, 0') andEl isearlierthanE2 b. Val(T,[1P ! PI ~ 1P2], S, R, E, 0 ) iffVal ( T, ! PI, S, RI , EI , 0 ) andVal( T, 1P2, S, R2, E2, 0 ) andEI is later thanE2 c. Val( T,[1P ! PI ~ 1P2], S, R, E, 0 ) iffVal ( T, ! PI, S, RI , EI , 0 ) andVal( T, 1P2, S, R2, E2, 0 ) andEI temporallyoverlapsE26 Thecrucialthing to seein theexpositionthusfar is thatthetruthconditionshave beenrestrictedto eventsor timesandto a linearbefore/afterrelationholdingbetween them. In this sensethetheoryreflectsa genuineB-theoryapproachto time. Canthe approachbe extendedfar enough? 5.2 Extending the Theory TheB-theoryaxiomsofferedsofar only hint at how variousB-theorysemantics of tensehavebeendeveloped.The researchin this areais vast, and to survey theentireliteratureherewould beimpracticable.Thereare, however,a few constructions that bear comment, if only becausethey have figured in disputes overA -theoryandB-theoryapproach esto tense. Aspect First, it is necessaryto seehow a theoryof aspectcanbe incorporatedinto this framework , andto seehow it will interactwith the rest of the theory. Herethe standardmove(see, e.g., Comrie 1976; Parsons1991; Kamp andReyle 1993) hasbeento regardaspectasa predicateof events. For example, Parsonsintroduces the predicatesCUL (for " culminates") andPROG(for " in progress") to accountfor the perfectiveand the imperfectiveaspect, respectively.This idea canbe incorporatedin a straightforwardway with the introductionof axioms like thosein (8).
(8) -at(e, E) Val(e, FROG , S, R, E, 0') iff in-progress Val(eicUL, S, R, E, 0') iff culminates at(e, E) This allows a single event variable to have both temporal and aspectual properties predicated of it . For example , with the axioms I have introduced thus
Chapter5
far, thederivationof aT-theoremsuchas(9) will bepossiblefor ' Smithswam ' yesterday} (9) Va1 CUL]][vp[y swims][ADV Smith][r[.[TNs ]]]], S, (T, [IP[NP PAST ] [ASP yesterday -at(e, E) R, E, 0') iff for someeS is laterthanRIB, andAt(e,E) andculminates ande is a swimmingby Smith, andR is thedaybeforeS,andSmithis the agentof e . Syntactically Othersortsof aspectual canbeaddedin similarfashion , predicates as can be can be housed in the Int1 I nodes ; semantically , they () regarded they . thattaketheeventvariableastheirargument predicates TensedNominal~ The secondextensiononecanmakehasto do with the fact that not only verbs but alsonounphraseshavebeenarguedto haveargumentpositionsfor times. For example, En~ ( 1986, 1987) hasobservedthat in ( 10) thereis goodreason ' means " to supposethat ' hostages somethinglike everyonewho wasa hostage
" duringtimet.
( 10) The hostagescameto theWhite House. frameworkbeingsketched Thewayto incorporatethisideain theReichenbachian hereis to havethe lexical axiom for the noun to introducea kind of reference eventof its own.9For example, we might try ( 11) . ( 11) ' ' Val(x, hosta&e, R') iff thereis ane', e' aneventof x beinga hostage , andAt (e , R ) Therearepuzzlessurroundingthe propertreatmentof suchcases(seethe En~ referencesfor discussion ), but no puzzlesthat appearto posespecialproblems for the introduction of this idea into an absolutesemanticslike the onebeing sketchedin this chapter. Sequenceof Tense Also of considerableinterestis the " sequenceof tense" phenomenondiscussed by Smith ( 1975, 1978), Ladusaw( 1977), Dowty ( 1982), Stowell ( 1994), En~ ( 1987), Higginbotham( 1995), Hornstein( 1990), Abusch( 1997), Giorgi and Pianesi( 1997), and many others. The basic idea is that in certain embedded environments thereferencetime getsshiftedfurtherinto thepast. Consider( 12) .
The B-TheorySemantics
( 12) Mary saidthat Biff wasill . Therearetwo readingsto this sentence . In one, Biff ' s illnessoverlapswith Mary' s remarks;in theother, Mary reportedthatBiffhad beenill . Thepuzzlelies in trying to accountfor both readings. To get the shiftedreading, onegeneralstrategyhasbeento takethe internal event(the eventof Biff beingill ) asbeingpastrelativeto the matrix event(the eventof Mary speaking). As Higginbotham( 1995) spellsout the idea, we want to saythat the eventof Mary speakingis pastrelativeto the utteranceeventas in ( 13) (where< express esthe relation of temporalprecedence , u is the utterance event, and W is the contentof Mary' s utterance), andthe eventof which Mary is speakingis previousto this evente. ( 13) (3e)[ say(Mary, We ) ande < u] ( 14) ' ' ' (3e )[ill ( Biff, e ) ande < e ] ' ' , andusing [ I. . .1] to indicateanILF, Higginbotham Putting( 13) and( 14) together proposesthe analysisshownin ( 15) for the shiftedreadingof ( 12). ( 15) ' (3e)[say( Mary, [ 1(3e )[ill ( Biff, e') ande < e'] I], e), ande < u] Thepuzzlefor this analysisthenbecomeshowto accountfor thenonshiftedreading . OnHigginbotham'sproposal( 1995,p. 235) "the appearance of thepasttense in a complementclausecanbeanappearance cross reference takesplace ; merely asin the first case, but the tenseof the complementis present, not past." That is, ' wasill ' only appearsto be past-tensemorphologyin thesecases- thus, the relationbetweene ande' is not temporalprecedencebut overlap. The result is shownin ( 16) . ( 16) ' ' (3e)[say( Mary, [ 1(3e)[ill (Biff , e ) ande' includese] I], e), ande < u] Of course, if this solutionis available, thensotoo is the optionof havingthe internal referencevariableR pick out somearbitrary pastreferenceevent. In the caseof ( 12), the identifiedeventmight be pastrelativeto Mary' s utterance or the speaker's utteranceor it might be contemporaneous with either of these events; all that is specifiedfor sureis that R is pastrelative to S and that E is . pastrelativeto S. The relationsbetweenE andR would remainundeterD1ined
ChapterS This idea, subsequently characterizedasthe " independenttheoryof embedded " tenses, IOwas initially suggestedby Smith ( 1975) and Ladusaw( 1977) and wasworkedout formally by Dowty ( 1982). Hthe approachis right,II thenit is " " of tense. really a mistaketo talk aboutgenuine ambiguity in casesof sequence What we arewitnessingis simply the phenomenonof havingmultiple internal thanto referenceeventsavailablefor R- somethingmuchcloserto vagueness ambiguity. H this story works for SOTcases,how canit be extendedto so-calleddouble accessreadingsin caseslike ' Biff said Mary is pregnant' ? In this case, ' Mary had to be pregnantat the time of BifI s utterance; and shemust remain pregnantat theutterancetime S, whenthe sentenceis spoken.How is this to be ?Presumablyonewantsto invokesomesort with B-theoryresources represented ' of referenceto an interval that includesboth the time of BifI s utteranceand ' the speakers time S. Of course, herewe areignoring the really interestingempiricalquestionsquestionsconcerningwhich environmentsadmit DAR readingsand concerning 12 themechanism by whichreferenceto therelevantintervalariseS. My interest hereis not to solvetheseproblems, but to give a sampleof someof the possible framework. Theideais to illustratethe within theReichenbachian responses toolbox. As we will kindsof theoreticalresourcesthat arein theReichenbachian see, theA -theoristhasa rathermorerestrictedsetof tools available, andsome of the solutionsto thesepuzzlesmustbe different. esto the This hasbeena very superficialwhirlwind tour of B-theoryapproach at of resources there are a number of tense . of the Clearly problem sequence thedisposalof theB-theorist(referenceevents,possiblerelativeorderings,etc.), so onemight supposethat a solutionis within graspevenif difficulties remain in eachof theseareas. But if the B-theorist encountersdifficulties here, one naturallysupposesthat theA -theorist(who doesnot haverecourseto reference events,to therelativeorderingof events,or to anyotherB-theoryresources ) will be in muchgreaterdifficulty. I will returnto this issuein chapter7.
of theTheory 5.3 MetaphysicalCommitments As I noted in section 5.1, this semantic theory for tense is tied to a certain metaphysical picture of time . If we take the metaphysical consequencesof semantic theory seriously, then we shall be committed to a metaphysics in which future and past temporal points can be referred to and in which they are, in some sense, just as real as the present.
The B-TheorySemantics
This is so, for, asarguedin chapter4, we will at minimumbe committedto thoseentitiesthatwe quantifyover, andmetalinguisticquantificationovertimes is ubiquitoushere. If this wasn't obvious, it is perhapsworth noting that each of the axiomsthat containvariablesfor utterance, reference, and eventtimes (S, R, andE) is boundby implicit metalinguisticquantifiers. More important, in the aboveT-theory 3)-( 7 it is supposedthat the valuesassignedto these variablesstandin certaintemporalrelationsto one another(earlier-than, laterthan, etc.). Thus, it seemsthat this semantictheory is committednot only to the existenceof timesbut alsoto their standingin certaintemporalrelationsto oneanother(howeverthoserelationsareultimately to be cashedout). , however,the valuesof R, S, andE neednot be supposedto be time Perhaps . It is points , to employeventsalone. In fact, possible, with minor adjustments the only portion of our truth definition that appearscommittedto the idea that R, S, andE aretime pointsis the clause'At (e,E) ' in the axiomsin (3). Revised axioms, makinguseof only events,couldbe written alongthe lines of (3'). ' (3 ) a. Val(T, [IPNP 1'], S, R, E) iff for somex, e, Val(e, 1', S, R, E) ande = E andfor somex, x is the agentof e andVal(x, NP, S, R, E) b. Val(e, [r I VP], S, R, E) iffVal (e, I , S, R, E) andVal(eVP S, R, E) , c. Val(e, [ypV ], S, R, E) iffVal (e, V, S, R, E) In this casethe commitmentis only to the existenceof eventsthat standin certain temporalrelationsto oneanother.TImescouldbeintroduced,of course,but they would consistof setsof temporallyoverlappingevents.13 Quiteapartfrom theobviouscommitmentsto eithereventsor temporalpoints (or both), theremay well be othermetaphysicalcommitmentslurking in a full developmentof thetheory. Oneinterestingexample,discussedin Parsons1991, concernsthe metaphysicalcommitmentsof the axiom I inb' oducedfor the progressive . Thecoreissue, perhapsfirst notedby Aristotle, concernsthe so-called imperfectiveparadox. It appearspossiblefor thereto be an eventof my drawing a circle thatin somesenseis in progressbut which is nevercompleted(perhaps I am run over by a busbeforeI finish the circle). However, it seemsthat on the proposalgivenin (8) thereis a circle suchthat therewasan in-progress eventof my drawingit . Cantherebe incompleteobjects? Cantherebeincompleteevents? Parsons( 1991) is inclinedto answerin the affinnative. Thereare
Chapter5
alternativestrategies(someof which Parsonsdiscusses), but it is interestingto note that, dependingon how the theory is axiomatized, we may well be committed to morethanvanilla-flavoredeventsandtemporalpoints. For now, I will haveto set asidethesetantalizing metaphysicalquestions andfocuson the relativemeritsof the B-theoryandtheA -theory. My surveyof the B-theorysemanticsof tensehasbeenbrief, but perhapsit hasbeendetailed enoughto supportan examinationof the basicphilosophicalpresuppositions of the B-theory(andhence, aswill be seenin the nextchapter,detailedenough for us to seejust how andwherethe B-theoryfounders).
Chapter
6
Problemswith the B-Theory Semantics
A numberof objectionsto theB-theoryhavebeenraisedin thephilosophicalliterature , someof which aregenuinelysemanticalin character . Chief amongthe cited is the that the B theorist is unable to accountfor the inproblems worry dexicalcharacterof temporaldiscourse.As we will see, however, indexicality is not the only semanticalproblem. Thereis alsothe long-standingquestionof how the B-theoristis to treatconstructionsthat refer to linguistic tokens(constructions ' . like ' Thereis no language'and ' Therewereno utterances ) Thesesemanticalobjectionsmayor may not be solvablein isolation. (I seriously doubtthattheindexicalityobjectionis solvable.) However,whentheobjections arecombinedtheyfonn a seriousdilemmafrom whichthereis no escape . ' basic line of will be the : The B theoristcan t solve My argument following the problemof indexicality, andin any casethe mostpromisingtheoryoffered by the B-theoristis onethat, if adopted,makesit impossibleto providesemantics for sentences aboutlinguistic tokens. After takingup theproblemsindividually, 1will showhow togethertheylead to the B-theorist's dilemma. 6.1 Problemswith the Indexical Nature of Temporal Discourse) Let us returnto thecasediscussedin the introduction, in which 1know that my fifth anniversaryis (this) March 12but do not know that todayis March 12. As the examplewassetup in the introduction, 1 initially think out loud " My fifth anniversaryis March 12. 1 shouldthink aboutbuying my wife an anniversary " ' present. 1thentakeout a calendarto find today s dateanddiscoverto my horror " thattodayis March 12. 1subsequently shout My fifth anniversaryis today!" As in the introduction, we have the following two distinct utterances(or 1languagetokenings) on the table.
Chapter6
( 1) is (this) March12. My fifth anniversary (2) My fifth anniversaryis today. And, arguably,( 1) and(2) do not expressthesamesemanticalknowledge.( They certainlyexpressdifferentkinds of knowledge.) The semanticaldifferencebetweenpropositionslike ( 1) and (2) seemsto evaporatein the semanticaltheoryI introducedin chapter5. In that theory, the axiomsutilized for ( 1) and(2) would be ( 1') and(2'). ' ( 1) Val(e, March 12, S, R, E, 0' ) iffR = March 12 ' (2 ) Val(e, ~ , S, R, E, 0' ) iffR = S And sinceMarch 12 is identical to the time of utteranceS, thesetwo axioms deliverthe sameevaluationfor ' March 12' and ' today' . Shortof disquotationallyintroducing' today' into the right-handsidesof the truth conditions, it appearsthat the axiom is going to fall shortin characteriz ' ' ing the semanticknowledgethat the agenthas. But if we try to add today into the right-handside, we havesold out the B-theory andadmittedthat A -theory resourcesarerequiredto give the semanticsfor temporaldiscourse. A relateddifficulty hasto do with (3), utteredafter a dreadedvisit to the dentist's office. (3) I ' m glad that' s over with. About preciselywhatam I glad? By hypothesis,' that' refersto the eventof my ' ' going to the dentist, and overwith meanspast(andculminated). On the standard B theoryanalysis, this amountsto my sayingthat I am glad that my visit to thedentist's office culminatedat sometime earlierthanS, the time of the utterance . If my utterancewas at 5 0' clock, this amountsto my sayingthat I ' m ' ' gladthevisit culminatedbefore5 0 clock. But is thatreallywhatI m gladabout? Evenif onecanpacksomemissingingredientinto the RHSsof the axioms, there is the dangerthat the missing ingredient will undenninethe B-theory program. The missingelementon the RHShasto be somethingthat aBtheory . What could that be? metaphysicscancountenance Theseare not trivial problems. Although they havebeenlargely ignoredin the linguisticsliterature, they haveexercisedsomeof the leadingadvocatesof B-theorymetaphysics .
Problemswith the B-TheorySemantics
Whatis thebestresponsefor theB-theoristhere?TheleadingB-theorystrategy , basedon suggestionsby Reichenbach( 1947), Kneale( 1949), and Smart ( 1963, 1966), hasbeento invokethe ideaof "token-reflexive" truth conditions, in which the missingingredientis an explicit referenceto the tokenitself. For that" ' now' meansthesameas' thetime example,Reichenbach (p. 284) suggests ' " at which this tokenis uttered. The token-reflexivepart is the part wherethere is talk of " this token." By makingreferenceto the very tokeningof a sentence like ' I am hungrynow' , onecan situateit relativeto one' s own temporalposition . Noticethatthis is a muchricherstorythanthestorythatsurvivesin thecontemporary Reichenbachian theoriesof tensesurveyedin chapter5. In the exampleof my fifth anniversary , thedifferencein therespectivetreatments of ' today' amountsto the differencebetween(4) and(5).
(4) ' An utterance is today'is trueiff my (tokening ) u atS of My fifth anniversary is atS (March12). anniversary ( 5) This utterance (tokening ) u at S of ' My fifth anniversary is today ' is true iff my anniversary is the day of S- the time of this very utterance, The treatment in (5) is the one that Reichenbach seemsto have ( informally ) proposed , The treatment in (4 ) appearsto be the one adopted by contemporary linguists who have executed Reichenbach' s general program for the semantics of tense in natural language} But clearly (4) is too austere, Reichenbach correctly saw that one needsto say more, Generalizing the token -reflexive strategy, Smart ( 1966, pp , 133- 134) proposed treating the basic tensesin a similar way3: Let usreplacethe words" is past" by the words" is earlierthanthis utterance," ( Notethe transitionto the tenseless" is," ) Similarly, let us replace" is present" and " now" by " is simultaneouswith this utterance," and " is future" by " is later than this utterance," , , , Notice that I am here talking of self-referential utterances, not self-referential sentences , (The samesentencecan be uttered on many occasions,) We can, following Reichenbach, call the utteranceitself a " token," and this sort of reflexivity " token" " " reflexivity, Tensescanalsobe eliminated, sincesucha sentenceas he will run canbe " he runsat somefuture time with " runs" andhence " he tenseless runs replacedby ( ) by later thanthis utterance," Similarly, " he runs" means" he runs (tenseless simultaneous ) with this utterance," and " he ran" means" he runs (tenseless ) earlier than this utterance ", All the jobs which canbe doneby tensescanbe doneby meansof the tenseless " " of way talking andthe self-referentialutterancethis utterance, This is all very well (or let us supposethat it is for now ), but theorems like ( 5) have to be generated with the help of axioms in the T- theory , Precisely where
Chapter6
doesthereferenceto theutterance(tokening) enterinto thepicture?As Yourgrau , thereareseriousdifficulties involvedin trying to ( 1987, note21) hasobserved constructa generaltheory that can introducethe neededdemonstrationof the utteranceevent. To seeYourgrau'spoint, consideranattemptto constructa generaltheorythat allowsdemonstrationof the utteranceevent. We might startby sayingthat the reflexiveclauseentersinto the RHS asin (6). (6) Val(e, ~ , S, R, E, 0' ) iffR is the sameday asthis very utterance(I -language tokening) So the RHS of the axiomintroducesa referenceto the utteranceitself (or the 1. If onelearnsthe axiom in languagetokeningitself). But this is not acceptable rule. If that is the case,thenwhatis the force to learn it as a 6 one has ( ), general ' ' ? If it is a in thisveryutterance of thedemonstrative , then genuinedemonstrative ' it shouldpick outthesameutterance(tokening) of (6). If it isn t a genuinedemonstrative . , thenit is not clearhow thetoken-reflexivetheoremcanbe generated This looks like a kind of a technical objection, but in fact it points to a very deepproblem with the token-reflexive response.Specifically, when the ' B-theoristattemptsto smuggleanindexicalexpression(suchas' this utterance) into theRHSof a theorem , theB-theoristis breakingfrom theusualpathof keeping the literal truth conditionsfree of suchelements.If the useof the indexical on the RHS is to makeany senseat all , it appearsthat ' this' hasto be treated as an indexical predicateof the sort that was discussedin section3.2. But if indexical predicatesare possiblein this case, why not allow them in the case of temporalindexicals? Still moreperplexingfor the B-theorist, the indexical elementin ' this utterance' looks an awful lot like a temporalindexical predicate . It certainly isn' t spatial; nothing in the perceptualenvironmentis being . It looks for all the world as if the extra indexical elementjust demonstrated ' meansnow, and asif the expression' this utterance meanssomethingakin to ' ' the utterance happeningnow ! ' I don t know if thesedifficulties canbe overcome,but a patternis beginning to emerge. The B-theorist can ladle more expressivecontentinto the RHS of the truth conditionsof a tensedsentence , but the theory beginsto capturethe of a tensed indexicalcharacter expressiononly at thepoint wheretheaddedcontent looks strikingly asif it containsan indexicalpredicate.If this is right, then it suggeststhat the B-theoristsimply cannotdo justice to the contentof tensed language.But evenif this patterncan be brokenandthe problemof indexical charactercanbe solvedby meansof a strategylike the token-reflexivetheory,
Problemswith the B-TheorySemantics the token -reflexive solution leads directly to an even more intractable puzzle about sentencesimputing that there are no utterances or linguistic tokens.
" 6.2 "ThereAre No Utterances In the previoussectionwe sawthat the bestchancethe B-theoristhasof capturing indexicalcontentis by utilizing token-reflexivetruth conditions. The resulting semanticalpicturefor the B-theoristis somethinglike (7) . (7) a. This utteranceof ' Fredis hungry' at time t is true iff Fredis hungryat t- the time of this very utterance. b. This utteranceof ' Fredwashungry' at time t is true iff Fredis hungryat time t ' earlierthant- the time of this very utterance. c. This utteranceof ' Fredwill be hungry' at time t is true iff Fredis hungryat time t' later thant- the time of this very utterance. The token-reflexive strategyis designedto allow B-theoriststo both haveand eattheircake. Thatis, thetheoryis designedto allowB-theoriststo situateevents in their past/present/futurewithoutinvokingA -theoryresources . Unfortunately, theB-theoristdoesnot getto eator to keepthis cake. The theoremsin (7) tell us that a specificutteranceu* of ' Fredis hungry' is true iff Fredis hungry at the very time that u* is uttered. This is the deviceby which (assumingI am the speaker ) the proposalcan situateeventsrelative to . , theB-theoristarguesthattheoremslike those mytemporalposition Furthennore in (7) givetenseless truthconditions(i.~., B-theorytruth conditions) becausethe "RHSof thebiconditionalis effectively" detensed only B-seriesresourcesare 4 calledupon. In the previoussectionI notedthat thereis alreadysomeslippagein the Btheoristposition, sincethe RHS of the theoremis not completelypurgedof indexicalcharacter . But, whetherthesetruth conditionsarereaily tenseless or not, it is most likely the casethat they simply fail to give the correcttruth conditions in a broad classof cases. If referenceto an utteranceis required, then therewill be caseswherethis extra requirement- this extra bit of contentthrowsa monkeywrenchinto the semantics . The basicproblemis not uniqueto token-reflexive treabnentsof temporal . Castefiada( 1967, p. 87) showedthat it alsoarosein treabnentsof inlanguage dexicalslike ' I ' :
Chapter6 " , ' Reichenbach , for instance,claimsthat the word of meansthe sameas the personwho " uttersthis token. This claim is, however, false. A statementformulatedthrougha normal useof the sentence" I am utteringnothing" is contingent: if a personuttersthis sentence he falsifies the correspondingstatement , but surelythe statementmight, evenin sucha case, havebeentrue. On the otherhand, the statementsformulatedby "The person " utteringthis tokenis utteringnothing areself-contradictory: evenif no oneasserts them, they simply cannotbe true.
Smith( 1993) andCraig( 1996a ) havesubsequently arguedthatthisargument of temporal . tothetokenreflexiveanalysis expressions appliesmutatismutandis -reflexivetruth-conditions of thisformfounderin caseslike (8). Token (8) . Thereis no spokenlanguage to SmithandCraig,onthetoken-reflexivetheorythetruthconditions According to comeoutasin (9). of (8) appear (9) ' is true iff thereis no This utteranceat time t of ' Thereis no spokenlanguage spokenlanguageat t- the time of this very utterance. Statement(8) is clearly false, but statement(9) allegedly incorrectly entails ' is thatthesentence'Thereis no spokenlanguage necessarilyfalseisThis appears to bea technicalobjection, but Smith( 1993) andCraig( 1996a) havearguedthat thereis no way to maneuveraroundit andthat, asa consequence , the objection tenseless of the token reflexive the undermines theory ultimately plausibility of time. For example, onemight hold that it is not crucial that onefix the relevant time by utteringa sentence , sinceit is equallypossibleto givethetruth conditions asin ( 10). ( 10) ' This tokening(verbalor mental) at time t of 'Thereis no spokenlanguageis true iff thereis no spokenlanguageat t- the time of this very tokening. like 'Thereareno mental But ( 10) will apparentlyfail whenwe considersentences ' or verbaltokensof language, which, thoughclearly false, do not appearto be necessarilyfalse. This objectioncanbe generalizedto anyproposedvehicle : written, spoken, thought, intuited, etc. for the tokeningof a sentence of certainsentencetypes6will not help either, utterances to Appeals possible for a possiblepasttenseutteranceby me will not be enoughto situatean event in my past. It is no comfortto me that my dentistvisit lies in the pastrelativeto a possibleutteranceby me at sometime. I haveto know thatthe eventlies in the pastrelativeto the time I amtokeningmy thought/expressionof relief.
Problemswith the B-TheorySemantics
Similar considerationsapplyto past-tensesentences like ( 11). ( 11) Therewasa time in which therewasno spokenlanguage(wereno tokens). Theworry is that ( 11) seemsto entailthatthereis sometime in thepastin which (8) holds. But if the truth conditionsof (8) areas in (9), how is this possible? Statement(9) suggeststhat (8) canbe true only if thereis an utteranceof a particular token, but thatwould be to underminethe very thing that (8) is asserting . Hereit is naturalto notethe similarity betweenexampleslike (9) andsimilar casesfrom Kaplan ( 1977, 1979) (e.g., ' I am herenow' ) . If we want to follow ' , thenwe wantto removetalk of theutteranceeventfrom theliteral Kaplans strategy truth conditions. Perhapsthe mostnaturalway to do this in our framework wouldbe to invokethe Larson-Segalmechanismof conditionalizedT-theorems discussedin section3.2 of thepresentbook.7As appliedhere, the ideawould be to conditionalize(8) asshownin ( 12). ( 12) If u is utteranceat time t of ' Thereis no spokenlanguage', thenu is true iff thereis no spokenlanguageat t. This fixes the evaluationto the time of utteranceof the token, but it keepsthe quantificationoutsideof theT-theorem. It alsoavoidsthe needto introducep0tential (possiblebut unactual) utterances .s Crucially, thetruth conditionsdeliveredin ( 12) do not make(9) a necessary ' truth. In similarfashion, thepast-tenseversion( 'Therewasno spokenlanguage ) would receivethe truth conditionsin ( 13). ( 13) ' thenu is true iff If u is utteranceat time t of 'Therewasno spokenlanguage , ' thereis a time t earlierthant suchthat thereis no spokenlanguageat t '. Thoughtheproblemof referenceto utterancescanthusbe resolved, noticethat we resolvedit by strippingthe referenceto the utteranceeventfrom the RHSs of the truth conditions. But we therebyundid the very movethat allowedus to accountfor the indexicality of utteranceslike ' My fifth anniversaryis today' ! This is the B-theorist's dilemma.
6.3 TheB-Theorist's Dilemma The strategy the B -theorist used in solving the problem of sentenceslike ' There is DOlanguage' was to say that the utterance that anchored the time S did not make it into the truth conditions proper. But the most promising solution we have
Chapter6
for the problemof temporalindexicals- the token-reflexivetheory- crucially reliesuponthe assumptionthat the utteranceanchoringSdoesmakeit into the truth conditions. Hence, the B-theoristfacesa rathernastydilemma: eitherremove talk of the utteranceeventfrom the truth conditions(in which casethe mostpromisinganswerto the problemof temporalindexicalshasto be abandoned ) or allow suchtalk (in which caseutteranceswhich claim that thereis no languageendup beingnecessarilyfalse) . The optionshereare limited. Higginbotham( 1995) hasattemptedto take the secondhorn of the dilemma, holding that talk of the utterance(tokening) must makeit into the truth conditionsbut that someof the information need not makeit into the truth conditionsif we evaluatethe sentenceundercounter" factualsituations.Wecan, accordingto Higginbotham,allow " modaldiscards; that is, we can allow information to be discardedwhen the sentenceis evaluatedin otherpossibleworlds. The generalidea behind using discardsis the following : The truth conditions for ' Thereis no language'arepreciselythosegivenin (8); however,if we evaluatethat very samesentencein certainpossibleworlds (in particularthe worlds in which thereis no language ), the truth conditionsarethe leanerones that we get in ( 12). The ideaof modaldiscardsmay havemerit in somecases(the lesscharitable ), might saythatthe mechanismamountsto discardingthe counterexamples but it is hardto seethat it helpswith the B-theorist's dilemma. Consider( 14). ( 14) If therewerecurrentlyno languageI would be relieved. ' Supposethatw is theactualworld andw is theworld in which I amrelievedthat . Thesenseof relief I feel in world w' cannotstemfrom there thereis no language being no languageat sometime t, which happensto be the time of my utterance of ( 14) in world w (the actualworld). Why would I careaboutthat in w' ! Even if we could find a way to makethe modal discardsolution wort, we haveto keepin mind that a numberof otherproblemswhich we setasideearlier still nagthegeneraltoken-reflexivetheory(with or withoutmodaldiscards). ' ' Thereremainstheproblemof theindexicalcharacterin drisutterance, andthere remainsthe problemof generalizingthe theory (andincorporatingit into a finitely ' axiomatizedT-theory) . Quitefrankly, the secondhorn of the B-theorists dilemmais not a very appealingoption. As far asI cansee, the only remainingoptionfor the B-theoristis to takethe ' first hornof thedilemmaandtry to arguethatwe don t needto keepthe indexical characterin the truth conditions at all- that our knowledgeof indexical
Problemswith the B-TheorySemantics
characteris extra-semanticalknowledge.The strategywould be to saythat we havetensedthoughts(for example, that my anniversaryis today), but that we ' t haveanA Deedn . This, asI understand theorysemanticsandanA -theorymetaphysics is the move advocatedby Mellor ( 1981). In short: if we needtense, it, we cankeepit in therealmof psychology(belief) andout of therealmof truthconditionalsemanticsandmetaphysics . In earlier chaptersI arguedthat metaphysics, semantics, and psychology arenot easilyseparated . This may be a goodtime to revisit that claim- particularly now that we know what is at stake.
6.4 On Mellor's WayOut Mellor ( 1981, chapter5) concedesthat tenseis indispensable , andthat indeed we rely on it to explainour actions(for example, my actingthe way I do when I discoverthattodayis March 12). But Mellor holdsthatit is enoughthatmy beliefs betensed.Onhis view, a commitmentto tensedbeliefsentailsnothingabout therebeingtensedtruth conditionsfor my tensedutterances , andcertainlynothing aboutreality' s beingtensed. The flaw in this line of reasoningis that beliefs too musthavea semantics , so it is hardto seewhy allegedproblemsthat occurat the level of languagedo not likewiseoccurat the level of belief.9That is, if thereis somecontradiction inherentin having temporalindexicalsin our semanticsfor naturallanguage, for belief?Indeed why is therenocontradictionif we allowthemin oursemantics , if theconsiderations in section1.3 aboveareon theright track, thenthelanguage of thought (i .e., the languagein which our beliefs are couched) is simply 1languageto begin with ! One can hardly say that I -languageis untensedand that the languageof thoughtis tensedif I -languageis the languageof thought. Evenif we reject the identity of I -languageandthe languageof thought, it seemsto me that Mellor' s strategyis fatally flawed. In the first place, we have 's beensupposingthat a semanticaltheoryis a theoryof the speaker knowledge ' of meaning.If thatis right- if thespeakersknowledgeis partof thepicture- - and if knowledgeandbelief aretensed,thenit is very difficult to seehow tense(and indexicalityin general) is goingto be expungedfrom the semanticaltheory. Mellor clearly is forcedto adopta piCtureof semanticaltheory that cannot allow notionssuchasknowledgeof truth conditions, butjust how adequatecan sucha semanticaltheorybe? It certainlycan' t serveasa theoryof meaning; it would fail to delivertruth conditionsin a way that displayedthe sensesof the sentences of the language , andthereforeit would fail oneof the minimum criteria of adequacyfor a semanticaltheoryasdiscussedin chapters2 and3.
Chapter6
( Mellor, afterall, maynot beinterested Quiteapartfrom theseconsiderations in the possibility that aT -theorymay be ableto serveasa theoryof meaning), whatwould it meanto saythatwe havetensedbeliefsbut aBtheory semantics ? If the world containsonly B-theoryresources andmetaphysics , thenprecisely how do we avoidhavingaBtheory psychology? The illusion of a possible way out here is fostered by thinking that there could be psychological concepts that are, as it were, disembodied- cut off from the actual world in important ways. How can a psychological property (call itfoo ) that bearsno relation to tensein the actual world have anything to do with tense? It is no good to say that our abstract property foo is tensed because it is grounded in our time consciousnessor temporal perception . That merely keeps the question one step removed. Then we must ask what it is about time consciousness and temporal perception that makes them tensed. Why do we call consciousnessor perception tensedif it doesnot correspondto something tensed in the actual world ? The weaknessin such appealsto psychology can be characterizedin a general way. Impressed by the ability of psychology to explain isolated caseswhere we are deceived ( sticks that appearbent when seen in water, etc.), we ask psychology to account for catastrophic mismatches between our philosophical theory of the world and the way we perceive the world to be. But psychology cannot shoulder this burden. As Burge ( 1986) has argued, psychological states ( particularly IO perceptualstates) are individuated in part by relations to the external world . In this case, that meansthat if the world is not tensedthen it is difficult to seehow our perception of the world could be tensed.II But if psychology cannot supply the missing tensed component, there is no avenue of retreat for the B - theorist. If tense is to be expunged from theseman tics , then it cannot be retrieved elsewhere. Here we have a clear case where both semantics and psychology place demands on our metaphysics, and aBtheory metaphysics appearsunable to meet the demand. Can the A theory alternative fare any better? This question will be addressedin the next chapter.
7 Chapter -. -The ATheory Semantics
7.1 The Appeal of the A -Theory Semantics The intuitive appealof theA -theorysemanticscomesfrom its ability to handle the indexicalnatureof temporaldiscourse.Unfetteredby the burdenof delivering tenselesstruth conditions, the A -theory can useindexical predicatesin the metalanguage to delivertensedtruth conditionsthat preservethe indexical characterof temporalphenomena . Consider,for example,the temporalmorphemePAST. On the Btheory, that expressionreferredto (or quantified over) pasteventsor times. In chapter6 we sawthe limitations of sucha strategy.What is the alternative? The ideais that, on thebasisof thediscussionof indexicalsin section3.2, theA theorycan treat PAST as an indexical predicatethat holds of a p,roposition-like object, effectively displaying the indexical senseof the past-tensemorphemeon the right-handsideof the T-theory axiom: ( 1) a. Val(x , PAST) iff x wastrue b. Val(x, PRES) iff x is true c. Val(x, PUT) iff x will be true Wecanhandletemporaladverbsin a similar fashion, takingthemas predicates of proposition-like objectsratherthanasreferringto timesor events: (2) a. Val(x, ~esterday ) iff x wastrue yesterday
98
Chapter7
b. Val(x, ~ ) iff x is true today c. Val(x, tomorrow) iff x will be true tomorrowd . iff x is true now Val(x , ~ Theremay alsobe waysto reducetheseaxiomsto a handfulof indexicalpredicates ' ' , or evenone. For example, tomorrow might receivetruth conditionslike ' ' Val(x , tomorrow) iff x is oneday after today. In turn, today might be defined ' ' 1 by useof the indexical predicate now . Nor is it necessarythat this decomposition take place in the truth conditions- it might just as well occur in the lexicon. The crucial point is that theremustbe at leastoneA -theory predicate ' to situatethe describedsituationin the speakers egocentricspace. " ' To illustrate, let mereturnto thecase,discussedin chapter6, whenI say I m " ' glad that s over with after a visit to the dentist. What I am glad aboutis that my visit to the dentistculminatedin the recentpast i.e., that it is no longerin ' or any other suchdevices reflexives token invoke to my present. We don t need to get this result; it falls out directly from the axiomsintroducedthusfar. , andreflect semantical The two theoremsshownin (3) display ratherdifferent senses . character different of knowledge substantially (3) a. Val(T, [sPAST[(3e)(e is the eventof my havinga root canalande culminates]]) iff [] [(3e)(e is the eventof my havinga root canalande culminates)] [] wastrue b. If u is an utterance, at t , of [s PAST[(3e)(e is the event of my having a root canalande culminates)]] , thenVal(Tu ) iff (3e)(e is the eventof my havinga root canalande culminatesande < t) ' WhenI amgladthatmy visit to thedentistis over, whatI m gladaboutis something that hasthe truth conditionsshownin (3a), sincethosetruth conditions effectivelydisplaythe indexicalcharacterof the describedstateof affairs and situateit in my past} I may be entirely indifferent to whethersomethingthat hasthe truth conditionsshownin (3b) shouldobtain. Now, it is certainly the casethat A -theory axiomslike thosein ( 1) and (2) are " modest" axiomsin the sensediscussedin section2.3, andit is fair to ask what the deeperanalysis- the elucidation- of the axiomsfor thesetemporal indexicalsis going to look like. That is, when we saythat a propositionwas true or will be true, exactlywhat arewe gettingat?
TheA -TheorySemantics
' As Dumrnett( 1969) hasargued,a semantictheorythataccountsfor an agents semanticknowledgemustshowhow portionsof the languagearelearnedfrom the evidenceavailableto the languagelearner. But now considerhow we learn to usepast-tenseexpressionssuchas(4) . (4) Dinosaursroamedthe Earth. Wedo not evaluatethis sentenceby imaginingsometime earlierthannow and detenniningwhetherat thattime (4) is true. Rather,we evaluate(4) by right now conductingthe sort of investigationthat is appropriatefor past-tensestatements like (4). (For example, we might studyfossil records.) Likewisefor any otherpast-tensestatement . Wehavecertainproceduresfor detenniningwhether a past-tensepropositionis true, andtheseproceduresdo not involve the evaluation of a propositionat sometime past; rather, we simply evaluatethe proposition in a particular way- a way which is independentof how we evaluate present-tenseandfuture-tensepropositions. Considerthe future-tenseproposition(5). (5) The economywill recoverin the third quarter. Clearlywe do not evaluatesucha propositionby picking sometime in the third quarteranddetenniningwhetherit is true at that time that the economyis recovering . Rather, we evaluateit by studyingthe currently availableeconomic data. Crucially, our evaluationof (5) canproceedwithout our ever attendingto a correspondingpresent-tensepropositionat somefuture time index. If this pictureof the underlyingrobusttheoryis correct, thenit immediately leadsto a secondadvantage for theA -theoryproposalunderdiscussion - in fact, a strikingepistemologicaladvantage . TheB-theoristis in theuntenableposition of assertingthatthereis actuallyreferenceto pastandfuturetimesand/ or events. However, this flies in the faceof everythingwe know aboutreference.We are in neithera perceptualrelationnor a causalrelationwith future events,andour causalconnectionwith mostpasteventsis tenuousat best. In regardto times, the ideathat therecould be referenceto suchabstractobjects surely requires majoradjustmentsto currentepistemologicalthinking. It is no goodto takethe standarddodgeandarguethattheB-theoristis using ' reference' in a looseand nonphilosophicalsense. Pastandfuture eventsand timesarequantifiedoverwith impunity in the B-theoryfragment; they serveas semanticvaluesin the theory, and, aswe sawin chapter4, thereis no escaping theontologicalcommitmentto semanticvalues.In short, thereis no escapingthe
100
Chapter7
metaphysicalcommitmentto theseentities, andwith thatcommitmentcomesa heavyepistemologicalburdenthat the B-theoristsimply mustown up to. Sofar, I havebeentalking abouttheA -theory(andits advantages ) at a pretty . abstractlevel. Perhapsit is time to go into a little moredetailaboutthesemantics
7.2 TheBasicA-TheorySemantics with thework of Prior( 1967, of tensearecorrectlyassociated A -theorysemantics . Ordinarily it is supposed 1968), but Prior' s position is still widely misunderstood ' ' ' that Prior took ' past' , ' present, and future to be quantifiersover past, ' present,andfuture times. For example, it is ordinarily supposedthat on Prior s ' ' Prior theory PAST[S] meansthatS wastrueat sometime earlierthannow. But ' nevergavesucha semanticsfor his tenselogic, nor couldhe. He didn t believe in future or pastevents. He endorseda kind of presentismsimilar to that discussed in the introductionto this book. The sourceof the confusionmay be that Prior neveractuallygavea semantics for his tenselogic. I will nowdo so, in a way thatis consistentwith Priorean in whichthetenses . My basicstrategywill beto developa semantics metaphysics . as like that take areindexicalpredicates objects their arguments proposition truth and In chapter5 we saw how the resourcesof eventquantification conditionalsemanticscouldbe exploitedasa frameworkfor a Reichenbachian semanticsfor tense. In this section, we shall makeuseof someresourcesintroduced S. Theideawill bethattense in chapter3- namely,interpretedlogicalfoOD . For example, canbe construedasa predicatethat takesll..Fs as its arguments we can takethe basictensemorphemesto havethe simple axiomsin (6), and ' ' we canhavethenontenninalaxiomfor sTNS S1 ] introducethepropositional object(in this case, an ll..F) . (6) a. Val(x, rAm iff x wastrue b. Val(x, : EUI) iff x will be true c. Val(x, : EUI) iff x will be true d. Val(T, [8TNS SI ]) iff , for somex, Val(x , TNS) andx = OSlO Of course, this analysisassumesa certainpicture aboutthe syntacticform of at which theseconstructions- specificallythatthereis a level of representation
TheA-Theory Semantics
101
thetensemorphemes takeclausalscope . Presumably thesetenseoperators originated in S-Internalposition(e.g., at infl) andadjoinedto Sat LF, sotherepresentations wewill beconsidering areassumed to beLF representations in the sense discussed in appendix TI . It is sometimes thatthecomplex tenses fallout argued fromthenesting 'PAST' 'PRES' naturally of theprimitivetenseexpressions , , and'PUT' . Forexample , it is naturalto suppose thatthefutureperfectcouldbeconstrued ashaving thesyntax(7), thatthepastperfectwouldreceivetheanalysis(8), andthatthe present perfectwouldreceivetheanalysis(9). (7) [sPUT[sPAST[S])] (8) [sPAST[sPAST[S]]) (9) [sPRES [sPAST[S])] If boththeauxiliary'had' andthepast-tensemorpheme riseandtakeclausal the result is the LF in 10 . , scope ( )
(10) [8had[8PAST[8Smithgoes ]]] As we will seea bit later, this proposalturnsout to be inadequate . It seemsthat someform of temporalanaphorais necessaryto accountfor genuinecasesof pastperfect, future perfect, etc. The additionalresourcesnecessaryfor temporal anaphorawill be introducedin chapter8. If we like, we cantreattemporalconnectivessuchas ' before' and ' when' disquotationally, asin ( 11).3 ( 11) a.
Val(T, [881~ 82]) iffVal(T, 81) whenVal(T, 82) b. Val(T, [881:b~ 82]) iffVal(T, 81) beforeVal(T, 82) c. Val(T, [881~ 82]) iffVal(T, 81) afterVal( T, 82) ' are , suchas' yesterday Temporaladverbs , simplypredicates takingI LFsas as in 12 . , arguments ( )
102
7 Chapter
( 12) a. Val( T, ~ ADV (temp) SID iff thereis anx, suchthatVal(x, ADV ) andx = [] SI [] b. ) iff x wastrue yesterday Val(x, ~esterday This is just an initial glossof anA -theory semantics.(I will flesh out the details in chapter8 andin appendixT5.) Beforedevelopingthe theory further, 1 needto dealwith a numberof objectionsthat havebeenraisedin the literature . of semantics 7.3 SomeObjections to the A -Theory Semantics In this section I will canvasssome of the semantical arguments against the Atheory semantics of tense in preparation for answering them in chapter 8. The arguments to be considered involve some alleged problems surrounding the nature of temporal anaphora and some alleged difficulties with the handling of nested temporal operators. Embedded Tenses and Nested Temporal Modifiers Hinrichs ( 1981) and Dowty ( 1982), considering earlier operator theories of tense ( principally that of Montague ( 1974 , observed that there are potential difficulties with a simple sentencelike ( 13) and its two possible LFs , ( 14) and ( 15) .
(13) . Smithleftyesterday (14) ]]] [5Smithleaves [5PAST[5yesterday (15) ]]] [5PAST[5Smithleaves [5yesterday On a theoryin which tensesaretakento be quantifiers, the LF in ( 14) is interpreted asassertingthat in the pastit wastrue that Smith departedon the previous day. This obviouslydoesnot yield the correcttruth conditions. On the other hand, ( 15) will be interpretedas sayingthat yesterdayit was true that Smith hadleft- clearly not consistentwith our understandingof ( 13). of a problemherestemsfrom the questionableassumption The appearance that tensemorphemesand temporaladverbialsarenestedandthereforemust take scopeover one another. But thereis really no reasonto supposethis. As
TheA-TheorySemantics
103
Dowty ( 1982) notes, it is equallyplausibleto supposethat the adverbandtense morphemesshouldbe sisternodes, asin the following structure:
S / " ", / / . . . .... TNS r "'...-.-S ADV 6 leaves PAST Smith yesterday
The axiom for this structurecanbe written as( 16).4
( 16) Val( T, sADV TNS S1 ]) iff , for somex, Val(x, ADV), Val(x, TNS), andx = OS1[] The UnconstrainedNature of the A -Theory Anotherimportantobjectionto theA -theoryhasbeenthatit is "unconstrained," ' meaningthat the devicein Prior s logic of nestingtemporaloperatorsseemsto wildly overgeneratethesetof possibletenses.sTo seethis, supposethatthenesting of tenses, as in FUT [PAST[ S]], could provide an accountof the present ' perfect. (As we will see, it doesnt evendo that, but let us setthis concernaside for the moment.) If tensescan be nestedin this way to generatenew complex ' tenses , thenwhy can t we nestthemarbitrarily deep? For example,why is there no tensecorresponding to PAST[ pAST[ pAST[S]]] or to PAST[ FUT[ pAST[S]]] ? This is of genuineconcernif we areinterestedin anaccountof tensethatgoes someway towardsexplaininglinguistic competence . It is altogethertoo easy to constructtheoriesthat overgenerate the setof possiblesentences . What we would like to haveis a theory that is constrainedenoughto generatethe sentences of our languageandonly thosesentences . the , unconstrained natureof the theory and Unfortunately despite apparent its propensityto overgenerate the possiblenatura1language sentences , the theory appearsto undergeneratetoo: it is allegedlyunableto accountfor a broad classof temporalphenomena . In particular, it hasno apparentresourcesto account for temporalanaphora . The Apparent Needfor Temporal Reference At lastwe cometo thecentralsemanticalobjectionto Prioreantheoriesof tense. In the introduction we consideredthe exampleshownhereas ( 17) (Partee 1973, 1984).
104
Chapter?
( 17) 1turnedoff the stove. Clearly, ( 17) doesnot merely meanthat at sometime in the past1 turnedoff the stove. Without a doubt there havebeenmany suchepisodesin my past. Accordingto Partee, ( 17) is informativebecausethereis implicit referenceto (or quantificationover) somespecifictime or referenceevent. 1might equally ' ' ' well haveuttered' I turnedoff the stovethen , with then servingasa temporal anaphorreferringto somesegmentof time or eventin the past. Similar considerationsapplyto ( 18) (Partee1973). ( 18) Smith did not turn off the stove. If we think of negationandthe past-tensemorphemeassimplesententialoperators , thentherearetwo possiblerelativescopesfor sentence( 18) : ( 19) [ not [ PAST[ Smith turnsoff the stove]] ] (20) [ PAST[ not [ Smith turnsoff the stove]]] But neitherof thesegetsthetruthconditionsright if we areutilizing thesemantics we suppliedfor theory A . The problem is that ( 19) would be given the truth conditionsthat it is not the casethat [] [sSmith turnsoff the stove] [] wastrue in otherwords, thathehasneverturnedoff the stove. (20) wouldbe assignedthe truth conditionsthat [] [s it is not the casethat Smith turnsoff the stove][] was true (which could havebeensatisfiedby any pasttime when Smith refrained from touchingthestove). Clearlyneitherof thesetruthconditionsis whatwe are after. Rather, thereis the sensein ( 18) that Smith failed to turn the stoveoff at somecrucial time - for example, whenhe left the housethis morning. This problem seemsto lie at the heart of another objection to Priorean theories: their allegedinability to accountfor complextenses.The objection is that, for example, [ PAST[ PAST[ S ]]] simply collapsesinto the simple " " past. Recallthat in chapter5 we calledthe minimum unit of time a chronon. Then, at best, [ PAST[ S ]] is true iff S wastrue at leastonechrononago. But then [ PAST[ PAST[ S ]]] is true iff S was true more than one chronon ago. But this doesn't seemto capturewhat we intendedto say by a past-perfect sentencelike ' I hadleft ' . Onemight try to get aroundthis difficulty by talking aboutdegreesof pastness ' ' ; however,aswasnotedabove, this movefails. I hadleft might be talking aboutan eventat any arbitrarydistancein the past. Plus, thereis the strongin-
The A -Theory Semantics
105
tuition thattherereally is a referenceeventhere- that we could very well continue ' I hadleft ' . . . with ' . . . whenSmith arrived' . How is that to be cashedout on a Prioreantheory if there is no way to avail ourselvesof temporal reference ? Beforewe look for solutions, we needto canvassonemore objectionto theA -theory- this onea philosophicalobjection. 7.4 The McTaggart Paradox: Is the A -Theory Contradictory ? ' Oneof theearliestandmostinfluentialcritiquesof theA theoryis McTaggarts ( 1908, 1927). Arguing for the unrealityof time, McTaggartbeginswith the observation that certainpairs of propertiesare suchthat it would be inconsistent for oneobjectto havebothproperties.For example, althougha tablecanbeboth roundandred, it cannotbebothroundandsquare,for roundnessand squareness areinconsistentproperties. Likewise, accordingto McTaggart, it would be inconsistent for somethingto be bothpastandfuture. Thus, for mostcases,when we affirm the truth of (21) we havestatedsomethingthat is inconsistentif not 6 contradictory. (21) future(cI & past(cI>) But accordingto McTaggartthis is exactlywhat theA theoryentails: an event E will at somepoint be past, at somepoint present, and at some point future} Thus, we havethe conjunction(22). (22) future( E) & past(E) & present(E) The initial reactionto this part of the argumentis often that it is absurd. Surely one is not sayingthat E is alwaysfuture and alwayspastand alwayspresent; rather, one is assertingthat E hasbeenfuture, that it will be past, andthat it is currentlypresent.As intuitive asthis answermay seem, McTaggart( 1927) argues that it is illegitimate: " and"will be" ?And . . . whatis meantby "hasbeen whatis meantby "is," when,ashere, it is usedwitha temporal , andnotsimplyfor predication ?WhenwesaythatX meaning hasbeenY, weareasserting X to beY atamoment of pasttime. WhenwesaythatX will beY, weareasserting X to beY ata moment of futuretime. WhenwesaythatX isY (in thetemporal sense of "is"), weareasserting X to beY ata moment of present time. ' s line of AccordingMcTaggart analysis, theproblemwith the stepto propositions like ' X is Y at a momentof pasttime' is that no momentof time is intrinsically past, present,or future. As a result, if it is now true that X is pastat a
106
Chapter?
momentof futuretime, it will alsohold at sometime laterthannowthatX is past at a momentof pasttime. Thus, againthereis a contradiction. Onemighttry to escapethecontradictionby suggestingthat(23) and(24) are nevertrue simultaneouslybut that (23) will be true and(24) hasbeentrue. (23) X is pastat a momentof pasttime. (24) X is pastat a momentof future time. However, McTaggartsuggeststhat this merely startsus upon an infinite (and vicious) regress. But thereareotherproblemsthat seemto stemfrom the rejectionof the Btheory. The A -theory and the Myth of Passage According to Williams ( 1951), thereis a deepconceptualproblemin any assumption " " " thattime " passesor flows . Despitethenaturalappealof metaphors " " thatmakereferenceto change, changeitself takesplacein time. Thus, whether one regardstime as a movementof an eventalonga time line or whetherone thinksof thefutureapproachingusandflowing pastuslike a river, themetaphor fails miserably. Williams ' s objection is aimedat a conceptionof the A -theory in which a particulareventis future, thenpresent,andthenpast, andin which the eventis thoughtof asundergoinga kind of change. This generalobjectionwas anticipatedby a numberof adherentsto the Atheory . Broad ( 1938), for example, arguedthat oneneednot recoil from such objectionsonceonegetsclearon thetruenatureof temporalbecoming.Thetraditional philosophicalnotionof temporalbecomingis thatit is a speciesof qualitative change.For example, it might be thoughtthat thereis a particularevent (say, World War ll ) that wasfuture, then waspresent, andfinally becamepast. But Broadthinksthatthis is a confusedway to think abouttemporalbecoming. , Broadnotes,thischange If temporalbecomingis reducedto qualitativechange take at least musttake placeat a certainrate (or must placein time). And if it takesplaceat a certainrate, it can hardly be maintainedthat one hasgiven a sensibleanalysisof temporalbecoming- one hasdoneso by appealingto the temporalnotion of change. ' According to Broad, eventsdon t changefrom future to presentto pastin thesamewaythatwaterchangesfrom hot to tepidto cold. Rather,whenanevent
TheA -TheorySemantics
107
becomespresentit becomesabsolutely.Accordingly, World War n wasnot an eventthat wasfuture, thenpresent,thenpast; it simply became absolutely. Whateverthe meritsof Broad' s suggestion it is clear that the , proposalasit standsdoesnot go far enough- at leastnotfar enoughto answerWilliams' s concerns . If " becomeabsolutely" simply means" pop into existence," then this sortof becomingmustalsotakeplacein time. Initially, a thing fails to ex.ist; then it comesinto existence;thenit ceasesto ex.ist. Theproblemis perhapsnot asobvious asin the caseof a fire pokerwhich is initially hot, then warm, andthen cool, but it is nevertheless a problemof the samegeneralcharacter . Broadmay " havesomethingelsein mind by his useof the phrase"become absolutely, but what it might be is difficult to see. Prior 's Defenseof the A -Theory Here, then, is the problem. TheA -theoryis allegedlycommittedto a notion of changein the temporalstateof events.Mellor ( 1981, p. 90) describesthe commitment asfollows: [for theA-theorist Change tense(A-serieslocation ] is basicallythechanging ) of things andevents movingfromfuturetopast. . . .Therealityof theclockhand'smovement consists thefigures' l ' and'2' becoming ultimatelyin theeventsof itspassing successively andthenpast;andsimilarlyfor all otherchanges present . Clearly it will not do to introduceaccountsthat tacitly appealto changeof this form. But what is the alternative? Onepossiblesolutionis to simply maintain that somepropositionsare past, someare present, and someare future, andthat is that. When we saythat an object or an eventunderwenta temporal change,we aresimply sayingthata propositiondescribingtheobjectwasfuture butis nowpast. Thatis, we aresayingthatthepropositionhasthepropertieswas future, andis past, andthat further analysisis impossible. This is essentiallythe proposalput forward by Prior ( 1968), who argued that for somethingto changeis just for it to fit the schema(C). (C) It wasthe casethatp , andis not now the casethatp . Problemsariseonly if we slide from this schemato taking its accompanying " " metaphorseriously. The flow of time, Prior argued(ibid., pp. 11- 12), is merelymetaphorical whatis meantby it isn' t a genuine , notonlybecause movement , 't a butfurtherbecause whatismeant it isn by ; butdieforceof themetaphor genuine change canstill beexplained- weusethemetaphor because whatwecalltheflowof timedoes fit theaboveformula.
108
Chapter
ConstructingMcTaggart-typeargumentsagainstthis versionof theA theory is going to be very delicate. Considerthe following part of the McTaggartargument , asI formulatedit above: , andat somepointfuture . . . aneventE will at somepointbepast, at somepointpresent . . . future(E) & past(E) & present (E). . Thus,wehavetheconjunction On the Prioreanversionof theA -theorythis premiseis flatly false, sinceE did ' not go from being future to beingpresentto beingpast; that simply isn t what to theclaimthatweareforcedto aninfiniteregress , changecomesto. Responding " " thatMcTaggarthasdrawna perverseconclusion: Prior( 1967,pp. 5--6) suggests whichis plainlywrong(that Wearepresented , tobeginwith(in step1), withatstatement whichis plainly to is corrected . This future and is event , something ) past,present every hasbeenfuture or and will be and future either is , event that past present ( every right andis past). Thisis then andwill bepast,or hasbeenfutureandpresent andis present intended which, in themeaning , is wrong.It is then (in step2) to something expanded whichis right.Thisisthenexpanded (in alittlemorecomplicated to something corrected the obvious this in correct if we told that we are and is which to 3 , wrong ) something step whichis againwrong,andif weare way, weshallhaveto expandit intosomething we shallhaveto go on ad infinitum. similar at nothappyto stopthere, or any point, thisway, weonlygetcontradictions in forward to move led Evenif wearesomehow compel halfthetime, andit is notobviouswhyweshouldgettheseratherthantheirrun-points.Butwhydowehavetomakethewrongmoves ningmatesasthecorrectstopping few times, whenwe've seenthetroubleit getsus the first after in anycase?At least ? into, whynotpassto thecorrectversionimmediately Swappingtalk of propositionsfor talk of events, the solution, roughly, is to saythatcertainpropositionshavebeenfuture, certainpropositionswill be past, andlet thematterrestthere. Thereis no point in sayingthattheremustbe a time atwhich a certainpropositionwasfutureor at whicha certainpropositionwill be past. To takethat stepis to blunderinto a confusedway of thinking abouttense. This generalanswermay not setwell with certainreaders,for thereis a natural that ' hasbeen pull (at leastamongthe philosophicallytrained) 'to suppose ' ' future mustbe reducibleto the moreprimitive form is future anda particular time index. But thereis no logical reasonwe areforcedto suchan analysis. " Mellor ( 1981) arguesthat Prior cannotsimply sweepaway the stopping ' " points that give rise to contradiction; that, in effect, Prior s havingto moveoff thesestoppingpoints showsthat he is trappedin a vicious regress. This is a very subtleissue, andI wantto setit asidefor now, sincewe needto equipourselves with a bettergraspof theA -theoryandthetechnicalexecutionof thattheory beforewe canproperlytacklethe question. I will return to the problemin section8.5.
TheA -TheorySemantics
109
Presentism If we follow Prior ' s solution to the A - series paradox es, what conception of time does the above proposal leave us with ? How can we think about the Atheory metaphysics once we have abandonedthe notion of time as change from future to present to past? One plausible answer is that, borrowing a phrase from Dummett , we are " immersed in time " : Whattherealistwouldlike to do is to standin thoughtoutsidethewholetemporalprocess and describethe world from a point which hasno temporalposition at all, but surveys all temporalpositionsin a single glance: from this standpoint- the standpointof the descriptionwhich the realistwantsto give- the different pointsof time havea relation of temporalprecedence betweenthemselves , but no temporalrelationto the standpoint of the description- i.e., they arenot beingconsideredaspast, present, or asfuture. The antirealist takesmoreseriouslythefact thatwe areimmersedin time: beingsoimmersed , we cannotframeany descriptionof the world asit would appearto one who wasnot in time, but we canonly describeit asit is, i .e., asit is now. (Dummett1969, p. 369) Notice that this version of the A - theory sharesa number of features with the antirealist position of Dummett . On both pictures we reject the reality of the future and the past (construed as future and past events) . We neverthelesshave a notion of temporality , albeit one that does not countenance a dynamic conception of temporality as change. Notice also that we have come to this conclusion via a rather different route than Dummett did. For him , the unreality of the past followed from the rejection of bivalence in semantical theory .8 Here we have found our way to the conclusion in our effort to find a consistent version of the A -theory. Our path did not involve the rejection of bivalence. All this having been said, we still have to rescue the A - theory of time . In particular it remains for us to show that the A -theory semantics is able to handle temporal anaphora. We take up this issue in the next chapter.
ChapterS Temporal Anaphora B - Series Resources
without
Themainpredicament for applyingPrioreantheoriesof tenseto naturaI language is thatsomehowoneneedsto makesenseof temporalanaphora withoutappealing to future and pastevents, times, etc. Can this be done? On the face of it , this demandlooks impossiblefor a true Prioreantheory to meet. The A -seriesapproach deniesus accessto pastandfuture events, aswell asto pastandfuture times. In effect, it deniesus the resourcesthat are the building blocks of all currentresearchinto tense.How canwe havetemporalanaphorato eventsin the future andthe pastif thereareno suchevents? . Oneof my goalsin this chapteris to do the impossible- to build a semantics of tensethat utilizes only A -seriesresources . More specifically, I shall try to constructa semanticsfor tensethat satisfiesthe following conditions: . It providesan accountof temporalanaphorawithout referenceto past and future eventsor times . It providesa way to build complextenseswithout the usualresources(reference events, etc.).
8.1 E-TypeTemporalAnaphora Basic Strategy The basic strategywill be to treat temporal anaphorsas a speciesof E-type . The differencebetweenthesecasesandE- typepronominal temporalanaphora anaphorawill bethatpronominalanaphorsstandproxy for descriptions , whereas stand for temporalanaphors proxy temporalconjunctions(e.g., when-clauses ) that might be extractedfrom previousdiscourse.For an example, consider( 1).
( 1) I turnedoff thestovethen.
112
ChapterS
' ' Following Partee( 1973), onemight hold that then refersto a momentor period . It is possiblethat ' then' is standing of time; however,this is not necessary ' ' proxy for a temporalconjunction. For example, in (2) then could be standing ' ' ' ' proxy for whenI left the house, or whenyou told me to . (2) [s [s not [s PAST[s I turn off the stove] [swhen. . .]]] ] Spinningout the strategya bit , the ideawill bethatby havinggeneralpropositions asthe bearersof tensewe can avoid commitmentto suchB-seriesresources astimesandpastandfutureevents.Temporalanaphoracanbeachieved ), by theintroductionof when-clauses(or, moregenerally,temporalconjunctions which expressgeneralpropositionsaboutthe world. " If this gambitis to work, ' when' cannotmean" at the sametime ; it mustbe takenasa kind of primitive, just asthe PAST, PRES, andPUT morphemesare. That is, ' when' must be understoodas being more fundamentalthan the Bseriesconceptionof simultaneity. ( Moreon this in a bit.)
Preliminaries Phllosophical Singular vs. General Propositions The basic idea here is that we shall want to distinguish propositions that are about some object from general propositions that, strictly speaking, are not about anyone or anything .) For example, we might ' ' supposethat the utterance He is tall , accompanied by a demonstration of some individual , would express a singular proposition about that individual . In contrast ' ' , the utterance No one lives forever , which is obviously not about anyone , would express a general claim about the world . Other casesare disputed. Does an utterance of a sentencewith a definite description ' ' (for example, The tallest man in the room is tired ) express a singular proposition , or a general proposition ? According to Russell ( 1910- 11), the answer would be that it expresses a general proposition - not a proposition about some particular individual , but rather a general claim about the world ( e.g ., that there is one tallest man in the room and that he is tall ) . Others have argued that such sentences in fact express singular propositions . How this and other casesfallout is not really important to the current discussion; I just want to introduce the distinction for later use.
Referencevs. Denotation A greatdealof semanticliteratureusesthe terms ' denotation'and ' reference'inter changeably. This is unfortunate,for it ignores - considerationsthatwerecentralin thedispute considerations someveryimportant betweenRussell( 1905) andStrawson( 1950).
TemporalAnaphorawithout B-SeriesResources
113
Oneuseful way to understandthe distinction betweenreferring anddenoting - exploited by Evans( 1982), by Neale( 1990), and by Ludlow andNeale ' ( 1991) - is to considerRussells theoryof psychology.Russell( 1910- 11) distinguished from knowledgeby description,arguing knowledgeby acquaintance that to haveknowledgeby acquaintanceof a certain object, one must be directly acquaintedwith the object. Alternatively, to haveknowledgeby description of a certainobjectoneneednot beacquaintedwith theobject; oneneedonly know the objectasthe uniquesatisfierof a certaindescription. On Russell's view, thesetwo speciesof knowledgeare in fact quite different . A belief aboutsomeobjectknownby acquaintance is a belief in a singular propositionwhich hasthat objectasa constituent.For example, my belief that my neighborJonesis tall is a singularpropositionhavingJonesasa constituent. It is a belief aboutJones. On the otherhand, I might believethat the thief who stolemy computeris tall, andthethief mayevenbeJones,but unlessI knowthat Jonesis the thief, or unlessI sawhim stealingmy computer, my belief will not be a singularpropositionbut rathera generalproposition. It will not be a belief aboutJones(or anyoneelse); rather, the object of my belief will be a general propositionaboutthe world (essentially, that the world is suchthat thereis a uniquethief of my computerandthat he or sheis tall). The samepoint canbe extendedeasilyto the semanticsof naturallanguage. Applied here, the questionis whetherthereis a semanticdifferencebetweena sentence containinga descriptionthatuniquelydeterminessomeindividual(e.g., 'thethird ' ' ' planet) anda sentence containinga referringexpression(e.g., Earth ). Of course, therehavebeenimportant discussionsas to whetherdescriptions cannotat timesbe referringexpressionsaswell,2andclaimsto the effectthat a numberof namesarein factdescriptionsin disguise,3but thesediscussionsneed not detainushere. The importantpoint is that genuinedescriptionsandgenuine referringexpressionsaretwo entirely differentthings. The distinctionis introduced herebecauseit will play an importantrole in our investigationof the two alternativemodelsfor tense. On the one hand, therewill be theoriesthat will introducereferenceto specifictimesor events.Obviousreasonsfor theintroduction of temporalreferenceincludeexpressionslike (3)--(5).
(3) I didn't knowthat~esterday . (4) I leaveonthe15th ofDecember .
(5) I'll leave at7o' clock .
114
Chapter8
On theotherhand, theparticularversionof A -theorysemanticsI will be developing will do without referenceto futuretime andevents.It will haveto account for what appearsto be temporalreferenceby claiming that (3)-( 5) merelyappear to containgenuinereferringexpressions . The problem(for theoriesof tensethat cannotavail themselvesof temporal
reference occurrence of temporalanaphora , ) is compounded bythewidespread asillusttatedin (6)-{ 8). (6) I didn't knowthat~
.
(7) I ' ll leavewhenSmithdoes. (8) I left beforeSmithdid. If anaphors thatprohibittemporal aresimplyreferringexpressions ref, theories erenceare in a difficult position. However, there is an alternative: Temporal " " anaphorsmight be E-type. E-TypeAnaphors Oneof the morehotly debatedquestionsin the recentliterature of semanticshasbeenthestatusof unbound,anaphoricpronouns.Bound pronounsarein the scopeof an operator, andthey behavelike boundvariables in first-orderlogic. Sentences (9) and( 10) containboundpronouns.
(9) [Everydogt hasitsjday. (10) [Nomanthateshisjdog. Unboundanaphoricpronouns, on the other hand, arepronounsthat do not (superficially) appearto be boundvariablesyet nevertheless appearto get their contentfrom someantecedentnounphrase.Consider( 11) and( 12) . ( 11) [A dog1cameinto the room. I ~ bit me. ( 12) [Somedogs1cameinto the room. TheYibarked. Evans( 1977) offeredseveralreasonsfor supposingthat examples( 11) and . His most ( 12) cannotbeconstruedasgenuinecasesof boundvariableanaphora in 12 for if the noun was that ( ), phrase example, quantified telling argument
withoutB-SeriesResources Temporal Anaphora
115
boundthe pronoun' theYi' , the truth conditionswould simply comeout wrong. Consider( 13). ( 13) [Somex: dogsx](x cameinto the room & x barked) Notice that ( 13) doesnot havethe sametruth conditionsthat ( 12) has, for ( 12) impliesthat all thedogsbarkedwhereas( 13) merelystatesthatthe intersection of the enteringdogsandthe barkersis nonzero. Anotherdifficulty with the bound-variableanalysisis that it assertsthe possibility of an operator's binding somethingoutsideof its scope. One test for this would be whethera quantifier like ' no x ' can bind a variablein the same circumstances . For example, it is pretty clearthat the operatorcannotbind the co-indexedvariablein ( 14), so one wonderswhy operatorslike thosein ( 13) shouldbe exceptionalin this regard.
( 14) . [No dogslcameintotheroom. TheYibarked One possible alternative analysis of these anaphoric pronouns would be that they are actually referring expressions, and that they refer to whatever object has been raised to salience by the previous discourse. A view like this has been proposed by Lewis ( 1979, p . 243) : I may say 'A cat is on the lawn' undercircumstancesin which it is apparentto all parties to theconversation thatthereis someoneparticularcatthatis responsiblefor thetruth of what I say, and for my sayingit . PerhapsI am looking out of the window, and you rightly presumethat I saidwhat I did becauseI sawa cat; andfurther (sinceI spokein the singular) that I sawonly one. What I said was an existentialquantification; hence it raisesthe strictly speaking,it involvesno referenceto anyparticularcat. Nevertheless salienceof the cat that mademe say it . . . . Thus althoughindefinite descriptions- that is, idioms of existentialquantification- are not themselvesreferring expressions , they may raisethe salienceof particularindividualsin sucha way asto pavethe way for referring expressionsthat follow. One objection to this view , discussed in detail by Helm ( 1982), suggests that certain facts undermine Lewis ' s idea that pronouns refer to objects raised to salience. As examples , Helm considers ( 15) and ( 16), where # indicates infelicity . ( 15) a. John has a spouse; she is very nice.
b. Johnis married : #she is very nice.
116
ChapterS
( 16) a. I droppedten marblesandfound themall exceptone; it mustbe underthe desk. b. I droppedten marblesandfound only nine; #it mustbe underthe desk. Similar examplesare notedby Geach( 1962), who considersthe contrastbetween pairs like ( 17a) and ( 17b), andby Evans( 1977), who notesthe contrast between( 18a) and( 18b). ( 17) a. Every manwho ownsa donkeybeatsit. b. #Everydonkeyownerbeatsit. ( 18) a. Johnownsa donkey. It is brown. b. #Johnis a donkeyowner. It is brown. Accordingto Helm, ( 15a) and( 15b) aretruth-conditionallyequivalent, and an utteranceof either would, on Lewis' s theory, result in John' s wife ' s being raisedto salience.Helm concludesthatthetwo occurrences of thepronoun' she' shouldthereforebe equallyfelicitous if saliencewerethe relevantnotion.4 In any case, it seemsfairly clear that the raised-to-saliencepicture cannot be universallytrue, sincetherearenumerouscasesin which we would wantto saythat a pronounis anaphoricon someexpressionyet we would haveno way of identifyinganobject- that is, we haveno singularobject-dependentthought. A sentencelike ( 19) would be a clearcaseof this. ( 19) A mancamein. & trippedoverthe chair. Oneattractivealternativeis that pronounsstandproxy for definite descriptions . The idea that unboundanaphoricpronounsmight standproxy for definite descriptionshasbeenproposedby Evans( 1977), Parsons( 1978), Cooper 5 ( 1979), Davies( 1981a), and Neale( 1990). The basicideais that the pronoun in ( 19) may standproxy for the underlineddefinitedescriptionin (20) . (20) A mancamein. The manwho camein trippedoverthe chair.6
TemporalAnaphorawithout B-SeriesResources
117
One of the key advantagesof such an analysis would be that it avoids the unwelcomeconclusionthat the anaphoricpronounis either a boundvariable or a referring expression.More to the point, the theory allows that one might understanda sentencecontaining an anaphoricpronounwithout therebeing someobject that is the referentof the pronoun. This is just a surfaceglossof the theoryof descriptivepronouns, of course. There havebeena numberof important objections to it , and the theory has beensubsequently in a varietyof interestingways. Fornow, it is enough developed that we placethe relevantmachineryon the table so that we canexploreways in which it might be incorporatedinto the theoryof temporalanaphoradeveloped later in this chapter. De Re/ DeDicto Distinction We will alsoneedto makeuseof the celebrated dere/dedicto distinction. Repeatingour earlierexamplefrom Quine, a sentence like (21) is ambiguousbetweena dere reading(211)anda dedictoreading(21"). (21) The numberof planetsis necessarilyodd. ' (21 ) de re: [the numberof planetsi necessarily [ej is odd] " (21 ) de dicto: necessarily [the numberof planetsis odd] The receivedview of theseconstructionsis that the de re/de dicto distinction is a reflex of a scopeambiguity, with the de re readingcorrespondingto a caseof quantifying into an intensionalenvironmentand the de dicto reading correspondingto a caseof quantification within the scopeof an intensional operator. We can generalizethis idea as in (22), where a OP is a determiner phrase(or, if oneprefers, a quantifiednounphrase) andej is a variablebound by the OP.
(22) dere: [DP1[Operator [. . . ej. . .]] dedicto: Operator[[DP]j[. . . ej. . .]] 8.2 Development of theTheory I beginby inttoducingwhat canbe called " absolutetense," which is not to be confusedwith thetenseof a naturallanguageutterance(thereasonbeingthatall naturallanguageutterances maywell beaccompanied by possiblyimplicit whenclauses ).
118
ChapterS
Let us take the absolutetensemorphemesto be PAST, PRES, and PUT as theyareappliedto a sentence(clause) in isolation. Thuswe havethe following ' ): syntaxfor the occurrenceof thesemorphemes(basicallyasin Prior s scheme
(23) AbsolutePresent : PRES [S] AbsolutePast:PAST[S] AbsoluteFuture : FUT[S] Relative Tense introduces the imnlicit when - clause. It narallel ~ the intrn . .
ductionof thereferenceevent/timein Reichenbachian I theories as shown in (24)-(26). (24) RelativePresent onReichenbach : E,R (simultaneous ) analysis on this : when . . . 5 [ ] syntax analysis[ ] (25) RelativePast onReichenbach analysisER : [5] before[. . .] syntaxonthisanalysis (26) RelativeFuture onReichenbach : RE analysis : [5] after[. . .] syntaxonthisanalysis Complex Tenses To get a handleon complextensesin naturallanguage , we needto makesome . Let usbeginwith two working conjecturesaboutthesb"uctureof tensedsentences : hypotheses (HI ) All naturallanguagesentences have(possiblyimplicit ) when-clauses. (82 ) The structureof animplicit when-clauseis the sameasanexplicit when-clause. Thebasicintuitiondriving( H2) is thatif we areto positimplicit tensedclauses our theorymustbe asconstrainedaspossible. Ideally, we do not want to be in the position of proposingnew grammaticalrules or principles to accountfor thebehaviorof theseclauses . If therereallyareimplicit when-clauses , thenprima
TemporalAnaphorawithout B-SeriesResources
119
facie we shouldexpectthemto be internally structuredjust like explicit whenclauses - the principle exceptionbeingthat they areunpronounced . If (H2) holds, we can extrapolatethe structureof implicit when-clausesby . 1Woobservations makingsomeobservationsabouttheir explicit counterparts areparticularlygermanehere: (01 ) All explicit when-clausesaretensed. (02 ) All explicit when-clausesarecoordinatedwith the tenseof the matrix clause. To illustrate (01 ), we would neverfind an explicit constructionof the form ' I left the room whenJohnto be hungry' . When-clausesare neverinfinitival (at leastin English). To illustrate(02 ), for example, we don' t find an Englishsentence like ' I will haveleft the room when Johnhad arrived' . The lead tenses ' will ' and ' had' fail to matchin this case. ( ) By (H2), it follows that implicit when-clauseswill havetheseproperties too. Accordingly, thelogicalform of theseconstructions(at a certainlevelof abstraction ) will be asfollows: Pluperfect on ReichenbachanalysisERS LF on A -theoryanalysis: PAST[S] beforePAST[ . . .] Futureperfect on ReichenbachanalysisS- E- R LF on A -theoryanalysis: FUT[S] beforeFUT[ . . .] Futurein future on ReichenbachanalysisS- RE LF on A -theoryanalysis: FUT[S] afterFUT[ . . .] Futurein past on Reichenbachanalysis: RE - S or R- SE LF on A -theoryanalysis: PAST[S] whenPAST[. . .]
or
By (H 1) it shouldfollow thatpast, present,andfuturetensesin naturallanguage donotconsistmerelyof thesimplePAST,PRES,andFUT morphemes but should be morecomplexconstructions : Past on Reichenbachanalysis: E,R- S LF on A -theoryanalysis: PAST[S] whenPAST[ . . .]
120
ChapterS
Present on ReichenbachanalysisS,E,R LF on A -theory: PRES[S] when/asPRES[. . .] Future on ReichenbachanalysisS- E,R LF on A -theory: FUT[S] whenFUT[. . .] The T-theoryaxiomsfor the absolutetensesarethe following:
(Tl ) Val(x, PAST , 0') iff x wastrue (T2) Val(x, PRES , 0') iff x istrue (T3) Val(x, FUT,0') iff x will betrue Theaxiomfor thetensephraseis asfollows: (TP) Val( T, [TP TNSS], 0') iff , for somex, Val(x, TNS, 0') andx = [] S[] areasfollows?: Thesemantics of thetemporalconnectives ( WI) Val( T, [8TPI ~
TP2] , 0') iffVal(T, TPl , 0') whenVal(T, TP2, 0')
( W2) Val(T, ~ TPI beforeTP2] , 0') iffVal(T, TPl , 0') beforeVal(T, TP2, 0') ( W3) Val(T, [8TPI ~
TP2] , 0') iffVal(T, TPl , 0') afterVal( T, TP2, 0')
: Example ' is trueiff 3e ' I hadeaten (le )][] wastruebefore [][( )(eating(e) & agent [] [ . . . ] [] wastrue. a theoryof aspectintothebroadertheory Finally, wewill wantto incorporate thatevents . 1havenoquarrelwithtalkaboutevents ; 1ammerelyconcerned . Eventsmaywellbethebearers of aspectual notbethebearers of tense properties for avoidingthisview). (althoughlaterwewill seethattherearereasons -based 1991 In event estosemantics ; Higginbotham (see,e.g., Parsons approach tohavethelogicalformshown like(27) areargued 1989 ; Schein1993 ), sentences in (28).
Temporal Anaphora without B -Series Resources
(27) A mankicked Bill . (28) (3e)(kicking(e) & [an x: manx] agent(x, e) & patient(Bill , e) & past(e Clearly this analysispresentsproblemsfor the A -theorist, sinceit arguesthat thereis an eventwhich is past. Is theresomeway to avoidthis consequence of the introductionof events? Recallthe discussionof the de dicto/de re distinctionearlierin this chapter. Eventdescriptionswill be innocuousso long asthey remainsafelywithin the scopeof TNS. The situationis parallel to that of modals. Dependingupon the relativescopeof anexistentialquantifierandthemodal, we might find ourselves committedto someunwelcomeentities. Consider(29) . (29) A unicornmay haveeatenmy vegetables . On theinnocentinterpretationof this example,theexistentialquantifierremains safelywithin the scopeof the modal, asin (30). (30) possibly[(3x)(x is a unicorn& x atemy vegetables )] In the problematiccase, illustratedin (31), the existentialquantifiertakeswide scopeoverthe modal; the resultis an ontological commitmentto unicorns.
(31) (3x: x is a unicom)possibly [x atp. my vegetables] Similarconsiderations applyto theintroductionof quantificationoverevents in caseslike (27). The innocentcasefinds the eventquantificationsafelywithin the scopeof the past-tenseoperator: (32) PAST[(3e)(kicking(e) & [an x: manx]agent(x, e) & patient( Bill, e ] The caseto be avoided (for the Priorean) is (28), where the event quantifier has scope outside of the temporal operator. However, so long as we exercise care, we can safely incorporate most of standard aspectual theories. Some modifica tions are necessary,however. Consider the following possible properties of events (Parsons 1991) :
(33) PROG(e,t) : e in progressat t
122
ChapterS
(34) CUL(e,t) : e culminatesat t
Canwegetrid of thetimereferences here? Easily. Suppose weadopt(35) and(36). (ThePROGrelationis actuallyredundant in theeventsemantics used here,andwewill notneedit in thefollowinganalyses .) (35) PROG (e): e in progress (36) CUL(e): e culminates Wethenhaveavailable theanalyses (37)-(42), wherethePRESandPROGmorphemes areomittedasredundant in theseparticularanalyses . (37) Smithis drawinga circle. (3e)(drawing (e) & agent (Smith, e) & theme (circle, e when[. . .] (38) Smithwasdrawinga circle. PAST[(3e)(drawing (e) & agent (Smith,e) & theme (circle, e ] whenPAST[. . .] (39) Smithdrewa circle. PAST[(3e)(drawing (e) & agent (Smith, e) & theme (circle, e) & CUL(e ] whenPAST[. . .] (40) Smithhadbeendrawinga circle. PAST[(3e)(drawing (e) & agent (Smith,e) & theme (circle, e ] beforePAST[. . .] (41) Smithhaddrawna circle. PAST[(3e)(drawing (e) & agent (Smith,e) & theme (circle, e) & CUL(e ] beforePAST[. . .]] (42) Smithwill havedrawna circle. FUT[(3e)(drawing (e) & agent (Smith, e) & theme (circle, e) & CUL(e ] beforeFUT[. . .]] in (39), (41) and(42) thatwedon't needto link theculmination Noticefromtheanalyses " time." timeto the reference
TemporalAnaphorawithout B-SeriesResources
123
Aktionsartencanbe incorporatedin the usualways. States, achievements , of events . , andactionscanbedistinguished accomplishments by internalpredicates For example,we canintroduce" HOW (e)" for states.If statesaredistinctfrom " " " " the progressive , we candistinguish PROG(e) from HOLD(e). Actions will " " ' ' " admit CUL( e), achievement verbslike win will have RESULT(e)," andsoon. Thereare, however,somereasonsfor supposingthataspectshouldbetreated not asa predicateof eventsbut ratherasa predicateof proposition-like objects. Chief amongtheseis the imperfectiveparadox, which wasdiscussedbriefly in chapter5. Consider(43). (43) Johnis drawinga circle. The " paradox" is that on the eventsemanticsjust given therewill be an existential quantificationovercircles( Thereis a circle x, suchthatJohnis drawing x), yet (43) canbeb11e evenif thereis no circle (say, if Johnis run overby ab11ck afterhecompletes only a 15- degreearc). Unlesswe arepreparedto follow Parsons " " 1991 and introduce ( ) incompleteobjects, our eventsemanticswill give incorrect b11thconditionsfor certainutterancesof (43). Notice that a tense-like analysisof the progressiveactuallycanfoil the imperfective paradox. If I take tenseto be a propertyof sentencesor of proposition -like entities,theanalysisof (43) mayrunasshownin (43' ), andthesemantics of PROG may be givenby the axiom (43* ). ' (43 ) PROG[Johndrawsa circle] (43* ) Val(x , PROG) iff x is in the processof becomingb11e Thus, if we go this route, (43') is b11eiff ' Johndrawsa circle' is in the process of becomingb11e . The existentialquantifierremainsburledsafelyin the scope of the PROGoperator. Recallthat one of the centralobjectionsto a Prioreansemanticsof tenseis that it is unconstrained . If complex tensesare derivedby nestingtensemorphemes , then there is potentially no limit to the number of possibletenses. That objectioncertainlydoesn't apply to this versionof a Prioreansemantics , for the simplereasonthat tensemorphemesaren't nestedat all! Theresultis a highly constrainedview of possibletenses , baseduponthepossible combinationsof PAST, PUT, andthetemporalconnectives . ( HereI assume that thereis no genuinepresenttense.)
124
Chapter8
matrix
conj -clause
conjunction
tense
PAST FUT PAST FUT PAST FUT
PAST FUT PAST FUT PAST FUT
when when before before after after
Past Future PastPerfect FuturePerfect Futurein Past Futurein Future
The picture is going to be a bit richer than this, sinceeverythingturns on the availableconstructionsin the systemof explicit temporalconjunctions.For example , aretheresomecaseswherematrix andwhen-clausetensescanbe mixed to , (say getfutureshiftedreadingof Futurein thePast)?It all dependson whether theexplicit systemof temporalconjunctionswill alsoallow ussuchmixing. For example, do we havethe following paradigm? matrix
conj-clause
conjunction
PAST FUT after ' e.g.: Johnwasleavingafterthe party next week'
tense Futurein Past
It is arguablethat this caseis really talking about a past intention of John' s (thathe wasintendingto leaveafterthe partynextweek), but whateveris going on in the explicit case, that is essentiallywhat accountsfor the apparentfuture shiftedreadingof ' Johnwasleaving' . The tensesystemmerelyrecapitulatesthe structuresmadeavailableby the , andinheritstheconstraintson thatsystem systemof explicit temporalconjunctions . That, in effect, is what constrainsthe tensesystemon this theory. Temporal Adverbs Again If thereare going to be difficulties in executingthis versionof a A -theory semantics , they aredoubtlessgoing to comein the analysisof temporaladverbs. Mattersbegin easily enough. As notedabove, temporaladverbslike ' yester' ' ' . At worst, we might run into day and tomorrow canbetreateddisquotationally a proliferationof suchadverbsandtheremight be someconcernabouta finite axiomatizationfor them- but theseworriesstemfrom adverbslike ' threedays ' ' ' ago and four daysago , andit is pretty clearthat we cancomeup with ways of decomposingthesewithout slipping into talk of B-seriesrelations.8 Onemight objectthat these" adverbs," evenwhentreateddisquotationally, are referring expressions . For example, whetherwe wish to call ' yesterday'
Temporal Anaphora without B - Series Resources
125
indexical or not, and whetherwe think we can display its senseor not, some would saythat it still refersto a day. That is, it refersto a time in the past, and hence(theargumentwould go) we do not escapetheneedfor theB-theoryconception of the past. The flaw in this objectionis that it simply begsthe critical questionby assuming that ' yesterday', ' tomorrow' , etc. refer to days (or to anything else) . On thetheorybeingdevelopedhere, theseadverbsaretreatedaspredicates(i.e., with axiomslike Val(x , ):esterda ):) iff x wastrueyesterday(or, alternatively,one day ago), and not axiomslike Val(x, ):esterday) iff x = yesterday). On the assumptions madeexplicit in section2.6, predicatesarenot referringexpressions on this theory. If this is right, a similar sort of disquotationaltreatmentshouldbe available for theothertemporaladverbsaswell. Forexample,we might introduceaxioms like thosein (44). (44) Val(x, ~ ~ Val(x, n~
) iff x is alwaystrue iff x is nevertrue
' and' alNow it might bearguedthat, evenif it is plausibleto treat ' yesterday ' as " " , it is suspiciousto try andtreat locatingadverbials in the ways predicates sameway, for theysurelydirectly refer to specificdates. Following the exposition - beginning by KampandReyle( 1994), let usconsidersomeof thesecases with calendarnames,asin (45). (45) The lastclassis July 4, 1995. Is it really possibleto avoid referenceto times in this case? I fail to seewhy not. Following the assumptionssetout above, this sentencemust havean implicit when-clause; hence, it is naturalto supposethat the dateitself is merelya constituentof that clause:
(46) Thelastclassis when[... July4, 1995 ]. The ellipsedpart of the clausecould indicatesomefonn of conventionaldating " system, suchas standardcalendarsystemsindicate. . . ." ' ' ' ' Similarconsiderations , as applyto before and after with NP complements in (47).
(47) Thelastclasswill bebeforeJuly5.
126
Chapter8
suchas(47) tooaresimplycases of partiallyellipsedtemporalconjunctions Examples . (48) Thelastclasswill bebefore[. . . July5] . , it seemsthattemporaladverbs(evenlocatingadverbials Accordingly ) need notberegarded asrefeningexpressions andhenceneednotposeconceptual difficulties for theA-theory. 'before'and' after' Wemightwishto treatthestandard B-theorypredicates ascomposed outof morebasicA-seriesrelations . Theideaherewouldbethat a sentence like (49) wouldhavea logicalformin which'before'is treatedas of a past-tensemorpheme anda simplewhen-clauseasin (50). composed (49) TNS[SI] beforeTNS[S2] (50) [sTNS[PAST[SI] when[S2]]] Noticethattheoriginaltensemorphemes arenowconstrued asa singlemorpheme thattakesscopeovertheentireconjunction . (Thesecondexplicitmorpheme asa kindof scopemarkerfor tense .) mightbeconstrued Thisis a little bit abstract , soconsiderthelogicalformthatwouldbegiven ' I atebeforeI left thehouse ' and' I will eatbeforeI leavethe for thesentences , ' house: (51) [sPAST[ pAST[thereis aneatingby me] when[thereis a leavingof thehouse by me]]] (52) [sFUT[pAST[thereis aneatingby me] when[thereis a leavingof thehouse by meill ' t introduce Noticethatthisstorydoesn nestedtenses , sincetheoutertenseapplies to theconjunction andtheinneroneappliesonlyto thefirstconjunct . Theanalysisfor ' after' wouldbeanalogous . A sentence the surface having form(53) wouldhavethelogicalform(54). (53) TNS[SI] afterTN S[S2] (54) [sTNS[ FUT[SI] when[S2]]]
TemporalAnaphorawithout B-SeriesResources
127
Onedifficulty for this generalstrategyis going to be handlingtemporaladverbs like ' seldom'or ' often' in termsof 'past' , 'present', and' future' . Technical difficulties beginto emergewith predicateslike ' since' , which hasthe following two senses : case1 I ' ve beenin EnglandsinceJanuary1 ( beentherecontinuously). case2 I ' ve beenin (to) EnglandsinceJanuary1 (beenthereonceor more). If we attempta reductivestrategy,then ' sinceJanuaryl ' might be part of a restricted substitutionalquantifierover when-clauses.(Seesection4.3 for a discussion of substitutionalquantification.) The glossfor case1 might be asin (55), the glossfor case2 asin (56). (55) ' '-' -" ' ( Forall S, sit. PAST[ S after January1. . .J] is T) ' I 've beenin England -' ' when S is T (56) ' '- ' - " ' (For someS, sit. PAST[ S afterJanuary1 . . .]] is T ) ' I ' ve beenin -' ' Englandwhen S is T So far so good, but when we moveto adverbsof temporalquantification, we soonrun into difficulties. Consider(57)-( 58), for example. (57) Wego to Londonoften.
(58) We'vebeento Londonsixtimes. Onemight supposethat thesecases(certainly(58 finally force us into the introduction of pastandfuturetimes; we say" six times," afterall. But if we were to look at translationsof (58) into otherlanguages of times , thequick acceptance would certainly seempremature- consider Italian, in which one would say " sei volte" six turns . ( ) Still , second-ordertemporaladverbsdo raiseinterestingissues.Note that in ' ' (57 ) and(58 ) it is not sufficientto quantify directly over when-clauses. ' (57 ) ' '(For manyS) we go to Londonwhen S is T ' (58 ) ' '(For six S) we went to Londonwhen S is T
128
Chapter8
As wasmentionedin note 14to chapter4, substitutionalquantificationbreaks down in thesecases: there can be many when-clausesthat describea single visit to London. But we do needsomeway of talking about" times," evenif we don' t wantto take them as being points of time in the senseof the B- seriesmetaphysics. Onceagain, it turnsout that we cansolvethe problemwith off-the-shelfphilo. Following a generalstrategyfor substitutionalquantification sophicalresources sketchedin Ludlow 1985, we cango secondorder andbuild times out of when-clauses,asin (59). (59) In a givencontextc, for eachclauseS, tensemorphemetns, thereis a unique " time" t S.t. t = Si: tns- ' ' - S- ' when ' - tns- ' ' - Si- ' ' is T , [ ] [ ] { } Intuitively, then, times arederivativeof our primitive notion of when. We say thatA andB happenat the sametime becausewe know thatA happenswhenB . Further, anythingthat happenswhenB happensalsomust, ipso facto, happens " " . happenat the same time thatA happens Furtherconstraintsarenecessary , sincethe when-clausemustuniquelyspecify whensomethinghappened . Forexample , it mightbetruethatI wentto London ' cannotbe a relevantsubstitution ' I hada headache but whenI hada headache , instance.So apparentlywe endup with (591). ' (59 ) In a givencontextc, for eachclauseS, tensemorphemetns, thereis a unique "time " t S.t. t = Si: tns- ' ' - S- ' when ' - tns- ' ' - Si- ' ' is T and , , [ ] [ ] { - tns ' [ ' Si ' ] only when ' tns ' [ ' S ' ] ' is true} This allowsthe extensionof t to include" anchoring" clauseslike " whenI celebrated " " ." my eighteenthbirthday, butnot whenI hadoneof my manyheadaches Noticethecontextualvariablehere. This is crucial, sincewhatcountsashappening " at thesametime" dependsuponour interests.Wemight meanthe same nanosecond we mean the same , or we might meanduring the brief day, might . history of the humanspecies Giventhis revisednotionof timesassetsof when-clauses,we cannow return to the non-first-ordertemporaladverbsandtreatthemasin (5T') and(58").9 " (57 ) For manyt , thereis an S, Set , S.t. ' we go to Londonwhen' S is T " (58 ) For six t , thereis an S, Set , S.t. ' we wentto Londonwhen' S is T
Temporal Anaphora without B - Series Resources
129
Thus, whenwe saythat we havebeento Londonsix timeswe arenot sayingthat our trips feUon six distinct time points (that wouldn' t work in any case; a single trip could overlapsix time points), nor arewe talking aboutsix events.We aretalking aboutsix non-overlappingeventdescriptions . Thus, I might characterize thesix timesI havebeento LondonasbeingwhenI cameto talk at Kings, whenI cameon my Nth birthday, andso on. 8.3 More on E-Type Temporal Anaphora AlthoughPartee( 1984) observeda numberof similaritiesbetweentemporaland nominal anaphora,shefailed to note one classof similarities: temporalconstructions thatmirror Evans's example' JohnownssomesheepandFrankshears ' them. The temporalanaloguesareasshownin (60). (60) Jackgoesup the hill sometimesandthenhe comestumbling down with Jill . The traditional analysesof anaphorafail here. If temporalanaphoraworks as boundvariableanaphoradoes, thenwe havesomethinglike (60'). ' (60 ) (sometimest)[Jackgoesup the hill at t andhe comestumbling down with Jill at t] But thatdoesn't getthetruth conditionsof (60) right, sinceit doesn't specifythat eachtime hegoesup hecomestumblingdown. Likewise, it doesn't work to suppose that eachof theseclauseshasan independenttemporalquantifier. That " merelygetsus to (60 ). " (60 ) (sometimest)[Jackgoesup the hill at t] and(sometimest )[he comestumbling downwith Jill at t] Of course, this doesn't get the truth conditionsright either. What standsout in this instanceis the fact that E-type temporal anaphoraseemsto work quite 1 smoothlyhere : (60* ) (sometimest)[Jackgoesup the hill at t] and [ hecomestumbling downwith Jill (after he goesup the hill )] Of course, it is alsopossibleto adoptthe treatmentof temporalanaphorain DiscourseRepresentationTheory (DRT), in particular as it is developedby
130
Chapter8
Hinrichs ( 1981, 1986) andby Kamp andReyle( 1993). This is not the placeto openup a debateaboutthe relativemeritsof E- type anaphoraand DRT (some of this debateis takenup in Ludlow 1994), much lessaboutthe extensionof . This is not a book aboutanaphora thosetheoriesto temporalanaphora ; my interest in temporalanaphorahereis philosophical, andit turnson preciselyone concern: that theA -theoryneedsto avoidtemporalreference.E-type anaphors presentonepossible(andworkable) way of meetingthat need. Theremay well be othersolutions, andit maybe that someof thosesolutionswill employDRT of GroenendijkandStokhof( 1991) resources , thedynamicsemantics (or, perhaps andChierchia( 1995 . At present,however, the philosophicalcontentof those escanhelpus to avoidtemporal theoriesis cloudyon this issue. If suchapproach ' ' reference(in thetechnicalsenseof referenceI amusinghere), I haveno objection to them. 8.4
Further Issues
In chapter 5, two of the more interesting extensions of the B -theory involved the " " incorporation of temporal reference in norninals and the phenomenon of sequence of tense. In this section I will show how these phenomena can be handled on the A -theory. I will then return to the philosophical objection raised by McTaggart and discussed in section 7.4 above.
Temporal Anaphora in Nominals In section5.2 I discussedan observation , dueto En~, that nominalsoften seem to havean implicit temporalreferenceof their own- possiblyquite independently " of thesentence asa whole. Theexamplefrom of the"temporalreference that discussionis repeatedhereas(61). (61) The hostagescameto theWhite House. There is an intuitive sensein which we wish the NP to pick out sometime of thevisit to theWhiteHouse- intuitively, thehostages framethatis independent at a certaintime andplace(say, the hostagestakenfrom the US Embassyduring the Iranianrevolution). It is sometimessupposedthat thesesortsof casespresentdifficulties for Atheory semantics , presumablybecausethey provideadditionalevidenceof the needfor temporalreference.As we haveseen,however,armedwith E-typetemporal anaphora,one can accountfor a numberof apparentcasesof temporal referencecaseswithout invokingreferenceat all. Canwe do that here?
Anaphorawithout B-SeriesResources Temporal
131
JustastheB-theoryhadto beaugmented , we mustsaysomethingmoreabout the A -theory if we areto accountfor the temporalcharacterof certainnominals . Clearly the when-clausesthat we haveintroducedthusfar will not do the job , sincetheyaremerelydescribingpropertiesof the matrix event(in the case of (61), thevisit to theWhite House). Whatcanbe saidabouttemporalanaphora in NPs? Oneattractivepossibility would be to developthe idea, suggestedby Bach andCooper( 1978), that therecan be implicit relative clausesin NPs. For example , in thecaseof (61) therewouldbeanimplicit relativeclausewith thepossible contentindicatedin (61'). ' (61 ) [NP[NPThe hostages ] [s (who werecapturedin the US Embassyduring the Iranianrevolution)]] cameto the White House. Thecontentof therelativeclausewouldbeextractedfrom previousdiscourseor from sharedbackgroundinformation in a manneranalogousto the treatment of E-type pronouns.II In sum, we can againavoid the moveto temporalreference if we arepreparedto accepta certaindegreeof abstractsyntacticStruCture .12 Sequenceof Tense As wasdiscussed in section5.2, sequence of tenseinvolvescaseslike (62), where ' Biff s illnessmayhaveoccurredsimultaneouslywith Mary' s reportor mayhave occurredat sometime previousto thereport. (62) Mary saidthat Biff wasill . Higginbotham( 1995) arguedthattheshiftedreadingwasthecorecase, andthat theunshiftedreadinginvolvedtheillusionof pasttensein thecomplementclause. In Higginbotham's words, "the appearance of the pasttensein a complement clausecanbe an appearance merely; cross-referencetakesplaceasin the first case, but the tenseof the complementis present, not past." Clearlythesamesolutionis availableto theA -theorist, butjust asclearlythere is somethingunsatisfactory aboutit. If thereis no past-tensemorphologyin these , why doesit soundasif thereis? More urgently, what are complementclauses the mechanisms which we get illusory pasttensein thesecases? by If this approachis unattractive,thereareothersolutionsavailablewith theAtheory " theoriesof framework. For example, the " independent SOTdiscussed in chapter5 havenaturalanalogueswithin theA -theory. To seethis, consider thatthestructureof a sentencelike (62) is actuallygoingto be somethingalong the linesof (63) .
132
Chapter8
(63) Biff was ill ( when) [ . . .]] when [ . . .]] [s Mary said that So Focusing our attention on the internal complement clause, we know (following 02 above) that the principal constraint on when-clauses is that their lead tenses match those of the clauses with which they are conjoined . Hence, in the above ' ' example, we know that the when-clause paired with Biff was ill must be past '' tense. But that is the only constraint. The clause might have a content like as " Mary was speaking (giving us the unshifted reading), or it might have a content " " like when she visited him last week (giving us the shifted reading) . In each case, the content of the when-clause is going to be extracted from previous discourseor from some form of common knowledge ) in fundamentally the same way that the content of E- type pronouns are reconstructed. " " If this picture is right , then it is a mistake to think of tense being shifted or " of tense at all . Rather what is " of there being a sequence , happening is that , situation reconstructed the described clause is on how the , implicit depending may be cotemporaneous with any described past tense situation . It might be ' cotemporaneous with the described event of the matrix clause (Mary s speaking ), or it might be cotemporaneous with any other contextually salient pasttenseevent description (e.g ., when Mary saw Biff last week, or when Biff missed his meeting with Napoleon , or whenever) . Interestingly , this parallels the last ( and in my view the most promising ) of the B -theory solutions that we looked at: simply letting R pick up any past- or future -tense event.13 In sum, the extensions to the B - theory that we looked at in section 5.2 have analogous extensions in the A - theory. Importantly , the extensions are possible without the appeal to temporal reference, and hence there is no conceptual cost to the basic A -theory program . Other extensions will proceed in like manner. Whereas the B -theory appeals to temporal reference , the A - theory looks for implicit clausal structure to carry the E type temporal anaphora. Whether this strategy can be carried out is an open empirical issue; perhaps the positing of this implicit clausal structure will collide with general principles of linguistic 14 theory. But notice how striking it is that every construction for which researchers are inclined to posit temporal reference happens also to be a construction for which an implicit -clause story seemsplausible . Indeed, it is even more striking that the case for temporal reference routinely involves some sort of explanation that relies upon the requisite clausal structure to identify the supposedreferent " (e.g ., hostageswho were captured in the US Embassy during the Iranian " . revolution ) Perhapsthis is a clue that reference was never involved at all .
TemporalAnaphorawithout B-SeriesResources
133
As wasnotedin chapter4, referencedoesnot comefor free. The epistemological burdeninvolvedin positing referencein thesecasesis great, andI see no efforts in the literatureto take responsibilityfor that burden. The positing of implicit clausesinvolvesa burdenof its own, and it will haveto be shouldered . But at leastwe havea model for how implicit clausescan be incorporated into current linguistic theory.ISCan the samebe said for the liberal postulationof reference?
8.5 McTaggartRevisited Having equippedourselveswith a betterunderstandingof the revisedAtheory ' , let us now returnto McTaggarts original objectionto theA -theory. Mellor ' ( 1981) hasclaimedthat Prior s answer(discussedin section7.4 above) fails to . But doesit reallyfail? Generatingtheregressrequires escapetheviciousregress that we be ableto movefrom (64) to sayingthatthereis a time at which x is future andthat thereis a time at which x is past (thus delivering the contradiction : x is future andx is past). (64) x wasfuture andwill be past. Prior' s responsewasthat we neednot havemadethis step, but we arenow in a . We don' t needto avoidthe step. In positionto makean evenstrongerresponse a certainsensewe could not makethe stepif we wantedto! Hereis the idea. (65) is shorthandfor anLF representation of the form given in (65'). (65) FUT[S] ' (65 ) FUT[S] whenFUT[ . . .] But thenthe semanticsdeliversthe following truth conditionsfor (65') :
(65*) [] S[] will betruewhen[] [. ..] [] will betrue
Similar considerations thetruthconditions shown applyto(66), whichwill have in (66*). (66) PAST [S]
134
Chapter8
(66* ) [] S[] wastrue when [] [ . . .] [] wastrue Wecannevergetto thepoint wherewe havea conjunctionof two conflictingAtheory tensedclaims. Clearly (65* ) and (66* ) are not incompatible, sincethe when clauseswill havedifferentcontents. To illustrate, takea propositionlike [] [(3e) e is the dying of QueenAnne] [] . That propositionwasfuture andis now past, but we can' t overlookthe temporal . Thereis an implicit when-clause, sothat whatwe actuallyhaveis anaphora that the propositionwasfuture (say, whenQueenAnne wasborn) andit is past (say, asI write thesewords). Thereis not eventhe illusion of a contradictionif we rememberto includethe temporalanaphora . The reasonfor the failure of the McTaggartargumenthereis not one usually given, and I believeit is not one anticipatedby Mellor. The claim is not merelythat the B-theoristhasattemptedto strip awaythe tensefrom (63) and (64) . TheB-theoristmustalsostrip awaythetemporalanaphorafrom theseconstructions . But, by hypothesis,all tensedsentences haveimplicit when-clauses . But if (63) and(64) comecomplete whichserveto do thework of temporalanaphora with when-clauses,then they can' t possiblycontradicteachother unless thosewhen-clauseshavethe samecontent. But they don' t. Nor will it help to attemptto takethis to a metalevel- for example,by suggesting that a givensetof A -theorytruth conditionslike (66* ) is future at some time, presentat sometime, andpastat sometime. If we try to formalize such claims, asin (67) and (68), we disguisethe actualstructureof suchclaims. (67) (66* ) is PAST (68) (66* ) is FUT Spelledout properlyaccordingto thetheorydevelopedin this chapter,the truth conditionsof (67) will be asin (67* ). (67* ) (66* ) wastrue when [ . . .] wastrue The B-theoristcanracealongthis pathindefinitely looking for a contradiction, but no level of embeddingis going to generatea contradiction.
TemporalAnaphorawithout B-SeriesResources
135
This is the sensein which the theorydevelopedhereis strongerthanPrior' s with respectto the McTaggartargument. Prior had no recourseto temporal , andhencehe had to allow that at everyother stepin the McTaggart anaphora argumentthere wasan apparentcontradiction. This gavethe impressionthat Prior had to keepmoving up a level to escapethesecontradictions,andhence that he wasforcedinto an infinite regress.ThoughI am not convincedthat this wasa genuineregress,the entire questionis now renderedmoot. The contradictionis nevergenerated . The McTaggartargumentfails.
Chapter
9
Broadening
the Investigation
As shouldbe evidentfrom the previoustwo chapters , I think thepreponderance of evidenceleansin favorof anA -theorysemanticsof tense.I hopeit is alsoclear that I considerthis semanticalevidenceto supportthe A -theoryconceptionof time. In this chapterI wantto broadenthe investigationsomewhatby taking up thequestionof whetherevidencefrom psycholinguisticsor evenfrom phenomonthetable, enologycanchangetheverdict. However,therearea lot of arguments andperhapsit is time to try andput themtogetherin a singlepackage . Let us review someof the reasonsfor favoring a A -theory conceptionof time andtense. First, we havegoodreasonsto supposethat thereis a close(if not isomorphic) relation betweenthe semanticsof tenseandthe metaphysics of time. As I rhetorically askedin previouschapters, what makessomething the semanticsof " tense" if not that it hassomereflex in the temporalcharacter of the externalworld? Moreover, appealsto the psychologyof time consciousness only delaytheinevitableconnectionbetweentenseandtime. Mter all, what makessomething" time" consciousness if not someconnectionbetweenthose mentalstatesandtemporalreality? Whatcanwe concludeaboutthe semanticsof tense?As we learnedin chapter 6, the B-theoristis going to haveprofounddifficulties in accountingfor the indexicalcharacterof temporaldiscourse.Moreover, therearereally only three optionsfor the B-theoristin this matter: . The B-theoristcan follow a token-reflexive strategyin which the utterance eventitself makesit into the contentof a tensedutterance / thought. But, as we sawin chapter6, that packstoo much information into the biconditionalsof the T-theory, andwe sooncollide with the problemof utteranceslike ' thereis no spokenlanguage'. . The B-theorist can try to cut the link betweenindexical meaningand the semanticsof naturallanguage.But, aswe sawin chapters3 and6, this strategy is a cheat.
138
9 Chapter
. TheB-theoristcandisquotationallyintroducetheindexicalpredicatesinto the axiomsof the T-theory. But this amountsto becominganA -theorist. It seemsto me that all the extantobjectionsto the A -theory semanticscan be dealt with. Insofarasthoseobjectionsturn on temporalanaphora , this does not forceusto accepta theoryof temporalreference ; to thecontrary,we canhave a theory of E-type temporalanaphora.Not only doessucha theory appearto coverthe facts; it also allows us to skirt someotherwiseembarrassingasymmetries betweentemporalreferenceandordinaryrefer~nce(for example, it allows us to avoid referenceto times or eventswith which we haveno causal connection) . Thus, the A -theory semanticshas epistemological advantagesas well as what I would argue to be an empirical advantage (namely, its ability to adequately handle the facts about temporal indexicals ) . But what if we throw out temporal indexicals ? This is a funny bit of evidence to throw out , since by some lights these facts constitute the very core of the temporal phenomena that we want to account for . Furthermore, if I am right , the temporal morphemesPAST and FUT are themselves indexical predicates, so ignoring indexicality amounts to ignoring the phenomenon of tense tout court . For the sake of argument, however, let us say that these facts can be ignored . Also for the sake of argument, let us dismiss whatever purchase epistemological arguments may have. ( Never mind the fact that epistemological concerns led to the invention of Russellian descriptions and E -type pronouns in the first place.) If we throw out all the evidence adduced so far on behalf of the A -theory, do we get a stalemate? Or is there further evidence that weighs in favor of one position over the other? As it turns out , even if we decide to ignore temporal in dexicals and epistemological considerations, there remains some very suggestive psycholinguistic researchthat supports the A theory semanticsdeveloped in this book.
Considerations 9.1 Psycholinguistic In section2.4 we considereda suggestionfrom Evans( 1985b) that supportfor oneof two competingsemantictheoriesmight be found by appealingto psychological evidence.In particular, we consideredthe possibility that evidence for a particularsemantictheorymight be found in factsaboutlanguageacquisition andin factsaboutacquiredlanguagedeficits. In this section, I will look at somesuggestivepsychologicalstudieson temporallanguage , andI will show that thesestudieslend supportto theA -theorysemanticsdevelopedin chapters 7 and8.
Broadeningthe Investigation
139
What psychologicalquestionswill be decisivein adjudicatingbetweenthe alternativesemantictheoriesfor tense? Intuitively, we will beinterestedinquestions thatdetenninewhetherthelanguageof theB-theoryor thelanguageof the A theoryis morefundamental . For example, accordingto the B-theory, the semantics given for the PAST, PRES, andPUT tensemorphemespresupposean of thebefore/after(earlier-than/later-than) relation. Alternatively, understanding thesemanticsfor theA -theorymight give the semanticsfor 'before' and ' after' in termsof thepredicates' past' , ' present', ' future' , andthe relationalpredicate ' when' . If it couldbe shownthatknowledgeof onesetof theserelationsemerged before the other, we would havestrongevidencefor the semantic significantly that takes thoserelationsasprimitive. theory Evidencealsomight be found in the studyof acquiredlanguagedeficits. As linguistic abilities erode, do we losethe ability to understandbasictensemorphemes first, or do we losetheability to understand termslike 'before' and' after' first, or do theseabilities alwaysdegradetogether? If oneability degradesearlier - th'([t it may , thatmaysuggestthatthatability wasnot themorefundamental havebeena higher-levellinguisticability requiringtheknowledgeof other, more primitive linguistic abilities. As it turns out, there hasbeenextensivepsycholinguisticresearchon the . acquisitionof temporallanguage,and a numberof the studiesare suggestive However, for the mostpart, the psychologicalstudieshavenot beendesigned to answerthekindsof questionsthatexerciseushere, andhencenot aUof them canhelp us choosebetweenour two semantictheories. This is not a criticism of thepsychologicalresearch . The focusof researchis inevitablytied to the interests of the field, and thereis no reasonthat theseinterestsshouldbe in the ultimateoutcomeof linguistic or metaphysicaldebates ? Still, someof theresearchhasbeenvery suggestive . For example,a greatdeal of researchhasgoneinto theinvestigationof our first question- whetherknowledge of the meaningof ' before' and ' after' emergessimultaneouswith or after our knowledgemeaningof ' past' , ' present', and ' future' . Here the evidence 'before'and' after' seemsto suggestthatthechild' s abilityto comprehend emerges significantlylater. Indeed, the acquisitionof simpletensesemergesquite early ( by age2), and, asClark ( 1973) reports, asa lower limit childrendo not learn the relationaltemporalterms' before' and ' after' until ageof5 !3 This orderof progressionis nicely illustratedby Weist( 1986), who specifically tracksthe developmentof temporallanguageusing the event-basedresources 's of Reichenbach theory. Accordingto Weist, childrenmovethrougha " seriesof four temporalsystemsduring the developmentof the capacityto express " increasingly complex configurations of temporal concepts (p. 357) .
140
Chapter9
' Utilizing Reichenbachs S (speechtime), R (referencetime), andE (eventtime), Weistproposesthat childrenproceedthroughthe following stagesof temporal . development stage1: R andE arefrozenat S; childrencanonly talk aboutpresentevents. stage2 (age18- 24 months): ChildrendistinguishE from S (usingsimpletense formslike pastandfuture), but R is frozenat S, sothechildrencannottalk about eventsoccurringrelativeto othertime points. stage3 (age3 years): R may be distinguishedfrom S, but when it is, E is restricted to R. (For example, a temporaladverblike ' yesterday' might modify R, yet E will still be fixed relativeto S.) stage4 (age4 years): R, S, andE canall be at separatetimes. ' theseresults Contraryto Weists analysis,however,it seemsmorenaturalto take . as supportingthe A -theory semanticsof tense. On the face of it , there is no , the orderof acquisitionshould principledreasonwhy, on B-theory semantics proceedin this order; accordingto theB theory, thePAST, PRES, andFUT tense to behigh-levelabilitiesrelativeto thebefore/afterrelation morphemesaresupposed . On theotherhand, thisorderof acquisitionseemsnaturalwithin theframework . The orderof acquisitionwould be ( 1) rule for of theA -theorysemantics -tensemorpheme PREStensemorpheme(if thereis a present ), (2) rulesfor PAST and FUT tensemorphemes , (3) rules for temporaladverbs, then (4) rules incorporating higher-level abilities involving temporalconjunctionsande-type . temporalanaphora Someof the interpretationsgivento dataof this sortin the psychologicalliterature areinterestingaswell. Forexample,Cromer( 1968) andMcNeill ( 1979), working within a broadly Piagetianframework, arguedthat beforeage4 children " themselves- that areunableto temporally" decenter theyaretemporally to mind the remarksin egocentricin a certain sense.This observationbrings chapter7 in which, following Dummett( 1969), I characterizedthe distinction betweenthe B-theory andthe A -theory as a distinction betweena perspective " in which we are standingoutsideof time, surveying(asDummettwrote) all " " temporalpositions in a single glance, and one where we are immersedin "4 '' " temporality andonly ableto describethe world asit is now. Acquisitiondatais not the only suggestiveevidence.It alsoappearsthat research on acquiredlinguisticdeficits(for example,amongParkinson's patients) showsthat comprehensionof ' before' and ' after' degradesmuchmorerapidly thanotherlinguistic forms (seeGoodglasset al. 1979). This might be takenas that ' before' and' after' arenotprimitivepredicatesat all but arerather suggesting built out of simplernotions, suchasthebasictensesandtemporalconjunctions.
Broadeningthe Investigation
Therearealsonarrowerquestionsto be investigated . TheA -theoryof tense, asdevelopedin chapter8, necessarilyincorporatesa notionof whenasa primitive to bedistinctfrom thenotionof occurringat the , andthis notionis supposed sametime. This is a verysubtledistinction, andmostpsychologicalstudiesarenot haveobserved , someresearchers designedto teaseapartthetwo notions. Nevertheless ' when' such a distinction and have noted that a sense of , just particular . emergesbeforethechild hasa notionof temporalorderandsimultaneity For example, in a remarkablepassage , Cromer( 1968, p. 110) notesthatfirst usesof ' when' (suchas " When its got a flat tire, its needago to the. . . to the station" ) aremorestatements of co- occurringeventsthanstatements of genuine 's " . Cromer conclusion is that perhapstheability to ' date' simultaneity fascinating ' an eventby a contemporaneous eventis more primitive' thanthe notion of serial ". ordering Interestingly,asKelier-Cohen( 1974) noted, similar observations aboutearly usesof ' when' (i.e., that it doesn't mean" at the sametime" ) canbe found in ChamberlainandChamberlain1904andin Jesperson1940. Clearlythis observationby Cromerandothersdoesnot by itself supportthe thesisthat thereis a nonsequentialuseof ' when' which is prior to the notion of two eventsoccurringat the sametime. The researchnecessaryto comeup with conclusiveresultsin this areawould be delicateindeedsYet Cromer's observation is promising, for it suggeststhatthequestionmayhaveanempiricalanswer (andindeedthat initial observationssupportoneof the key suppositionsof the A -theorysemantics ). I hopethatthekindsof questionsraisedherewill helpto focusfutureresearch on the acquisitionof eventstructure, of tense, andof the generalconceptionof temporalorder. Thereis certainlymuchat stakein theanswerto properlyframed questionsaboutrelativeorderof acquisition.First, thesemanticsof tensefor natural languagemaywell beilluminatedby theanswer; second,if thegeneralprogram followed in this book is correct, ultimatelythe metaphysicsof time could be illuminatedby the answerswe get to thesevery low-level questions.It may seemsurprisingthatanswersto grandmetaphysicalquestionsshouldturn on answers to questionsabouttheacquisitionof languageby children, but I hopethe considerationsoutlined in this book will makesucha conclusionseemmore still surprising. plausible, if nevertheless
9.2 Savingthe Phenomenology6 TheA -theoryis often supposedto be morefaithful to our intuitions abouttime thanthe B-theory is, but in the theory I developedin chapters7 and 8 this advantage may appearto be lost. After all, like the B-theorist, I havegivenup the
142
9 Chapter
notion of genuinetemporalchange.Perhapsmoresignificantly, I haverejected the reality of future andpastevents. Surely, thesemovesalsoconflict with our intuitions aboutthe natureof time! Or do they? A lot turnson whatwe meanby ' intuitions' . Therearethoseintuitionswhich wearetrainedto havein ourphilosophyclasses , andthentherearewhatwe might " " -induced call untutoredintuitions. Obviously,I amunimpressed by philosopher intuitions in this domain. This leavesour untutoredintuitions. It is fair to ask whethertherearesuchthings, and, if thereare, what they might be. At a minimum, we might expecta theorythatcomportswith our intuitionsto be consistentwith the way we experiencethe world. That is, if we experience somethingsasfuture andothersaspast, we would saythat a theorywhich reverses that" direction" clasheswith our intuitions. Likewise, if we experience the world ashavinga genuinefuture" outthere," then( barringsomeexplanation ) the sortof theoryadvocatedheremight well countasconflictingwith our intuitions abouttime. This talk of " experiencingthe world," of course, landsus thick in phenome the method is controversial it seems to me , nology, Although phenomenological that it canbe instructivein caseslike this, becauseit showsthat eventhe notion of how we experiencethe world hasto be handledwith delicacy- particularly in thecaseof the phenomenologyof time. How do we experiencetime, then? Onemight supposethat the philosophically untutoredway we experiencetime is asa dynamicsystemin which events movefrom the future to the presentandon into the past. That is certainlyhow a number of philosophershave written about our experienceof time. The metaphorson this scoreareendless.Williams ( 1951, pp. 461- 462) catalogued themasfollows: Timeflowsor fliesor marches . Moreexplicitlywemayspeakas , yearsroll, hourspass if theperceiving mindwerestationary whiletimeflowsby likea river, withtheflotsam of eventsuponit; or asif presentness werea fixedpointerunderwhichthetapeof happenings slides;or asif thetimesequence werea movingpicturefilm, unwindingfrom thedarkreelof thefuture,andrewound intothedarkcanof thepast.Sometimes , again, wespeakasif thetimesequence werea stationary onwhichwevoyage , plainor ocean or in a variegated rivergorgedownwhichwedrift; or, in Broad's analogy , asif it were " arowof house frontsalongwhichthespotlight of thepresent of nowness plays. Theessence " " "runslike fire the fuse of time . , Santayana , says along However,thesemay all be examplesof badphenomenology . A numberof phehavearguedthatour experienceof time is not like this at all, and, nomenologists more significantly, that our experienceof time doesnot supportan independent future andpast. Onecasein point is the work of Merleau-Ponty( 1962, p.
Broadeningthe Investigation
143
411), who categorically rejects the idea that we experience time as flowing or that we experience an independent future and past: We saythat time passesor flows by. We speakof the courseof time. . . .If time is similar to a river, it flows from the pasttowardsthe presentandthe future. The presentis the of the past, andthe future of the present.But this oftenrepeatedmetaphor consequence is in a way extremelyconfused.For, lookingat thethingsthemselves , the meltingof the snowsandwhatresultsfrom this arenot successive events.. . . The point is that if we think of time as a series of successiveevents, we have to think of an observer of the earlier and later events. For example , standing on the banks of a river today and watching the water passby , I might say that there was an event of this water coming down from the mountains. But that supposes a witness up in the mountains , and our taking the perspective of that witness. ' Phenomenologically speaking, I really can t take that position . This blunder, in ' " Merleau - Ponty s view , has persisted from the time of Heraclitus to our own " " day, and the root of the blunder is our surreptitiously putting into the river a witness of its course" (ibid ., p . 411 ) . That move is simply out of court if we are interested in how we experience the world . If we concentrate on our experience of the world , we have to reject the idea that time is a process or that it involves reference to independent future and past events. Rather, we have to think of the future and past being , as it were, in the present. "Time ," writes Merleau - Ponty (ibid ., p . 412), is, therefore, not an actualprocess,not an actualsuccessionthat I amcontentto record. It arisesfrom myrelationto things. Within thingsthemselves , thefuture andthe pastare in a kind of eternalstateof preexistenceandsurvival; the waterwhich will flow by tomorrow is at this momentat its source, what hasjust past is now a little further downstream in the valley. What is pastor future for me is presentin the world. Note how smoothly this comports with the claim in chapter 7 that we evaluate past- tense and future - tense utterancesby considering current evidence. Indeed, how could we not evaluatetensedpropositions this way if the future and the past are " present in the world " ? Merleau - Ponty (ibid ., p . 413 ) also cautions against appeals to memory to try and make senseof time as a sequenceof events: . . . it haspassedunnoticedthat our bestreasonfor rejecting the physiologicalpreservation ' and of the pastis equally a reasonfor rejectingits ' psychologicalpreservation , ' ' that reasonis that no preservation,no physiologicalor psychic trace of the pastcan makeconsciousness of the pastunderstandable . This tablebearstracesof my pastlife , for I havecarvedmy initials on it and spilt ink on it . But thesetracesin themselvesdo not refer to the past: they are present. . . . Nor can one, a fortiori , constructthe future
144
9 Chapter
out of contentsof consciousness : no actualcontentcan be taken, evenequivocally, as evidenceconcerningthe future. . . . My appeal to Merleau - Ponty here is admittedly an appeal to authority , but the point is not to nail down one certain picture of how we experience time . Rather, the point is to illustrate that the picture of time as involving an independent future and past is by no means considered faithful to our experience by those who purport to be interested in the phenomenology of time . Accordingly , it cannot be said without sustained (and no doubt subtle) argumentation that the theory presented in chapters 7 and 8 clashes with the untutored way in which we experience the world . Very simply , the burden of proof rests with those who would advance the priority of one particular set of intuitions . 9. 3 Conclusion
Let us takestock. In chapter6 we sawthat therewereseriousproblemsfor any versionof the B-theorythat hopedto accountfor the phenomenonof temporal indexicals.As wasshownin earlierchapters , this limitation of theB-theorycannot " be dismissedout of hand. If the world is " tenseless , then one cannotappeal to semanticsor psychologyto deliver the missing tensedingredient, for - they cannotadd semanticsand psychologyare not disembodiedenterprises tenseto a tenselessworld. Only a full A -theorysemantics /metaphysicsseemsto be capableof makingsenseof our temporaldiscourse. In chapter7 we sawthattheA -theorysemanticsandmetaphysicscouldclaim a numberof strongepistemologicaladvantages . For example,theA -theoryfrees B that we can actuallyrefer to times us from the embarrassing theoreticclaim andeventsin the future and in the causallydisconnectedpast. It further frees us from the claim that to evaluatepast(future) tensesentenceswe must, as it were, travelinto the past(future) andevaluatethe sentenceat that past(future) tensetime. In short, theA -theorystoryseemsmuchmoreplausibleasanaccount of our semanticalknowledge. In thischapterwe haveseenthatbroadeningour investigationleavesthis conclusion intact. If we turn to cognitivepsychologygenerallyandto the theoryof conceptJIanguageacquisitionspecifically,we find thattheevidenceweighsheavily in favor of theA -theoryof tense. Moreover, by someaccountstheA -theory of time. I concludethat doesa betterjob of comportingwith thephenomenology theA -theory (of both tenseandtime) is the moreplausiblegeneraltheory and that it shouldbe adoptedin future researchon the natureof tenseandtime.
Broadeningthe Investigation
145
This is a strong conclusion in a number of respects. As regards metaphysics the A - theory is certainly a minority view , and as regards semantics the Atheoryis virtually ignored . But the conclusion is also a strong one in that it entails a number of radical and far -reaching consequences for other domains of philosophy and linguistics . In the next chapter, I will outline some of these consequences and propose some directions for future research within this general framework .
Chapter
10
Consequences
So far , this investigation has been limited to some of the more immediate consequences of the A -theory. In particular , we have been interested in the consequences for semantical theory, and to a lesser extent for psychology . But , as was noted in the introduction , the doctrine of time is interwoven with virtually every branch of philosophy and indeed nearly every branch of human inquiry . For the most part these other forms of inquiry have been predicated on B -theory assumptions( either tacitly or explicitly ) . If those B -theory assumptionsare abandoned and replaced by an A -theory metaphysics, it is natural to wonder what some of the consequenceswill be. Because the consequenceswill be felt in areas ranging from the theory of causality and the philosophy of action to the philosophy of spaceto the philos ophy of religion , it would be impossible to do justice to all of the potential areasaffected.! Accordingly , I think it might be more useful for me to pick out one string of philosophical consequences, and one string of linguistic consequences , and follow them for a while in an effort to illustrate just how rich and complex they will be. 10.1
Philosophical Consequences
Logic and 1ruth -Value Links2 One of the most pressingissuesfor this particular defenseof the A -theory is the apparentlossof truth-valuelinks, andhencethe lossof an apparentlynatural accountof certainlogicalinferences . As anillustrationof theproblem, consider 1 below discussed in , already ( ) chapter6.
(1) Dinosaurs roamed theEarth .
148
10 Chapter
On the presentistversionof theA -theoryadoptedhere, we do not evaluatethis sentenceby "traveling" to sometime earlierthannow anddeterminingwhether, atthattime, ' DinosaursroamtheEarth' is true. Rather,thetruth of it is grounded by currentfacts- for example, existingfossils. This generalline of responseis, of course, similar to onethat hasbeenpursued by Dummett(mostnotablyin his 1969paper) andother antirealists, but thereis arguablya difficulty for the position. It canbe held that thereareimportant truth-valuelinks betweenstatements aboutthepastandstatements about the present, andthat antirealists (andA -theorists) mustprovidesomealternative accountof theselinks. For example, we routinely makeinferenceslike (2). (2) I am hungry. Next Tuesdayit will be true that I washungry. On aBtheory semantictheorylike thatsketchedin chapter5, it is clearhow this sort of inferencecanbe made. If I am hungrynow, thenthereis a time t = now suchthat I am hungryat t. But thenfor anydatel' later thant it will be true at l' that I was hungry. Sincenext Tuesdayis later than t , next Tuesdayit will be true that I washungry. The A -theorist, on the other hand, will needto articulatealternative(antirealist ) truth-value links that can be drawn on for theseinferences. To illustrate , Wright ( 1993, chapter5) formalizesthe truth-valuelinks groundingthe B theoryinferences asfollows:
for tl < t2 A ('v'tJ [PastiSis trueatt2iff (3tl)(S wastrueat tJ] B ('v'tl)[Fut:S is trueat t) iff (3tJ(S will betrueattJ] C ('v'tJ [S is trueat t2iff ('v'tJ(FutiSwastrueattJ ] D ('v't) [S is trueat t) iff ('v'tJ(PastiSwill betrueat tJ] andsoon.3 Thus, if FUT[S] is trueonMay 1, 1998,thereis a laterdateatwhichS will be true(byB), andsoforth.
Consequences
149
Hereis thedifficulty for theA -theorist: Sincefuture- andpast-tensedsentences areto be evaluatedon the basisof the present,it is possibleto envisiona situation in which evidencethat may havebeenpresentat 1. is erasedor eliminated andis hencenot availableat 12 ' That is, supposethat today (1.) I commit some crime(say, I kill ColonelPlumwith a candlestick ) andtheneraseall evidenceand takeamnesiapills so asto alsoforget the cover-up.4Now we appearto havea ' situation where ' I am (success fully ) killing Col. Plum with a candlestick is true. The problemis this: If my cover-up is good, at 12therewill be absolutely no factsto supportthetruthof ' I killed Col. Plumwith a candlestick'. Thus, if the 12statementis to be evaluatedby facts holding at 12 , the statementwill not be true at 12 , Then, by (D ), it follows that ' I am (success fully ) killing Col. Plum with a candlestick'is false, contradictingour initial assumption . The movethat looks fishy hereis the one wherewe saythat at 12therewill beno factsto groundthetruth of the statementin question. Doesn't this amount to traveling into the future and evaluatingthe situation- and isn' t this precisely the sortof situationthat is supposedto be blocked? Indeed, all the A -theoristis compelled to admit is the following : We begin with thetruth (now) of ' Therewill beno factsto groundthetruth of ' I killed Col. Plum with a candlestick'. From this and our truth-valuelinks, all that we can ' infer is that the following will be true at 12 : there are no facts to ground the ' truth of I killed Col. Plum with a candlestick'. But this isn' t enoughto move backto a contradiction. Gettingthe contradictionrequiresthe following additionalpostulate: E ' ' (V1)[ Thereareno factsto groundthe truth of S is/was/will be true at tiff S is/was/will not be true at t)] ' Usingthis, we caninfer thatit will not betruethat I killed Col. Plumwith a can' ' dlestick at 12 , Now ( D) carriesusto thefalsity of I am(success fully ) killing Col. Plum with a candlestick', andwe haveour contradictionis It appearsthat theA -theoristmustfind anotherway to escapecontradiction here. Onepossibleansweris that this worry abouttruth-valuelinks hasthings , asMcDowell ( 1978) seemsto suggest upsidedown. Perhaps , the real issueis that the realist aboutthe pastneedstruth-valUelinks in orderto supporthis or her realism. Accordingly, perhapsthe burdenis on the realistto justify the appeal to theselinks. McDowell (p. 132, note 10) thinksthat no suchjustification canbe forthcoming: Thetruth-valuelink realist's viewof whatit is, say, for something to haveoccurredis . Heconceals thatfromhimselfwith a confusedthoughtof beingwith unintelligible
150
Chapter10
' knowledge acquiringpowersdifferentfromours. Thustherealists viewof thereality of thepastcanbedescribed , with onlymildcaricature , astheideaof anotherplace,in whichpasteventsarestill occurring , by God. , watched , perhaps Like Wright ( 1993, chapter3), I am inclined to think that this criticism is unfair . The realist aboutthe future andthe pastis underno compulsionto claim specialknowledgeacquiringpowers. We aretalking logic here, not epistemology, andthe realistis quite happyto cut the link betweenthe two. Of course, by parity of reasoning,an antirealistis not compelled to accept truth-valuelinks asgroundinglogical reasoninglike that in (2). Still , onewonders whatcouldgroundsuchreasoningif not the usualtruth-valuelinks. Would it evenbe possibleto dispensewith suchlinks? Wright ( 1993, p. 179) considers the prospectsdim: tothosefeatures of, say,classical Thetruth-valuelinksarenottobecompared logicwhich the in tuition ists. The intuitionistic modifications are modification deep -reaching undergo by with respect to ourordinarynotionof validinference ; buttheyarerelativelyconservative .Wholesale of thetruth-valuelinks, in contrast , wouldbeboundtoleave rejection to workat of howtensedlanguage wassupposed us, it seems with no clear , conception ' to betaken 'It will rainatnoontomorrow whoasserts all. What,for instance , is someone of hisutterance arenotheldtocoincidewiththatof any assayingif thetruth-conditions ?Maybetherecouldbeacoherent utterance of 'It israining'atnoontomorrow revisionary for tensed discourse , pivotingaroundrejectionof thetruth-vauelinksasthe programme intuitionistrevisions of classical ; andperhaps logicpivotaroundrejectionof Bivalence amountto say.Butit doesnotseemlikely. weshouldbeleft witha surprising . Is Let us suppose , then, that someform of truth-valuelinks areindispensable therea way out for the antirealist here? Wright considerstwo possibilities. The first way out- the oneadvocatedby Wright- is to refiguretruth-value links in a way that is moresympatheticto the antirealist aboutthe pastandfuture . Accordingto Wright, the antirealist shouldreject(A )-( D) andopt instead for links like (A * )-( D* ) . A* esat t2is true iff ( 3tl)(what S expressedat tl is true)] (~ tJ [ WhatPastiSexpress B* esat tl is trueiff (3tJ (whatS will expressat t2istrue)] (~ tl )[ WhatFutiS express C* esat t2is true iff (~ t .>(what FutiS expressedat tl is true)] (~ tJ [ WhatS express D* esat tl is trueiff (~ tJ (whatPastiSwill e~pressat t2istrue)] (~ tl )[ WhatS express andsoon.
Consequences
The difference is that the proposed * links supplant the tense inflection in ' be true ' with tenseinflections in the verb ' to express' . Wright ' s idea is that by doing this we can retain the truth -value links by giving up the stability of the truth predicate - - that is, by giving up on the timelessnessof truth . To illustrate with our example of Col . Plum: even if our evidence erasureprogram is successful, what ' I killed Col . Plum with a candlestick ' will express at ' 12is (currently ) true ,. It follows unproblematically that what I am ( success' fully ) killing Col . Plum with a candlestick now expresses is true. No contradiction need arise. ' Wright s solution then is to allow that what we express by an utterance (tokening ) of S will not be eternally true in the traditional sense (whatever that comes to once we have given up the reality of the future and the past), but rather that what is said might flip - flop between true and false over time . This is a pretty strong conclusion . As Wright ( 1993, p . 201) puts the matter, "the prospect of rehabilitating something importantly akin to the classical notion of truth is , in effect , finally dashed." Though I have not been reluctant to embrace ' strong conclusions in this book , one naturally wonders if there isn t a less radical way out. One alternative idea, which Wright attributes to Durnmett ( 1969), is that what S expresses will shift over time . Truth could remain stable if we allowed that the content of an utterance (tokening ) was temporally unstable. Put another way, perhaps an utterance at 11has a certain content C at I . but that very same 11utterance has another content at 12, Applied to the current problem of my killing Col . Plum , we might accept that what ' I killed Col . Plum with a candlestick ' will express at 12is true but argue that it will express something rather different at 12than it does as uttered now (at 11), and we are in no position now to know what the content of that utterance might be at 12, Wright ( 1993, p . 194) rejects this solution , not becauseof its paradoxical feel , but becauseof its alleged inability to cope with diachronic disagreements: . . . if Dummett' s antirealist hasindeedfound a way to explain how he can avoid dismissing the views of his earlierand later self, the threatenedcost is an unability to explain how therecan be sucha thing asconflicting views held by protagonistswho are sufficientlyseparatedin time. Dummett's antirealistis thereforeopento a simpledilemma. Doeshis positionpermit thereconstructionof somesortof generalnotionof diachronicinconsistency , or does it not? If it doesnot, that merely furnishes the realist with a further powerful objection - whataccountarewe nowto giveof thegrowthof human knowledge,thehard-won defeat of and error in which we are gradual superstition encouragedto believe, etc. etc.? But if somesort of notion of diachronicinconsistencycan be saved, the task will still
152
10 Chapter
of t11lth doesallow remainof showing thatthekindof clashwhichanindexical conception in timedoesnotobtainbetween hispresent to holdbetween statements widelyseparated . andhislater, andearlieropinions It seemsto me that if onewantsto breakthe dilemmaandpursuethe shiftingarepossible contentstrategy,then one shouldhold that diachronicdisagreements butrejecttheideathattheycanbethoughtof asconflictsbetweenpositions held at I . andIrRather , the ideawould be that the relevanthistorical disagreements are all " in the present." Thus, if I now dispute Plato' s doctrine of the forms, it doesnot meanthat the view I hold at 12conflicts with the contentof somethingthat Plato said at 1., but ratherthat the conflict betweenour views musttakeplaceastheyarecouchedat IrAlthough this subtleandcomplexgambit obviously must be playedwith care, I seeno reasonto supposethat it inevitably fails. In my view, then, we are left with two solutions to the problem of truthvaluelinks, bothof whichneedto beexploredin moredetailbeforetheyarefully . Still, in theinterim, it wouldbeinterestingto seewhattheconsequences secured of thesegambitsmight be, sincethe A -theory will haveto makegood on at leastone of the two. In the next sectionwe will beginexploring someof these . consequences Externalism, Self. Knowledge, and Memory Thetwo solutionsthatWright discussesfor theproblemof truth-valuelinks turn that ripple into debatesaboutthe natureof memory, out to haveconsequences as it relates to the authorityof our self-knowledgeandthe doctrine particularly of externalismaboutmentalcontent. If (asPutnam( 1975), Burge( 1979), and manyothershaveargued) the contentsof our mentalstatesdependuponsocial for our ability andenvironmentalfactors, thenthereareallegedto beconsequences to know the contentof our mentalstateswithout first investigatingthe environment " . That is, if (as Putnamsays) " meaningsain' t in the head, it is not clearhowwe canhaveauthoritativeself-knowledge.This wouldunderminebasic assumptionsaboutthe natureof self-knowledgethat havebeenwidely held at . leastsinceDescartes But Davidson( 1987), Burge( 1988), andothershaveobservedthat thereis -order notensionin externalismandauthoritativeself-knowledge , sincethesecond knowledgeof our mentalstatesmustbe environmentallydeterminedas well. -orderthoughtthatmightbeexpressed as" I amthinkingthatwater Thus, a second " -orderthoughtinvolvingwateror twater, dependingupon is wet will bea second . in whetherthe agentis a waterenvironmentor a twaterenvironment
Consequences
153
This movehasled to a furtherdiscussionaboutexternalismandmemory, driven in partby arguments madeby Boghossian ( 1989) andLudlow( 1995). Memory comesin asfollows. Boghossianarguesfor the incompatibilityof externalism andself-knowledgebasedon the so-calledslow-switchingthoughtexperiments of Burge ( 1988). We are askedto imaginean agentwho unknowingly moves betweenEarthand1\ vin-Earthandis laterinformedthat switcheshadbeentaking ? place.Wouldsuchanagenthaveauthoritativeknowledgeof his mentalstates , the agentwould not, sinceit appearspossiblefor an Accordingto Boghossian agentS to know his thoughtsat time 1, forget nothing, yet at sometime later than1( havingbeeninformedof the possibilityor prevalenceof slow-switching) be unableto saywhatthe contentsof his thoughtswereat 1. Boghossianargues: Theonlyexplanation , I ventureto suggest , for whyS will notknowtomorrowwhathe is saidto knowtoday,is notthathehasforgottenbutthatheneverknew.Burge's selfverifying do notconstitutegenuineknowledge . Whatotherreasonis there judgments for whyourslowlytransported thinkerwill notknowtomorrowwhatheis saidto know directlyandauthoritatively today? Ludlow ( 1995) reconstructsthis argumentasfollows: ( 1) (2) (3) (4)
IrS forgetsnothing, thenwhat S knowsat 1\, S knowsat 12 . S forgot nothing. S doesnot know that P at 12 . ThereforeS did not know that P at 1\.
At which premiseshouldthe externalisttakeaim?Accordingto Burge( 1998), 's theweaklink in Boghossian argumentis premise3. Accordingto Burge, memory " " providessomethinglike anaphoriclinks to eventsin the pastthat allow us accessto pastcontents. Thus we do haveresourcesthat allow us accessto the contentP at time 12 . Of course,if we rejectthereality of futureandpastevents,thentheBurgeposition is a non-starter. It cannotbe thejob of memoryto " reachback" into the pastfor a content, for the simplereasonthat thereis literally no pastto reach back to. The picture of memorythat we are left with is somewhatHumeanin - not in the sensethat memoriesaresimply vaguesenseimpressions character , but in the sensethat the contentof a memoryis fixed entirely by presentcircumstances andnot by its linking us to thepast. Boghossiancannotbe engaged on premise3. Alternatively, Ludlow ( 1995b) holdsthat the weakpremisein the argument is the first one. The reasoningis that if the contentsof our mentalstatesaredetermined , it is naturalto supposethat the contentsof by our socialenvironment
154
Chapter10
our memorieswill dependuponour socialenvironment . As we movefrom one environmentalconditionto another(perhapswithout evennoticingthe environmental change), the contentsof our memorieswill shift accordingly. Ludlow ' ibid. ( ) appliesthis line of reasoningto Boghossians argumentin the following manner: . . .let's saythatattimet . I knowthatI amthinkingthatarugulais bitter. Suppose thatat timet2laterdiant I' I recalldIatinitialdIoughtaboutarugula , but, dueto undetected changes in my linguistic community, the contentof my thoughtsabout arugulahave shifted. Boghossianis arguablycorrectin assertingthat I do not know at t2what I knewat t I' but he is incorrectin supposingthat "the only explanation" for this is that I " neverknew" my thoughtsin the first place. It is entirelyconsistentwith dIe socialextemalistview of memorythat I forgot nothing, but that the contentsof my memorieshavenonetheless shifted. Hofmann ( 1995) points out some apparentdifficulties with this move. It appears that memory fails to reflect the content of memory targets, and further that it is not reliable regarding the truth of my earlier thoughts . Hofmann puts the first problem this way : . . . if [thecircumstances of recollection] detenninememorycontent, thenmemoryturns into an empty, absurdfaculty. This is so, sincememoryno longercando what it is supposed to do, namely, to recall the very samethoughtsone earlieron had entertained . If Peterhadcometo believeat I . somehowthat arugulaonly growsin Mediterraneanclimate , and now, at 12 , recalls this thought, then what he recollectswill not be the same thoughtaboutarugula, but someother thoughtabout, say, tarugula- accordingto the newly adoptedindividuatingconditions. Hofmann puts the second problem as follows : . . . equally embarrassing , the truth valuesof the memorieswill havechanged. So, for ' s belief at that Peter , example arugulaonly growsin Mediterraneanclimatewill now, II whenrecollected . . . . For, at time 12it will bea thoughtabouttarugula , turn into a falsehood which is a vegetable(let' s assume ) that growsonly in tropical climates. Evenworse, if Peterrecollectsthathehadhad, at I I' somearugulaexperience , thenalsothis memoryhas turnedby now into a falsehood, sinceit hasbecomea memory-thoughtabouttarugula which (let' s assume ) Peterneverhashadany direct encounterwith. And, all of this has occurredwithout Peterbeing able to introspectivelybecomeawareof it . Memory, as socialexternalismwill haveit, has' turnedpseudo'. It is no longera sourceof knowledge. In response, Ludlow ( 1996) arguesthat, although memory is not faithful in certain respects, in most important respectsit will be. For example, the veracity of most of my beliefs will survive the switching cases. If I believe at t I that water is wet , then at t2I will have a belief that twater is wet. If my earlier beliefs were on the whole reliable , then my later beliefs should be as well . As for the fact that
155
Consequences
my recollection of the initial thought is that it was a twater thought, we simply neededto rethink the point of having memories in the first place: . . . is thereanyreasonto supposethat it is thejob of memoryto " recordthe contents" of ?On thefaceof it, this assumptionbegsthecentralquestionat issue. pastmentalepisodes According to the extemalistconceptionof memorythat I haveproposed, it is not the job of memoryto recordcontents,but ratherto provideinformationaboutpastepisodes relativeto currentenvironmental conditions.Evenif therewerea mechanism whichcould, asit were, freezethe contentsof an initial mentalepisodeandcarry it in memoryindefinitely ' , I m not surethat it would haveany utility . Indeed, it would be a way of preserving the contentof a thoughtwhich we could no longerhave. (ibid., p. 316) However, the analysis of truth -value links proposedby Wright (the one that gives up the timelessnessof truth ) allows an even bolder response. If the switch is such as to make events on Earth in principle inaccessible to us, then it is simply no longer true that my original memory was a water memory . In a sense, then, a conception of memory that is sensitive to the current environment is crucial for preventing error , since otherwise it would be preserving contents that are no longer true. The second objection thus collapses. Alternatively , supposewe adopt the alternative way out canvassed( and then dismissed) by Wright : supposewe hold that the timelessnessof truth is preserved, but that the contents of our statementsshift over time. This too gives us a way out. In the latter case, the episode of my thought E that occurs at to has water content when I am at tobut twater content when I am at tr Thus, the thought might not be identified with a single content , but rather with a series of contents at different times. What I was thinking at to would depend upon when I happen to recollect my thought- or , more generally, where I happen to be at later points of time . That is , a given episode of thought E occurring at to might be assigned different contents at different times , as follows . E ' s content
to
t.
~ o
~ O
t2 XYZ
t3 XYZ
t4 XYZ
ts H2O
Returning to the original Boghossian argument, we find that the Ludlow ( 1995b) reconstruction suppressessome important details. The full argument should have been as follows : (0) (I ) ( 2) ( 3) (4 )
If a first -order thought E has content P at t ., then it has content P at tr . If S forgets nothing , then what S knows at t . S knows at tr S forgot nothing . At t2 S does not know the content of E to be P. Therefore , at t . S did not know the content of E to be P.
156
Chapter10
Obviously, the previouslyhiddenpremise(0) is seriouslyflawed, sinceE may well havethe contentP at time I . but somethingelsealtogether(say, Q), at 12 . Premise1 is flawedtoo, of course, sincewhatI know at I . is E' s I . content, and whatI know at 12is E' s 12content. No longerknowingthe 11contentof E doesn't count asforgetting any more than the fact that I can no longer know that E is happeningnow. But the crucial thing to seeis that, althoughmy second-order knowledgeof my thoughtsshifts overtime, that shift is crucial, sinceit allows me to track the shifting contentsof my first-orderthoughts. Memoriesthat are sensitiveto environmentalconditionsare crucial for us to keeptrack of what we werethinking. This may seema surprising conclusion. Episodesof belief are no longer stableobjects; they arenow highly dynamicandsensitiveto the communities that we inhabit over time. On the other hand, if the conclusionis surprisingit alsohasa certainelegance . Our revisedtheory of truth-valuelinks wasdriven our of the by rejection reality of the future andthe past. It is remarkablethat thenewway of thinkingabouttruth-valuelinks shouldalsohelpto secureanapproach to memorythat is intimately interwovenwith widely held views about mentalcontentandself-knowledge.But, in view of the tight connectionsholding betweentimeandmemory,betweenmemoryandmentalcontent,andbetween mentalcontentandself-knowledge , therealsurprisewouldhavebeenif themove to theA -theorywasphilosophicallyinert. Clearly, it is anythingbut that.
10.2 LinguisticConsequences The linguistic consequences for this proposalare, of course, vast. We havealready seenthat the proposalsketchedherehaswide-rangingconsequences for theanalysisof (implicit) temporalconjunctions relative clauses etc. Moreover , , , the consequences flowing from the analysisof temporalanaphorawill not be insignificant. Ratherthandwell at lengthon someof theseobviousconsequences , in this sectionI want to pursuea much strongerand more provocativepossibleconsequence : the possibility of eliminating the notion of tenseas a grammatical categoryaltogether. In this bookwe haveseenthatthereis no referenceto futureor pasteventsthat(for example) theEnglishtensesystem,suchasit is, doesnot needreference timesin thefutureor in thepastin orderto makesenseof temporaladverbs , temporal , complextenses , etc. Whatwehavearebasictemporalmorphemes anaphora (pAST,PRES,FUT) thatarepredicates takingproposition-like objectsastheirarguments . Perhapswe cango onestepfurtherandexorcisethetalk of temporality
Consequences
157
from what we sloppily call ' temporal adverbs' , ' temporal anaphora' , ' tensemor' phemes , and so on. This might sound crazy, but in a senseit is entirely natural , since that many natural languagesdon ' t have tense morphemes anyway. We need not look to un-
familiar languagessuchas Hopi. English doesn't havea genuinefuture-tense ; rather, it relieson modalsto do (or sowe think) thework of afuturemorpheme tensemorpheme.As we move from English to other languages , we find that - in thoserarelanguageswherethey are future-tensemorphemes purportedto exist- are alwayssuspiciouslooking. In Romancelanguages , they appearto havemodal elementspackedwithin them. Purportedpast-tensemorphemes areno lesssuspicious , usuallybeingnothingmorethanaspectualmarkers. The standardview supposesthat we are using modalsandaspectualmarkers to expressfuture tenseandpasttense(hence,to expressthingsaboutthe future andthe past),6but why shouldwe supposethat? Why not supposewe are just usingmodalsto expressmodality(potentialityor probability, for example) andaspectualmarkersto expressaspect(perfectaspect,for example)? Perhaps somewhere(maybeit wasAristotle' s fault) bad philosophyinfectedlinguistic theorizing. Wemovedfrom theassumptionthattheremustbea futureanda past to the conclusionthat theremustbe linguistic elementsthat allow us to speak aboutsuchthings. Perhapsthis philosophicalassumptionhasbecomea procrustean bed in which we categorizethings as temporalelements, when to a Martian linguist with no knowledgeof westernphilosophyof time, theseelements would look like ordinarymodalsandaspectualmarkers. No Future To makethis discussiona little lessabstract,let ustakeup thequestionof thefuture tensein Romancelanguages : What we aretaughtwhenwe learnSpanish, for example, is that therearefuture-tensemorphemesthat in the regularform ' ' conjugateasin (3) for a verb like hablar (to speak). (3) hablare hablaras hablara hablaremos hablareis hablaran And we aretaught that the future - tense morphemes are as shown in (4 ) .
158
Chapter10
(4) - are - aras - ara - aremos - areis - aran
Orperhaps wearetaughtthatthefutureendingis attached to aninfinitivestem, so that the actual morphemes are as shown in ( 5) .
(5) -e -as -a -emos -eis - an
(6) he has ha hemos habeis8 han In the caseof (5) thereis nearidentity with (6) . In the caseof (4) the picture that emergesis that ratherthan a single unbrokenfuture-tensemorphemewe havea modalelementandan irrealis marker' ar' , so that the actualstructureis somethinglike (7). (7) habl - ar - e habl - ar - as habl - ar - a habl - ar - emos habl - ar - eis habl - ar - an
Consequences
159
Someversionof this paradigmholds in eachof the Romancelanguagesthat hasan apparentfuture tense. (It is worth noting that manyRomance" dialects" do not havea future tenseat all.9) Now it might be objectedthat, althoughthis story tells us somethingabout the origin of the future tense, it doesn't saymuch abouthow tenseis actually representedin I -language.That is, it may be that the structureproposedin (7) haslong sincevanished, and we now representthe future as a single unstructured morpheme.As it turnsout, however,thereis goodreasonto supposethat this sort of multiple-morphemeapproachpersiststo this day. One pieceof evidenceon this scorecomesfrom EuropeanPortuguese , in which clitic pronounscanbe insertedbetweenwhatI havecharacterizedasthe modalelementandthe irrealis element. That suggeststhat thesetwo elements arein fact distinct representations andarenot merelypart of a singleunbroken future-tenserepresentation . Furthennore, in other Romancedialectsthe very sameelements(e.g. theauxiliary ' have') do not appearattachedto theverbstem at all- as in certain southernItalian and Sardiniandialectsin which, e.g., ' I will speak' can be expressedas ' Ho a parlare' , utilizing the auxiliary ' avere' (have) followed by a prepositionandan infinitive. Here we loseeventhe illusion of a genuineunbrokenfuture-tensemorpheme . this we can Following generalstrategy, proceedapacewith all the Indo. All the future tensefonDSappearto be modalin origin. I Europeanlanguages wouldmerelysuggestthatperhapstheyremainmodals(or somethingverymuch like modals). This is certainlyconsistentwith theinterpretations commonlygiven to theseelementsin spokenlanguage . Indeed, in spokenstandardItalianpresent tenseis generallynot what one usesto expressthe future. For example, if one wantsto say" I am going to th eatertomorrow," onesays(8) andnot (9). ./~ (8) Vadoal teatro domani (go- ISO-PRES to the theater tomorrow) (9) Androal teatro domani (go- ISO-FUT to the theater tomorrow) Whenthe future is used, it is most likely being usedto expresspossibility or " uncertainty,aswhenone says( 10) (which hasthe senseof It mustbe around ' " 8 o clock ). ( 10) Saranno Ie otto (be 3PL Fut eight)
160
10 Chapter
In short, whatgetscalledfuturetensenotonly lookslike modality; in manycases it actslike modality. But, onemight ask, what exactlyarethesemodalsreferringto, if not to the future? Oneideathat canbe found in the linguisticsliteratureis that modalssuchas ' will ' do not refer to the futurebut rather specifyanepistemicnotion similar to 10 , however predictability. This ideacanbeonly partiallysatisfactoryfor ourpurposes , sinceoneis temptedto takethe talk of predictionor predictability as involving predictionof thefuture, in which casewe landback wherewe started. Alternatively, we might considermodalsasreferring to dispositions, construed asreal propertiesof the world. So if I saythat ' I will leave' I am talking not aboutsomedistinct future out therebut abouta dispositionof the present world- a certainpotentiality, asit were. It might be useful to think for a momentaboutsimple, relatively primitive clocks. WhentheWickedWitch of theWestturnsoverthehourglass , theamount of sandleft on top reflectshow muchtime Dorothy hasremaining. But rather than think aboutthe remainingtime being spooledup out there in Dorothy' s future, perhapsit would be betterto think of the hourglassashaving a certain dispositionalpropertywhich reflectssomethingaboutthe temporaldisposition of Dorothy. Whenthe top of the hourglassis full of sand, Dorothy' s positionis only somewhatprecarious. As the sanddrainsinto the bottom, as the system losesits potentialenergy, Dorothy' s positionis rathermoreprecarious.It is not precariousbecauseof someeventwaiting for her out therein the future. Rather, it is precariousbecauseof the presentdispositionof eventsin the world. The hourglass merelysharescertaindispositionalpropertieswith Dorothy. . ' in thistreatmentof dispositions Cautionis necessary , of course.Wecan t turn aroundandsaythat a dispositionof this sortis somethingthat will comeabout undercertaincircumstances(asAristotle seemedto sayin places). The notion of dispositionhasto be groundedwithout appealto temporalnotions. But there is no apparentreasonto doubtthat dispositionscanbe so grounded. No Past The situationonly seemsmorecomplexwhenwe moveto thepasttense.Again, in mostnon-Indo-Europeanlanguagesthe so-calledpastis generallyjust some ? The form of aspectualmarker. Is the sametrue for Indo-Europeanlanguages caseis certainlygoodfor English, in which our so-calledpast-tensemorphemes ' ' are deadringersfor perfectaspectualmarkers. (A prime exampleis the -ed , which is takento showthat the eventin questionhasculminated.) morpheme
Consequences
Indeed,just astherearenumerousreasonsto besuspiciousof thefuturetense, thereareplenty of reasonsto be equallysuspiciousof the past. For example, it is a notoriousfact that pasttensedoesnot behavelike pasttensein counterfactuals. Consider( 11) . ( 11) If I hada million dollars. . . . This led Isard ( 1974) to speculatethat the PASTmorphemedoesnot refer to thepastbut hasa moregeneralsenseof " remote." (In thecaseof ( 11), the sense wouldbe" remotefrom reality." ) A somewhatsimilarideais pursuedby latridou ( 1996), who developsa generalnotionof " exclusion" that coversbothpossible worlds andtemporalintervals. In both casesthereis a sensethat somedeeper third elementunderliesbothtenseandcounterfactualmodality- thattensecan' t simply be a primitive elementthat refersto the past.II es help to showthat thereis some~hing more fundamental Although theseapproach than the notion of tense, they probablydon' t go far enoughfor present . Obviously, evenif an eventis remote, it is eitherremotein time purposes or possibility, andwhatwe really wantto do is avoidthenotionof temporalorder . Ideally, what we would like to sayabout (andtemporalremoteness ) altogether so-calledpast-tensemorphologyis thatit is really telling ussomethingaboutthe kind of evidencethat we currentlyhavefor our claims. On this score,we mightbenefitfrom thestudyof languages with evidentials.12 In theselanguages(whichrangefrom NativeAmericanlanguages to Bulgarian), therearemorphemeswhich havethefunctionof indicatingsomethingaboutthe sourceof the infonnationthat we havefor our claim. For example, a particular morphememightindicatethatwe havefirst-handevidencefor our claim. Another might indicatethat our evidenceis basedon second-handtestimony. It is interesting to note that in somecasesthesemorphemesare found in complementary distributionwith whateverresourcestheselanguageshavefor expressing the past, perhapssuggestingthat what we aretaking to be tensemorphemesor aspectualmarkersmight actuallybe evidentials. Hereis the idea. In the caseof modalswe haveso-called " root modals," or modalsof obligation, whichareoftentakento bethecorecasesof modality; then therearealsomoresophisticatedfonDSof modality (epistemic, metaphysical , etc.). Perhapsan analogousstory holds for evidentials. That is, perhapsevidentials indicatingwhetherthesourceof informationis experienceor testimony arethe root evidentials,andthentherearemoreabstractfonDSthat includeaspectual markers.A languagelike Bulgarian, then, which is often takento have
162
Chapter10
, wouldsimmorphemes past-tense morphemes, perfect aspect, and evidential of evidentials. ply have three kinds with currentworkontenseandonevidentiality Thisideais certainlyconsistent . Izvorski( 1997) hasarguedthatthepresentperfectin manylanguages eswhatshe to Bulgarian ) in factexpress (rangingfromTurkishto Norwegian " " in 12 the . For of callsthe perfect evidentiality example , examples ( ) all express " ." /evidentlyarrived akinto I apparently a meaning ( 12) Turkish gel -mi -im comePERFISO Bulgarian do aJ Az sam 1 be Isg,PREScome-P.PART Norwegian kommet leg har PRES come-P.PART 1 haveISO, "the is observed As Izvorskinotes(ibid., p. I ), because perfectof evidentiality relatedor geographic thatarenotall genetically in languages , it ally proximate a is case connection evidential that the is quiteunlikely simply present perfect
" of accidental syncretism. What I am suggestingis consistentwith theseclaims , but it is much more general . Why not suppose that all past-tense morphology is simply a kind of evidential ? We have already seenthat , philosophically speaking, past- tense claims really amount to claims about evidence that is currently available in the world . Why not make the obvious next step and take so- called past tense morphology to , in fact , be a kind of evidential marker? All of this is extremely sketchy, of course. My point is not that any of this is an inevitable consequenceof the A -theory, but rather that this is a possible avenue of investigation that has beenopenedup . Whether this particular avenuewill prove successful is far from clear, and I for one would be hesitant to speculate. If this general strategy for the elimination of tense does pan out , however, it appearsto leave us in a paradoxical situation. Throughout this book I have argued from the semantics of tense to the nature of time , and now I am suggesting that the grammatical category of tense can be dispensedwith altogether. But this is not a genuine paradox; I am not kicking away the ladder that got us this far. There is a real grammatical phenomenon (or class of phenomena) that we sloppily call tense and which we supposeto be connected to temporal reference.
Consequences
163
Whatwe really haveon our handsis mostlikely not a singlephenomenonbut a mixtureof modality andevidentiality. Likewise, what I havebeencalling temporal anaphoracan perhapsmore accuratelybe considereda combinationof propositionalanaphoraand implicit modal and evidentialconjunctions(e.g., when-clauseconjunctions). But whateverwe chooseto call thesegrammatical , we haveseenhow theyallow us to exorcisethe future andthe past phenomena from semanticaltheoryandconsequentlyfrom metaphysics . If we sweepawaythe grammaticalcategoryof tensein this fashion, we do " world- at not therebyend up with a " detensed leastnot in the senseadvocated by Mellor andotherB-theorists.To thecontrary, by reducingtalk of tense to evidentialityand aspectwe arefinally ableto locatethe indexicalcharacter of so-calledtemporaldiscoursein the world: it residesin the natureof evidentiality , as well as in the dispositionsand aspectualpropertiesthat furnish the world.13 Justasmetaphysicshasno room for the future andthe past, linguisticsmay haveno call for a grammaticalcategoryof tense. Still , if the grammaticalcategory of tenseis dissolvedinto morebasicandfruitful concepts,thoseconcepts will haveto preservethebasicfeaturesof theA -theory(includingthe indexical natureof so-calledtemporaldiscourse ). Whateverthe outcome, however,tense andtime will remainintimately linked. And if they areultimately dissolved, it standsto reasonthat they will be dissolvedsimultaneously , andthat the resulting semanticalconceptswill in largemeasuremirrC?r the resultingmetaphysical concepts. 10. 3 Conclusion
I hopethat the illustrations I havegiven in this chapterhaveshownjust how rich andfar-reachingtheconsequences of adoptingtheA -theorywill be. I realize that a numberof philosophersandlinguistsmay look at the emergingquestions , looseends, andphilosophicalpuzzlesasgroundsfor retreat. On the other hand, I seea newandlargelyunexploredintellectualterritory openingup before us- one without nicely pavedroads, andwith dangersandlogical conundrums at nearlyeveryturn, but also one with new ways of thinking aboutold (often stalled) philosophicalprojects,andcertainlyonein whichformerlydisconnected philosophicalpuzzlesbeginto dovetailin marvelousways. Whetherphilosophersand linguists chooseto pursuethis coursemay depend in large measureon intellectual temperament . Do we stay on the well chartedB theoryapproachin which we ignorethe obviousbig problems(like failure to accountfor indexicality ) and are rewardedwith a stable class of
164
Chapter10
tractablemini-problems? Or do we leavethatpathfor thewild andwoolly world thattheA -theoryoffersus?At this point I canspeakonly subjectively: I would preferto leavethe beatenpath to explorethe rich andcomplexconsequences of theA -theory. This is not anindictmentof theideaof a communityof scholars.To the contrary , I am hoping that otherswill join in this investigationof the Atheory in part becauseonecanseehow vastandintricatethis new world is. This book is a partial testimony to that. We havehad to move through linguistics, the , andon into investigationsof selfknowledge philosophyof mind, logic, andepistemology andmemory. In view of the breadthandcomplexityof this undertaking , it is really impossiblefor anyone personto work through the details or evenanticipatewhat the final outcomewill be. More important, this needfor a communityeffort is evidenceof theprogress thathasbeenmadein philosophy,andof howphilosophyhasoutgrownthe stage wherean individual or a small groupcansynthesizethe field andwork out the details. No longer is it possiblefor an individual with the statureof Descartes or Leibniz to singlehandedlyarguethrougha philosophicalposition. The problems havebecomeso subtle, andtheir interrelationshipsso complex, that the lonegeniusmustyield to collaborativeinterdisciplinaryresearchteams.Among other things, this tells us that the ageof giantshaspassedin philosophy; and while that is sad, perhapsit is alsofor the best.
PI Appendix Is I Language of Thought ?
the Language
In section 1.3, I briefly entertained the possibility that I -language might be the language of thought ( LOT) . Here I return to that theme and examine in more detail some arguments against the thesis that have been offered by Ray Jackendoff and by Steven Pinker. Once again I will be unable to draw strong conclusions , but perhaps it is enough to see that the thesis is not out of the question, and indeed that the extant argufnents against the thesis are very weak indeed. I will ' ' begin with Jackendoff s arguments and then move on to some of Pinker s. Jackendoff ( 1993, p . 185) argues that if natural language were the language of thought it would be a mystery why it is possible to translate from one language to another- say from Japaneseto English : The basicreasonfor keepinglanguageand meaningseparateis that pretty much anything we cansayin onelanguagecanbetranslatedinto anyother, preservingthethought that the original languageconveys.This meansthat thoughtscan' t be embalmedin the form of any single language- they mustbe neutralas to what languagethey are to be expressedin. Jackendoff concludes ( ibid ., p . 186) that a single independent thought " must be distinct from the linguistic garb in which it is clothed." Let us setasidefor the moment the question whether such meaning-preserving translations are possible. If they are possible, it is arguably becausethe LF representations of different languages (say, English and Chinese ) do not vary . Indeed , there is some speculation that the LFs for the two language markedly be identical , so it is hard to see what ice this argument is going to cut.) might Even less compelling , in my view , is Jackendoff ' s argument (ibid ., p . 187) to the effect that some thinking is not present to consciousnessbut is intuitive . It is no part of an I -language/ thought identity thesis that all I - language thoughts be conscious. Why should they be?
166
AppendixPI
Oneof Jackendoff's argumentsappearsto speakdirectlyto thepossibleidentification of I -languagewith whathecallsl- concepts . Accordingto thatargument "the ibid. . 186 structure of is ,p ( ), syntactic language built out of thingslike nouns " " ' andverbs, prepositionalphrasesandtenses , but thought isn t built out of such units thoughtconcernsthingslike objects, actions,properties,andtimes." That is, thereis supposedto be a mismatchbetweenlanguage(evenI -language ) and . I to I have confess that find the . thought argumentunpersuasive , asJackendoffsays, that languageis built out of thingslike tenses, Suppose and that thoughtconcernsthings like times. Wouldn' t our naturalsupposition be that the tensesof languageconcernthings like times? So what prevents1language(with its tenses) from being the languageof thought? Of course, we havelong knownthat thereareapparentmismatchesbetween . Forexample,Williams ( 1983) hasarguedthat syntacticandsemanticcategories in certain environmentsindefinite NPs like ' a lawyer' are not quantified expressions but predicates,as, for example, in ' Mary is a lawyer' . Supposethis story is right} All that follows is that the interpretationof an NP will depend . This is no offenseto the LOT uponthe syntacticposition in which it appears thesis, for the LOT thesisis not committedto the view that somethingof acertain syntacticcategorymust alwayshavethe sameinterpretation; it is merely committedto the view that we can statethe principlesof interpretationfor the languagein question. It is a trivial matterto havethe semanticsreflect adifference in syntacticposition. But perhapsJackendoffhassomethingmore in mind. Observingthat some nounsdenoteobjectsandothersdenoteevents,hearguesthat" objectnounslike ' ' chair' and eventnounslike ' , which are not distinguishedin syntax earthquake " , mustbe radically distinguishedin thoughts. Supposethat the thoughtsinvolving ' ' the noun chair are radically different from thoughts involving ' ' earthquake. Exactly why arethesenot radically distinguishedin the syntax? Jackendoff,afterall, proposesradicallydifferentlexicalconceptsfor thesesorts of nouns. But what is the differencebetweena lexical conceptand the sort of lexical entry commonlyassumedto occurin syntax? It hasbeensupposedsincethe earliestdaysof generativegrammarthat the lexicon and structuredlexical entriesarepartsof syntactictheory. In current ) bepartof the theory, it is evenmoreurgentthatthelexicon(robustlyconstrued e. . . ? Because in current 1995b ) syntacticforms theory( g , Chomsky syntax Why virtue of severalvery are projectedfrom richly structuredlexical entriesby simplesyntacticprinciples(greed, last resort, etc.). Without the lexicon, there
Is I -Languagethe Languageof Thought?
167
simply areno syntacticforms. Nor is theremuchpoint in sayingthat the conceptual of a nounor a verbis morerichly structuredthanthecorresponding representation representationthat we get in the linguistic lexicon. It is not clear " that arenot relevant that thereareany featuresof " conceptualrepresentation to thederivationof structurein the syntax. But in the secondplace, supposethat therewas somepropertyfoo that a given conceptualrepresentationC hadbut which thecolTesponding lex.ical entryL did not requirefor thederivationoflinguisticforms in the syntax. Is thereanyreasonto saythat L couldn't or shouldn 't containthe additionalinformation? That is, is thereany reasonwhy L and C mustbe distinct? Jackendoff( 1992) appearsto think thattherichly structured" conceptualrepresentations " that he " " gives are not part of straight syntax. For example(p. " 30), we should be clear. . . that [conceptualstructure] is as different from " " straight syntax (the grammarof NPs, V Ps, etc.) as straight syntax is from " " " phonology. But what is straightsyntax ? In currentlinguistic theorywe have two levelsof representation : PFandLF. Accordingly, if " straightsyntax" means . If PFis partof straight anything, it mustreferto thesetwo levelof representation then syntax, why is straightsyntaxdistinctfrom phonology? Likewise, is there any non-question-beggingway to arguethat LF (which alreadyencodesscope distinctionsandat leastasmuchlexicalstructureasis necessary to projectphrase structure) cannotencodethe informationthat Jackendoffattributesto conceptual structure? I can' t imaginewhat suchan argumentwould look like. So muchfor Jackendoff's arguments . The argumentscataloguedby Pinker ( 1994, p. 78ff.) fare no better. Let me considerthoseargumentsin turn before offering criticism. First, Pinkernotesthat thoughtsare ambiguous, and arguesthat linguistic forms supposedlyarenot. He givesseveralamusingnewspaperheadlines(e.g., ' StudTiresOut' as ) examplesof this phenomenon , but other sortsof examples from written text arecertainlylegion. .Second,Pinkersuggeststhat naturallanguagerepresentations lack " logical " explicitness, meaningthat certaininferencesthat intuitively makesenseto us do not follow on the basisof naturallanguageform: " Englishsentences do not the information that a needs to out common ." sense embody processor carry Third, Pinkerarguesthat naturallanguageis unableto accountfor coreference. Wemight introducea subjectin our discourseas" the tall blond manwith oneblack shoe" but subsequentlyrefer to the individual as " him." How do we know that we aretalking aboutthe sameindividual? Allegedly, nothing in the Englishlanguagetells us.
168
AppendixPI
Fourth, Pinkerarguesthat naturallanguageis unableto handleconversation ' ' ' ' ' ' ' ' specificwords suchas I , you , here , and now , sincetheir meaningsvary from contextto context. Supposedlythesecasesdo not posea problemfor a distinct languageof thought. Fifth andfinally, Pinkerarguesthatnaturallanguages , unlikethelanguageof fail to account for . Thus , , thought synonymy they allegedlycannotaccountfor 'John ' and ' John ' the wall with why sprayed paint sprayedpaint onto the wall meanthe samething. Pinker' s conclusionis that " peopledo not think in English or Chineseor " Apache;theythink in a languageof thought. If EnglishandChineseandApache are construedasE-languages , I do not disputethe conclusion(thoughI doubt theseargumentssupportthat conclusion3 ). On the otherhand, if we taketalk of theselanguagesto be standingin for talk of variouspossibleI -languagestates, the argumentsfail miserably. We havealreadyseenthat I -languagerepresentations arerichly structured. The structureof LF representations relevant whenwe consider seemsparticularly ' Pinker s line of reasoning. Considerthe argumentfrom ambiguity. The ' ' inscription Every man loves somewoman hastwo distinct LF representations , which intuitively encodetherelevantscopeambiguity. But of courseother kinds of ambiguitiesare structurally encodedtoo - a famousexamplebeing ' ' ' ' Chomskys Visiting relativescan be boring . Is the expression visiting rela' tives gerundiveor a nounphrase? Clearly that makesall the difference. Thereis no barrier to our sayingthat LF representations alsoreflect lexical ' Johnwentto the bank' there . If I be several , say may possibleinterpretations ambiguity of what I say, dependingupon which lexical entry for ' bank' is inserted . I assumethatin the I -languagerepresentation thereis a fact aboutwhich lexical entry hasbeeninserted. Likewise, Pinker' s example' StudTires Out' reflectsa combinationof lexical andstructuralambiguity. Is theLF representation structuredas( 1), or as(2)? ( 1) [NPStudtires] [(are) out] (2) [NPStud][tires out] Again, I assumethat thesetwo meaningscorrespondto two distinct LF representations . . Similarconsiderations applyto thelogicalexplicitnessof LF representations in an AI than offered LF are no less anything explicit Typical representations
Is I -Languagethe Languageof Thought?
169
inferenceengine, andthereis certainlyno barrierto defining logical inference rulesdirectly for LF representations .4 WouldLF representations fail to accountfor pronominalcoreference?Again it is hardto seewhy. Indeed, much of the most interestingwork indiscoursehas assumed representation someform of the LF hypothesisandhasdefined theory the relevantanaphoricrelationsfor LF representations . Likewise for the contextdependencyof indexicalsandlike expressions . Finally , it is far from clear why two structurally diverse syntactic forms can' t havethe samemeaning. Henceit is unclear why synonymypresentsan argumentagainstthe identification of I -language(or evenE-language) with the languageof thought. It is entirelypossiblethat thereshouldbe two distinct LOT representations with thesameinterpretation , sowhy shouldn't therebedistinct I -languagerepresentations with the sameinterpretation? Now, it may well be that Pinkerdoesnot intendtheseargumentsto cut againstthe candidacyof I -languagesfor the languageof thought. Onething is clear, however. If so intended , the argumentsfail miserablys
Appendix P2 Language / World } Isomorphism ?
In chapter4 I arguedfor a strong connection betweensemanticsand metaphysics, but I also hinted that there might be the possibility of an even stronger relation one that takes the structure of language and the structure of the world to be iso' morphic to each other. Let s call this thesis the language/world isomorphism ( LWI) hypothesis. The LWI label is a little bit crude, since the key idea is that there is an iso morphism holding between logical forms and the world . The basic idea is not restricted to the analytic tradition in philosophy . It has been at the root of the " semantic tradition " (in the senseof Coffa 1991) from Kant , through Bolzano and the early Wittgenstein , up to the present. As early asKant we find some suggestion that there is an isomorphism between a representation and the thing represented: [Representation ] is that determinationof the spirit (Bestimmungder Seele) thatrefersto other things. What I call referring (Beziehen ) is when its featuresconfonn to thoseof the externalthings! [ Therepresentation ] is composedout of its componentconceptsin thesamewayin which theentirerepresented thing is composedout of its parts. Justas, for example,onecansay that the notesof a musicalpiecearea representation of the hannonicconnectionof the notes, not becauseeachnoteis similar to eachtonebut becausethe notesareconnected to eachotherjust asthe tonesthemselves .3 The most celebrated version of this general idea is surely the picture theory of meaning advanced by Wittgenstein in the Tractatus. On the view articulated there, just as " what a picture must have in common with reality , in order to be " able to depict it- correctly or incorrectly - in the way it does, is its pictoral form 1961 b 2.17 a , ( ), proposition must be isomorphic in structure to a state of affairs in order to represent or be about that state of affairs.
172
AppendixP2
" " Of course,we arenottalkingaboutthe surfacefonn of a proposition(whatever " " that might be). We aretalking aboutthe proposition under analysis- in other words, we are talking about its logical fonn. For Wittgenstein, only if onewereto getdownto thecompleteanalysisof propositionwould it reflectthe structureof the world.4 ' Chomsky s Arguments against LWI
In Chomsky's recentwritings containa numberof argumentsdesignedto show that if we wereto adopta referentialsemanticsthe kind of ontologywe would predictdoesnot appearto track our intuitions aboutthe kinds of thingswe are really talking about. Suchtheoriesallegedlycommitus to thingsthat we would neveracknowledgeasexisting. Theyalsoallegedlycommitusto typesof things which aredifferent from the typesof thingsthat we ordinarily supposewe are talking about. And thethingswe talk aboutareallegedlytoo unruly for suchtheories to handle. To havelabels for all thesepossibilities, let us call them the " " " " , and , the type-mismatchargument argumentfrom implausiblecommitments ' " " -objectargument. I will reviewall threeof Chomskys arguments the misbehaving beforeoffering what I taketo be the mostnaturalreply to them. The Argument from Implausible Commitments Chomsky ( 1981, p . 324) draws attention to the fact that a referential semantics (apparently in conjunction with the LWI hypothesis) commits us to some apparently implausible entities (flaws for example) and suggests that we really ' t take can seriously a theory that commits us to such entities: " ' If I say"the flaw in the argumentis obvious, but it escapedJohn s attention, I am not committedto the absurdview that amongthings in the world areflaws, oneof them in , the NP the flaw in the ar2UInent behavesin all the argumentin question. Nevertheless relevantrespectsin themannerof thetmly referentialexpressionthecoatin thecloset. . . . Pursuing a similar line of attack, Hornstein ( 1984, p . 58) has drawn attention to constructions like ( 1) .
( 1) abouthisweight. manis concerned Theaverage " Hornsteincontendsthat no onewishesto arguethat thereareobjectsthat are " averagemenin anymeaningfulsense. ' ' The generalconcernintroducedby exampleslike flaw in the argument ' and ' the averageman is that the doctrinethat theremustbe a tight connection
Language/ WorldIsomorphism?
173
betweensemanticsandontologyis suspect , for it commitsus to apparentlyabsurd views. This argumentneedsto be fixed up a bit beforeit is fully functional. As it stands,the argumentmakesthe assumptionthat the logical form of theseconstructions is transparent - e .g., that the expression' the flaw ' really is nothing morethana simpleNP containinga determineranda noun. This is not suchan innocentassumption . After all, theLWI hypothesisdoesnot hold thatthereis an between surfacelinguisticform andtheworld, butratherthatthere isomorphism is an isomorphismbetweensome ultimate logical form and the world. As Higginbotham( 1985) hasargued,beforeadmittinga commitmentto objectslike flaws andaveragemen we would haveto look moreclosely at the underlying . logical form of theseconstructions ' flaw in the ' ' ' that Higginbothamsuggests argumentis parallelto badsinger . ' ' The semanticsof bad singer is not suchthat we saysomeoneis a bad singer iff he is badandis a singer. Rather, we understandthat someoneis a singerand that his singing is bad. Likewise, we shouldnot say that ' that is a flaw in the ' argumentis true iff thatis a flaw andthat is in the argument.Rather, we should saythatthe phraseis true iff somethingis anargumentandit is flawed. Despite ' ' , flaw may not be a nominal; it may be a modifier like 'bad' . appearances Higginbothamoffersa similaranalysisfor thecaseof 'theaverageman' . First, he notesthat the expressionis actuallyambiguousbetweenthe caseof an individual who hastypical propertiesandthe sensein which we saythat the average family has2.3 children. It is presumablythe lattersensethat is problematic ' here. In this latter case, Higginbothamsuggeststhat ' average , despiteappearing to be an adjective, is functioningasa kind of adverbial. The constructionis parallelto exampleslike (2), discussedby HaiK( 1983). (2) Let' s havea quick cup of coffee. Clearly, in (2) we arenot suggestingthattherearecupsof coffeethatarein some sensequick. Rather, ' quick' is behavingas an adverbial, modifying the activity . Higginbothamproposesthat in a constructionlike (3) the analysiswill be ' somethingalongthe linesof (3 ). (3) The averagefamily has2.3 children. ' (3 ) On average , a family has2.3 children.
174
AppendixP2
more thesomewhat Chomsky mightreplyin a numberof wayshereisConsider 4 . sentence ( ) complex (4) 6 Yourreportontheaverage familyfailsto makeit clearthatit has2.3 children. canbeworkedout, theywill technicalspellouts Evenif suitableglosses , of of their own . But the difficulties course , andthesespellouts glosses mayintroduce innocent . As Chomsky(personalcommunication arenotnecessarily themselves like (5)? for examples ) asks,areweto continuewith thestrategy (5) Thathisincomeis fallingbothersJohnDoe. ' man', which Thatis, arewetotakeJohnDoetobestanding proxyfor anaverage ' ' ? asakindof adverbial with averageoperating in (5) is to befurtherunpacked thesisabout averystrongandpossiblycounterintuitive Thisbeginstoconstitute . thelogicalformof theseconstructions ' therearelikewise to HigginIn thecaseof 'flawin theargument , responses communication . Forexample bothamavailable ) asks,how , Chomsky (personal ' ' like (6)? in anexample couldonetreat flaw asakindof predicate (6) ourefforts. Wefixedthreeof theflawsyoufoundbuttherestof themresisted Onepossibleansweris to saythat(6) canbeglossedasin (6'), whereit is the thatareintuitivelyflawed. stepsin theargument ' (6 ) our Wefixedthreeof theflawedstepsyoufoundbuttherestof themresisted efforts. Againglossesof this naturehaveto bejustifiedat somepoint, andthereis a hiddenstructure heavyburdento showthattheintroductionof thisproposed about the we know of what with the rest well syntacticformof these comports shown that it has to be more . Still , constructions , generalrulesareavailable here course theHigginbotham topursue notonlycaseby casefixes.Anyattempt . will notbea trivialexercise The Type-Mismatch Argument Even if we are comfortablewith potentially counterintuitiveentities suchas flaws, Chomsky( 1995a) notesthat thereis an apparentmismatchbetweenthe type individuation that objects and substancesintuitively haveand the type
Language/ WorldIsomorphism?
175
individuationthata referentialsemanticswill provide. To getclearon the issue, a tenninologicaldistinction will be helpful. Supposewe distinguishan I -substance from a P-substance , wherea P-substance(if thereis sucha thing) is the sort of stuff that would playa role in physicaltheory (H2O, for example) and an I -substance(if thereis such) is the stuff that we areintuitively talking about whenwe uselanguage(the intuitive referentof ' water' ). To put the point in a moretheory-neutralway, theI -substance is whatit appearswe aretalking about baseduponour useof language . The allegedproblemis that P-substances and ' I -substances just don t matchup right. If , followingPutnam( 1975), 'water' refersto ~ O, thena referentialsemantics will assigna P-substance(H2O) asthe semanticvalueof ' water' . But the problem is thatthestuffwe areactuallytalkingaboutwhenwe usetheterm ' water'- is somethingelse altogether. To seethis, considerthe fact the I -substance that what we find in the HudsonRiver is called ' water' thoughit could hardly be consideredH2O. Also problematicis the fact that thereare substances (e.g., icedtea) thatchemicallyapproximate~ O muchmorecloselythanHudsonRiver waterdoes, yet we don' t call them ' water' . According to Chomsky, the situation is evenmoreproblematicthanthis. If someoneat the watercompanywere to pour tealeavesinto the system, so that what cameout of the tap waschemically identicalto Lipton Iced Tea, we would still call it ' water' - aI thoughwe might complainaboutits impurity. Thus, what we are talking about when we use the term ' water' - the 1substance - dependsupon the social settingin which we find that substance . But accordingto referentialsemantics , the meaningof the term is supposedto dependuponthe chemicalcompositionof the substancereferredto - it is supposed to be a P-substance . Conclusion: Referentialsemantics(if respectingthe LWI hypothesis) will not track the intuitive notion of meaning. Onemight think it possibleto getoff thehookby appealingto socialtheories of external(referential) content(in the senseof Burge 1979). RatherthanP-substances : substances , we might posit S-substances that areindividuatedaccording to communitynorms. Thus, while my conceptof watermightnot accordwith H2O, it might still accordwith a certainsociallydetenninedobjectthat hasthe propertyof beingwaterwhenit comesfrom the faucetbut not whenit is served at a restaurant . Theproblemhereis thatthereareplentyof exampleswheresuch S substances (if therecouldbe suchthings) would not trackour intuitionsabout the extensionof terms- Burge' s own examplesabouttharthritis and brisket are casesin point. That is, there is a mismatchbetweenI -substancesand Ssubstances . We will comebackto this point a bit later.
176
P2 Appendix
The Misbehaving - Object Argument There are a number of interesting features to the water/tea story, one of which can be broken out as a separateobjection to referential semantics. ' We have already seenin the type- mismatch argument that I - substancesdon t track P- substances, but there is another problem . The water/tea story also seems ' ' to show that the I - substancewe are talking about when we use the term water is a most ill -behaved sort of substance. Something may ceaseto be water even if no internal physical changeshave taken place. For example, the same chemical compound is water when it comes from the tap but ceasesto be water when it is served at a restaurant. If that is the intuitive character of I - substances, then " " there is really little hope that referential semantics can give the reference of ' ' ' ' what we talk about when we talk about water and tea , since referential semantics is supposedly going to say that the content of these terms is ~ O in the first case and ~ O plus certain other elements in the latter. That is , I - substances are so unruly that it is wildly implausible to suppose that they could have any counterparts in the physical world . Hence, they have no counterparts that a referential semantics could utilize as their referents. Chomsky ( 1975, p. 203) makesa related if somewhatmore generalpoint when notes that the very notion of whether we are talking about a single object or a collection of objects turns on any number of social and institutional factors: We do not regarda herdof cattleasa physicalobject, but ratherasa collection, though there would be no logical incoherencein the notion of a scatteredobject, as Quine, Goodman, and othershavemadeclear. But even spatiotemporalcontiguity doesnot sufficeasa generalcondition. Onewing of anairplaneis anobject, but its left half, though equallycontinuous, is not. . . . FurtherDiore, scatteredentitiescan be takento be single : considera picketfencewith breaks, or a Calder physicalobjectsundersomeconditions mobile. The latter is a "thing ," whereasa collection of leaveson a tree is not. The reason , is that the mobile is createdby an act of humanwill . If this is correct, , apparently thenbeliefs abouthumanwill and actionand intentionplaya crucial role indetermining eventhe mostsimpleandelementaryof concepts. Moving that discussion into the current debate, we might say that it is implausible for even such simple semantic concepts as object and collection to correspond in any interesting sensewith P- substances. Analysis of the Arguments : Some Replies
The conclusionthat we candraw from exampleslike the aboveis that anyreferential semanticspurportingto respectthe LWI hypothesisis going to misfire ' since it is boundto utilize asreferentsP-substances , whichjust don t track badly . the intuitive meaningsof naturallanguageexpressions
Language/ WorldIsomorphism?
177
Onereply for the semanticistis to simply reject the LWI hypothesis.That is, onecould say: Yes, a semantictheoryis concernedaboutthe connectionbetween languageand the world (aspart of a four-placerelation also involving speakerandcontext), butthis saysnothingabouttherebeinganisomorphicmapping betweenprimitive linguistic expressions(underanalysis) and things (or kinds of things) in the world. If thereis a singlelexical entry correspondingto ' water' it , doesnot maponto any singlesubstance ; rather, it mapsonto various things, dependingon circumstances , discourseparticipants, etc. Suchtheories areno lessreferentialthanthosethat respectLWI. After all, they arestill in the businessof articulating language/world relations, albeit througha very complicated story, anddespitethe fact that theserelationsare often dynamiconeto-manyrelations. Is thereanotherway out? ? That is, why not bite the metaphysical Why not give up the P- and S-substances bulletandacknowledge thattherereallyareI -substances - thatis, things like flaws. They are clearly not logically absurdentities, and it neednot be concededthattheyareparticularlyoddentities. In thecaseof flaws, at least, one mightsaythattheyarealtogethercommonin theargumentsonerunsacross,and onemight wonderwhy they shouldbe consideredanylessreal than, , tables say andchairs. Likewise, coatshangingin the closetneednot have any particular ontological priority over averagefamilies andflaws. But is this a metaphysicalbullet that we canbite? Can our ontology admit suchthings? Here (finally) we cometo the issueof " aspectsof the world" and the kinds of thingsthat we canbe realistsabout. Aspect ~ of theWorld ' Operating throughout all three of Chomsky s arguments cited above is the assumption ( shared with Putnam and others) that P- substancesare the kinds of substancesthat a referential semantics is going to favor. For example, Chomsky ( 1993) remarks: To be an Intentional Realist, it would seem, is aboutas plausibleas being a Desk- or - or Cat- or Matter-Realist; not thatthereareno such Sound-of-Language thingsasdesks, etc., but that in the domainwherequestionsof realismarisein a seriousway, in the context of the searchfor laws of nature, objectsare not conceivedfrom the peculiar perspectives providedby the conceptsof commonsense. But why should we make the assumption that " the domain where questions of realism arise in a serious way " is " in the context of the searchfor laws of nature "? One possible answer would be that physical theory gets to say what is real, and that hence if we are to have a genuine referential semantics in which
178
AppendixP2
" the referentsare" real existingentitiesthenwe arestuckwith the kinds of entities andsubstances positedby physicaltheory. But this answermakesa strong realism- an assumptionthat is controversialto scientific about assumption most and likely falsein my view. saythe least, A greatdealof literaturein thephilosophyof science(e.g., vanFraassen1980) holdsthat the entitiespositedby sciencedo not existin the samesenseasmidsize earthboundobjectslike tablesandchairs. Pursuingthis line of thinking, we might saythat scientifictheories, despitetheir greatinterestandutility , are not the arbitersof what is real. If we setasidethe exclusiveclaim of the physicalscienceson our ontology i .e. ( , if we dismiss P- substancesas our semanticvalues), then we may well areentirely plausiblecandidatesfor the referentsof a semantic find that I -substances ' ' theory. For example, it may be that the semanticvalueof water just is water- the complicatedI -substancethat movesthroughpipesinto our homes. If we takethat route, then it is far from clearthat the LWI hypothesismustbe : surrendered or perhapsit is entirelyparasitic is a cheatiS But perhapstheappealto I -substances . What is an I -substanceif not on the notion of I languagerepresentations " " whatever to a particularI languagerepresentation? Seen corresponds simply ' ' in this light, isn t the talk of I -substancevacuous? At best, isn t it a misleading way of talking aboutI -language? Herewe needto headoff a confusion.Evenif it shouldturnoutthatI -language arein somesenseprior to I -substances (perhapslogically prior representations our that not follow it does or metaphysicallyprior), pathasempiricalinvestigators The . will beginwith the linguistic representations ideais that the linguistic will indeedunderwriteour metaphysicalintuitions, but that because representations of this we canexpectour metaphysicalintuitions to shedsomelight on ' ' . A concreteexample:Thelexicalentryfor water , when thenatureof I -language (at least, if it fully fleshedout, is boundto be a very complicatedrepresentation ). Our knowledge is to interactin complexwayswith otherI languagerepresentations of this fact doesnot comefrom directinvestigationof thelexicon; rather, it intuitionsaboutwaterlike thosethatChomskyevinces is guidedby metaphysical in his articles- intuitionsthattakeinto accountrich contextualinformationabout whetherthe materialin questionis comingfrom a faucetor whetherit is being . servedin a restaurant ' It will not do to arguethatChomskys waterintuitionsarenot metaphysical " " that they areonly abouthis I -concept of water, or aboutthe phenomenology of water. If thereis no world beyondtheseconceptsor beyondthe phenomenology, then oncewe havemadethe stepfrom talking aboutlinguistic repre
Language/ WorldIsomorphism?
179
sentationsto talk of concepts , or phenomena , or sensedata, we havebasically steppedinto the world- or asfar into it aswe areevergoing to get.9 Of course, this picture is as Kantianasit canbe. We have metaphysicalintuitions , and we want to know what underwritesthoseintuitions. The first departure from Kant lies in the answergiven- not "the categoriesof reason" but ratherthe structureof I -language.The seconddeparturelies in the fact that our approachneednot be entirely transcendental . We do havesubstantialindependent of the knowledge languagefaculty, andwe canusethatknowledgeto gain insight into the natureof reality.
Tl Appendix -A BasicQuantiticational Fragment
" is in fact an Following Chomsky( 1995b), supposethat a " sentence ordered pair of representations(PF, LF) , wherePF representationsinterfacewith the " " perceptual-articulatory componentand LF representationsinterface with " the conceptual-intensional" system(or, as I would prefer to put it , LF is the level of representationthat is visible to the semantictheory). Sinceour principal interestwill be with LF representations , a few wordsare in orderabouttheir nature(for presentpurposeswe can skip over their derivation ). I will assume , following work datingbackfour decadesnow, thattheform of a linguistic representation canbe characterizedasa branchingtreestructure like the following.'
A " " " " " ' " ' " " ' ~ " BE " " / " ' / ,/ ""D F "C / "//"/,G ~ " ' /H / ""I
In this particularcasewe havea binary branchingstructure, but thereis an empirical questionas to whetherother forms are possiblein natural language.2 The tenninalnodesof sucha structurewill bethelexical items, andthebranching nodeswill belabelsfor grammaticalcategoriesS (clause), NP (nounphrase), VP (verbphrase), PP(prepositionalphrase), andsoon. Of particularinterestto us are the geometricrelationsthat hold betweenthe nodesof suchstructures,
AppendixTl
182
andtherole thatthoserelationsplay in allowingus to formally characterizethe notion of scopein naturallanguage! Recall, first, that in formal logics suchasthe propositionalcalculuswe can ' ' utilize parenthesesas scopemarkers. For example, in - (AvB )& (CvD ) we ' ' understandthat the negationis to havescopeover the first conjunct ( A v B ) but not the second.Many philosophershavedespaired(andsomehaverejoiced) . In fact, however,very thatmattersareapparentlynot sotidy in naturallanguage utilized kinds of for the representations tight definitionsof scopearepossible in linguistic theory. Let us saythat a nodeA dominatesB iff thereis a path moving uniformly downthe treefrom A to B. So, for example, in thediagramabove, A dominates all othernodes, andB dominatesonly C andD. Following Reinhart( 1976), we can also define a relation of c-commandasfollows: Xc -commandsY iff neither of X andY dominatesthe other andeverybranchingnodedominatingX alsodominatesY. For example, accordingto this definition, B will c-command I (becausethe first branchingnodedominatingB , that is A , alsodominatesI ), but C will not c-commandI . C-commandis a relativelysimplegeometricalrelation on phrase-structurerepresentations , but it is alsoa very goodcandidate for specifyingscoperelationsin naturallanguage.That is, we can assumefor .4 purposesof this book that c-commandjust is SCOpe Wecanfurtherassume , following work datingbackto Chomsky( 1976) and ) by movementof May ( 1977), that LF representationsare generated(in part " surfaceform" shown with the sentence a For . quantifiedexpressions example, ' shownin (2), where' e is a co-indexed in ( 1) will havethe ( LF) representation traceof movement(later we will seehow thesetracescanbe treatedasbound variables).S
)loves J(1 ohn everyo " ' /John " , VP NP / " '"NP /V , /loves everyone
s " " ~ " " ' -(eNP S " / / ' / " NP )jvery vp A Joh V NP lov ej
A BasicQuantificationalFragment
183
Wecanalso" linearize" representations like (2) asin (3).
(3) John][vp[y loves][NP s [NP ] j s [NP everyone eJ]]] It is often supposedthat certainambiguitiesin naturallanguagereflect multiple LF representations for a givenutterance.Thus, an utteranceof (4) will have two differentpossibleLF representations , reflectingtwo possiblescoperelations that might hold betweenthe nounphrases . (4) Every manlovessomewoman
s""" "" " " " " " "" """"" " " --..-VP .. NP //A 6 / """,NP V every ~ loves6 some wo ~
Tl Appendix
184 ' (4 )
S -----""" " """ "' """'-"""" -......... ........S NP / "" " / ' / / " " " / 6 S NP every ~ "" / "" / 6 VP NP Worn some ~ A V
NP
loves
I ej
eI.
(4")
~S....-.. """ --, , , ....... -....---, , ! s-" NP' / / " ' / " " / / 6 S NP Worn some ~ /' 6 every ~
/ NP eI.
"""" VP A V
NP
loves ej between thedistinction 2.2incorporated insection introduced predicates Thefragment Ludlow and Larson . Here ) (1993 andreferring , following expressions first-order basic thatfragment andSegal andLarson byadding ),weexpand (1995 . noun formore resources basic phrases complex , aswellassome quantification
A BasicQuantificationalFragment
185
Syntaxof Ll 8 - + 8 I and/ or 82 8 - + it is not the case that 8 8 - + NPVP VP - + V2 NP VP - + VI VI - + ~ ,~ V2 - + ~ , ~ NP - + eve~ N NP - + ~ N NP - + Smith , ~ N - + mDn, woman , ~ , dQ& In addition to these simple phrase- structure rules, we can assume(following our discussion above) that there is a rule that maps the product of these rules onto
otherphrasemarkers(whichwill becalledLF representations ). Thisrule - which, " " following Chomsky( 1976) andMay ( 1977, 1985), we cancall QR , for quantifier raising- statesthat a quantifiednounphrasecan be Chomsky-adjoined to S, leavingbehinda co-indexedtrace. TheresultingLF representation is analogous to a restrictedquantifierform of first-orderlogic. As wasnotedabove,QR will alsoprovide a way of representingcertainscopeambiguities. (Recall the ' ' ' " LF representations givenfor Every manlovessomewoman in (4 ) and(4 ).) But how do thesetwo representations , giveriseto two distinctinterpretations andpreciselywhatis therelationbetweenthequantifiednounphrasesandtheircoindexedtraces? Forthatmatter,preciselyhowcanc-commandfunctionasscope? The shortanswerto thesequestionsis that the valuationpredicateVal is defined in sucha way that the co-indexedtracesaretreatedasboundvariablesin the sensefamiliar from first-order logic, andthe compositionalaxiomsof the T-theoryensurethatc-commandrelationswill reflectscoperelations. Thebasic ideais to incorporateTarski' s idea of sequences into the axiomsof our T-theory . Specifically, we can introducethe predicateVal(A , B, 0 ) to be readas"A is the valueof B with respectto assignment0 ," where0 is an assignmentof valuesto all the variablesof the language . If we think of thevariablesasbeingof theform Xi' wherei is a numericalsubscript , then a sequence0 will assignan object to eachvariable Xi. This idea will bereflectedin our basicaxiomfor theinterpretationof traceasin (5), where o (i ) is the ith positionof the sequence .
(5) Val(x, ~, 0) iff x = o(i ) for i ~ 1.
186
Tl Appendix
The crucial axiomsfor the interpretationof the quantifiersare as shownin 0" differs from 0' at mostin the ith (6), where0" =j 0' indicatesthat the sequence position. (6) a. Val( T, [s [NPI N ], 51 ], 0' ) iff for every0" =j 0' suchthatVal(O"(i ), N , 0' ), ~ Val( T, 51, 0" ) b. Val( T, [s [NPI .sQme . N ], 51 ], s) iff for some0" =j 0' suchthatVal(O"(i ), N , 0' ), Val( T, 51, 0" ) c. Val( T, [sNPj 51 ], 0 ) iff for 0" =j 0' suchthatVal(O"(i ), NP, 0' ), Val(t , 51, 0" ) With the axiomsandrulesintroducedthusfar, the theoremderivedfor ' Every ' dog barks would be asshownin (7). (7) Val( T, [s [NP eve~ [NdQi] 1 [s [NP eJ [vp[VIhmJU ]) iff for every0" =j 0' such that O"(i ) is a dog, O"(i ) barks It is alsopossibleto incorporategeneralizedquantifiertheoryinto this framework (seechapter8 of Larsonand5egal I995); however, for currentpurposes we canmakedo with thetwo standardfirst-orderquantifiers' every' and' some'. We arenow readyto seehow the c-commandrelationcashesout the notion of scope.No stipulationis necessary ; asLarsonand5egal(ibid., p. 252ff .) note, the situationis exactlyparallel to the way scopeandquantifier/variablebinding arerelatedin the standardpredicatecalculus. As an illustration, considerthe sort of structurethat would arisein the multiply ' ' quantified sentence every man loves somewoman . Abstracting from detail, the structurewill be asfollows:
S / ' / / " " , * NP S 'det /"",Nt
Given the way the axioms are defined for this structure , the interpretation of S* will be relativized to the sequenceintroduced in the resolution of the S node.
A BasicQuantificationalFragment
187
Thus, LarsonandSegaloffer thefollowingcontrasting theorems , reflecting thedifferentinterpretations of thetwoLFs: (8) Val(T, ~ [NP [NmBnJ1 [s' [NP ~ ~ [Nwoman ] ] k [NP eJ [vp[V2~ ] [NP ej] ll ]) iff for every0" =j 0' suchthatO"(i ) is a man, for some0'" =j 0" suchthat O"' (j ) is a woman , O"' (i ) lovesO"' (j ) (9) Val( T, [s[NP eve~ [NmBn] ] [So ~ [Nwomanllj[s' [NP [NP ~ [NP eJ [vp[V21 ej] = iff for some 0 " 0 ' such that i O " ll ]) ( ) is a woman , for every0'" =j 0" suchthat j O"' (j) is a man, O"' (i) lovesO "' (j ) In thewordsof LarsonandSegal(ibid., p. 253): "Thecorrelation [between scope andc-command of thewayin whichthesemantics usesthe ] is asimpleoutcome object-language , Corn syntaxto constructtheT sentence positionally building ' ' partsof thelatterusingthe skeletonprovidedby theformer." Semantics of Ll
I nowintroduce a three-placeValpredicate Val(A, B, C), to bereadas"A is the value of B with respect to assignment
TerminalNodes
Val(x, Smith, 0) iff x = Smith Val(x, ~ , 0) iff x = Jones Val(x, ~, 0) iff x = o(i ) for i ~ 1 b. Val(x, dQg,0) iff x is a dog Val(x, ~ , 0) iff x is a cat Val(x, ~ , 0) iff x is a man Val(x, woman , 0) iff x is a woman c. Val(x, bm , 0) iff x barks Val(x, ~ , 0) iff x walks d. Val x,y), ~ , 0) iff x seesy Val x,y),,~ , 0) iff x likesy
188
AppendixTl
Nontenuinal Nodes
a. Val( T, sNP VP], a ) iff for somex, Val(x , NP, a ) andVal(x , VP, a ) b. Val(x, [vpV NP], a ) iff for somey, Val((x,y), V, a ) andVal(y, NP, a ) c. Val(x, ra~], a ) iffVal (x, ~, a ) (wherea rangesovercategories , and ~ ranges overcategoriesandlexical items) (3) a. Val( T, sSI ~ S2], a ) iff it is both the casethatVal( T, SI , a ) andVal( T, S2, a ) b. Val( T, [sSI QI S2 ], a ) iff eitherVal( T, SI , a ) orVal(T, S2, a ) c. Val( T, [sit is not the casethat SI ], a ) iff it is not the casethatVal(T, SI , a ) (4) a. ' ' Val( T, [s [NPi evea N], SI ], a ) iff for everya =j a suchthatVal(a (i ), N , a ), ' Val( T, SI , a ) b. Val( T, s [NPi ~ N] , SI ], s) iff for somea ' =j a suchthatVal(a '(i ), N, a ), Val( T, SI , a ') c. Val( T, sNPj SI ], a ) iff for a ' =j a suchthatVal(a '(i ), NP, a ), Val(T, SI , a ') Definitions I. For any sequencea , a (i ) is the ith elementofa . ii . For any sequences a anda ', a ' =j a iff a ' differs from a at moston a '(i ) . iii . Val(T, S) iffVal (T, S, a ) for all sequences 0' . Finally, let us againmakeuseof two derivationrulesthat wereintroducedwith the languageL .
A BasicQuantificationalFragment
189
DerivationRules (SoE) ...a ...
aiff (i therefore
. . . p. . .
(Soi)
iff for some x , x = a and . . . x . . .
therefore cI>iff . . . a . . . Recallthat thesearenot logical rules, andthat the stepsin eachderivationare muchmoretightly constrainedthanthey would be if the full resourcesof logic wereat our disposal. Theseare simply rulesfor derivingT-theorems. This fragmentis significantlymoreinvolvedthanthelanguageL introducedearlier , andit might be usefulto work througha derivationin orderto more fully atplay. Considera derivationfor the sentence' Everydog graspthemechanisms ' barks . Recallthat the truth definition will apply to a structuraldescriptionof a sentence : rs[NP , andin thiscasewill applyto theLF of thesentence [NdQ&] 1 ~ . The derivation will as [S[NP ej] [vp[VIbmJ ]]] proceed follows:
[s[NP ~ [N.dQ &]l s[NP eJ[vp[VIbmJ ])] is trueiff (1) forevery a' =j 0' suchthatVal(O " (i), [NdQgl , 0'),Val(T, fs[NP eJ[vp[VIb~ ]]], a' ) of (4a)] [instance (2) forevery0" =j 0' suchthatVal(O " (i), [NdQ x, Val(x, [NP &} , 0'), forsome eJ, 0")
andVal(x, [vp[VIhm] ], 0') [from( 1), by (2a), SoB] (3) for every0" =j 0' suchthatVa1 " (i), dQi, 0'), for somex, Va1 (O (x, ej, 0" ) and Val(x, bm , 0" ) [from(2), by (2c), SoE] (4) for every0" =j 0' suchthatO " (i) is a dog, for somex, x = O " (i ) andx barks from 3 1 a 1 b 1 c E So [ ( ), by ( ), ( ), ( ), ] (5) for every0" =j 0' suchthatO " (i ) is a dog, O " (i) barks from 4 [ ( ) by Soi)
Appendix T2 A Quantificational Fragment with Events (from Larson and Sega11995)
Following a proposaldue to Davidson( 1967b), we can hold that there is an implicit quantificationover eventsin English action sentenceslike 'Johnbuttered the toastslowly' . Subsequentwriting in this vein hassuggestedthat the logical form is evenmorecomplexthan this, perhapsthat it is alongthe lines of ' There is an evente, suchthat e is a buttering, John is the agentof e, the toastis thepatientof e, e is past, ande is slow' .2 The standardassumptionis that the Davidsonianeventanalysiswill require us to introducean explicit referenceeventquantificationinto the syntaxitself. Wecanprobablydo this, but, asLarsonandSegal( 1995) haveshown, it is unnecessary . Supposethat we havea simple extensionto the languagediscussedin appendix Tl , so that we haveadverbsandprepositionsintroducedasfollows.
VP-+V ADV VP-+VPP P-+PNP ADV - + slowly, quickly P - + in, with V - + swims, kicks NP - + the lake Then the task is to showhow the resultingstructurescanbe interpretedas involving a kind of quantificationoverevents.As with the simplerfragmentin appendix TI , the axioms for the adverbsand prepositionswill essentially be . Theaxiomsfor theverbswill likewisebedisquotationalwith the disquotational twist thattheyaretakento bepredicates of events,ratherthanindividuals.Axioms for NPswill be asin the quantificationalfragmentintroducedin appendixTI . The crucial stepfor introducingthe eventstructurewill comein the rule for the interpretationof S and the rule for the interpretationof the VP. The first
AppendixT2
192
rule introducesthe agencythematicrole, the secondintroducesthe theme(or someotherrelevantthematicrole, dependingupon a numberof factors).3The rulesfor theADV andPPcategorieswill be straightforward. Syntax
ofL2
Add the following to the syntax of L I :
VP -+V ADV VP-+VPP P-+PNP ADV - + slowly, quickly P - + in , with V - + swims, kicks
NP-+ thelake for L24 Semantics TerminalSymbols ( 1) a. ' Val(e, lQ. wlx, 0' ) iffe is slow (for somethingofe s kind) of e'skind) Val(e, guickl~, 0 ) iffe is quick (for ~omething b. Vale ,x), in, 0 ) iffe is in x
c. Val(e, swims,0) iff e is a (eventof) swllnlnlng Val(e, ~ , 0) iffe is a kicking Val(e, ~ , 0) iff e is a barking Val(e, ~ , 0) iff e is a walking Val(e, ~ , 0) iff e is a seeing Val(e, ~ , 0) iff e is a liking d. Val(x, Smith, 0) iff x = Smith Val(x, ~ , 0) iff x = Jones Val(x, thelake, 0) iff x = thelake Val(x, ~, 0) iff x = o(i) for i ~ 1
A Quantificational withEvents Fragment
193
e. Va1(x , dQi, 0 ) iff x is a dog Va1(x, ~ , 0 ) iff x is a man Va1(x, woman, 0 ) iff x is a woman
NonterminalNodes (2) a. Va1 ( T, [sNPVP], 0) iff , for somee, Va1 (e, VP, 0) and, for somex, x is dIe agentof e andVa1 (x, NP, 0) b. Va1 (e, [ypVNP], 0) iffVa1(e, V, 0) and, for somey, y is dIedtemeofe and Va1 (y, NP, 0) c. Va1 (e, [ypV ADV], 0) iffVa1(e, V, 0) andVa1 (e, ADV, 0) d. Va1 (e, [ypV PP], 0) iffVa1(e, V, 0) andVa1 (e, PP,0) e. Va1 (e, [ppP NP], 0) iff , for somez, Va1e,z), P, 0) andVa1 (z, NP, 0) f. Val(x, fa13 ] , 0') iffVal(x, 13 , 0') (wherea rangesovercategories , and13ranges overcategories andlexicalitems) (3) a.
Va1 thatVa1 (T, [s51~ 52], 0) iff it isboththecase (T, 51, 0) andVa1 (T, 52, 0) b. Va1 (T, [s51Q[ 52], 0) iff eitherVa1 (T, 51, 0) orVa1 ( T, 52, 0) c. Va1 that51], 0) iff it isnotthecase thatVa1 (T, [sitisnotthecase (T, 51, 0) (4 ) a.
Va1 ( T, [5[NPi ~ Va1 ( T, SI, 0" ) b. Va1 ~ ( T, [5[NPi Va1 (T, SI, 0" )
N ], S1 ], 0') iff for every0" ~i 0' suchthatVa1 (O"(i ), N, 0'), N], SI], 0') iff , for some0" ~i 0' suchthatVa1 (O"(i ), N, 0'),
194
Appendix1' 2
c. Va1(T, sNPj SI ], a ) iff , for a ' =ja suchthatVa1(a '(i ), NP, a ), Va1( T, SI , a ') (5) derivationrules(repeatedfrom Ll ) (SoE) . . . a. . . . a. iff ~ therefore . . . ~ . . . (Soi) cj)iff for somex , x = a. and . . . x . . .
therefore
Val(T, ~ [NP .smi1h ~~ [ADVS ~ W)]) iff . . . ] [vp[Vis (1) x, x isthe thereisane, Val(e, [vp[V~ im] [ADV .si.Q ::wW )], 0), andforsome Val x of e and 0 , ( , [NPSmitbJ ) agent , of (2a)] [instance (2) x, thereis ane, Val(e, [v.s~~ , 0) andVal(e, [ADV .si.Q ::wW , 0), and,forsome x is theagentore, andVa1 (x, [NPSmi ! h] , 0) [from( 1) by (2c), SoB] (3) thereis ane, e is a swimmingandVa1 , 0'), and, for somex, x is . } QwW (e, [ADV
of e, andVal(x, [NPSmi theagent !h] , 0) ] [from(2) by(lc), SoB
A QuantificationalFragmentwith Events
195
(4) thereis ane, e is a swimmingande is slow, and, for somex, x is theagentof e, andVa1 (x, [NPSmid1 J, 0) [from(3) by ( la), SoE] (5) thereis ane, e is a swimmingande is slow, and, for somex, x is theagentof e, andx = Smith [from(4) by ( ld ), SoE] (6) thereis an e, e is a swimmingande is slow, andSmith is the agentof e [from (5) by Soi] In this fragment, prepositionalphrasesfunctionbasicallylike adverbs,although the derivationsfor sentences containingPPsareslightly morecomplex. To see this, considerthederivationof thetruthconditionsfor thesentence' Smithswims
in thelake'.
Val( T, [8[NPSmi lake]]]]), iff . . . :th] [vp[v. ~~ [pp[pin] [NPthe (I ) thereis ane, Val(e, [vpk. ~~ [pp[pin] [NPthe lake]]], 0), and, for somex, x is theagentofe, andVal(x, [NPSmithJ, 0) of (2a)] [instance (2) thereis ane, Val(e, k ..s~im ] , 0 ) andVal(e, [pp[pin] [NP thelake]], 0) andfor somex, x is theagentof e, andVal(x, [NPSmi :th], 0) [from( I ) by (2d), SoE] (3) thereis ane, Val(e, k . ~~ , 0) and, for somez, Vale ,z), [pin] , 0) andVal(z, thelake]], 0), andfor somex, x is theagentof e, andVal(x, [NPSmi [NP :th], 0) [from(2) by (2e), SoE] (4) thereis ane, Val(e, k .s~im ] , 0) and, for somez, e is in z andVal(z, [NP ~ lm ] ], 0), andfor somex, x is theagentof e, andVal(x, [NPSmid1J, 0) [from(3) by ( Ib), SoE] (5) thereis ane, e is a swimmingandfor somez, e is in z andVal(z, [NP thelake]], 0), andfor somex, x is theagentof e, andVal(x, [NP . smi1hJ , 0) [from(4) by ( Ic), SoE]
196
T2 Appendix
(6) thereis ane, e is a swimmingandfor somez, e is in z andz = thelake, and for somex, x is theagentof e, andx = Smith [from(5) by ( 1d), SoB] (7) thereis ane, e is a swimmingande is in thelake, andSmithis theagentof e [from(6) by Soi]
Appendix
T3
A Fragment with I LFs for PropositionalAttitudes (from LarsonandLudlow 1993)
Syntaxof L3 ~Amp; as LI , with addition
of thefollowing:
VP-+Vin'S
Vinl- + believes , claims,thinks Semantics of L3 Ten Dina ) Nodes
a. Va1 (x, Smith, 0') iff x = Smith Va1 (x, ~ , 0') iff x = Jones Va1 (x, ~, 0') iff x = O'(i ) for i ~ 1 b. Va1 (x, g , 0') iff x is a dog Va1 (x, w, 0') iff x is a cat Va1 (x, mIn, 0') iff x is a man Va1 , 0') iff x is a woman (x, woman c. Va1 .rb, 0') iff x barks (x, .b.@ Va1 x , ~~ , 0') iff x walks ( d. Va1 ({x,y}, ~ , 0') iff x seesy Va1 ({x,y},~ , 0') iff x likesy Va1 , 0') iff x believes ({x,y}, believes y Va1 x 0 ({ ,y}, tbiDk~, ' ) iff x thinksy
198
AppendixT3
Va1 ({x,y), claims.0) iff x claimsy NonterminalNodes (2) a. Val( T, [sNPVP], a) iff , for somex, Val(x, NP, a) andVal(x, VP, a) b. Val(x, [vpV NP], a) iff , for somey, Val x,y), V, a) andVal(y, NP, a) c. Val(x, [a ri ], a) iffVal(x, ri, a) (where<Xrangesovercategories , andri ranges overcategories andlexicalitems) d. Val(x, [vpVintS ] , a) iff , for somey, Val x,y), Vint ' a) andy = [] S[] w.rit. a (3) a. Val( T, [sSI .a.1ldS2], a) iff it is boththecasethatVal( T, SI, a) andVal( T, S2, a) b. Val(T, [sSI Q[ S2], a) iffeitherVal( T, SI, a) orVal(T, S2, a) c. Val(T, [sit is notthecasethatSI], a) iff it is notthecasethatVal( T, SI, a) (4)
a. Val(T, ~ [NPI N] , SI ], a) iff , for everya' =j a suchthatVal(a' (i), N, a), Val( T, ~ ' 51, a ) b. Val(T, ~ [NPI a' =j a suchthatVal(a' (i), N, a),Val(T, ~ N] , 51], a) iff, forsome ' 51, a ) c. Val(T, [8NPj51], a) iff, fora' =ja suchthatVal(a'(i), NP,a), Val( T, 51, a') DeftnitiODS (repeated from Ll ) I. Foranysequence 0', O ' (i ) is theith elementof 0'. 11 . Foranysequences 0', 0" , 0" =j 0' iff 0" differsfrom0' atmostonO"(i ). iii . Va1 0'. ( T, S) iffVa1(T, S, 0') for all sequences
A Fragment withn..FsforPropositional Attitudes
199
Definition:Let a bea phrasemarkerwith rootS, let 0' bea sequence , andfor eachnodePof a , let x bethesemantic valueassigned to p, assuming Val( T, a , 0). Then: 1. If Pis atenninalnodeandVal(x, p, 0'), thenDPD=
(a) .6 is a tenninalnode,then[]6D --. -- = (6) -. -
= [('J)D51D is [15152 . . . 50]forn~ I , then0130 (b) 13 0520. . . D50rn rulesweusedin LI andL2. , weshallusetwoderivation Finally (5) DerivationRules (SoE) ...a ... aiffp derive . . . p . . .
(Soi)
forsome x, x = a and. . . x . . . derive . ..a . . . '. Consider asample derivation for'Smith believes Jones walks Val(T, S[NPSmid1 ] [vp[vbelieves ] S[NPJ ~ [vp[~ S~ ]]], 0) iff (I ) forsome x,Val(x, [NPSmi :thJ,0) andVal(x, [vp[~ believes ] S[NPJ ~ [vp[v ~~ ]]], 0) of(2a)] [instance (2) forsome x, Val(x, [NPSmith ], 0) andforsome ], 0) and y, Val({x,y}, kbelieves ][vp[~ ~~ ]][] w.rit. 0 y = [][s[NPJones [from(I ) by(2d), SoE]
200
AppendixT3
(3) forsome x, x = Smithandforsome believes 1, 0) andy = [][s ((x,y), [Vim y, Va1
[NPJ ~ [vp[v:w:Blk~ ]][] W.fit. 0from [ (2) by ( la), SoB] (4) for somex, x = Smithandfor somey, x believes Jones ] [vp[y y, andy = [][ s[NP
~~ ]][] W.fit. 0[from(3) by(ld), SoE]
(5) Smithbelieves[] s[NPJones][vp[v~ Wb] ]] [] w.rit. 0[from (4) by Soi] Constructionof the ll...F proceedsasfollows: First, the truth definition mustbe appliedto the embeddedclauseto determinethe semanticvaluesof eachnode of theembedded p- marker. Recallthattheembedded p- markerin this caseisjust the following.
s / """VP NP I V I Jones I ~
We canget the semanticvalueof eachnodefrom the following (partial) derivation . VaI( T, [s [NPJ~ [vp V:wI1U ] )], 0) iff . . . ( 1) for somex, Val(x, [ NPJ~ , 0) andVaI(x, [vp V~DJg]], 0) of (2a)] [instance (2) for somex, Val(x, ~ , 0) andVaI(x, ~~ , 0) [from( 1) by (2c), SoB] (3) for somex, x = Jonesandx walks [from(2), by ( Ia), ( Ic), SoB]
A Fragmentwith ll ..Fsfor PropositionalAttitudes
201
Given the calculation of these semantic values, we can construct the ILF fol -
thedefinition introduced above . lowing [][S[NPJones ] [vp[v~~ ]]D =[(5,ThIe ~ DD[yp [y~~ ]]D] )D[NPJ [by(3ii)]
= [(So True IIe S )[
S True , ( ) " / / / " " , N P Jones ,Jones V p (a~ ) ( ) II) ,IJones ,Jones ) (V 8 True , ( ) " / / " " " Jones P Jones ,~ , (~N ) ( V P ) I,Jones I ,Jones ) (V ) I
In a more familiar tree representation, we have the following .
~ ~ , Jones } then : , wehavederivedthefollowingtheorem Finally, Smithbelieves Joneswalksis trueiff Smithbelieves
lb, Jones ~ .@ )
T4 Appendix
204 Semantics
for LB
TerminalSymbols ( 1) a. Va1 , S, R, E) iffR is thedaybeforeS (e, ~esterday Va1 (e, ~ , S, R, E) iff R is thesamedayasS Va1 (e, tomorrowS, R, E) iff R is thedayafterS b. Va1 ({e,x), in, S, R, E) iffe is in x Val(e, swims,S, R, E) iff e is a swimming Val(e, ~ , S, R, E) iff e is a walking d. Val(x, Smith, S, R, E) iff x = Smith Val(x, ~ , S, R, E) iff x = Jones Val(x, TheLake, S, R, E) iff x = TheLake
Va1 , S, R, E) iffS islaterthanRIB (e, PAST Va1 , S, R, E) iff S, R, andE temporally (e, PRES overlap Va1 (e, FUT,S, R, E) iff SisearlierthanRIB Va1 PERFECTS , RE ) iffE isearlierthanSIR (e, PRES Va1 , RE ) iffE isearlierthanRandRisearlierthanS (e, PASTPERFECTS Va1 , RE ) iff SisearlierthanE, E isearlierthanR (e, FUTPERFECTS Va1 , RE ) iff E isearlierthanS, SisearlierthanR (e, FUTPERFECTS Va1 e FUT PER F Ecr S isearlier thanR , , R, E) iff EandStemporally (, overlapS Va1 , S, R, E) iffR isearlierthanS, RisearlierthanE (e, FUTIN PAST f. -at(e, E) Va1 , S, R, E) iff in-progress (e, PROG Va1 e S R E iff culminates at ( ,~ , , , ) (e, E) g. Val({x,y}, ~ fQm, S, R, E) iff x is earlierthany Val({x,y}, ~ , S, R, E) iff x is later thany Val({x,y}, ~ , S, R, E) iff x temporallyoverlapswith y
A B-TheoryTechnicalFragment
205
Axioms for NonterminA ISymbols (2) a.
Val(T, [IPNP1'], S, R, E) iff, forsome x, e, Val(e, 1', S, R, E) andat(e, E) and, forsome x, x istheagent of eandVal(x, NP,S, R, E) b. Val(e, [r I VP], S, R, E) iffVal(e, I, S, R, E) andVal(eVP, S, R, E) c. Val(e, [vpV] , S, R, E) iffVal(e, V, S, R, E) Val(e, [vpV PP], S, R, E) iffVal(e, V, S, R, E) andVal(y, PP,S, R, E) Val(e, [vpV ADV], S, R, E) iffVal(e, V, S, R, E) andVal(e, ADV,S, R, E) Val
( e , [ vpV
Val
( e , ADV
PPADV
( e , V , S , R , E ) and
] , S , R , E ) iffVal
Val
( e , PP , S , R , E ) and
, S , R , E )
d . Val
( e , [ pp P NP
] , S , R , E ) iff
, for
some
z , Vale
, z ) , P , S , R , E ) and
Val
( z , NP , S ,
R , E ) e . Val
TNS
( eL
] , S , R , E ) iffVal
ASP
( e , TNS
Val
, S , R , E ) and
( e , ASP
, S , R,
E )
f . Val R2
( T , [ IP IPI , E2 ) and
IP2
CON Val
] , S , R , E ) iffVal
, E2 ) , CON
EI
, S , R,
( T , IPI
, S , RI
, EI
) and
Val
( T , IP2
, S ,
E )
g . Val
(x ,
over
S , R , E ) iff a P ] , and
categories
Val
lexical
( x , P ) where items
a
ranges
over
categories
and
p ranges
fragment
would
.
Rules
Production
( So E ) . . . a
. . .
aiffp
. . . p
therefore
( Soi
)
therefore
To
. . .
illustrate
proceed
.
iff
for
.
iff
. . . a
, the as
shown
some
a
and
. . . x
Smith
swam
. . .
. . .
derivation in
x , x =
( 1) .
for
'
' yesterday
in
this
206
T4 Appendix
Val(T, [IP[NP PAST CUL))[vp[v.s~g [ADV Smith ] [r r[TNS ] [ASP ~esterda ~)))), S, R, E) iff . . . (1) forsome x,e, Val(e, [r r[TNS CUL))[vpV.s~g [ADV PAST ] [ASP ~esterda ~))], S, R, E) andAt(e,E) andx istheagent of eandVal(x, [NP Smith ] , S, R, E) of (2a)] [instance iff (2) forsome x,e, Va1 CUL]], S, R, E) andVa1 (e, [I[TNS PAST ] [ASP (e, [vpk S R E and At E e and x is the .s~ s][ADV ofe andVa1 , , , , ) ~esterda ( ) ~]] (x, agent S R E [NP Smi .d1 ], , , ) [from(1), by(2b), SoB ] iff (3) forsome x,e, Va1 CUL], S, R, E) and (e, [TNS PAST ] , S, R, E) andVa1 (e, [ASP Va1 S R E and At e E ofe (e, [vpk.sM~ [ADV , , ]], , ) ~esterday ( ) andx istheagent
andVa1 (x, [NP Smidl ], S, R, E) from 2 2e So [ ( ), by( ), E]
iff (4) for somex,e, Val(e, [TNS PAST], S, R, E) andVal(e, [ASP CUL], S, R, E) and Val(e, k swims], S, R, E) andVal(e, [ADV S , ], ~esterday R, E) andAt(e,E) and x is theagentof e andVal(x, [NP S Smith], , R, E) from 3 2c SoB [ ( ), by ( ), ] iff (5) for somex,e, Val(e, PAST , S, R, E) andVal(e, CUL, S, R, E) andVal(e, .s~ , S, R, E) andVal(e, ~esterday , S, R, E) andAt(e,E) andx is theagentof
e andVal(x, Smith , S, R, E) from 4 [ ( ), by(2g), SoB ] iff
(6) for somex, e, Val(e, PAST , S, R, E) andVal(e, CUL, S, R, E) andVal(e, ~ , S, R, E) andR is thedaybeforeS,andAt(e,E) andx is theagentof e
andVa1 , S, R, E) (X, Smith from 5 1 a [ ( ), by( ), SoB ]
A B-Theory Technical Fragment
207
iff (7) for somex,e, Va1 , S, R, E) andVa1 (e, PAST (e, CUL, S, R, E) ande is a swimand R is the beforeS and At e E , ( , ) andx is theagentof e andVa1 (x, ming, day Smith, S, R, E) [from(6), by ( Ic), SoE] iff (8) for somex,e, Va1 , S, R, E) andVa1 (e, PAST (e, CUL, S, R, E) ande is a swimand R is the beforeS and At e E , ( , ) andx is theagentof e andx = ming, day Smith [from(7), by ( Id), SoE] iff (9) for somex,eS is laterthanRIB, andVa1 (e, CUL, S, R, E) ande is a swimming, andR is thedaybeforeS,andAt(e,E) andx is theagentof e andx = Smith [from(8), by ( Ie), SoB] iff ( 10) -at(e, E) ande is a swimming for somex,eS is laterthanR/ E, andculminates , andR is thedaybeforeS,andAt(e,E) andx is theagentof e andx = Smith [from(9), by (It), SoB] iff ( 11) -at(e, E) ande is a swimming for someeS is laterthanR/ E, andculminates by Smith, andR is thedaybeforeS,andAt(e,E) andSmithis theagentof e [from( 10), by Soi]
Appendix
TS
A Basic A - Theory
Fragment
This appendix, which borrowsfrom the ILF fragmentof Larson and Ludlow 1993, combinesan ILF theorywith an event-basedanalysisof aspect.To keep matterssimpleandto keepthe crucial axiomsat centerstage, basicquantifiers arenot introducedhere(althoughtheycouldbeintroducedalongthelinesof appendix Tl without difficulty). Propositionalattitudeverbsarealsoomittedhere. Syntax of LA Terminal Symbols
As in LB (appendix T4). NonterminalSymbols S ~ IP when/before /afterIP IP ~ TNSIP (ADV) IP~ NPI' I' ~ IVP VP -+V (PP) PP-+PNP I -+ASP Semantics of LA
TerminalNodes ( 1) a. Va1 <{x,y), ~ iff x seesy Va1 <{x,y), bib) iff x hit y
210
TS Appendix
b. Vale ,x), in) iffe is in x. c. Val(e, ~ ) iffe is a swimming Val(e, ~~ iff e is a walking d. Val(x, Smith) iff x = Smith Val(x, ~ V iff x = Jones Val(x, thelake) iff x = thelake e. Val(x, PAST ) iff x wastrue Val(x, PUT) iff x will betrue Val(x, PRES ) iff x is true f. Val(e, CUL) iff e culminates Val(e, PROG ) iff e is in progress . g Va1 (x, ~esterda ~) iff x washueyesterday Va1 ) iff x will behuetomorrow (x, tomorrow Va1 x iff ( , ~ ) x is huetoday Va1 (x, ~ iff x is huenow NonterminalNodes (2) a. Va1 1P2 ]) iffVa1(T, IPI ) whenVa1 ( T, 1P2 ) (T, SIPI ~ IPI 1P2 Va1 T IPI 1P2 iffVa1 before Va1 , , ]) (T ) (T ) ( , [s ~ ~ Va1 ] ) iffVa1(T, IPI ) afterVa1 ( T, 1P2 ) (T, [sIPI ~ 1P2 b. Va1 (x, TNS), andx = [] IPI [] ( T, [IPTNSIPI ]) iff , for somex, Va1 c. Va1 (x, TNS), Va1 (x, ADV), and ( T, [IPTNSIPI ADV]) iff , for somex, Va1 = x []IPI [] d. Va1 (e, 1') andfor somex, x is theagentof e and (T, [IPNP1']) iff for somee, Va1 x NP Va1 ( , )
A BasicA-Theory Fragment
211
e. Val(e, [1'1VP]) iffVal(e, I ) andVal(eVP ) f. Val(e, [vpV PP]) iffVal(e, V) andVal(e, PP) g. Val(e, [ppP NP]) iff for somez, Vale ,z}, P) andVal(z, NP) i. Val(x, [uP ]) iff Val(x, P) wherea rangesovercategories andp rangesover lexicalitems. Definition Let a bea phrasemarkerwith rootS andfor eachnodeIi of a, let x bethesemantic valueassigned to p, assuming Val( T, a). Then:
=
(SoB) . ..a .. . aiffp ...p... (Soi) ~x , x = a and . . . x . . . . iff for some cj)iff . . . a . . .
212
AppendixT5
( RR) Reduction Redundancy
(1) Va1 PAST ] [1P2 [NP Smi :tbJ[rL[As (T, [IPI[TNS PCUL]][vpkswims]]]]]) whenVa1(T, . . . [1P3 [TNS PAST ] [1P4 ][ADV ]) ):esterdayl of(2a)] [instance iff (2) forsome x, yVal(x, [1NS [NP Smid1 ] [rL[As PAST ] ) andx = [] [IP2 PCUL]][vP[v 000 Val(y, [1NS and swimsl PAST ]),Val(y, [AD ) ] [] ]]] [] when \' ~esterdayl y =[] [1P4 from 1 2b 2c SoB [ ( ), by( ), ( ), ] iff (3) CUL]][vp[v forsome x, yVal(x, PAST S.mid1 [NP ][rL[ASP ] ) andx = [] [1P2 000] [] Val(y, PAST swims ),Val(y, ~esterday ) andy = [] [1P4 ]]]] [] when ] [from(2), by(2i), SoB iff (4) for somex, y, x wastrueandx = (] [1P2 ]][vp [vswimsl [NP SmidlJ l[rL[A~pCUL = . . . true and [] wheny wastrue, y was yesterday y [] [1P4 ] [] [from(3), by ( Ia), ( Ie), ( Ig), SoE] iff
A Basic A-Theory Fragment
213
(5) for somex, y, x wastrueandx = (] [1P2 [NPSmit Il ][rL[AsPCUL)][vpkswims])]] [] wheny wastrueyesterday andy = [] [1P4 . . .] [] [from(4), by RR] iff (6) CUL]] Vp [] [1P2 ... [] was [NPSmid1 ] [rL[ASP [v swims]]]] [] wastruewhen[] [1P4 trueyesterday [from(5), by Soi] Thetreatment of theI LFsproceeds asin appendix T3.
Notes
Introduction 1. Seechapter3 of Sorabji 1983for discussion . Kretzman( 1976) arguesthat the issue is alsotakenup by Aristotle, althoughthis claim is somewhatmorecontroversial. 2. In anycase, thatis thereceivedview of Einstein's position. But seep. 287of Shimony 1993for dissensionon this point. It is worth noting that direct evidenceof Einstein' s positionon the matteris hardto comeby; oneof the bestpiecesof evidencefor his Btheory sympathiesis a letter of condolencewritten to the family of a friend (Einstein 1955). 3. This pagereferenceis to the versionin Yourgrau1990. 4. This in factechoesa pointmadeseveraldecades earlierby Reichenbach ( 1956, p. xiii ) : " If the oriented of our time deniesto contemporaryscienceits speculatively philosophy philosophicalcharacter, if it calls contributionssuchas the theory of relativity or the theoryof setsunphilosophicalandbelongingin the specialsciences , thisjudgementexpressesonly the inability to perceivethe philosophical contentof modem scientific " thought. 5. See, e.g., Smith 1993andCraig 1990. Thefollowing passage from Craig(p.339) illustrates this point of view nicely: " I find it surprisingthat anyonereadingEinstein's 1905 that absolutesimultaneitydoesnot exist and papercanthink that Einsteindemonstrated that time is thereforerelativeto a referenceframe. For the entire theory dependsupon ' acceptanceof Einstein s arbitrary (and, indeed, highly counterintuitive) definition of simultaneity, coupledwith a philosophicalpositivism of Machianprovenanceaccording to whichabsolutesimultaneityis meaningless if it is empiricallyundetectable . . . . One who is not a positivistandwho thereforerejectsEinstein's definitionswouldregardthese relativelymovingobserversasdeceiveddueto the natureof their measurements , which fail to detecttruetime. In a real sense , he would not regardEinstein's theoryasa theory abouttime and spaceat all , but, as Frank put it , ' as a systemof hypothesesaboutthe behaviorof light rays, rigid bodies, andmechanisms , from which newresultsaboutthis behaviorcan be derived.'" But seeShimony1993for further discussion(andcriticism) of this view of relativity theory. 6. This view alsoappearsto havebeenheld by Godel. SeeYourgrau1991.
216
Notesto pp. 3- 20
7. I don' t meanthat physicistscan' t do philosophy.The point is that the questionscan' t beansweredin isolationof philosophicalinquiry- whethercarriedoutby personsin phi. losophydepartmentsor in physicsdepartments . 8. Strawson1959is a classicexampleof this generalapproach 9. A caveatis necessaryhere. Someso called Prioreantenselogics are not genuine . As van Benthem Prioreantheories, becausetheir semanticsimport B-seriesresources " ' ' ' ' " ' ( 1982) notes, on many Priorean theoriesthe semanticsfor past , present, and fu' ture are often given in termsof the relationsbeforeand after. For example, it is typically assertedthat x satisfies' past' iff x is earlierthany, wherey is the utterancetime. It seemsto me that this way of giving the semanticsfor a Prioreantenselogic is quite out of the spirit of Prior' s generalprogram, which happensto eschewfutureandpastevents aswell asthe B-seriestime line. 10. Of courseneitherKaplannor Perryadvocatestruth-conditionalsemantics , but the key point here is independentof the semanticalframework we chooseto work with. This way of thinking aboutthe Kaplan-Perryview of indexicalswithin the truth-conditional frameworkis dueto LarsonandSegal( 1995) . 11. Althoughtheobjectionturnson a theoryin whichtheoperatorsapplyto eventsrather thanto propositions,the objectionis readily adaptedto propositions.The ideawould be that a givenpropositionis future, thenpresent,andthenpast. Chapterl I . Portionsof this sectionaredrawnfrom Ludlow (forthcoming). 2. Chomskywould addthat I -languageis " individualistic," meaningthat the properties of the systemdependon the individual in isolationratherthanon relationsbetweenthe individual andthe externalenvironment.I don' t needto makethat assumptionhere, and I ' ll rejectthe thesisfor psychologyin generala bit later. ' 3. The hedgeabout" co-opting" (more accurately, exaptation) is to cover Chomskys view that the languagefaculty may not havebeenthe productof gradualevolution. For moreon this, seesection1.2. 4. SeeLewis 1975afor an articulationof this view of language.SeeLewis 1969for a discussionof the natureof convention. 5. Strictly speaking, the differencebetweenI -languageandE-languagemay not even countasanactualdistinction, sinceon Chomsky's view thereis I -languageandthenthere is a collection of poorly defined if not incoherentviews which might be lumped together as" E-languagetheories." 6. Davidson( 1986) reachesa similar conclusion, thoughby a muchdifferent route. In a funny sense,althoughneitherwould admit it , Davidsonand Chomskyare in accord on onepoint: that thereis no suchthing asa language(construedasan E-language ). 7. An alternativewould be to regardlanguagesasabstractobjectsin the senseof Katz 1981. For reasonsoutlinedby Higginbotham( 1983), I will setasidethatpossibilityhere. 8. For a generaldiscussionof this issue, seeBeakley1991.
Notesto pp. 21- 39
217
9. SeeGould andLewontin 1979. For someof Gnuld' ~popularwritings on this theme, see
Gould
10 .
For
II
The
.
this
12 to that
a
In
one
no
of
the
Fodor
phenomena
, see
( 1975
Carruthers
) and
1995
Hayes
( 1996
.
) will
be
evident
in
of
portions
.
amusing
episode
certain
manipulate he
of
description
influence
section
.
.
1980
longer
, Russell
geometric had
names
shapes for
shapes
, while , asked
taking if
he
an could
intelligence terminate
test the
that exam
required on
the
him ground
.
13. This concessionis controversial. For a sampleof the literatureon whethermental imagesarefundamentallypictorial or discursive, seeBeakleyandLudlow 1992. Chapter2 1. For an exampleof someonewhosealarmhasgoneoff , seeDevitt 1993. 2. An interestingpuzzlearisesif onewantsto holdto thisextemalistconceptionof know1. edgeandto the ideathat onehasa priori knowledgeof the meaningsof the expressions Somehavearguedd1ata priori self-knowledgeandexternalismof thischaracterareincompatible . However, it is fair to saythat the issuesare very subtle. This is not the placeto take up thesequestions.For a surveyof the.currentliteratureon the topic, seeLudlow andMartin 1998. 3. See, e.g., Chomsky1986. 4. For discussionsof this possibility, seep. 209 of Montague1974. 5. Durnrnett( 1991) hasstressed thatmodel-theoreticsemantics cannotprovidesuchconnections but that it nonetheless is usefulfor characterizingentailmentrelationsbetween , . Whethermodel-theoreticsemantics canevenaccountfor entailmentrelations expressions also turns out to be controversial. For skepticismaboutmodel-theoretictreatmentsof logicalentailment,seeEtchemendy1990andField 1991. 6. Having an interpretiveT-theory without knowing that it is interpretive, or without knowingto what useit may be put, is like havinga mapwithout knowing whetherit is to scale, or without knowing what it is a mapof. For a discussionof this concern, see Foster1976. 7. Therewould be no harmin writing a fragmentthat utilized the semanticalnotionsof satisfactionandreferenceinstead.In thatcase, ' x = Sally' might give way to ' Sally refers to x ' , and ' x walks' might give way to ' x satisfies' walks" . 8. Reificationof "the True" canbe avoidedif we wish. I usethe notion here'merelyfor convenienceandnot out of Fregeanintuitions that thereis sucha thing asThe True. 9. Werethis constraintto be relaxed, we would endup provingtheoremslike " ' Snowis white' is trueiff snowis white andthe number7 is prime." For a discussionof this issue, seeDavies1981a. 10. For somecriticismof this line of investigation , seeFodorandLepore1997andFodor andLepore1998. This criticism appearsto be drivenby the assumptionthat lexical decomposition andthe like drive us into holism. I fail to seewhy this musthappen.
218
Notes to pp . 41- 57
11. For further discussionof Evans's argument,seeWright 1986; Davies 1981b, 1987; Peacocke1986. 12. See, e.g., McDowell 1980; Evans1982; Bilgrami 1987; LeporeandLoewer 1987. Theseauthorsareinfluencedby remarksin Dummett 1973andDummett1975. 13. I' m not sureif LeporeandLoewersawthis. 14. SeeLudlow 1993. 15. Onemightsupposethat, within a broadlyDavidsonianprogram , T-theoriesaretreated in a " deflationary" way, so that they do not refer to objects, statesof affairs, etc. Lepore andLoewer( 1987, p. 103) argueasfollows: " If onethinks. . . that a truth theoryassigns then we can seewhy [certain] possiblestatesof affairs or factsto indicativesentences truth theoriesassignthe sametruth conditionsto the two sentences' Cicerois bald' and ' ' Tully is bald . But it is not necessaryto think of truth theoriesin this way. Davidsonrejects the reificationof truth conditions. . . ." Contraryto this line of thinking, I fail to see how we can stipulateT-theoriesto be deflationary. The fundamentalproblem is that, becauseoneis using(e.g.) Englishon the right-handsideof a T-sentence , whatis stated on the right-handsidecannotbe moreor lessthanwhatis statedby that Englishexpression . Oneway to get a handleon this issueis to think of the axiomsandtheoremsin a T. Onemaynot wantto reify talk aboutplanets theoryasakinto thelawsin anyotherscience andquasars,but it is hard to seehow, short of generalscientific antirealism, one can escapecommitmentto them. The problemis not uniqueto certainkinds ofT -theoriesin which it is stipulatedthat the right-handsideof the T-sentencesrefersto a stateof affairs . Rather, the problemstemsfrom the fact that naturallanguageis usedto statethe truth conditions, andthusthat theT-theorycan" state" no moreor lessthanwhat is said by the naturallanguageexpressionusedto give the truth conditions. 16. This referencewasfirst broughtto my attentionby RichardLarson.
Chapter3 1. See,e.g., Cresswell 1985,1990;Hintikka1962 , 1969a , 1975;Lewis1972;Stalnaker 1984. 2. See,e.g., Richard1990;Salmon1986;Soames 1987. 3. Seeappendix T3 for technical details . 4. Thephilosophers includeBuridan(1966 ); the ), Montague (1960 ), andQuine(1960 Karttunen and Ross . include 1974 1976 1976 ), ( ), ( ) linguists McCawley( 5. Forattempts , seedenDikken,Larson , andLudlow1996andLarson , denDikken,and Ludlow1997. I departfromHigginbotham 6. Hereapparently (1986a ). 7. Thefollowingfewparagraphs aredrawnfromLarsonandLudlow1993. 8. Reiber( 1997,p. 289) objectsto ILF theoriesat preciselythis point, arguingthat suchtheories- particularlytheversionadvanced by LarsonandLudlow- still facethe " " "content [Larson problem: The pointI wishto makeis thatnomatterhowsuccessful . andLudlow] maybein this project, it will leavethecentralobjectionunanswered Explaining(i ) howtwobeliefreportscandifferin contentbutreportthesameattitude
Notesto pp. 57- 65
219
maybe a first stepin explaining(ii ) why ordinarythinkersmistakenlythink theyhavethe samecontent,but explaining(i ) doesnot sufficefor explaining(ii ). A 'pragmatic'explanation of (ii ) would haveto go beyondLarsonandLudlow' s; it would needto be an ' error ' " theory thattells us why ourjudgementsabutcontentaremistaken. This objectionfails becauseit assumes thatwehaveintuitionsaboutcontents(apartfrom thoseintuitionswe are taughtto havein philosophygraduateseminars ) . Clearly we haveintuitions about whentwo sentences saythe samething, but it is quite anothermatterto supposethat we havenaturalintuitionsaboutcontents . The situationis somewhatrelatedto thatin syntax, whereChomskyhasarguedthat we haveintuitionsaboutthe acceptabilityof a sentence but not aboutits grammaticality.The latter notion is a theoreticalconstruct(just like, I submit, the notion of content) . Accordingly, no error theory is necessary ; we haveno naturalcontentintuitions aboutwhich to be in error. 9. For what it ' s worth, I find this a plausiblereconstructionof what Fregewastrying to say. It is certainlya reasonableway of taking remarkslike the following (quotedearlier in this chapter): "Although the thoughtis the same, the verbalexpressionmust be different so that the sense , which would otherwisebe affectedby the differing timesof utterance " , is readjusted. In other words, the verbal expressionmust change with circumstances in orderto expressthe samesense . 10. Doesit follow thattheremustbean independentlyexistingsense , or is talk of sense just a handyway of talking aboutour ability to keeptrack of when two indexical expressions or attitudeattributionssaythe samething? I aminclinedto supposethatthelatter is thecase,althoughof courseit wouldbepossibleto saythatsuchabilitiesunderwrite any epistemicrelation we might haveto sensesconstruedas abstractobjects. Sucha possibility will be discussedin moredetail (if obliquely) in chapter4. 11. In recentyearstherehasbeena greatdealof work on thesesortsof simulationtheories . For surveysof this literature, seeCarruthersand Smith 1986; Daviesand Stone 1995a,b. 12. For a proposalin this vein, seeNolan 1970. 13. Divide yes, but not wall off. Thuswe havea proliferationof disciplinesin thenatural sciences(molecularbiology, primateecology, etc.) which ignoretraditionaldisciplinary boundariesandwhich supposethat the varioussciencesareconnectedat somelevel and thatpostulatingtheorieswhich rely on theseconnectionsis entirelyappropriate . Chapter4 1. Consider,for example, the following fragmentfrom Parmenides( 1984): " It mustbe that what is therefor speakingof and thinking of is; for it is thereto be." The idea is alsoclearlyarticulatedin Plato'sSophist(237DE ), wherea s1l'anger pointsoutto Theatetus that if we fail to refer whenwe usethe expression' something'we havefailed to express a determinateproposition: ' is SIr: Surelywe canseethatthisexpression' something alwaysusedof a thing thatexists . Wecannotuseit just by itself in nakedisolationfrom everythingthatexists, canwe? Theatetus : No. SIr: Is your assentdueto thereflectionthatto speakof ' something'is to speakof ' some onething' ?
220
Notesto pp. 65- 71
11reatetus : Yes. SIr: Becauseyou will admit that ' something' standsfor one thing, as ' somethings' standsfor two or more. : Certainly. 11reatetus ' is to SIT: Soit seemsto follow necessarilythatto speakof whatis not ' something speak of no thing at all. 11reatetus : Necessarily . SIr: Must we not even ' refuseto allow that in such a casea personis saying something , though he may be speakingof nothing? Must we not assertthat he is not even ' ' sayinganythingwhenhe setsaboututteringthe sounds a thing that is not ? Aristotle broke with Platoon a numberof metaphysicalissues, but not on the connection betweenlanguageandmetaphysics . In chapter1 of Book Z of the Metaphysics , for esthe questionof whetherpredicatesmust refer to independent example,Aristotle address forms: "And so onemight evenraisethe questionwhetherthe words ' to walk, ' ' to be healthy,' ' to sit,' imply that eachof thesethings is existent, and similarly in other casesof this sort; for noneof them is either self- subsistentor capableof being separated from substance , but rather, if anything, it is that which walks or sits or is healthy that is an existentthing. Now theseareseento be morereal becausethereis something definite which underliesthem (i .e., the substanceof individual) which is implied in sucha predicate; for we neverusethe word ' good' of ' sitting' without implying this." HereAristotle disagreeswith Platoon thequestionof whethertheremustbe independent forms, but crucially baseshis argumenton the linguistic fact that predicatescannot .) standalone. (Seep. 2 of Alston 1964for a brief discussionof this passage 2. The ideathat the domainshouldbe stipulatedin advancemaybe groundedin certain assumptionsaboutthe desirabilityof formal rigor in semantictheory (andthe sciences generally); however,asis arguedin Ludlow 1992, sucha desideratumreflectsa misunderstanding aboutthe role of formal rigor in sciencegenerally. 3. It may well be that either (2) or ( 3) is inadequatefor certain psycholinguisticreasons , but this would be a questionfor investigation. 4. For a discussionof a fragmentof this kind, seesubsection4.3.2 of LarsonandSegal 1995. 5. A similar argumenthasbeenraisedagainstusing singulareventanaphoraasan argument for the existenceof events. 6. For a lesspessimisticview, seeBealer 1982, ChierchiaandThrner 1988, and many of the essaysin ChierchiaandThrner 1989. 7. Somephilosophershavesuggestedthat we can escapethe apparentcommitmentto . Thefirst assumption if we makea coupleof basicassumptions , standard objectslike Pegasus in free logics, is that we neednot assumeexistentialgeneralizationholds, so that from (i ), for example, we neednot infer (ii ). (i ) Pegasus= Pegasus (ii ) . for somex, x = Pegasus
Notesto pp. 71- 76
221
We cangrantthis point, sinceour concernis whetherreferringexpressionsthemselves raiseontological commitments(quite independentlyof the involvement.of quantification ). The questionis whetheran axiom like ( 17) by itself mustcommit us to an object like Pegasus . The issueis complicatedby the fact that (i ) appearsto be contentfulin a way that (iii ) doesnot. (iii ) Val(x, Egret) iff x refersto Egref. Oneway of distinguishingaxiomslike ( 17) from axiomslike (iii ) is by supposingthat they invokeconcepts.So, for example, ( 17) is contentfulbecausewe havea conceptof , but (iii ) is not becausewe haveno conceptcorrespondingto Egref. The prima Pegasus facie problemwith this solutionis that axiomslike ( 17) do not appearto be aboutconcepts , but ratheraboutobjects. So, for example, a philosopherwho believedin the existence of Pegasusmight well distinguishhis beliefsaboutPegasusfrom his beliefsabout the conceptof Pegasus , and in turn may distinguish the two in his speech. In sucha case, ( 17) mayplaya role in giving the semanticsfor this philosopher's utterancesabout . That would requirea second , but not his utterancesaboutthe conceptof Pegasus Pegasus axiom referring specificallyto the concept. In any case, following considerations raisedin section1.3, I am highly skepticalof any analysisof conceptsasbeingdistinct from their I -languagemanifestations . 8. See, e.g., Kripke 1980. 9. There is probably also room to wonder about the relevanceof rigidity intuitions here. Thoughit maybeconcededto Kripke thatnamesrigidly designatein a way thatdescriptions do not, Stanley( 1997a) hasarguedthat Kripke merelyassumesthat rigidity is relevantto the theory of meaning. Pointing to Kripke ' s ( 1980, p. 14) remarksthat " we havea direct intuition of the rigidity of names, exhibitedin our understandingof " thetruth conditionsof particularsentences , StanleyholdsthatKripke begsthequestion. The intuitions aboutrigidity , if thereare suchthings, may haveno bearingon our understanding of truth conditions. 10. Note the parallelbetweenthis andmy discussionofindexicals in section3.2. 11. I am indebtedto Ernie Leporefor discussionof this point. 12. For a statementof the concerns,seeWallace1972. For criticism, seeKripke 1976. 13. SeeDreyfus 1982for just a sample. 14. In this sectionI havefocusedexclusivelyon the questionof whethersubstitutional quantificationin fact freesus from ontological commitment. Thereare also numerous technicaldifficulties that arisefor substitutionalquantificationasa thesisaboutnatural , someof which bearmentioning- for example, the questionof whethersubstitutional language ' ' ' ' quantificationcan be extendedto quantifierslike most and few (Davies 1981a; Ludlow 1985, section4.B ; Lewis 1985). For example, if the truth conditionsof ' Most ' spiesarebored aregivenasin the following example, the truth conditionscome out wrong, sincespieswill go by manynames,andit is possiblethat onespyin particular of names, (say, Jones) maybetheonly boredspy, but by virtueof his havingthousands the following will be true: For mosttermst , suchthat t- ' is a spy' is true andt- ' is bored' is true.
222
Notesto pp. 76- 85
Althoughthesetechnicaldifficulties areinteresting,I havesetthemasidebecauseit must first bedemonstrated that substitutionalquantificationfreesusfrom all ontological commitments . I will returnto this problem(underanotherguise) in chapter8.
ChapterS I. Thisisbasically theapproach totense in veryearlyeffortswithinMontague Grammar in particular of Montague 1974.It hasalsobeenadopted in truth, in thePTQfragment conditional semantics ; seeLeporeandLudwig1998. 2. Variations onthisthemearepossible . SeeVickner1985for anexample . 3. A morecomprehensive T theorycanbefoundin appendix T4. 4. SeeVickner1985andchapters 2 and3 of Hornstein1990for proposals onderiving fromsimpleronesin theReichenbachian . framework complextenses thataquantificational 5. Othershaveobserved treatment of theseconnectives is in order. SeeBonomi1997foronesuchproposal andfor asurveyof someof therelevant literature . of temporal 6. Thistreatment connectives reliesuponasyncategormatic treatment of the . Thatis, theyareassigned connectives no specificcategoryandareimmediate daughters of S. If wemakethenaturalassumption thattheyshouldbeassigned acategory CONJ, thena setof axiomsalongthelinesof (7') couldbeemployed . ' (7 ) a. Val( T, ~ SI CONJ1 , s, R, E, 0') iffVal( T, SIS , R, E, 0') andVal( T, CONJP , S, R, E, 0') b. Val(x, [CONI CONIS2], S, R, E, s) iff thereis a y, sit. Val x,y), CONI, S, R, E, s) and Val(y, S, S, R, E, s) c. Val x,y), [CONIa], S, R, E, 0') iffVal x,y), a., S, R, E, 0') (wherea is 'before', ' after', or ' when' ) d. Val x,y), 'beforeS , R, E, 0') iff x is earlierthany e. Va1 x,y}, ' after' , S, R, E, 0' ) iff x is later thany f. Val x,y}, ' when' , S, R, E, 0') iff x temporallyoverlapswith y. 7. SeeappendixT4 for detailsof the derivation. 8. Here ' Infl ' standsfor verbalinflection. 9. SeeHigginbotham 1989for further motivation for the introduction of eventpositions into nouns. 10. The label is from Abusch1997. 11. Abusch( 1997; seeespeciallyp. 4) arguesthat it doesnot work, but I don' t find her . judgementson the crucial datapersuasive 12. SeeGiorgi andPianesi1998for discussion. 13. Proposalsfor the reductionof times to setsof eventsor propositionsarecommon in the philosophicalliteratureon time. See, e.g., Quine 1960.
Notesto pp. 87- 93 Chapter 6 1. The discussionthat follows owesmuchto Smith ( 1993) . 2. It alsoappearsto be the coursetakenby Mellor ( 1981) . However, aswe will seein a bit, Mellor proposesanotherway out. ' s andSmart' 3. Smith ( 1993) suggeststhat thereis a differencebetweenReichenbach s strategiesin that, whereasReichenbachultimately makesuseof the notion of the time or eventof this very utterance , Smartdispenseswith talk of timesor events. 4. HereI am ignoring anotherseriousdifficulty for the B-theorist: Preciselywhat does it meanto give detensedtruth conditions? If the truth conditionsareto be expressedin the metalanguage is English, andall expressionsof English are , and the metalanguage tensed,by virtue of whatpowerarewe supposedto strip the metalanguage of its tensed of suchvocabulary vocabulary? And if we can strip the metalanguage , is the remaining ? In addition, quite apartfrom our interestin languageone that we actuallyunderstand , thereis the very real questionwhethera tenseless giving a truth-conditionalsemantics languageis evenpossible. (SeeTeichmann1998for a discussion.) 5. Smith( 1993) putsthis asanepistemologicalpoint ratherthana logical one; heclaims we canseea priori that (9) is false, but not so with (8) . 6. SeeDavidson1967b, Evans1985a, andButterfield 1985for variousversionsof this idea. 7. I am gratefulto RichardLarsonandErnie Leporehere. An alternativestory(Ludlow 1997b) suggests thatwethink of (8) ascontaininga strictbiconditional- or , if oneprefers, asbeingin the scopeof a modaloperator, asin (i ) . (i ) o (An utteranceat time t of 'Thereis no spokenlanguage'is true iff thereis no spoken languageat t) ( Evenif we supposethatthetheoremsthemselvesareextensional , asin Davidson1967a andin LarsonandSegal1995, we might want to allow suchoperators , sincewe takethe theoremsto be derivedfrom whatwe know to be a trueT-theoryfor thelanguage . In that case, the hiddenoperatoris somethinglike "we know it is a theoremof trueT-theoryfor ' Englishthat. . .. ) . Accordingly, we needto be on alert for scopeambiguities.Intuitively, what is amisshereis that we want the actualtokeningto simply fix the referencetime, but we don' t want to be committedto the absurdview that necessarilythereis an utterance of the tokenin questionat the time that thereis no spokenlanguage.If we want to capturethis intuition we needto makesurethat the eventquantificationhasscopeover the modaloperatorin (i ). Payingattentionto the relative scopeof the utterance-event quantificationandthe modaloperator, we get the following truth conditionsfor (8) : (ii ) ' is suchthat o u is true iff An utteranceu at time t of ' Thereis no spokenlanguage there ( is no spokenlanguageat t) As above, this fixes the evaluationto the time of utteranceof the token, but it keepsthe quantificationoutsidethescopeof themodaloperator.Therestof thereasoningproceeds asabove. 8. It does, however,entail someconsequences for how we think aboutthe T-theoryintroduced in chapter5. Now it is no longersentences thatareto beevaluablebututterances ,
224
Notesto pp. 93- 103
andthe structuraldescriptionof the sentencein questionwill not be on the LHS of the ! Consideraxiom (i ) : biconditionalbut will be outsidethebiconditionalaltogether (i ) Val( T, LpNP 1'], S, R, E, 0' ) iff for somex , e, Val(e, 1', S, R, E, 0' ) andx is the agentof e andVal(x, NP, S, R, E, 0' ) We will haveto rewrite this asthe conditionalizedaxiom in (ii ) if we areto incorporate the lessonslearnedfrom our discussionof token-reflexives. (ii ) If u is utteranceat S, of [IPNP 1'], thenVal( Tu , S, R, E, 0' ) iff for somex,e, Val(e, 1', S, R, E, 0' ) andx is the agentof e andVal(x, NP, S, R, E, 0' ). 9. The proposalin Higginbotham1995, after all, is for tensedthoughtsandnot merely . tensedutterances 10. SeealsoHigginbotham1986b. 11. I realizethatappealsto psychologyarecommonenoughamongB-theorists(andhave beensinceRussell1915), but suchproposalsseldomgetbeyondhandwaving. Constructive proposalsarefew andfar between,but whentheyaremadeit becomesclearthattheyface conceptualdifficulties in their ownright. Atkins ( 1986, p. 98), for example,suggeststhat is simply our experienceof entropy: " We havelookedthrough our time consciousness the window on to the world providedby the SecondLaw, andhaveseenthe nakedpurof nature. The deepstructureof changeis decay; the spring of changein poselessness all its formsis thecorruptionof thequalityof energyasit spreadschaotically,irreversibly, andpurposelesslyin time. All change, and time' s arrow, point in the direction of corruption . The experienceof time is the gearingof the electrochemicalprocesses in our " brainsto this purposelessdrift into chaosas we sink into equilibrium and the grave. ' more detail. to be out in rather spelled Obviously the core of Atkins s position needs Justhowis it thatwe areconsciousof entropy? How doesentropyleaveanelectrochemical of time, and in imprint on our brains? How doesany of this give rise to consciousness particularto theexperienceof time asbeingfundamentallytensed?Needlessto say, these questionsare profoundly difficult , andAtkins has not provided any suggestionas to . Nor, frankly, hasanyoneelse. how they may be answered Chapter7 1. 1 am indebtedto Ed Keenanfor discussionon this matter. 2. Nor would this simply imply that 1am glad that a certainpropositionwasno longer true. My stateof relief is not aboutthe characterof a certainabstractproposition. A better " way of glossingthis is as I am in a stateof relief, and"the contentof the stateis that the event of [][(3e)(e is my havinga root canal)][] is past. . llitimately onewouldwant 3. Here1treatthetemporalconnectives syncategonnatically a categonnatictreatment.1amtakingthefonner approachfor expositorypurposesonly. 4. Of coursethis strategyrunsafoul of assumptionsthat syntacticstructuremustbe binary branching, but the samegeneralideacanbe cashedout in binary branchingstructures . Oneway of doing this (perhapsnot thebestway) is to supposethat thereareboth adverbphrases(AdvPs) and tensephrases( TPs) and to hypothesizestructureslike [TP Tense[Adv PAdv S]], andto write axiomslike the following:
Notesto pp. 103- 124 Val( T, [TPTenseAdvP]) iff thereis an x, suchthatVal(x , Tense) andVal(x, AdvP) Val(x, [Adv PAdv S]]) iff thereis an x, suchthatVal(x, Adv) andx=[] S[] 5. For a clearstatementof this objection, seeComrie 1985. 6. Obviously, if ' is an eternal(e.g. mathematical ) propositionthenit may well be both future andpast. Herewe areconcernedwith boundedevents. 7. Although the objection turns on a theory in which the operatorsapply to events rather than to propositions, the objection is readily adaptedto propositions. The idea would be that a givenpropositionis future, thenpresent,andthenpast. 8. That is, from the rejectionof the idea that thereare only two possibletruth values. Truth-valuegapsarealsopossible. ChapterS 1. For discussionseePrior 1971; Evans1982; Neale1990; Ludlow andNeale 1991. 2. For example, by Strawson( 1950) and Donellan ( 1966) . For criticism of the view, seeKripke 1976andNeale 1990. 3. Of courseRussell( 1905) wasthe first to introducethe possibility of namesas descriptions in disguise. Numerousothershaveadvocatedsucha view. SeeKripke 1980 for criticism. 4. Of course, thoughthey often areinfelicitous, exampleslike ( 15b), ( 16b), ( 17b), and ( 18b) canon occasionbe perfectlynatural. In fact, in the computersciencecommunity thetaskis oftennot to explaintheinfelicity of suchexamplesbut ratherto find techniques ' " " by which the subsequentpronouns intendedspecification canbe detennined. For a discussionof this point seep. 317 of Sidner 1983. 5. The proposalof Evans( 1977) differs from the othersin that, while suchpronounsdo not standproxy for descriptions , theyhavetheir referentsfixed (rigidly ) by description. 6. Amongthe varioustheoriesof descriptivepronouns,therearea numberof proposals as to how descriptivecontent is to be recovered. In somecasesit is to be recovered pragmatically; in othersit is to be recoveredby a generalsyntacticalgorithm. 7. Axioms canalsobe written for binary branchingstructureslike the following:
8. Oneideawould beto introducenumericalsubscriptsinto theseoperators(e.g., to indicate "N " daysago ) andto developthe semanticsaccordingly.
226
Notesto pp. 128- 139
for discussionof the material 9. I am grateful to GregRay and Murali Ramachandran in this section. ? Partee( 1984) suggeststhat (i ) 10. Are therealsocasesof temporaldonkeyanaphora . below would be an exampleof E-temporaldonkeyanaphora (i ) WheneverMary telephonedon a Friday, Samwasasleep. It seemsto me that this is mistaken.The problemis that the tensevariablein ' [Samwas ' ' ' ] seemsto bestraightforwardlyboundby wheneverMary telephonedon a Friday , asleep ' ' not by someembeddedtemporalquantifier (i.e., not by a Friday ) . The differenceis that donkey anaphorahasthe pattern shownin (ii ) below, where bracesindicate the . donkeyanaphor
(ii) ' [All x: R(x, [some y: Py])]R(x, {y}) 's of (iii). ButPartee followsthepattern example iii ( ) t': Pt'])]R'(y, {t}) [All t: R(x, y, [some The problemof courseis that the temporalanaphorin (iii ) is anaphoricon t ratherthan e. It is simply a caseof bound variableanaphora.I haveyet to discovera convincing . caseof temporaldonkeyanaphora " 11. Notice that in (61), aswith stand-alonesentences , thereis a when-clause( during the Iranianrevolution" ) in the relativeclause. 12. I amprobablyconcedingtoo muchhereby callingtheneededstructureabstract , since I doubtthat it is any moreabstractthanthe sort of structurewe posit everyday- beginning . Strictly speaking,thoseboundariesarenot pronounced with sentenceandwordboundaries either. It remainsa mysteryto me why the positing of suchstructuresshould meetresistancein certainquarterswhile the liberal positingof referenceto abstracta is allowedto passwithout complaint- asthoughreferencecamefor free! 13. Thesewerethe solutionsinitially exploredby Smith ( 1975), Ladusaw( 1977), and Dowty ( 1982). 14. This is achallengewhichI takeseriouslyandwhich will haveto bemetat somepoint. Furthermore, arguingfor the plausibility of thesestructureswithin current linguistic , theoryis not a trivial enterprise.As an illustrationof the kind of effort that is necessary seethe treatmentof implicit complementclausesin Larsonet at. 1997. The arguments for implicit when-clausesandimplicit relativeclauseswill be no lesssubtleandno less intertwinedwith the principlesof currentlinguistic theory. 15. Here I am thinking of the argumentfor implicit complementclausesfound in den Dikken et at. 1996andin Larsonet at. 1997. Chapter9 ' ' ' ' I . Not necessarily , since, aswe sawin chapter7, it is possibleto treat before and after ' before' and ' after' that the A theories is . The critical for simply thing disquotationally not be cashedout in terms of a precedencerelation holding betweeneventson aBseries time line.
Notesto pp. 139- 147
227
2. For example, a greatdealof psychologicalresearchhasbeendevotedto the question of whether ' before' or ' after' is acquiredfirst. This researchmay initially appearunlikely to sheda greatdealof light on whethertheB-theoryor theA -theoryis correct; however . Forexample,it is possibleto arguethatif ' before' , this appearance maybedeceptive ' ' and after essentiallyexpressthesamerelation(as, for example,theydo accordingto the B-theory), thenthe acquisitionof thesetwo termsshouldoccur at the sametime. If so, thenthe questionof whetheronemight be moreeasilylearnedthanthe othermight become significant. It might suggestthatthebefore/afterrelationis not a primitive afterall, and that the meaningsof ' before' and ' after' have rather different fine structure. Unfortunately,theresultshereareinconclusive.Early studiesby EveClark ( 1970, 1971) ' ' ' ' suggestedthat childrenlearned before prior to learningthe meaningof after . Clarke further hypothesizedthat this could be accountedfor if one supposedthat thesepredicates weredecomposibleinto moreprimitive semanticfeatures(including, on her theory , +/- simultaneousand+/- prior). ThoughClarke' s findingshavebeenconfinned by somestudies(e.g. We'll and Stenning 1978; Munro and Wales 1982), a great deal of subsequentresearch(Amidon and Carey 1972; Kelier-Cohen 1974, 1987; Frenchand Brown 1977; Coker 1978; Goodz 1982; Trosborg1982) hasshownno consistentorder of acquisition. Thesestudiesshowedthat otherfactorsmay be involvedin the order of acquisition- for example, whetherthe clausesare arbitrarily or logically ordered, or the orderof mentionof the eventsdescribed , or somecombinationof thesefactors. 3. Subsequent research(seee.g. French1986) hasreportedotherresults, althoughcaution is necessaryin interpretingthe results. Crucially, in the experimentsreportedby French, childrenshowedfacility with suchtermswhenreportingcausalor logical connections betweenevents. But this is not the sameassuggestingthat childrenhavefacility with thetemporalusesof ' before' and ' after' (which is of coursewhatis at issuehere). 4. Presumablythe psychologistsregardedthe ability to decenteras a certain form of achievement , thoughone can imaginephilosophicalcritiques suggestingthat it really amountsto our adoptinga certainalienperspectiveon the world. s. Among other matters, thereis the thorny problem of separatingout causalusesof ' when' asin " whenI touchthestoveI bum ' ( myself ) . As Kelier-Cohen( 1974) found, the causalandtemporalusesarenot well differentiatedin early child language. 6. I owethis sectionheadingto BrianBeakley.Thematerialon Merleau-Pontywasdrawn to my attentionby at leasthalf a dozencolleaguesandstudentsat StonyBrook. Chapter 10 1. Eachof thesequestionswould requirea book-length treatment. On the philosophy of space,for example,it shouldbeapparentthattheargumentfrom indexicalityemployed in chapter6 appliesjust aswell to spatialindexicalslike ' here' and ' there' . Accordingly, onemight anticipatethe collapseof spaceinto a point, or perhapsone might find support for a Leibnizianpictureof space.The issuesarevery subtle, however,anddemand , Mellor ( 1981) hasarguedthat the fact that no one very careful study. Parenthetically hasadvocatedspatialindexical predicatestells againsttheir applicationin the caseof time. Calling this the B-theorist' s tu quoqueargument, Craig ( 1996b) hasarguedthat spaceandtime arefundamentallydifferent in this respect.That is certainly a plausible response,but it seemsto me that it would at least be an interesting exerciseto treat
228
Notesto pp. 147- 149
spacein an A -theory-like manner(that is, building spatialco-ordinatesout of indexical ' ' ' ' predicateslike here , there , etc.). The obviousproblemis that wherewe all sharethe sametemporalperspective(modulorelativisticeffectsit is currentlynow for all of us), we do not sharethe samespatialposition, so ' here' , ' there' etc. allegedlycan' t be real featuresof physicalspace.The challengefor anA -theoryof spaceis thereforeto carry out the projectwithout collapsinginto a kind of extremesolipsism. 2. I am indebtedto Michael Potterfor discussionof this matter. 3. It appearsthat the A -theorist cannotappealto the sametruth-value links without , sinceall theselinks employwhat appearsto be quantificationovertimes. This changes problemis only superficial, however, sincetheA -theoristcould reinterpretthe quantifiershereassubstitutionalquantifiersovercalendardates.Theresultmightbe something like the following (wherethe substituendsof t aredates): AI ('v't2)[Past:S is true when [ . . .a ] iff (3tl )(S wastrue when [ . . .tl ])] B' ('v'tl )[ Fut:S is true when [ . . . tl ] iff (3a )(S will be true when [ .. . a ])] C' ('v't2)[S is true when [ . . . a ] iff ('v'tl )( Fut:S wastrue when [ . .. II ])] D' ('v'tl )[S is true when [ .. . tl ] iff ('v'a )( past:s will be true when [ . . . a ])] andso on. 4. Canall recordsof aneventbe erased? Couldn' t onearguethat it is physicallyimpossible to eraseall recordsof a crime? Imagine, for example, an explosionin deepspace. What would it taketo eraseall recordsof suchan event? The criminalscould certainly eliminateall witnessesneargroundzero, andthe criminalscould takememory- erasing pills to striketheir memoriesof the witness-eliminationprogramaswell asthe original explosion. But of coursethe problemis that informationaboutthe explosionis departing the sceneat the speedof light. The criminalscan' t catchup to that signal, soin principle the informationis out there. Now it might be objectedthat while the criminalscan' t catchup to the information, barringhelpfrom someonewhoreflectstheinformationbackto usneithercanwe, sothe evidenceis in principle out of our grasp. That might be a problemfor the thoroughgoing idealist, but we might try to arguethat we areonly committedto antirealism about the future and the past. The presentis anothermatter. Unlesswe moveto somesort of verificationistsemanticsit shouldbe enoughto say that the information is therein the present(hereI am overlookingrelativistictroublesinvolving the notion of simultaneity, aswell asthe epistemologicalburdensthis might placeon our semantictheory). What holdsfor eventsin deepspacecould thenbe arguedto hold for eventsthat take placein zippered-up roomsaswell. The criminalsmight soundproofthe room andfind a way of blocking electromagneticemissions , but thereis a genuinequestionof physics aboutwhethertheycanthereby" erase" the informationof what took placein that room. It is arguablethat the walls of the room cannotannihilatethe informationpresentin the room. At bestthey canabsorbit asa statechangeor bury it in noise. Reconstructingthe information would no doubt be a tall order, but it appearsto be possiblein principle.
Notesto pp. 149- 161
229
And if the room is later destroyed , well thenthe infonnation is still not annihilated, but now releasedinto the environmentalbeit with yet morenoise. I am not sureabouthow plausiblethis line of argumentis with respectto physical theory. At somepoint the whole universemay collapseinto a giant black hole. Do we really want to say that the evidenceof the crime is preservedeventhen? If we extend the questionto quantumphysicsthen the preservationof all information seemseven lesslikely, sincemeasurements takenbetweenthe time of the crime andthe time of our to reconstruct the crime will introducean elementof indeterminacy . attempt Moreover, evenif it couldbe shownthat all informationis preserveduntil theendof thephysicaluniverse,it certainlyseemsconceivablethatthereshouldbepossibleworlds wherethis is not the case. But thenour conceptof logical inferenceseemsfar too brittle . It shouldsurvivein worldswith differentphysicalproperties,shouldit not? Of course we aretalking aboutextremelygeneralphysicalpropertiessothematteris far from clear. As I haveno ideahow to tackle thesequestionsI passover them here. However, they arenodoubtwell worthpursuing.(I amindebtedto EmilianoTrizio andEmilianoBoccardi for discussionof thesematters.) 5. Thereis a question, however,asto whethertheA -theoristis compelled to accept( E) . Onesupposes thattheansweris yes, sincegiving it up appearsto entailgiving up thegeneral thesisthat currentfactsdeterminetruth. The issueis subtle, however, and perhaps there are ways for the A -theorist to block (E) . Until we have somejustification for blocking ( E), however, we needto proceedon the assumptionthat the truth-valuelinks (A )-( D) will drive theA -theoristto contradictionin situationswhereall evidenceof an actioncanbe erased. 6. For defensesof the standardview, seeWekker1976andDavidsen-Nielsen 1988. 7. For a very interestingdiscussionof the origin of futuretensein Romancelanguages , seeFleischman1982. 8. This is the only casewherethe identity breaksdown, althoughnot by much. Notice that onedifferencehereis that the intervocalic" b" hasdroppedout- a familiar process in the history of Romancelanguages . 9. ExamplesincludeSicilian andSouthernItalian dialectsandspokenPortuguese . 10. For discussions , seeJenkins1972(p. 73ft'.); Huddleston1976(p. 69ft'.); Coates1983 (p. 177ff.) . Thereis alsoa brief reviewof this literaturein Palmer1990(p. 163ff.) . 11. For the record, in Romancelanguagesthe conditionalmorphemesand the perfect (preterite) morphemescomefrom the sameLatin forms, and in many casesthe common origin is obvious. Considerthe following paradigmfrom Italian: passatoremoto cred -ei -esti -e (-ette) -emmo -esteer -erono 12. SeeChafeandNichols 1986; Palmer1986; Willett 1988.
condizionale cred -er-ei -er-esti -er-ebbe -er-emmo -este -er-ebbero
230
Notesto pp. 163- 174
13. Theindexicalcharacterof evidentialsshouldbeobvious. In theroot case, something that is first-handevidencefor you might be second-handevidencefor me. Likewise, I might haveperceptualevidencefor somethingin the beginning, but archivalevidence at anothertime. A similar storycanbetold for modaldispositionsin the world. Eachtime we return to the Wicked Witch ' s hourglass, it has a different dispositional property. Obviouslytheseremarksareloose, sinceI am usingsequentialvisits andtimesto illustrate the indexicalcharacterof evidentialsand dispositions. Strictly speaking, evidentiality and modaldispositionswill be the fundamentalnotionsthat underwriteour talk of timesandtemporalsequence . Appendix PI 1. Still , therearedifferencesin the parametricsettingsof differentI -languages , and, although we shouldnot exaggerate thesedifferences,I think it is flatly falseto supposethat everythingis in principle translatable- at least, if preservationof senseand shadeof ' meaningare importantto us. On this scoreJackendoffs claimsare completelyat odds with whatprofessionaltranslatorsknow. For a historicaldiscussionof translationtheory, seeVenuti 1995. 2. In point of fact, asis arguedin Ludlow 1985, I think the storyis false. For the sakeof argument,however, let us supposeotherwise. 3. In fact, theseargumentsare at bestargumentsagainsta thesisthat would deny any structureto a sentenceapartfrom that visible in the written form of the sentence . 4. Indeed, doingjust that is a pet projectof mine. SeeLudlow 1998. 5. I havealreadyhinted that I doubt they tell againstthe thesisthat E-languagecould be the languageof thought. The ideais that onecanquite readily hold that E-language sentences haveLF representations with all the structurethat I -languagerepresentations aresupposedto have. Indeed, a numberof individualshold preciselythis thesis.
AppendixP2 1. Thematerialin thissectionis drawnfromLudlow(forthcoming ). 2. Kant1910 - 1983 volume 15 . 76-77 translation from Coffa 1991(p. 31). , , pp ; 3. Kant1910 - 1983 fromCoffa1991(p. 31). , volume15, p. 78; translation " 4. Wittgenstein the matter as follows in his 1915notebooks , p. 46): It (1961a explains is clearthattheconstituents of ourstatements canandshouldbeanalyzed of by means definitions to therealstructureof thestatement , andmustbe, if we wantto approximate . . . .Theanalyzed . Theanalysis propositionmentionsmorethantheunanalyzed makestheproposition morecomplicated thanit was,butit cannotandmustnotmakeit morecomplicated thatits reference wasfromthefirst. Whentheproposition isjust as ." , theit is completely complexasits reference analyzed I amindebted 5. In thefollowingdiscussion topersonal communication withChomsky . ' 6. Onepossible for Higginbotham hereis to render(4) alongthelinesof (4 ). response ' 4 ( ) Yourreportfailsto makeit clearthatonaverage a familyhas2.3 children .
topp. 174 Notes - 182
231
Thereare, however,potentialpitfalls in the (4') gambit. For starters,thereportdescribed in (4') neednot be aboutthe averagefamily at all ; it could very well be a reporton milk " consumptionin urbanareas.Is (4 ) a possibility? " (4 ) Your reporton the averagefamily fails to makeit clearthat on averagea family has2.3 children. Obviouslynot, sincethis bringsus backto the apparentquantificationover the average family. But perhapsthe upstairsNP canbe analyzedawayin anotherfashion, alongthe linesof (4"' ) . " (41) Yourreporton what, on average , the stateof familiesis, fails to makeit clearthat on average a family has2.3 children. But this might not do either, since(4) doesnot saythat the reportis on the generalstate of families. 7. Somemight claim to seea paradoxemerginghere. I am claiming that sciencehasno specialclaim on ontology, yet I -languagedoes. The apparentparadoxis that the nature of I -languageis supposedto be the product of scientific investigation. I am not sure thereis anythingmoreto the tensionherethanoneexpectsto find in naturalizationprojects - whether it is epistemologythat we are naturalizing or the philosophy of language andmetaphysics . ( Forsurveysof someissuesin naturalizingphilosophy,seeQuine 1969; Komblith 1994; ShimonyandNails 1987; Papineau1993.) 8. Both PaulHorwich andNoamChomskyhavesuggestedasmuchto me. 9. Nor doesit help to arguethat the conceptsmediatebetweenlanguageandthe world. In that case, the sameconsiderationsapply to the concept/world relation. If conceptual structureunderwritesour metaphysicalintuitions about I -substances , then our intuitions aboutI -substances will provideinsightsinto our conceptualstructure,andour conceptual structurewill provideinsightsinto the structureof I -language.For the record, I considerconceptualstructureto be a dispensable" middleman" here. Appendix Tl 1. Here I ignore issuesaboutthe proper formalism for representinglinguistic phrase markers.Among efforts to addressthis issue, seeLasnik andKupin 1977. 2. SeeKayne 1984for a defenseof the thesisthat only binary branchingstructuresare possiblein naturallanguage.SeeKayne 1994for a defenseof the view that they must be rightwardbranching. 3. SeeNealeandLudlow (forthcoming) for further discussionof this point. 4. Thereis an empiricalquestionasto whetherit is the optimal way of characterizing c-commandin naturallanguage . Othercandidatetheoriesareofferedby Aoun, Hornstein, andSportiche( 1981) andby May ( 1985) . The differencesbetweentheseproposalswill not be consequentialto anyclaimsmadein tJiisbook. 5. Theserepresentations area bit retrofrom theperspectiveof the" minimalistprogram" (Chomsky1995b). For a gooddiscussionof L F in theminimalistprogram , seeHornstein 1995.
232
Notesto pp. 191- 192
AppendixT2 I . Thenotationandthespecificexecution of theeventanalysis herearefromthe(EC) of LarsonandSegal1995. fragment 2. See,e.g., Higginbotham 1989,Parsons 1985 1991. , andParsons 3. Clearlythisis anoversimplification . In particular , anNP in subjectpositionis not , is notto givea full account alwaystheagent(at leastnotin English). Ourgoal, however of English,butmerelyto introducethebasicmechanisms of theeventcalculus , to if see be useful in the semantics of tense . ultimately theymay 4. Followingthe(EC) fragmentof LarsonandSegal1995.
Bibliography
Abusch,D. 1997.TheSequence of TenseandTemporalDeRe. LinguisticsandPhilosophy 20: 1- 50. Alston, W. 1964. Philosophyof Language.Prentice-Hall. Amidon, A ., and Carey, P. 1972. Why Five-Year01dsCannotUnderstandBefor~ and After. Journal of VerbalLearningand VerbalBehavior 11: 417- 423. Aoun, J., N. Hornstein, and D. Sportiche. 1981. Some Aspects of Wide Scope Quantification.Journal of Linguistic Research1: 69- 95. Aqvist , L. 1976. Formal Semanticsof Verb TensesasAnalyzed by Reichenbach. In Pragmaticsof Languageand Literature, ed. T. vanDijk . North-Holland. Atkins, P. W. 1986. TlID.e andDispersal: The SecondLaw. In TheNatureof7ime, ed. R. Flood andM . Lockwood. Blackwell. Bach, E. 1981. OnTlID.e, Tense,andAspect: An Essayin EnglishMetaphysics . In Radical Pragmatics,ed. P. Cole. AcademicPress. Bach, E. 1986. NaturalLanguageMetaphysics.In Logic, Methodologyand Philosophy of ScienceVII, ed. R. Marcuset al. North-Holland. Bach, E., andR. Cooper. 1978.TheNP-SAnalysisof RelativeClausesandCompositional Semantics . Linguisticsand Philosophy2: 145- 150. B. Beakley, 1991. The Structureof Scientific Explanationin CognitiveScience. PhiD. thesis, SUNY StonyBrook. Beakley, B., and P. Ludlow, eds. 1992. The Philosophy of Mind: Classical Problems / ContemporaryIssues. MIT Press. Bealer, G. 1982. Quality and Concept.Oxford UniversityPress. Content.PhilosophicalTopics Bilgrami, A. 1987.An ExternalistAccountof Psychological 15:191- 226. , P. 1989. ContentandSelf-Knowledge. PhilosophicalTopics17: 5- 26. Boghossian , B., andT. Briscoe,ed. 1989. Computational Boguraev Lexicography for NaturalLanguage . Longman. Processing Bonomi, A. 1997. Aspect, QuantificationandWhen-Clausesin Italian. Linguisticsand Philosophy20: 469- 514.
234
Bibliography
Brennan, S., and H. Clark. 1992. Lexical Choice and Local Convention. Manuscript, SUNY StonyBrook andStanfordUniversity. Broad, C. D. 1938. An Examinationof McTaggart's Philosophy. CambridgeUniversity Press. Burge, T. 1979. Individualism and the Mental. In Studiesin Epistemology, volume4, ed. P. Frenchet al. Universityof MinnesotaPress. . PhilosophicalReview95: 3- 45. Burge, T. 1986. IndividualismandPsychology Burge, T. 1988. IndividualismandSelf Knowledge.Journalof Philosophy85: 649- 663. Burge, T. 1998. Self-KnowledgeandMemory. In Externalismand Self-Knowledge,ed. P. Ludlow andN. Martin. CSLI Publications(CambridgeUniversityPress). Buridan, J. 1966. Sophismson Meaningand Truth. Appleton-Century-Crofts. Butterfield , J. 1985. Indexicals and Tense. In Exercisesin Analysis, ed. I . Hacking. CambridgeUniversityPress. : Finding Sensein Wingenstein's Tractatus. Carruthers, P. 1989. TractarianSemantics Blackwell. : An Essayin Philosophical Carruthers , P. 1996. Language,Thought, and Consciousness . . Press PsychologyCambridgeUniversity Camlthers , P., andP. Smith, eds. 1996. Theoriesof Theoriesof Mind. CambridgeUniversity Press. . AmericanPhilosophicalQuarterly Castenada , H.-N. 1967. IndicatorsandQuasi-Indicators 4: 85- 100. . Chafe,W., andJ. Nichols, eds. 1986.Evidentiality: TheLinguisticCodingof Epistemology Ablex. Chamberlain , I. 1904. Studiesof a Child. PedagogicalSeminary , A., andChamberlain 11: 264- 291. Chierchia , G. 1984. Topicsin theSyntaxandSemanticsof InfinitivesandGerunds.PhiD. thesis, Departmentof Linguistics, Universityof Massachusetts , Amherst. Chierchia, G. 1995. Dynamicsof Meaning: Anaphora, Presupposition , and the Theory of Grammar. Universityof ChicagoPress. Chierchia, G., andMcConnell-Ginet, S. 1990. Meaningand Grammar: an Introduction . MIT Press. to Semantics Chierchia, G and R. Turner. 1988. Semanticsand PropertyTheory. Linguistics and Philosophy11: 261- 302. Chierchia, G., and R. Turner, eds. 1989. Properties, Types, and Meaning, volume 1: FoundationalIssues. Kluwer. . Chomsky, N. 1975. Reflectionson Language.Pantheon . on Rules of Grammar . N. 1976 Conditions LinguisticAnalysis2: 303- 351. Chomsky, -Holland, 1977). and in on Form ( North Essays Interpretation Reprinted and Chomsky, N. 1981. Lectureson Government Binding. Foris. . Chomsky, N. 1986. Knowledgeof Language.Praeger
Bibliography
235
Chomsky, N. 1993. ExplainingLanguageUse. PhilosophicalTopics20: 205- 231. Chomsky, N. 1995a.LanguageandNature. Mind 104: 1- 61. Chomsky,N. 1995b. TheMinimalist Program. MIT Press. . In Syntax: Chomsky,N., andH. Lasnik. 1993. TheTheoryof PrinciplesandParameters An International Handbookof ContemporaryResearch,ed. J. Jacobset al. Walter de Gruyter. Reprintedaschapter1 of Chomsky1995b. Clark, E. 1970. How Young Children Describe Events in Time. In Advances in ' , ed. G. Floresd Arcais andW. Levelt. North-Holland. Psycholinguistics Clark, E. 1971. On theAcquisitionof theMeaningof BeforeandAfter. Journalof Verbal Learningand VerbalBehavior 10: 266- 275. ' Clark, E. 1973. What s in a Word? On the Child' s Acquisitionof Semanticsin His First , ed. T. Moore. AcademicPress. Language.In CognitiveDevelopment Coates,J. 1983. TheSemanticsof the Modal Auxiliaries. CroomHelm. Coffa, J. A. 1991. TheSemanticTraditionfrom Kant to Carnap. CambridgeUniversity Press. Coker, P. 1978. Syntacticand SemanticFactorsin the Acquisition of BeforeandAfter. Journal of Child Language5: 261- 277. Comrie, B. 1976. Aspect. CambridgeUniversityPress. Comrie, B. 1985. Tense . CambridgeUniversityPress. Cooper, R. 1979. The Interpretationof Pronouns. In Syntaxand Semantics10., ed. E. HenyandH. Schnelle.AcademicPress. Craig, W. L. 1990. God andRealTime. ReligiousStudies26: 335- 347. Craig, W. L. 1996a.Tenseandthe New B-Theoryof Language.Philosophy71: 5- 26. ' Craig, W. L. 1996b.TheNew B-Theory s TuQuoqueArgument. Synthese107: 249- 269. Cresswell, M . 1985. StructuredMeanings. MIT Press. Cresswell, M . 1990. Entitiesand Indices. Kluwer. Cromer, R. F. 1968. The Developmentof TemporalReferenceDuring the Acquisition of Language.PhiD. thesis, HarvardUniversity. Davidsen-Nielsen, N. 1988. HasEnglisha Future? Remarkson the FutureTense.Acta UnguisticaHa.fniensia21: 5- 20. Davidson, D. 1967a.Truth andMeaning. Synthese17: 304- 323. Reprintedin Inquiries Into Truthand Interpretation(Oxford UniversityPress, 1984). Davidson, D. 1967b. The Logical Form of Action Sentences . Reprintedin Essayson Actionsand Events(Oxford UniversityPress, 1980). Davidson, D. 1986. A Nice Derangementof Epitaphs. In Truth and Interpretation: on the Philosophyof Donald Davidson, ed. E. Lepore. Blackwell. Perspectives ' Davidson , D. 1987.KnowingOnes OwnMind. Proceedings of theAmericanPhilosophical Association60: 441- 458. Davies, M. 1981a.Meaning, Quantification . RoutledgeandKeganPaul. , Necessity
236
Bibliography
Davies, M. 1981b. Meaning, Structure, andUnderstanding . Synthese48: 135- 161. Davies, M. 1987. TacitKnowledgeandSemanticTheory: Cana FivePerCentDifference Matter? Mind 96: 441- 462. Davies, M ., andT. Stone, eds. 1995a. Folk Psychology: The Theoryof Mind Debate. Blackwell. Davies, M., andT. Stone, eds. 1995b.MentalSimulation: EvaluationsandApplications. Blackwell. den Dikken, MR . Larson, and P. Ludlow. 1996. IntensionalTransitiveVerbs. Rivista di Linguistica 8. Abridgedversionreprintedin Readingsin thePhilosophyof Language, ed. P. Ludlow ( MIT Press, 1997). Devitt, M. 1993. Comingto Our Senses . CambridgeUniversityPress. Dieks, D. 1988. SpecialRelativity and the Flow of Time. Philosophyof Science55: 456- 460. Donnellan, K. S. 1966. ReferenceandDefinite Descriptions.PhilosophicalReview77: 281- 304. , TimeAdverbs, andCompositionalSemanticTheory. Linguistics Dowty, D. 1982. Tenses and Philosophy5: 23- 55. Dreyfus, H., ed. 1982. Husserl, Intentionality, and CognitiveScience.MIT Press. Dummett,M. 1964.BringingAboutthePast.PhilosophicalReview73: 338- 359. Reprinted in Dummett 1978. Dummett, M. 1969. The Reality of the Past. Proceedingsof theAristotelian Society69: 239- 258. Reprintedin Dummett 1978. Dummett, M. 1973. Frege: Philosophyof Language.HarvardUniversityPress. Dummett, M . 1975. What Is a Theory of Meaning? In Mind and Language, ed. S. Guttenplan.Oxford UniversityPress. Dummett, M. 1978. Truthand OtherEnigmas. HarvardUniversityPress. Dummett, M. 1991. TheLogical Basisof Metaphysics . HarvardUniversityPress. Einstein, A. 1955. A Letter of Condolenceto the Family of M. Besso. Reprintedin H.D. Zeh, ThePhysicalBasisfor the Direction of1ime (Springer-Verlag, 1989) . En~, M . 1986. Towardsa ReferentialAnalysis of TemporalExpressions. Linguistics and Philosophy9: 405- 426. En~, M. 1987. AnchoringConditionsfor Tense.Linguistic Inquiry 18: 633- 657. . HarvardUniversityPress. , J. 1990. TheConceptof Logical Consequence Etchemendy Evans, G. 1977. Pronouns,Quantifiers, andRelativeClauses(I) . CanadianJournal of Philosophy7: 467- 536. Reprintedin Evans1985. . In Meaning and Understanding,ed. Evans, G. 1981. UnderstandingDemonstratives H. ParretandJ. Bouveresse . Walterde Gruyter. Reprintedin CollectedPapers(Oxford UniversityPress). Evans, G. 1982. TheVarietiesof Reference . Oxford UniversityPress.
237 Evans, G. 1985a.DoesTenseLogic RestUpon a Mistake? In CollectedPapers. Oxford UniversityPress. Evans, G. 1985b. SemanticTheory andTacit Knowledge. In CollectedPapers. Oxford UniversityPress. Fanner, D. 1990. Beingin lime: TheNatureof1ime in theLight of McTaggart's Paradox. UniversityPressof America. Field, H. 1991. MetalogicandModality. PhilosophicalStudies62: 1- 22. Fleischman , S. 1982. TheFuturein ThoughtandLanguage.CambridgeUniversityPress. Fodor, J. 1975. TheLanguageof Thought. Crowell. Fodor, J., andE. Lepore. 1997. MorphemesMatter: theContinuingCaseAgainstLexical . Draft manuscript,RutgersUniversity. Decomposition Fodor, J., and E. Lepore. 1998. The Emptinessof the Lexicon. Linguistic Inquiry 29, no. 2: 269-288 Forbes,G. 1996. SubstitutivityandtheCoherence of QuantifyingIn. PhilosophicalReview 105: 337- 372. Foster, J. 1976. MeaningandTruth Theory. In Truth and Meaning, ed. G. EvansandJ. McDowell. Oxford UniversityPress. Frege, G. 1956. The Thought. Mind 65: 289- 311. French,L. A. 1986.TheLanguageof Events.In EventKnowledge : StructureandFunction in Development , ed. K. Nelson. Erlbaum. French, L. A., and Brown, A . L . 1977. Comprehensionof beforeand after in Logical andArbitrary Sequences . Journal of Child Language4: 247- 256. Gale, R., ed. 1967. ThePhilosophyof1ime. Doubleday. Geach, P. 1962. Referenceand Generality. Cornell UniversityPress. Giorgi, A., andF. Pianesi. 1997. TenseandAspect. Oxford UniversityPress. Giorgi , A ., and F. Pianesi. 1998. Sequenceof Tense Phenomenonin Italian: A MorphosyntacticAnalysis. Manuscript. Gt) del, K. 1949.A RemarkAbouttheRelationship BetweenRelativityTheoryandIdealistic Philosophy. In Albert Einstein- Philosopher-Scientist, ed. P. Schilpp. Open Court. Reprintedin Yourgrau1990. . Hyde, E. Green,andS. Statlender . , H., S. E. Blumstein , J. Berko-Gleason , MR Goodglass 1979. The Effect of SyntacticEncodingon SentenceComprehension in Aphasia. Brain and Language7: 201- 209. Goodz, N. S. 1982. Is beforeReallyEasierto Understandthanafter? Child Development 47: 822- 825. Gottlieb, D. 1980. OntologicalEconomy : SubstitutionalQuantificationandMathematics . Oxford UniversityPress. Gould, S. J. 1980. ThePanda's Thumb. Norton. Gould, S. J., andR. Lewontin. 1979. The Spandrelsof SanMarco andthe Panglossian . Proceedingsof the Royal Society Program: A Critique of the AdaptionistProgramme London 205: 281 288. of
238
Bibliography
, J., andM. Stokhof. 1991.DynamicPredicate Groenendijk Logic. In LogicsinAl , European
-Verlag . WorhhopJEELlA 1990,ed. J. vanEijk. Springer Grimshaw Structure . MIT Press . , J. 1990.Argument
Guenthner . , F. 1979. Time Schemes , TenseLogic, and the Analysis of English Tenses In Formal Semanticsand Pragmaticsfor Natural Languages , ed. F. Guenthnerand S. Schmidt. Reidel. Haik, I. 1983. Syncategonnatic Adjectives. Manuscript, MIT. Hale, K., andKeyser,J. 1987. A View from theMiddle. LexiconProjectWorkingPapers 1O, Centerfor CognitiveScience,MIT. Hale, K., andKeyser,J. 1993. ArgumentStructure.In TheViewfrom Building 20: Essays in Linguisticsin Honor of SylvainBromberger , ed. K. Hale andJ. Keyser. MIT Press. Hall, H. 1982 . WasHusser . aRealistor anIdealist?In Hussert , Intentionality , andCognitive
Science . Mit Press . , ed. H. Dreyfus
Hayes, B. 1995. Metrical StressTheory: Principles and CaseStudies. University of Chicago Press. Helm , I . 1982. The Semantics of Definite and Indefinite Noun Phrases. PhiD . thesis , University of Massachusetts. Higginbotham , J. 1983. Is Grammar Psychological ? In How Many Questions: Essays in Honor of Sydney Morgenbesser, ed. L . Cauman et al. Hackett . Higginbotham , J. 1985. On Semantics. Linguistic Inquiry 16: 547- 594. ' Higginbotham , J. 1986a. Linguistic Theory and Davidson s Program in Semantics. In Truth and Interpretation : Perspectives on the Philosophy of Donald Davidson , ed. E. Lepore . Blackwell . Higginbotham , J. 1986b. Discussion , Peacockeon Explanation in Psychology . Mind and Language I : 358- 361. Higginbotham, J. 1989. Elucidations of Meaning . Linguistics and Philosophy 12: 465- 418. Higginbotham, J. 1990. Contexts, Models , and Meanings: A Note on the Data of Semantics. In Mental Representations: The Interface BetweenLanguage and Reality , ed. R. Kempson. Cambridge University Press. Higginbotham , J. 1991. Belief and Logical Fonn . Mind and Language 6: 344- 369. Higginbotham , J. 1995. Tensed Thoughts . Mind and Language 10: 226- 249. Hinrichs , E. 1981. Temporale Anaphora im Englishen . Staatsexamenthesis, University of Tiibingen .
Hinrichs, E. 1986. Temporal Anaphora in Discoursesof English. Linguistics and Philosophy9: 63- 82. Hinrichs, E. 1988. Tense,Quantifiers,andContexts.ComputationalLinguistics14: 3- 14. Hintikka, J. 1962. Knowledgeand Belief Cornell UniversityPress. Hintikka, J. 1969a.Semanticsfor PropositionalAttitudes. In PhilosophicalLogic, ed. J. Daviset aI. Kluwer. ' ' Hintikka, J. 1969b.On Kant s Notionof Intuition (Anschauung ) .in Kant s First Critique, ed. T. PenelhumandJ. Macintosh. Wadsworth.
Bibliography
239
Hintikka. J. 1972. KantianIntuitions. Inquiry 17: 341- 345. Hintikka, J. 1975.IntentionsforIntentionalityandOtherNewModelsforModality. Reidel. Hofmann, F. 1995. Externalismand Memory. Manuscript, Departmentof Philosophy, Universityof Ttibingen. Hornstein, N. 1984. Logic as Grammar. MIT Press. Hornstein, N. 1990. As lime GoesBy. MIT Press. Hornstein, N. 1995. Logical Fonn: from GB to Minimalism. Blackwell. Huddleston,R. 1976. An Introductionto EnglishTransformationalSyntax. Longman. Husserl, E. 1972. Ideas: GeneralIntroductionto Pure Phenomenology . Collier. latridou, S. 1996. The GrammaticalIngredientsof Counterfactuality . Manuscript, MIT. loup, G. 1977.SpecificityandtheIntelpretationof Quantifiers.LinguisticsandPhilosophy 1: 233- 245. Isard, S. 1974. WhatWouldYou HaveDoneIf . . . TheoreticalLinguistics 1: 233- 255. Izvorski, R. 1997. The PresentPerfectasan EpistemicModal. Proceedingsof SALT7, StanfordUniversity. Jackendoff,R. 1972. SemanticInterpretationin GenerativeGrammar. MIT Press. Jackendoff,R. 1983. Semanticsand Cognition. MIT Press. Jackendoff . l.i Jnguages . MIT Press. , R. 1992 of theMind: EssaysonMentalRepresentation Jackendoff, R. 1993. Patternsin the Mind: Languageand HumanNature. Harvester Wheatsheaf . Jenkins, L. 1972. Modality in EnglishSyntax. PhiD. thesis, MIT. : , O. 1940.A ModernEnglishGrammaron Historical Principles. Copenhagen Jesperson . Ejnar Munksgaard -Laird, P. 1983. Mental Models. CambridgeUniversityPress. Johnson ' ' Kamp, H. 1971. Fomlal Propertiesof Now . Theoria37: 227- 273. . In Truth, Interpretation , Kamp, H. 1984.A Theoryof TruthandSemanticRepresentation and Infonnation, ed. J. Groendijket al. Foris. Kamp, H., andU. Reyle. 1993. FromDiscourseto Logic. Kluwer. Kant, I. 1910- 1983. Kantsgesammelte Schriften, volumes1- 29. De Gruyter& Reimer. . Manuscript,UCLA. Reprintedin Themes Kaplan, D. 1977. Demonstratives from Kaplan, ed. J. Almog et al. (Cornell UniversityPress, 1989) . . Journal of PhilosophicalLogic 8: Kaplan, D. 1979. On the Logic of Demonstratives 81- 98. . In Yourgrau1990. Kaplan, D. 1990. Thoughtson Demonstratives Karttunen, L . 1976. DiscourseReferents. In Syntaxand Semantics7: Notesfrom the , ed. J. McCawley. AcademicPress. Linguistic Underground Katz, J. 1972. SemanticTheory. Harper. Katz, J. 1981. Languageand OtherAbstractObjects. RowmanandLittle field. 39: 170- 210. Katz, J., andFodor, J.A. 1963.TheStructureof a SemanticTheory.l.i Jnguage
240
Bibliography
Katz, J., andPostal, P. 1964. An IntegratedTheoryof LinguisticDescription. MIT Press. and Binary Branching. Foris. Kayne, R. 1984. Connectedness Kayne, R. 1994. TheAntisymmetryof Syntax. MIT Press. Kelier-Cohen, D. 1974. TheAcquisitionof TemporalReferencein PreSchoolChildren. Ph. D. thesis, SUNY Buffalo. Kelier- Cohen . Journal , D. 1987. ContextandStrategyin AcquiringTemporalConnectives of PsycholinguisticResearch16: 165- 183. Kneale, W. 1949. Is Existencea Predicate ?In Readingsin PhilosophicalAnalysis, ed. H. FeigelandW. Sellars. Appleton-Century-Crofts. Komblith, H., ed. 1994. NaturalizingEpistemology , secondedition. MIT Press. Kratzer, A. 1989. An Investigationof theLumpsof Thought. LinguisticsandPhilosophy 12: 607- 653. Kretzman, N. 1976. Aristotle on the Instantof Change.Proceedingsof theAristotelean ) : 91- 114. Society50 (supplement Kripke, S. 1976. Is Therea ProblemAbout SubstitutionalQuantification? In Truthand Meaning, ed. G. EvansandJ. McDowell. Oxford UniversityPress. Kripke, S. 1979. A PuzzleAbout Belief. In Meaningand Use, ed. A. Margalit. Reidel. . HarvardUniversityPress. Kripke, S. 1980. Namingand Necessity Ladusaw , W. 1977.SomeProblemswith Tensein PfQ . TexasLinguisticsForum6: 89- 102. Larson, R., andP. Ludlow. 1993. InterpretedLogical Forms. Synthese95: 305- 356. Larson, R., andG. Segal. 1995. Knowledgeof Meaning. MIT Press. Larson, R., M. denDillin , andP. Ludlow. 1997. IntensionalTransitiveVerbsandAbstract ClausalComplementation . Manuscript Amsterdam . , SUNYStonyBrookandFreiUniversiteit Lasnik, H., and J. Kupin. 1977. A RestrictiveTheory of TransformationalGrammar. TheoreticalLinguistics4: 173- 196. Lepore, E. 1983. What Model TheoreticSemanticsCannotDo. Synthese54: 167- 187. . Synthese48: 121- 133. Lepore, E., andB. Loewer. 1981. TranslationalSemantics E. B. and Loewer . 1983 . Three Trivial Truth Theories . CanadianJournal of Lepore, , 13: 433 447. Philosophy . In NewDirectionsin Semantics , Lepore, E. andB. Loewer. 1987. DualAspectSemantics ed. E. Lepore. AcademicPress. , Lepore, E., and K. Ludwig . 1998. Truth Conditional Semanticsfor Tense. In Tense 1imeand Reference , ed. Q. Smith. Oxford UniversityPress. Lewis, D. 1969. Convention . HarvardUniversityPress. Lewis, D. 1972. GeneralSemantics . In Semanticsof Natural Language,ed. D. Davidson andG. Harman. Reidel. Lewis, D. 1975a.LanguageandLanguages . In Language , Mind and Knowledge,ed. K. Gunderson . Universityof MinnesotaPress. Lewis, D. 1979. Score-keepingin a LanguageGame. Journal of PhilosophicalLogic 8: 339- 359.
Bibliography
241
Lewis, H. A. 1985. SubstitutionalQuantificationand NonstandardQuantifiers. Nous 19: 447- 451. Ludlow, P. 1985. The SyntaxandSemanticsof ReferentialAttitude Reports. PhiD. thesis , ColumbiaUniversity. . Technical Ludlow, P. 1989.Psychology , Semantics , andMentalEventsUnderDescriptions of and Information 135 Center for the CSLI 89 , Stanford , Language Study Report University. Ludlow, P. 1992. FormalRigor andLinguisticTheory. Natural LanguageandLinguistic Theory10: 335- 344. ?ElectronicJournalof AnalyticPhilosophy Ludlow, P. 1993. Do T-theoriesDisplaySenses 1: 4. . Linguae Stile29: 165- 183. Ludlow, P. 1994.Conditionals , Events,andUnboundPronouns and Self Ludlow, P. 1995.SocialExternalism , , Knowledge Memory. Analysis55: 157- 159. Ludlow, P. I996. SocialExtemalismandMemory: a Problem?ActaAnalytica 14. Reprinted in Ludlow andMartin 1998. Ludlow, P. ed. 1997a.Readingsin the Philosophyof Language.MIT Press. TruthConditionsfor TokenLudlow, P. 1997b.Semantics , Tense,andTune: OnTenseless . Protosociology10: 190- 195. ReflexiveTensedSentences Ludlow, P. 1998. LF andNaturalLogic. Manuscript, SUNY StonyBrook. ? In Chomskyand His Ludlow, P., forthcoming. ReferentialSemanticsfor I -Languages Critics, ed. N. HornsteinandL. Antony. Blackwell. . . CSU Publications Ludlow, P., andN. Martin, eds. 1998. ExternalismandSelf-Knowledge : In Defenseof Russell.Linguistics Ludlow, P., andS. Neale. 1991. IndefiniteDescriptions and Philosophy14: 171- 202. Malotki, E. 1983. Hopi lime : a Linguistic Analysisof TemporalConceptsin the Hopi Language.Mouton. May, R. 1977. The Grammarof Quantification. PhiD. thesis, MIT. May, R. 1985. Logical Form: Its Structureand Derivation. MIT Press. . LanguageResearch Clauses McCawley,J. 1974. OnIdentifyingtheRemainsof Deceased 9: 73- 85. Reprintedin Adverbs, Vowels, and Other Objectsof Wonder(University of ChicagoPress, 1979). " McDowell, J. 1978. On "The Reality of the Past. In Action and Interpretation, ed. C. HookwayandP. Pettit. CambridgeUniversityPress. McDowell, J. 1980. On the Senseand Referenceof a ProperName. In Truth Reality and Reference , ed. M . Platts. RoutledgeandKeganPaul. McNeill , D. 1979. TheConceptualBasisof Language.Erlbaum. McTaggart,J. 1908. The Unreality of Time. Mind 68: 457- 474. , volume2. CambridgeUniversityPress. McTaggart,J. 1927. TheNatureof Existence
242
Bibliography
. TranslatedasThe Theory of Objectsin Meinong, A. 1904. Uber Gegenstandstheorie Realismand the Backgroundof Phenomenology , ed. R. Chisolm( FreePress, 1960) . Mellor, D. H. 1981. Real1ime. lt ..nTPress. Merleau-Ponty, M . 1962 . ThePhenomenology . RoutledgeandKeganPaul. of Perception Monk, R. 1996. BertrandRussell: TheSpirit of Solitude. FreePress. Montague,R. 1960. OntheNatureof CertainPhilosophicalEntities. Monist53: 159- 194. Reprintedin FormalPhilosophy. Montague, R. 1974. FormalPhilosophy. YaleUniversityPress. Munro, J., andWales, R. 1982. Changesin the Child' s Comprehension of Simultaneity andSequence . Journal of VerbalLearningand VerbalBehavior21: 175- 185. Neale, S. 1990. Descriptions. lt ..nTPress. Neale, S., and P. Ludlow. Forthcoming. Syntax. In The CambridgeEncyclopediaof Philosophy. CambridgeUniversityPress. Nirenberg, S., andV. Raskin. 1987. The SubworldConceptLexicon and the Lexicon ManagementSystem. ComputationalLinguistics 13: 276- 289. Nolan, R. 1970. Foundationsforan AdequateCriterion of Paraphrase . Mouton. Palmer, F. R. 1986. Mood and Modality. CambridgeUniversityPress. Palmer, F. R. 1990. Modality and theEnglishModals, secondedition. Longman. , D. 1993. PhilosophicalNaturalism. Blackwell. Papineau Parmenides . 1984. Parmenidesof Elea. Universityof TorontoPress. Parsons , C. 1971a.A Pleafor SubstitutionalQuantification. Journal of Philosophy68: 231- 237. Parsons . PhilosophicalReview80: 158- 159. , C. 1971b.OntologyandMathematics Parsons . Manuscript, Universityof Mass. atAmherst. , T. 1978. PronounsasParaphrases Parsons T. . , 1980 NonexistentObjects.YaleUniversityPress. Parsons , T. 1985. UnderlyingEventsin the Logical Analysisof English. In Actionsand Events: Perspectiveson the Philosophy of Donald Davidson, ed. E. Lepore and B. McLaughlin. Blackwell. Parsons , T. 1991. Eventsin the Semanticsof English. ~ Press. Partee, B. 1973. SomeStructuralAnalogiesBetweenTensesand Pronounsin English. Journal of Philosophy70: 601- 609. Partee, B. 1974. Opacity and Scope. In Semanticsand Philosophy, ed. M . Munitz and P. Unger. NYU Press. Partee , B. 1984.NominalandTemporalAnaphora.linguistics andPhilosophy7: 243- 286. Peacocke , C. 1986. Explanationin ComputationalPsychology: Language,Perception, andLevel 1.5. Mind and Language1: 101- 123. Perry, J. 1969. The Problemof the EssentialIndexical. Nous 13: 3- 21. . PhilosophicalReview86: 474- 497. Perry, J. 1977. Fregeon Demonstratives
Bibliography
243
Pinker, S. 1994. TheLanguageInstinct. William Morrow. Pinker, S., andP. Bloom. 1990. NaturalLanguageandNaturalSelection.Behavioraland Brain Sciences13: 707- 784. Prior, A. N. 1967. Past, Presentand Future. Oxford UniversityPress. . Oxford UniversityPress. Prior, A. N. 1968. lime and Tense Prior, A. N. 1971. Objectsof Thought. Oxford UniversityPress. , J. 1995. TheGenerativeLexicon. MIT Press. Pustejovsky Pustejovsky, J., and S. Bergier, eds. 1991. Lexical Semantics and Knowledge . Springer-Verlag. Representation Putnam, H. 1967. Time andPhysicalGeometry.Journal of Philosophy64: 240- 247. Putnam, H. 1975. The Meaningof ' Meaning' . In Language , Mind and Knowledge,ed. K. Gunderson . Universityof MinnesotaPress. Quine, W. V. O. 1953. OnWhatThereIs. In W. V. O. Quine, Froma Logical Pointof View. Harper& Row. Quine, W. V. O. 1960. Wordand Object. MIT Press. . ColumbiaUniversityPress. Quine,W. V. O. 1969. OntologicalRelativityandOtherEssays . PhilosophicalStudies Solutionto SubstitutionPuzzles Reiber,S. 1997.A Semiquotational 86: 267- 301. Reichenbach , H. 1947. Elementsof SymbolicLogic. Macmillan. Reichenbach , H. 1956. TheDirection of1ime. Universityof California Press. Reichenbach , H. 1958. ThePhilosophyof Spaceand lime . Dover. Reinhart, T. 1976. The SyntacticDomainof Anaphora. PhiD. dissertation,MIT. Richard, M . 1990. PropositionalAttitudes. CambridgeUniversityPress. of Knowledgeand Belief, ed. M. Rips, L. 1986. Mental Muddles. In TheRepresentation BrandandR. Harnish. Universityof Arizona Press. Ross, J. R. 1976. To Haveandto HaveNot. In Linguistic and Literary Studiesin Honor of ArchibaldHill , ed. E. Polomeet al. Peterde Ridder. ' Routley, R. 1980. ExploringMeinongs Jungleand Beyond. ResearchSchoolof Social Sciences , AustralianNationalUniversity, Canberra. Russell, B. 1905. On Denoting. Mind 14: 479- 493. Russell, B. 1910- 11. Knowledgeby Acquaintanceand Knowledge by Description. Proceedingsof the AristoteleanSociety. Reprintedin Mysticismand Logic (Allen and Unwin, 1917; Doubleday, 1957). Russell, B. 1915. On the Experienceof Time. Monist 25: 212- 233. Salmon, N. 1986. Frege's Puzzle. MIT Press. Schein, B. 1993. Plurals and Events. MIT Press. . Journalof Philosophy86: 73- 89. Segal, B. 1989. A Preferencefor SenseandReference Sellars, W. 1963. AbstractEntities. Reviewof Metaphysics16: 627- 671.
244
Bibliography
Semenza , C., andM. Zetin. 1989. Evidencefrom Aphasiafor theRole of ProperNames asPureReferringExpressions . Nature342: 678- 679. Shimony,A. 1993. TheTransientNow. In A. Shimony,Searchfor theNaturalisticWorldView, volumeII . CambridgeUniversityPress. : a Symposiumof Two Shimony, A., andD. Nails, eds. 1987. NaturalizingEpistemology Decades.Reidel. Sidner, C. 1983. Focusingin theComprehension of DefiniteAnaphora.In Computational Modelsof Discourse, ed. M. Brady andR. Berwick. MIT Press. Sklar, L. 1974. Space,lime , and Spacetime . Universityof California Press. Sklar, L. 1981.Time, Reality, andRelativity. In Reduction , TimeandReality, ed. R. Healey. CambridgeUniversity Press. Reprintedin Demonstratives, ed. P. Yourgrau(Oxford UniversityPress, 1990). Smart, J. J. C. 1963. Philosophyand ScientificRealism. Routledge& KeganPaul. Smart, J. J. C. 1966. The River of Time. In Essaysin ConceptualAnalysis, ed. A. Flew. RoutledgeandKeganPaul. Smith, C. 1975. TheAnalysisof Tensein English. TexasLinguistic Forum I : 71- 89. of TemporalExpressions Smith, C. 1978.TheSyntaxandSemantics in English. Linguistics and Philosophy2: 43- 99.
Smith . Kantifying In. Synthese 54: 261-274. , D. W. 1983 Smith, Q. 1993. Languageand lime . Oxford UniversityPres~. Soames, S. 1987. Direct Reference, PropositionalAttitudes, and SemanticContent.
Philosophical Topics15: 47- 87. , R. 1983.nme, Creation . CornellUniversityPress , andtheContinuum . Sorabji Stalnaker . , R. 1984.Inquiry. MIT Press . RigidityandContent . In Language , J. 1997a in Honor Stanley , Truth,andLogic: Essays . , ed. R. Heck. OxfordUniversityPress of MichaelDummen . NamesandRigidDesignation . In A Companion , J. 1977b to thePhilosophy Stanley of , ed. B. HaleandC. Wright. Blackwell Language -MinkowskiSpace -Tinle. Journalof Philosophy Stein,H. 1968.OnEinstein 65: 5- 23. Stowell . Manuscript , T. 1994.Syntaxof Tense of Linguistics , Department , UCLA. Strawson . Mind59: 320- 344. , P. F. 1950.OnReferring Strawson . Methuen . , P. 1959.Individuals Teichmann Possible ? Philosophical , R. 1998.Is a Tenseless Language Quarterly48: 176- 188. ' . Review Thompson,M. 1972- 73. SingularTennsandIntuitionsin Kant s Epistemology 26: 314 343. of Metaphysics ' ' ' ' ' . Trosborg, A. 1982. Children s Comprehensionof before and after Reinvestigated Journalof Child Language9: 381 - 402.
245 van Benthem, J. 1982. Later than Late: On the Logical Origin of the TemporalOrder. PacificPhilosophicalQuarterly63: 193- 203. vanFraassen , B. 1980. TheScientificImage. Oxford UniversityPress. Venuti, L . 1995. TheTranslator's Invisibility: A History of Translation. Routledge. Revisited: One, 1\' Io, or ThreeTemporalRelations.Acta Vickner, S. 1985. Reichenbach Linguistica Hafniensia19: 81- 98. . In Semanticsof Natural Language,ed. D. Wallace, J. 1972. On the Frameof Reference DavidsonandG. Hannan. Reidel. We'll, J., and Stenning, K. 1978. A Comparisonof YoungChildren' s Comprehension of TemporalRelations.In RecentAdvancesin thePsychology andMemoryfor Statements R. ed. of Language, CampbellandP. Smith. Plenum. Weist, R. M. 1986. TenseandAspect. In LanguageAcquisition, secondedition, ed. P. FletcherandM . Gannan. CambridgeUniversityPress. British English. NorthWekker,H. 1976. TheExpressionof Futurelime in Contemporary Holland. Wettstein, H. 1986. Has SemanticsRestedon a Mistake? Journal of Philosophy 83: 185- 209. Whorf, B. 1956. Language , Thought, and Reality. MIT Press. Willett , T. 1988. A Cross-Linguistic Surveyof the Grammaticizationof Evidentiality. Studiesin Language12: 51- 97. . Journal of Philosophy48: 457- 472. WilliamsD . C. 1951. The Myth of Passage Williams, E. 1983. Semanticvs. SyntacticCategories.Linguistics and Philosophy6: 423- 446. , 1914- 1916, ed. G. von Wright and G. Anscombe. Wittgenstein, L . 1961a. Notebooks Harper& Row. . RoutledgeandKeganPaul. Wittgenstein,L. 1961b. TractatusLogico-Philosophicus ' Wright, C. 1983. Freges Conceptionof Numbersas Objects.AberdeenUniversityPress. ' Wright, C. 1986.Theoriesof MeaningandSpeakersKnowledge.In Philosophyin Britain . SUNY Press. Today, ed. S. Shanker , Meaning, and Truth, secondedition. Blackwell. Wright, C. 1993. Realism Yourgrau,P. 1987. The Dead. Journal of Philosophy84: 84- 101. . Oxford UniversityPress. Yourgrau,P., ed. 1990. Demonstratives Yourgrau,P. 1991. TheDisappearanceof1ime: Kurt Godeland the Idealistic Tradition in Philosophy. CambridgeUniversityPress. . Reidel. Zalta, E. 1983. AbstractObjects: an Introductionto AxiomaticMetaphysics Zalta, E. 1988. IntensionalLogic and theMetaphysicsof Intensionality. ~
Press.
Index
Names Aqvist, L., 78 Bach,E., 65, 131 , S., 39 Bergier , G., 25 Berkeley Bloom,P., 21 , P., 153, 154 Boghossian , B., 39 Boguraev Bolzano , B., 171 , F. H., 46 Bradley Brennan , S., 56 Brentano , F., 22 Briscoe , T., 39 Broad , C. D., 106 , T., 96, 152, 153, 175 Burge Ca111lthers , P., 45, 46 Casteiiada , H.-N., 91 Chamberlain , A., 141 Chamberlain , I., 141 Chierchia , G., 37, 130 Chomsky , N., 13, 14, 17- 23, 38, 166, 168,
- 182 172 , 185 Clark , E., 139 Clark , H., 56 Coffa , J. A., 171 Comrie , B., 81 , 131 , R., 116 Cooper , W. L., 92 Craig Cromer , R. F., 140 , 141
Davidson , 191 , 31, 38, 73, 75, 152 , Do Davies , Mo,116 Descartes , Ro , 25, 152 Dieks , Do ,3 , 103 , D., 82, 84, 102 Dowty Dummett , , 3, 5, 38, 39, 44, 71, 76, 99, 109 140 , 148 , 151
Einstein , A., 1, 2 En~, M., 82, 130 Evans , G., 13; 14, 40, 44, 60, 61, 113, 114, 116, 129, 138 Fodor , J. A., 26, 28 , G., 44, 58- 61 Frege Gale,R., 2, 77 Geach , P., 116 Giorgi,A., 78, 8 , H., 140 Goodglass Gould.S. J... 21 Gottlieb , D., 74 Grimshaw , J., 39 , J., 130 Groenendijk Guenthner , F., 78
Hai'k, I., 173 Hale , K., 39 Hall,H., 75 Helm . -I... lIS.. 116 , J., 13,30, 39, 49. 59. 82. 83. Higginbotham 94, 120 , 131 , 174 , 173 Hinrichs , E., 78, 102 , 130 Hintikka , J., 72 Hofmann , F., 154 Hornstein , N., 78, 82, 172 Husserl , E., 75 Iamblichus ,1 Iabidou,S., 161 Ioup, G., 74 Isard,S., 161 Izvorski , R., 162
- 167 Jackendoff , R., 23, 28, 165 -Laird,P., 26 Johnson , 130 , H., 81, 125 Kamp
248 Kant , I., 4, 171 , 179 Kaplan , D., 7, 8, 13. 14,57-60.64.93 Katz , J., 28- 30 -Cohen Kelier , D., 141 , S. J., 39 Keyser Kneale , W., 89 , S., 71, 72 Kripke Ladusaw , W., 82, 84 Larson , R., 32,33,42,49, 51,54-58,61,62,69, 70, 93, 184 , 186 , 187 , 191 , 197 , 203,209 Lasnik , H., 21 , E., 29- 31, 36, 43 Lepore Lewis , D., 29, 115 , 116 Loewer , B., 36, 43 Ludlow , P., 49, 51, 54-56, 61, 62, 64, 113 , - 155 128 , 130 , 153 , 184 , 197 , 209 , 185 May,R., 182 -Ginet McConnell . S., 37 McDowell , J., 149 McNeill , D., 140 . 14.15.105 . , J. M. E., 1, 2, 8, 9, 12 McTaggart 108 - 135 , 133 , A., 71 Meinong Mellor , H., 1, 95, 96, 107 . 134 . 163 , 108 , 133 -Ponty - 144 Merleau , M., 142 , R., 102 Montague Neale , S., 113 , 116 Nirenberg , S., 39 Pannenides , 65 Parsons , C.. 75, 76 Parsons , T., 71, 81, 85, 116 , 120 , 121 . 123 Partee , B., 9, 103 , 104 , 112 , 129 , J., 7, 8, 13.57, 58, 60 Perry Pianesi , F., 78, 82 - 169 Pinker , S.. 21, 23-25. 165 Plato , 46 Postal , P., 28 - 109 Prior , A. N., 4, 5, 15,78. 100 , 107 . 133 , 135 , J., 39 Pustejovsky Putnam , H., 2, 3, 152, 175, 177 , W. V. 0 ., 66, 75, 117 Quine
Raskin , V., 39 Reichenbach . H.. 1.4. 14.78-80.89. 139 . 140 Reinhart , T., 182 , U., 81, 125 , 130 Reyle , L., 26 Rips Rout Iey,R., 71 Russell , B., I , 11,25, 112 , 113 , 217
Index
Schein , B., 120 , G., 32, 33, 42, 49, 54, 57, 58, 69, 70, Segal 93, 184 , 186 , 187 , 191 , 203 Sellars , W., 74 Semenza , C., 42 , R., 25 Shepard Sklar , L., 3 Smart , J. J. C., 89 Smith , C., 82, 84 Smith . D. W., 72 SmithQ., 92 Stein , H., 3 Stokhof , M., 130 Stowell , T., 82 Strawson , P. F., 112 Tarski , A., 185 vanFraassen , B., 178 Weist , R. M., 139 , 140 Wettstein , H., 64 Whorf , B. L., 65 WilliamsD . C., 106 , 142 Williams , E., 166 , L., 44-46, 171 , 172 Wittgenstein , C., 68, 148 , 150 , 151 , 155 Wright , P., 90 Yourgrau Zalta, E., 71 Zettin, M., 42 Subjects Adverbs andattachment , 53 ambiguity in disguise , 173, 174 , 125 locating aspredicates , 125 , 97, 98, 102, 103, 124- 129, 156, temporal 157 Aktionsarten , 123 , 84, 168, 173, 183, 184 Ambiguity Anaphora E-type, 10- 12, 114- 117, 131 andmemory , 153 plural, 69, 114 , 10, 11, 68, 69, Ill , 114- 117 pronominal , 8- 15, 101, 102, 111- 124, 129- 135, temporal 138, 140, 141, 157, 162 Antirealism, 109, 148, 150 A-series , 1- 15, 40, 77, 95, 96, 141, 144, 147- 149, 152, 162, 163 , 80, 81, 85, 86, 120- 123, 161 Aspect tenseto, 157 reducing
Index Behavior of , 56 , explanation Belief -sense common theoryof, 55, 56 of, 56 inferringmodels semantics of, 95 andslow-switching , 154 stabilityof, 156 natureof, 95 tensed Beliefatbibution , 43, 49-51, 54- 57, 61 fine-grainedmodelsof, 56 logisticsof, 55, 62 referential of, 55 component of, 55, 56, 61 syntactic component tacittheoryof, 55- 57, 61, 62 Beliefs , trackingabilitiesfor, 57, 62 B-series , 1- 8, 12, 15, 40, 77, 141, 147, 163 B-seriesrelations , 2, 77, 81, 126, 139 's dilemma B-theorist , 87, 93- 96
249 Evolution , 14, 20- 22 Externalism , 152- 156, 175 Future of, 157- 160 , elimination Generative grammar , 166
Gradualism . 21
I-concepts , 166, 178 Idealism , 75, 76. SeealsoAntirealism ILFs. SeeInterpreted LogicalFonns , 85, 86, 123 Imperfective paradox , II , 12, 53, 131- 133 Implicitclauses Indexicality andsemantics , 8, 87- 96, 137 anddescriptive content , 58 Indexical , 47, 90, 97- 102, 138 predicates Indexicals , 7, 13, 14, 20, 47, 168 character of, 7, 20, 57- 60 andcommunication , 2, 106, 107 , 59 Change contextsensitivity of, 57, 169 , 58, 64, 71 Cognitive significance of scholars andcounterfactuals , 164 , 59, 60 Community of, 14, 97- 102 , 185- 187 disquotational analysis Compositionality -Intensional andmodality , 23, 181 , 8, 14, 59, 60 system Conceptual Content senses of, 57- 61, 97, 98 , 6- 8, 13, 14, 15, 47, 80, 81, 87- 102, , 58 descriptive temporal 22, 57, 131, 152, 155, 156 137, 138 propositional token-reflexiveanalysis of, 87- 96 Co-referringtenns,48, 167, 169 Counterfactuals , 59, 60, 161 trackingabilitiesfor, 14, 61- 64 Inference , 69, 73, 147- 152, 164 DAR, 84 Intended model , 31 Davidsonian Intensional transitive verbs , 75 , 52- 54 program Dedicto/deredistinction Intentions , 117, 121 , 20, 22 research Demonstratives , 58, 59, 90. SeealsoIndexicals Interdisciplinary , 164 , 47- 57, 61, 62 , 10, 11, 60, 71, 113 Interpreted LogicalFonns Descriptions asarguments of tense Diachronic , 51, 100- 102 , 151, 152 disagreement construction of, 200 Dialects , 17, 18, 159 fonnaldefinitionof, 199 boundaries , 64 Disciplinary with 197-201 Discourse fonnalfragment , 129, 130 Theory Representation intoA-theoryformalfragment , , 160 incorporated Dispositions Doubleaccess 209- 213 , 84 readings DRT, 129, 130 I-substance , 175- 179 Entailment . 27 , 29 Knowledge of. 27 Cartesian Entrainment , 56 conception Environmentalism . SeeExternalism growthof. 151 Environments . 153- 156 preserved by memory second -order . 45. 152. 156 , 47 hyperintensional self-. 152- 156. 164 intensional , 53 semantic . 27- 29. 31. 38. 43- 45. 66. 70. 71. , 47, 48 opaque 88, 144 , 164 , 2, 72, 99, 100,138,144, ISO Epistemology shared Events , 61, 131 , 73, 74, Ill , 120- 123, 166 tacit, 27, 28, 44, 45, 55- 57 in B-series , 2, 77- 86, 203- 207 fonnalfragment for, 197-201 futureandpast,3, 4, 84- 86, 99, 100 Language Evidentials , 140 , 161- 163 acquireddeficits in, 41, 42, 138, 139
250 Language(cont.) acquisitionof , 15, 41, 42, 99, 138- 141, 144 communicativefunction of, 8, 13, 14, 17- 22,
59, 63 ascomputational , 17, 19 system E-, 17- 19, 169 facultyfor, 179 1-, 14, 17- 26, 159, 165- 171, 178 andidiolect, 19 asnaturalobject , 19 natureof, 8, 13, 14, 17- 26, 63 aspartof biologicalendowment , 17- 22 associalobject, 18 of thought , 14, 21- 28, 54, 95, 165- 169 andworld, 22, 28- 32, 46 - 179 I World , 171 Language Isomorphism Lexicon , 39, 71, 72, 166, 167, 177, 178, 181 , 156- 163, 173, 174, 181- 184 Linguistics LF, 73, 119, 126, 171- 173, 191 Logic. SeeInference Logicalfonn, 73, 119, 126, 171- 173, 191
Index -like, 50, 100, 156 proposition social, 18 , 67, 70, 71, 74, 75, 172, 173 Ontology , mythof, 106, 107 Passage Past of, 160- 163 , elimination -articulatory , 23, 181 Perceptual system states , 96 Perceptual PF, 23, 167, 181 , 15, 137, 141- 144, 178 Phenomenology
, 145 Philosophy ofaction . 147
, 171 analytic ofmind, 2, 164 of ageof giantsin, 164 passing of religion,2, 147 of science , 2, 178 of space , 147 Western , 157 , 167 Phonology , 2- 4, 177, 178 Physics Possible utterances , 92 Markerese ,' 28, 29 Possible worlds,60, 94 sparadox , 8- 12, 105 108, 133 135 Predicate calculus McTaggart , 186 Predication Meaning , 45, 46 externalist of, 152 Prediction conception intuitivenotionof, 175 of behavior , 55 of, 139 of future, 160 knowledge and, 165, 166 Presentation language , 50 Presentism picturetheoryof, 171 , 100, 109, 143, 148 semantics and, 64 Pronouns T-theoryand,43 , 10, II , 116 descriptive andreference , 15, 143, 144, 152- 156, 164 to properties Memory , 68, 69 intuitions calculus , 40, 71, 141- 144, 172, Propositional Metaphysical , 182 175, 178, 179 , 45, 46, 49, 67- 70 Property theory , 1- 6 attitudes Metaphysics , 10, 14, 20, 47- 57, 62 Propositional andsemantics , 4, 5, 15, 65-76, 95, 96, 141,163 Propositions , 108 -up, 4, 5 bottom of, 50 eternality Mebicaltheory , 21 , 10, 112 general Mind/ brain,54 asInterpreted , 50 LogicalFonns Modaldiscards , 94 singular, 10, 112 modals surface fonDof, 172 , 8, 10, 14, 71, 121 anddispositions P-substance , 160 , 175- 179 futuretenseto, 157- 160, 163 reducing , 137- 141 Psycholinguistics reference of, 160 Psychology andbeliefattribution , 56, 62 Names andsemantic , 39, 54 competence calendar andtheexperience of time, 2 , 125 andsemantics , 71, 72 rigidityand , 64
, 70-72 proper
Objects , 123 incomplete , 172, 176 misbehaving natural , 19 nonexistent , 71
Quantification basicT-theoryfragmentwith, 181- 189 andconditionalizedT-theorems , 93 generalizedquantifiertheory, 186
implicit, 74, 85 , 66 metalinguistic
Index overevents, 73, 74, 77, 84- 86, 121 overproperties,69 oversemanticvalues, 66, 76 overtimes, 77, 78, 84- 86, 100 substitutional , 74- 76, 127, 128 Quantifierraising(QR), 182- 184
Realism , 75, 149, 150, 177 scientific , 178 Reason of, 4, 179 , categories Reference vs. denotation , 112,113 to futureandpastevents , 15, 77- 86, 99, 104,
113 , 156 ,9 implicit ofmodals , 160
to properties , 46 , 103- 106, III , 130 temporal andtemporal adverbs , 124, 125 to times, 113, 140, 144, 156 to utterances , 89-93 , 47 Referring expressions ,3 RelativityTheory , 171, 172 Representation Rigidity,71, 72 Russellian , 49 propositions , 167 Scope asc-command , 182, 186, 187 of descriptions andmodals , 60, 71, 72 of quantified , 115, 117, 182- 184 expressions of tensemorphemes , 101, 104, 121 Self-referential , 89 expressions markers Semantic , 28, 29 Semantics , 13, 20, 27- 64 absolute , 14, 28, 31- 38, 57, 65, 82 A-theory , 15, 97- 141, 144,209-213 in, 54 austerity B-theory , 6- 8, 15, 77- 96, 99, 132, 137, 138, 203- 207 andepistemology , 72, 99, 100 , 58, 59 Fregean andlanguage /worldconnections , 27- 31, 36-38, 171- 173 andmetaphysics , 4, 5, 14, 64-76, 95, 84, 85, 95, 96, 99, 100, 141, 163, 171 -theoretic model , 28, 30, 31 -world, 50 possible andpsycholinguistics , 39- 42 referential , 14, 172- 179 structural , 28- 31 of Tense , 4, 47, 77- 135 truth-conditional , 14, 28, 31- 46, 50- 52, 57, 64- 76 values Semantic , 32- 36, 45, 46, 48-52, 66- 71, 74
251 Senses , 42- 46, 57- 59, 61, 97, 98 , 185 Sequences Simulation , 63 theory Slow-switching , 153- 156 S-substance , 175, 177 Stoffen , 75, 76 , 168, 169 Synonymy , 166, 167, 169, 181- 185, 191, 192, Syntax 197,203, 209
. 20-22 Teleological Explanation . 1- 7. 77. 106 Temporal becoming Temporal conjunctions . 11- 12.80.81. 101 . III . .
118- 135, 139- 141, 156 discourse , 6, 87- 91, 97, 98, 137 Temporal , 140 Temporal egocenbism , 96 Temporal perception Tense (s) absolute , 117, 118, 120 of, 15, 138- 141 acquisition asaspect , I57, 162 , 9, 78-80, 101, 104, 111- 124, 156 complex asconstituent of language , 166 elimination of, 15, 95, 96, 156- 163 embedded (nested ), 82- 84, 102, 103, 126 asevidentiality , 161- 163 asmodality , 157- 160 -likeobjects aspredicate of proposition , 50, 100, 156 of, 82- 84, 131- 133 sequence Tensed natureof world, 96 Tensed Nominals , 82, 130, 131 , 4, 166 Thought in animals , 24, 54 , 25, 26 imagistic of, 14, 21- 28, 54, 95, 165- 169 language andmentalmodels , 26 , 25, 26, 166- 169 nonlinguistic andsense , 58, 59 tensed natureof, 95, 96 Time A-seriesconception of. SeeA-series B-seriesconception of. SeeB-series of, 96 consciousness of, 141- 144 experience flowof, 106- 109, 142- 144 futureandpast, 151 of, 15, 137, 141- 144 phenomenology asphysicalproperty , 2- 4 tensed of, 2 conception untensed of, 1, 2, 77, (tenseless ) conception 78, 144, 163 -reflexives Token . SeeTruth-conditions Translation , 165 -worldIdentity Trans , 72
252 Truth indexical of, 152 conception timelessness of, 151, 155 Truthconditions , 7, 27- 46, 58, 59, 77- 96, 104, 129 tenseless , 77, 78, 91 token-reflexive , 89-96, 137 Truth-valuelinks, 147- 152, 155, 156 T-schema , 32 T-theories , 7, 14, 31- 46, 63, 64, 181-213 of, 38 compositionality conditionalized , 57, 58, 64 derivation rulesfor, 34, 35 natureof, 36, 37, 62, 124 disquotational of, 181-213 fragments , 32, 41 interpretive -worldconnections of, 36- 38, 65- 76, language 84- 86 modest vs. robust , 38, 39, 54, 62, 71, 98, 99 - 141, evidence for, 39- 42, 137 psycholinguistic 144 -reflexives andtoken , 89-93 , 172- 176 lYpe-mismatch argument , 84 Vagueness When-clauses . SeeTemporalconjunctions
Index