On Travis cases Agustin Vicente Linguistics and Philosophy ISSN 0165-0157 Volume 35 Number 1 Linguist and Philos (2012) 35:3-19 DOI 10.1007/s10988-012-9109-x 1 23 Your article is protected by copyright and all rights are held exclusively by Springer Science+Business Media B.V.. This e-offprint is for personal use only and shall not be self- archived in electronic repositories. If you wish to self-archive your work, please use the accepted author’s version for posting to your own website or your institution’s repository. You may further deposit the accepted author’s version on a funder’s repository at a funder’s request, provided it is not made publicly available until 12 months after publication. 1 23 Author's personal copy LinguistandPhilos(2012)35:3–19 DOI10.1007/s10988-012-9109-x RESEARCH ARTICLE On Travis cases Agustin Vicente Publishedonline:8May2012 (cid:2)SpringerScience+BusinessMediaB.V.2012 Abstract Charles Travis has been forcefully arguing that meaning does not determine truth-conditions for more than two decades now. To this end, he has devised ingenious examples whereby different utterances of the same prima facie non-ambiguous and non-indexical expression type have different truth-conditions depending on the occasion on which they are delivered. However, Travis does not argue that meaning varies with circumstances; only that truth-conditions do. He assumes that meaning is a stable feature of both words and sentences. After sur- veyingsomeoftheexplanationsthatsemanticistsandpragmaticianshaveproduced in order to account for Travis cases, I propose a view which differs substantially fromallofthem.Iarguethatthevariabilityinthetruth-conditionsthatanutterance typecanhaveisduetomeaningfactsalone.Tosupportmyargument,Isuggestthat wethinkaboutthemeaningsofwords(inparticular,themeaningsofnouns)asrich conceptualstructures;sorichthatthewayinwhichapropertyconceptappliestoan object concept is not determined. Keywords Meaning (cid:2) Truth-conditions (cid:2) Occasion-sensitivity (cid:2) Conceptual semantics (cid:2) Semantic knowledge (cid:2) World knowledge 1 Introduction Charles Travis has been forcefully arguing that meaning does not determine truth- conditions for more than two decades now. To this end, he has devised ingenious A.Vicente(&) DepartmentofLogicandPhilosophyofScience,UniversityoftheBasqueCountry,UPV/EHU, Avda.Tolosa,20080SanSebastia´n,Spain e-mail:[email protected] A.Vicente Ikerbasque:BasqueFoundationforScience,AlamedaUrquijo,36-5,PlazaBizkaia,48011Bilbao, Spain 123 Author's personal copy 4 A.Vicente examples where different utterances of the same prima facie non-ambiguous and (in the given context) indexically non-problematic expression type have different truth-conditionsdependingoncontextualfactors,orashewouldhaveit,dependingon theoccasiononwhichtheyareuttered.1Thishasledhimtoarguethattruth-conditions areoccasionsensitive.Travis’singenuousexamplescanbefoundalmosteverywhere inrecentdebatesinthephilosophyoflanguage,oratleastinthosedebatesthatfocus on what kind of content (if any) is carried by sentences in context. Here I want to formulate a rough taxonomy of the responses given to Travis’s challenges and to propose my own diagnosis. The first part of the paper briefly introduces some of Travis’sexamplesandstateswhatcanbetakentobehisargumentagainstmeaning determiningtruth-conditions.Thesecondsectionpresentsthedifferentreactionsthat theexamplesandtheargumenthavemotivated.Inthefinalsections,Iargueforaview whichdifferssubstantiallyfromallofthosereactions. 2 The examples and the argument Travis’sexamplesareverywellknownbynow.Theytypicallyinvolveverysimple non-ambiguous and indexically non-problematic utterances, that turn out to have different truth-conditions on different occasions, or in different contexts or scenarios. Probably the example that has attracted most attention is that of the green leaves (Travis 1996). Travis asks us to think of two different occasions on whichtheexpressiontype‘theleavesaregreen’isutteredbysomeonecalled‘Pia’. Onthefirstoccasion,Piaistalkingtoaphotographerwhoneedssomegreenleaves forherpicture.Piafindssomeleavesthathavebeenpaintedgreen,thoughtheyare originallyred.Still,inthatcontext,herutteranceof‘theleavesaregreen’isjudged tobetrue.However,thenalongcomesabotanistlookingforgreenleaves.Referring to the very same leaves, Pia says again: ‘the leaves are green’. This time her utteranceisjudgedtobefalse.Nothinghaschangedeitherregardingtheleavesorin the meaning of the utterance, and yet, what was true before is now false. Inanother example, we are asked toimagine a kettle that has beenburnt all over (Travis1985).Uponfindingit,Piasays,alarmed:‘thatkettleisblack!’Weseemtojudge thatshehastoldthetruth.Butthedayspass,andPiaandherboyfriendseethekettleina shop.Again,Piatellshim:‘thatkettleisblack’.Herboyfriend,however,responds:‘no,it seemsitwasburntandtheyhavenotbotheredtocleanitup.See’,headdspassinghis fingeracrossitssurface,‘thekettleissilvercoloured’.Inthisscenario,weareinvitedto thinkthatwhatPiasaidwasfalse.Athirdexample,developedatlengthinTravis(2000), revolvesaroundtheutterance‘theinkisblue’.HereTravisclaimsthattherearetwo waysinwhichtheinkcanbejudgedtobeblue,andso,thattherearetwodifferentsetsof truth-conditionsthatanutteranceof‘theinkisblue’canhave,dependingontheoccasion whenitisuttered.Inonecontext,theinkcanproperlybesaidtobeblueifandonlyifit producesbluewritingwhenused;butinanothercontext,itcanonlybesaidthattheinkis 1 By ‘indexically non-problematic’ (without qualification, from now on), I mean either that the expressiontypecontainsnoindexicalsatallorthatthevalueoftheindexicalswithintheexpressiontype donotvarywiththerelevantcontextualvariations(i.e.,withtherelevantscenarios).Theonlyexpression typeconsideredherewillbesententialexpressions. 123 Author's personal copy OnTraviscases 5 blueifandonlyifitlooksblue(irrespectiveofwhetheritproducesbluewritingwhen used).Perhapstherearemorepossiblereadingsof‘theinkisblue’,butTravisseems contentwiththesetwo. Famously,Travisconcludesfromallthisthat‘‘[w]hatwordsmeanplaysarolein fixingwhentheywouldbetrue;butnotanexhaustiveone’’(1996,p.451).Thatis,he takeshisexamplestoshowthat,thoughmeaninghastoconstraintruth-conditionsin someway,itfallsshortofdeterminingthem.ItisnoteasytoseewhyTravisinterprets his examples in this way. In principle, he could take them to show that both word meaningandsentencemeaningareoccasionsensitiveorunstable.Thatis,hecoulduse hisexamplestoarguefortheviewthatonlyutterances(ofwordsandofsentences)are bearersofmeaning,denyingthatthereissuchathingasthemeaningofasentencetype orofawordtype.Ifhewentdownthatpath,hecouldpreservetheideathatthemeaning ofatokenofasentencedetermines(orisidenticalto)itstruth-conditions.Yet,hedoes notwanttogothatway.Rather,hethinksthattwofactorsinterveneinthedetermination of the truth-conditions of an utterance: the meaning and the circumstances. It is certainlydifficulttogetaclearideaaboutwhatmeaningconsistsofaccordingtoTravis, butheapparentlyassumesthatmeaningisquiteastablefeatureofwords,i.e.,thatthere issomethingthatcouldbesaidtobetheword’stypemeaning(seethequoteaboveand alsoTravis2000,pp.221–227).TheminimalinterpretationofwhatTravisholdswould bethatthemeaningofawordisjustasetofconstraintsonwhatthatwordcanbeused for.2Theroleofmeaning,heclaims,istosetsomeconstraintsonwhatcanbesaid:a word cannot simply be used to refer to anything (Travis 2000). That is, there is variabilityinworduse,butthereisalsostability,anormativestability. So, we may think that word types have meaning in a very thin sense: all the utterances of a word have something in common, namely, all the utterances of a wordsetthesamegeneralconstraintsonwhatcanbesaidwhenusingthatword.If we move from words to sentences, we may assume that non-ambiguous sentence typesalsohavemeaning,accordingtoTravis.Thisfurtherassumptiononlyrequires that compositional rules are stable, something which Travis, on the face of it, does not call into question. Now,ifthisinterpretationofwhatTravishasinmindisright(Idonotclaimthat it is, only that it might be), his claim that meaning does not determine truth- conditions may stem from his endorsement of the following two theses: (i) Any two non-ambiguous, indexically non-problematic, utterances, u and u0, of the same linguistic type, can have different truth-conditions on different occasions; 2 I find it difficult to go further than this minimal interpretation of Travis. That meanings somehow embodyconstraintsonthecorrectnessofworduseissomethingheclearlyholds:e.g.‘‘[s]uchconstraints aremeaningcontributionstothecollaborativeenterpriseofmakingthecontentofourstatementswhatit is’’(Travis2000,p.222).However,heapparentlywantstoexplainwheretheseconstraintscomefromas well:‘‘Meaning’scontributiontothecooperativeenterpriseoffixingcontentistomakethecorrectnessof one’swordsamatterofhowthethingsonespeaksoffitintothatsystem[ofcategories]—amatterofthe waythesystemonethusinvokes,soinvoked,wouldcategorize’’(ibid.,p.224).Idonotdaretosuggest aninterpretationofwhatTravissayshereandinothersimilarpassages.However,thepurposeofthis paperisnotexegetical.WhatI’minterestedinisnotsomuchTravis’saccount—hewouldsurelyreject therepresentationalistapproachIamtaking—ashisexamples. 123 Author's personal copy 6 A.Vicente (ii) Any two non-ambiguous, indexically non-problematic, utterances of the same linguistic type have the same meaning. The examples seem to show that the first thesis is true. It is perhaps more controversial to ascribe the second thesis to Travis. Yet, as I have just argued, if meaningsareindeedstablefeaturesofwords,andcompositionalrulesarestableas well,(ii)hastobetrue.Itisimportant,however,nottoplacetoomuchcontentinto (ii). The thesis should be interpreted as the claim that any two non-ambiguous utterancesofthesametypesetthesameconstraintsontheconditionsthatcanmake them true. The assumption that both (i) and (ii) are true leads us to conclude that meaningdoesnotdeterminetruth-conditions.Thatis,theconstraintsbroughtabout by the meaning of a sentential utterance are not sufficient to determine its truth- conditions. In the next section, I briefly present the main reactions to this argument developed in recent years by other authors, before I develop and defend a new approach. 3 Minimalism, contextualism, indexicalism and relativism As it happens, many philosophers of language seem to find the argument unobjectionable, even though there is radical divergence as to what it actually shows.MinimalistssuchasBorg(2004)orCappelenandLepore(2005)clearlyreject the argument, trying to block it by denying one of its premises. For, according to minimalists,(i)isfalse.Ifwehavetwonon-ambiguousutteranceswhichcontainno overtindexicals,orcontainindexicalswhosevaluesdonotchangewiththeproposed shift of context, then those two utterances must have the same truth-conditions. In theirterms,theymustexpressthesameminimalproposition(whicheitherisjustthe setoftruth-conditionsordeterminesthosetruth-conditions).Accordingtominimal- ists,thereasonwhywemaythinkotherwiseisthatwemistake‘‘knowledgeoftruth- conditions’’for‘‘knowledgeofwhenthosetruth-conditionsaresatisfied’’(seeBorg 2004). Knowledge of truth-conditions is a purely semantic knowledge and can be capturedinasimpleT-schema.Intheexampleofgreenleaves,suchknowledgetells usthat‘theleavesaregreen’istrueifandonlyifthecontextuallysalientleavesare green:thatisall.Thetruth-conditionsTravistalksaboutarenottruth-conditions,but rather,theylooklikeverificationconditions. Indexicalistsareonboardwiththeminimalists,astheywouldalsorejectpremise (i).Incontrasttominimalists,indexicalistscanconcedethatthetruth-conditionsof the different utterances of ‘the leaves are green’ are those that the contextualist prefers.Whattheywanttodefendisthatallvariationshavetobetracedbackeither to some indexical expression (Rothschild and Segal 2009) or to some covert indexical attached as a hidden variable to some expression (Szabo´ 2001; Stanley 2002). Thus, if two utterances of the same linguistic type differ in their truth- conditions,itmustbe duetothepresence ofsomeindexical constituentorother:it cannot be that indexically non-problematic expressions differ in their truth- conditions. 123 Author's personal copy OnTraviscases 7 As I mention above, many other philosophers of language take the argument to be good. Only perhaps Recanati (2004) could be taken as objecting to premise (ii). Recanati distinguishes two layers of meaning: standing meaning and modulated meaning. Standing meaning is not sensitive to contextual shifts and does not determine truth-conditions, whereas modulated meaning does determine truth- conditions. Modulated meaning is the meaning interpreters obtain after taking contextual(pragmatic)informationintoaccount.Itseems,moreover,thataccording toRecanati’sview,themeaningofanutteranceisitsmodulatedmeaning.Thus,(ii) would be false on this account. However, if by meaning we understand semantics, Recanatiwouldnotbeobjectingto(ii):heagreesthatthesemanticsofanutterance does not fix its modulated meaning and, a fortiori, that it does not fix its truth- conditions. Relevance theorists share this overall view. They also distinguish between encodedmeaningandexplicatures:theexplicaturehasdeterminatetruth-conditions, but it incorporates pragmatic knowledge. The level of encoded meaning, on the other hand, is purely semantic, but it falls short of determining truth-conditions. There are two strands in Relevance theory, though. According to the first (Sperber and Wilson 1986/1995), (categorematic) words encode concepts. According to the second (Carston 2008), word meaning is not conceptual; rather, the semantic informationattachedtoawordisthin,anditsroleissimplytoputsomeconstraints on what concepts the word may express, or to indicate a part of conceptual space (seeCarston2002).Thesemanticlevel,followingthisconstrual,isveryclosetothe syntactic level. Now,eventhoughbothstrandsinRelevancetheorywantencodedmeaningtobe non-propositional, the ‘‘conceptualist’’ strand may have more problems accounting forTraviscasesintheirpreferredway.Thereseemstobenodifferenceatthelevel ofencodedconceptualmeaningbetweentwodifferent(indexicallyfixed)utterances of ‘the leaves are green’. So, the conceptual meaning of the expression type ‘the leaves are green’ does not change from context to context. However, if its conceptual meaning does not change, then it seems to follow that its propositional meaning will not change either (a proposition being, according to this view, an adequately structured combination of concepts). Thus, if we assume that proposi- tions fix truth-conditions, it means that two different utterances of ‘the leaves are green’ have to have the same truth-conditions, contrary to premise (i) of the argument. In short, it seems that if Relevance theorists want to endorse Travis’s argument (instead of sharing the minimalist take on it), they had better opt for Carston’s thin construal of semantics. The last group of authors that would not object to the argument is the relativist group. There are profound disagreements between those authors and radical contextualists such as Travis, to be sure. However, the disagreements do not touch thetruthofthepremisesorthecorrectnessoftheargument.Relativistswouldhave no quarrel with the premises at all; and they would agree that meaning does not determine truth-conditions. Their disagreement with contextualists lies in the lessons the latter purport to extract from the argument. As Predelli (2005) puts it, what they deny is that there is something exciting about all this, something that should make us rethink the semantics enterprise. For, according to the relativist, 123 Author's personal copy 8 A.Vicente Travis cases can be tackled perfectly well with the apparatus of formal semantics. Traviscasesshowthatintensionsalonedonotdeterminetruth-conditions.However, intensions plus points of application, or circumstances of evaluation, do determine truth-conditions. Formal semantics, thus, runs no risk from acceptance of the argument. Inwhatfollows,IdevelopadifferentapproachtoTraviscases,whichresortstoa ‘‘thick’’conceptualsemanticsforlexicalitems.Inthelastsection,Iwilldiscussthe virtues of this approach vis-a-vis the responses to Travis cases that I have just reviewed. To advance: the thick semantics approach is, overall, clearer and more conservative than its rivals. It is clearer than some of its rivals in that it does not invokeanythingotherthanconceptualmeaningsforconstituentsandsemanticrules that have such meanings as their inputs. In this, it contrasts with the ‘‘thin’’ semantics approach adopted by Carston (2008) and suggested by Travis himself, where meanings are not concepts, but some other kind of representations which encodeconstraintsonconceptualexpressions.Thinsemanticistshaveyettoexplain whattheserepresentationsareandhowtheywork.Theapproachisalsoclearerthan indexicalism, which accounts for Travis cases in an unnecessarily complicated manner (and is most probably unsuccessful: see Clapp forthcoming). Furthermore, asIwillsuggest,theindexicalistwouldhavetopostulatethepresenceofindexicals inallsortsoflexicalitems(i.e.,notonlypredicates,butalsoverbs)tocoverallthe data. The approach I advocate is more conservative than other approaches in that it takes itsexplanandatobe‘‘intuitive’’truth-conditions, andnotsomearbitraryand/ or flat truth-conditions, the so-called ‘minimal’ truth-conditions. It is also more conservative than the thin semantics approach in that it assumes that utterance meaning is truth-conditional. Finally: thick semantics forms part of an explanatory programme, conceptual semantics, which, among other things, has provided successfulaccountsofphenomenaasdiverseasargumentstructureand(somecases of) polysemy. Minimalists, indexicalists and thin semanticists have yet to provide explanationsforthosephenomena.(Itseemsthatminimalismisespeciallyillsuited to that task.) 4 Denying (ii) TheviewofmeaningthatIwanttoargueforprovidesanotherwayofrejectingthe argumentabove;byblocking itatpremise(ii).Thus,Iproposethatweworkunder the assumptionthat (i)is true; thatis, that neither minimalists are rightinclaiming thatcontextualistsmisfireintheirattributionoftruth-conditionsinTraviscases,nor is it plausible to postulate the existence of as many indexical constituents as the indexicalist programme would require.3 In short, I take it for granted that two indexically non-problematic, non-ambiguous utterances of the same linguistic type 3 For an effective critique of the indexicalist programme as applied to colour predicates, see Clapp (forthcoming). Minimalism, in contrast, besides refusing to take intuitive truth-conditions as its explananda,haspoorexplanationsformanyotherlinguisticphenomenawhichwillbementionedhere, suchaspolysemy,argumentstructure,impossiblewords,orcoercion. 123 Author's personal copy OnTraviscases 9 can have different truth-conditions. My claim, however, is that this does not imply that meaning does not determine truth-conditions, nor that the meaning of a sentential utterance is not fully propositional. In order to argue for this claim, I suggest that, instead of embracing a thin semantics, as Carston (2008) proposes, we should go for quite a thick semantics, in the spirit of what has been labelled as ‘‘conceptual semantics’’ (Jackendoff 2002). Before going into the details of the proposal, there is a precision to be made: Ihave been speakingfreely about non-ambiguous utterances.Ambiguity, however, is not a totally regimented concept. In the present context, it is usually taken to cover both scope, and, in general, syntactic, ambiguities and homonymy. Thus, an utterancecanbeambiguouseitherbecauseitissyntacticallyambiguousorbecause itcontainsahomonymousterm.Inwhatfollows,Ionlyusetheterm‘ambiguity’to refer tosyntacticambiguityandhomonymy.Inaccordancewith thisunderstanding of ‘ambiguity’, my proposal is rightly seen as a denial of premise (ii) of the argument.Yet,onabroaderunderstandingoftheterm,itisprobablybestseenasa rejectionof(i):justasindexicalistsareabletoreject(i)bypostulatingtheexistence ofmanymoreindexicalsthantheclassicones,itispossibletorejectthepremiseby postulatingtheexistenceofmoresourcesofambiguity,broadlyunderstood,thanthe classic ones.4 Another caveat: premise (ii) states that any two indexically non-problematic, non-ambiguousutterancesofthesamelinguistictypemusthavethesamemeaning. Onecouldthinkthatthisisobviouslyfalse.Takeasimplesentencesuchas:‘thecat isonthemat’.Icanusethissentencetomeanthatagivenanimalisonthemat,but alsotomeanthatacertaintoyisonthemat.Referentialexpressionscanbeusedto refereithertoobjects(orsetsofobjects)ortotoyswhichrepresentthoseobjectsor sets of objects. That is, all referential expressions are polysemous, so there is no reason to think that two utterances of the same sentence type must have the same meaning.5 Alternatively, think about metaphors as sources of polysemy: as is well known, many metaphors become conventional and some even become dead (the metaphorical element of the meaning is not even retrievable). So, there is a point when a metaphor is properly treated simply as another meaning of a certain expression(whichhassomekindofresemblancetothehistoricallyformermeaning of that expression). That is, the retrieval of the metaphoric content is no longer a pragmatic issue, but a matter of polysemy resolution. This shows again that two utterances of the same expression can have different meanings (in this case, one ‘‘literal’’, the other metaphorical). 4 Infact,ifwetake‘ambiguity’inthisbroad,unregimented,sense,itseemsthatitistautologicalthatthe variabilityinthetruth-conditionsofindexicallynon-problematiclinguistictypesisduetoambiguity. 5 Infact,itseemsthatreferentialexpressionscanbeusedtorefereithertoobjectsortojustanyiconic representation of those objects. Culicover and Jackendoff (2004) tell us that there is a ‘‘statue rule’’, accordingtowhichanyNPcanrefereithertosomedeterminatexortoaphysical/iconicrepresentationof x.Theirtalksuggeststhatthis‘‘statuerule’’isaconventionallinguisticrule.Cognitivelinguistswould opposethiswayofthinkingaboutwhatgoesoninsuchcases:theytakeitthatwethinkofrepresentations intermsofrepresentedthings(PantherandThornburg,2003).Bethatasitmay,thiskindof(cognitiveor linguistic)ruleisapermanentsourceofpolysemy. 123 Author's personal copy 10 A.Vicente However, this obviously misses the point. Polysemy is a deep problem for a literalist,thatis,someonewhoholdsthatthereisaone-to-onerelationshipbetween words and concepts and that sentences have one, and only one, propositional content.Thus,polysemyisaproblemforthosewhoplacemuchmoreintoclaim(ii) than is actually asserted. For instance, it is a problem for minimalists (witness Fodor’squarrelwithpolysemy:Fodor1998;FodorandLepore2002);butitisnota problem for Travis or, in general, for any of the authors who hold that linguistic meaning is thin. A polysemic expression can share the same meaning across its different uses if one thinks about meaning in terms of constraints. In fact, what distinguishes polysemy from homonymy is that the different uses of a polysemic expression have something in common: one of the constituents of polysemy is that there is some stability across uses. A thin semanticist would identify the stable feature of a polysemic expression with its meaning. Inanycase,polysemycouldeventuallygiveusreasonstosuspectthatclaim(ii) is not true, but polysemy is not what seems to be at issue in Travis cases. The variation in truth-conditions detected by Travis is not due to ambiguity, but seems not to be due to polysemy either: prima facie, ‘leaves’, ‘are’ and ‘green’ do not behaveaspolysemicexpressionsacrossthecontextsweareinvitedtoconsider.So, just what is going on here, then? Here is a story a conceptual semanticist can tell. First of all, let us accept that lexical entries are conceptually rich, as conceptual semanticists tend to think. For instance, Pustejovsky (1995) has famously argued that lexical entries are highly complex representations which include information about things such as the argumentandeventstructure(inverbs)andthequaliastructure—astructurewhich specifiesknowledgeabouttheconstitution,thespecificproperties,thecausalorigin andthepurpose(ifthereisone)ofthedesignatum.Thus,alexicalentryforanoun, accordingtohisview,givesusasinformationabouttheoriginsofitsreferent,about its purpose, about its constitution and about its distinguishing properties, i.e. those propertiesthathelptosingleitoutwithinasuperordinatecategory.Forinstance,the meaning of the noun ‘novel’ tells us that novels are constituted by a narrative, that theyarebooks,thattheyareforreadingandthattheyarewrittenbyagents.Allthis information is meaning-constitutive, according to Pustejovsky. Pustejovsky’s approach has famously been put to service to explain polysemy (seebelow);whichmanyauthorsseeasafailedattempt(see,e.g.,Carston2010).It is not my purpose here to defend the attempt against its critics, but only to borrow theverygeneralideathatwordmeaningsmaybericherthanpreviouslythought.In particular,thattheycanbesorichthattheyincludeknowledgethatmanywouldsee as world (and not semantic) knowledge (more on this below). So,Iproposetoborrow agoodpartoftheoverallideaandclaimthat,typically, alexicalentryfor,e.g.anoun,consistsofacomplexconceptualrepresentationthat collects rich information about whatever entity the noun is intended to designate, such conceptual representation being its meaning. For instance, if the noun is intended to designate a set of objects such as pencils, its lexical entry will include information such as: that they are used for writing or drawing, that they typically produce grey writing, that they have a certain stereotypical shape, that they are composedoftwoparts,thattheycanbesharpened,etc.Also,thelexicalentrymay 123