strategies of deception - arbeitsbereichemfranke/...2017_strategies_of_deception.pdf · keywords:...
TRANSCRIPT
0
TitlePage
Title:
Strategiesofdeception:under-informativity,uninformativityandlies–misleadingwithdifferentkindsofimplicature
Authors:
GiulioDulcinati,UniversityCollegeLondon,[email protected]
MichaelFranke,UniversityofTübingen,[email protected]
NausicaaPouscoulous,UniversityCollegeLondon,[email protected]
Keywords:lying;implicature;signallinggame;cooperation;Grice
1
Strategiesofdeception:under-informativity,uninformativityandlies–
misleadingwithdifferentkindsofimplicature
Abstract
Conversationisoftencastasacooperativeeffort,andsomeaspectsofit,suchasimplicatures,have
beenclaimedtodependonanassumptionofcooperation(Grice,1989).Butanysystematicclassof
inferencederivedfromassumptionsofcooperation,suchasimplicaturescouldalsobe,onoccasion,
usedtodeceivelistenersstrategically.Hereweexploretheextenttowhichspeakersmightchoose
differentkindsof implicaturetriggers inanuncooperativegameofcommunication.Concretely,we
present a production study in the form of a cooperative or competitive signalling gamewith the
potentialofexploitingthreekindsofimplicatures,namelyexactreadingofnumeralexpressions,scalar
implicaturesandparticularisedAdhoc implicatures.Wefindthatwhile theexact interpretationof
numerals isused similarly to truth-conditional content, scalarandparticularised implicatureselicit
different strategies.We also observe heterogeneity in individual strategic behaviour. Expecting a
distrustfulreceiver,someparticipantsusedhighratesofuninformativehints,andequalratesoftrue
andfalsehints.Otherparticipantsusedahigherrateof liesandfalse implicatures,suggestingthat
they were expecting their interlocutor to infer implicatures as if they came from a cooperative
speaker.
Introduction
Grice(1989)famouslypresentsconversationasacooperativeactivity inwhichparticipantsabidea
cooperativeprinciple,whichbindsthemtomakeappropriatecontributionstotheconversation.From
thisprinciplefollowmorespecificmaximssuchasthefirstmaximofquantity:“Makeyourcontribution
as informative as is required” (Grice, 1989, p. 45). Speakers can exploit the maxims in order to
communicate implicit propositions (implicatures) of various types. For example, the speaker can
violatethefirstmaximofqualitytocommunicateaquantityimplicature.IfItellyouthatIusedsome
ofyournewshampooinacontextwhereitwouldberelevantandmoreinformativetoknowwhether
2
Iusedallofyourshampoo,youmayinferthatthereasonwhyIamviolatingthefirstmaximofquantity
isthatthemoreinformativestatementisnottrueandthereforeinfertheimplicaturethatIdidnot
useallofyourshampoo.Thelasttwodecadeswitnessedawaveofexperimentalinvestigationofhow
differenttypesofquantityimplicaturesareprocessedandinterpreted;andinharmonywithGrice’s
account,theseinvestigationshavefocusedonsituationswherethecooperationandhonestyofthe
speaker is taken for granted. However, conversation also takes place in non-cooperative or
competitivesituations,wherethespeakermaybedeceptiveoruninformative.Politiciansareoften
goodexamplesofunhelpfulinterlocutors.Forinstance,considerthisevasiveanswerthatTheresaMay
gave in2016whenaskedwhethertheUKshouldhaveaccesstotheEUsinglemarketafterBrexit:
“WhatIwanttoseeisthebestpossibledealfortheUnitedKingdomintradeingoodsandservices”
(Bull, 2016). The use and comprehension of implicatures in non-cooperative settings is a vastly
understudiedtopic.Theveryfewexistingcomprehensionstudiesonthistopicsuggestthatlisteners
facedwith an uncooperative speaker tend to infer less implicatures than if they are facedwith a
cooperativespeaker(Pryslopska,2013;Dulcinati&Pouscoulous,2017).Additionally,tothebestof
our knowledge no published experiments investigate implicature production in uncooperative
contexts.Thisispreciselywhatthisstudywillfocusonandwewilllookattheproductionofdifferent
typesofquantityimplicaturesinanon-cooperativesetting.Itisparticularlyinterestingandtimelyto
fillthisgapintheliterature,firstlybecausethismethodallowsustocomparehowspeakersuseexplicit
andimplicitcommunicationstrategically,andsecondlybecauseitmayofferanewperspectiveonthe
differencesbetweenwellstudiedtypesofquantityimplicatures.
A non-cooperative speakermaydiffer froma cooperative one in that theymaybemore likely to
deceiveortobeuninformative.AlthoughGrice(1989)presentsconversationasacooperativeeffort,
he contemplates both the possibility that speakers may be uninformative by opting out of the
cooperativeprincipleorofamaximinanovertway,forexamplebysaying“Ican’ttellyouthat”,and
thepossibilitythattheymaybedeceitfulbycovertlyviolatingamaxim.Theparamountexampleof
covertviolationsofmaximsislying,wheretheliarcovertlyviolatesthefirstmaximofquality(i.e.“Do
3
notsaywhatyoubelievetobefalse”;Grice,1989,p.46)andintendstheaudiencetoremainunaware
oftheviolation.Besideslying,therealmofverbaldeceptionincludesfalselyimplicating.Whiletolie,
atleastaccordingtotraditionaldefinitions(Isenberg,1973;Primoratz,1984),istosaysomethingthat
thespeakerbelievestobefalsewiththeintentiontodeceive;tofalselyimplicateistocommunicate
something believed to be false by means of a conversational implicature (Meibauer, 2014). For
example,ifIsaidthatIusedsomeofyournewshampoowheninfactIbelievethatIusedallofit,I
couldbefalselyimplicatingthatIdidnotuseallofyournewshampoo.Althoughthereisanongoing
conceptualdebateonwhetherfalseimplicaturesshouldbeconsideredlies(Meibauer,2005,2014)or
not(Dynel2011,2015)herewewilltreatthemasseparateforthepurposesofexperimentdesignand
analysis.Weletthedataspeakaboutanypotentialdifferencebetweenfalse implicaturesandlies.
Thereforethephenomenawhichweexpecttoobservewithpossiblydifferentbehaviouralsignatures
inourstudyareuninformativityoroptingout,liesandfalseimplicatures.
Wemayexpectthatexplicitandimplicitcommunication1(seeCartson,2002,2009andRecanati,2004
for a review of this distinction), which include lies and false implicatures respectively, are used
differently in non-cooperative contexts. One reason for this expectation is that in cooperative
conversationthesetwomodesofcommunicationareoftennotinterchangeableandinagivencontext
we usually have clear preferences as to whether a piece of information should be asserted or
communicatedimplicitly.ConsiderthefollowingexamplesadaptedfromRussel(2012):
1. Careful!Someofthemushroomsarepoisonous!
2. #Careful!Notallofthemushroomsarepoisonous!
Theutterancesin1and2carrythesamecontentexceptthattheexplicitcontentoftheutterancein
1isimplicitintheutterancein2andviceversa(seealsovanTiel,2014foradiscussionofasimilar
example). Sincewe infer from theexpression ‘careful’ thatbothutterances arewarnings, the key
informationthatthespeakerpresumablywantstoconveyisthatatleastsomeofthemushroomsare
1WewillassumethattheexplicitimplicitdistinctioncorrespondstothedistinctionbetweenGrice’swhatissaidandwhatisimplicated.However,seeCartson(2002,2009)andRecenati(2004)fordifferentperspectives.
4
poisonous,andnotthatsomeofthemareharmless.Conveyingthekeyinformationasimplicitcontent
ratherthanexplicitcontentmakestheutterancein2soundodd.Intuitively,thismaybebecausein
thiscontextwewouldprefertocommunicatekeyinformationthroughtherelatively‘secure’channel
of explicit information as implicit communication is arguably more prone to misunderstanding
(Reboul,2017).Anotherreasonwhywemayexpectexplicitandimplicitcommunicationtobeused
differentlyinourstudyisthatsomefeaturesofimplicitcommunicationcouldbeadvantageousinnon-
cooperative contexts. For example, implicit communication offers the advantage of plausible
deniability (Pinker,Nowak& Lee, 2010). Since implicatures are cancellable the communicator can
deny having intended to communicate them. For example, after saying that I used some of your
shampoo,IcanclaimthatIdidnotmeanthatIdidnotfinishit,whichIcouldnotdoifIexplicitlysaid
thatIdidn’tuseallofit.Thisfeatureofimplicitcommunicationisusefulincaseswherethespeaker
wantstocommunicatesomethingthatmayincurthemsomepenalty,suchasproposingabribeor
communicatingfalseinformation.Reboul(2017)proposesthatimplicitcommunicationmayalsooffer
anotheradvantageinthatitmaybeacceptedmoreeasilybythehearerthanexplicitlycommunicated
content.Firstlysheclaimsthathearersaremorevigilanttowardscontentthatthespeakerisstrongly
committed to, and explicit content carries a higher degree of speaker commitment compared to
implicitcontent(Morency,Oswald&deSaussure,2008).Secondlysheclaimsthathearersareless
vigilanttowardscontentthatisthefruitoftheirowninferences,whichisthecaseforimplicaturesbut
notforassertedcontent.
Thepreviousstudiesthatareclosesttotheonewearepresentingarerecentstudieslookingatnon-
verbaldeceptioninthecontextofsignallinggameswheresignallershavetogivenon-verbalhints(e.g.,
images,maps)toareceiverplayerwhohastomakechoicesbasedontheinformationprovidedinthe
hints.Cruciallyinsomecasesthegameiscompetitiveandthesignallerbenefitsfromthereceiver’s
wrong choices, which provides motivation to deceive. Signallers can give true hints, false hints,
uninformativehints andmisleadinghints,which like false implicatures consist in conveying a true
piece of informationwhich leads the receiver to infer something false.Montague and colleagues
5
(2011)foundthattheirplayerspreferredgivingmisleadinghintsratherthanfalsehints.Intheirgame
thereceiverdidnotknowwhetherthesignallerwascooperativeorcompetitiveandtheycouldchoose
tocheckwhetherthehintswerefalseandcalibratetheirtrustaccordingly,whichwasanincentivefor
thesignallernottobecaughtlyingasitwouldhavereputationconsequencesfortherestofthegame.
InasimilarcompetitivegameRansom,Voorspoels,PerforsandNavarro(2017)gavetheirparticipants
theoptionstogivetothereceivertrue,misleadingoruninformativevisualhints,butnotfalsehints,
andtheymanipulatedthesignaller’sexpectationsregardinghowsuspiciousortrustfulthereceiver
wouldbe.Because the receiverdidnot knowwhether their signallerwashonestordeceitful, the
signaller could pretend to be helping the receiver while in fact feeding them false ormisleading
information.Theyfoundthatwhensignallersexpectedatrustfulreceivertheyweremore likelyto
mislead,whereaswhentheyexpectedasuspiciousreceivertheyweremorelikelytobeuninformative.
Thesestudiesaresimilartooursfirstlybecausewealsoemployedacompetitivesignallinggameand
secondlybecausethetypesofdeceptiontheystudiedfollowthesamefundamentalmechanismsof
thekindsofverbaldeceptionweareinterestedin,whicharetocausesomeonetohaveafalsebelief
(Mahon,2007)eitherbycommunicatingsomethingfalse(i.e.falsehints,lies)orbycommunicating
somethingtrue(i.e.misleadinghints,falseimplicatures).
Oneimportantcomplicationofstudyingexplicitandimplicitcasesofverbaldeceptionisthatwhile
thestudiesonnon-verbaldeceptionthatwementionedcoulddrawacleardistinctionbetweenfalse
andmisleadinghints;drawingadistinctionbetweenliesandfalseimplicatureisnotstraightforward.
In two studies (Coleman&Kay1981,Hardin, 2010)whereparticipantswereasked to ratea false
implicatureonascalethatrangedfromanutterancebeingalietoanutterancenotbeingaliethe
averageratingwasnearthemiddleofthescale.Inparalleltotheseresults,studiesontheexplicit-
implicitdistinctionincomprehensionfoundthatlaypeoplearelikelytoconsiderimplicaturespartof
whatissaidundersomecircumstances(Nicolle&Clark,1999;Doran,Baker,McNabb,Larson&Ward,
2009;Doran,Ward,Larson,McNabb&Baker,2012).Doranandcolleagues(2012)askedparticipants
tojudgewhethersentencesthatcouldgiverisetoanimplicatureweretrueorfalseinthelightofa
6
factthatcontradictedtheimplicature(e.g.judgingwhetherthesentenceIusedsomeofyourshampoo
istruegiventhatIusedallofit).Theyfoundthatparticipantsincorporatedscalarimplicaturesarising
fromquantifierssuchassomeandmostintothetruthconditionalmeaningofthesentence32%of
thetimeandimplicaturearisingfromcardinalnumbers(e.g.IhavethreecatsimplicatingthatIdon’t
havefour)53%ofthetime.Becausedifferenttypesofimplicaturesmaybedifferinwhethertheyare
consideredpartofwhatissaid,andthereforeinwhethertheywouldbeconsideredtobeliesifused
deceptively,inourstudyweaimedtogainamorecomprehensiveperspectivebyusingthreedifferent
typesofimplicatures.
We used three types of quantity implicatures: implicatures arising from numerals, the scalar
implicaturearisingfromthequantifiermostandadhocorparticularisedquantityimplicatures.The
quantityimplicaturesorupper-boundinterpretationsassociatedwithscalartermsandnumeralsare
drawnbynegatinganalternativeutterancewhereastrongertermonthesamelexicalscaleasthe
scalar/numeraltermisused.Forexample,theimplicatureof‘Iusedsomeofyourshampoo’arises
fromnegatingthealternativethat‘Iusedallofyourshampoo’.Andtheimplicatureorupper-bound
interpretationof‘Ihavethreecats’arisesfromnegatingthealternativethat‘Ihavefourcats’.This
contrasts with ad hoc (sometimes called particularised) quantity implicatures where the stronger
alternativecanonlyarisefromthecontextandnotfromthelexicon.Forexample,inthecontextwhere
Ann andRose have their birthdays together someonemight say ‘I bought a present forAnn’ and
implicatethattheydidnotbuyapresentforRosealso.Thestrongeralternativenegatedisthatthey
boughtapresentforAnnandRosebutitcanonlyariseinthiscontextas‘Ann’and‘AnnandRose’are
not on a semantic scale. Both scalar terms and numerals have been at the centre of theoretical
controversies concerning whether their upper-bound interpretation is an actual implicature or
whetheritispartoftheirsemanticordefaultmeaning(Levinson,2000;Geurts,2010).
Although some theorists have proposed that the upper-bound interpretation of scalar terms (e.g.
someandnotall)istheirdefaultmeaning(e.g.Levinson,2000;Chierchia,2004),recentexperimental
evidencesuggeststhatthescalarimplicaturesofquantifiers(i.e.someandnotall)arederivedinthe
7
samewayasparticularisedquantity implicatures (seeKatsos&Cummins, 2010 for a review). The
distinction between scalar and particularised implicatures has received particular attention in the
acquisitionliterature,whereexperimentalstudiesoffermixedresults:somestudiessuggestthatpre-
schoolchildrenhavemoretroublecalculatingscalar implicaturescomparedtoadhoc implicatures
(Barner, Brooks & Bale, 2010; Stiller, Goodman & Frank, 2011) and others suggesting that they
calculate these two types of implicatures to the same extent (Katsos, 2009). Katsos (2009) asked
participantstoevaluateutterancesthatcouldgiverisetoeitherscalarorparticularisedimplicatures
incontextswherethecontentoftheimplicatureisfalse.Hefoundthatbothadultsandchildrenreject
scalar implicature utterances and ad hoc implicature utterances to the same extent, but adults
consider an under-informative scalar-implicature utterance to be a more serious violation of
informativitythananunder-informativeparticularisedimplicatureutterance.
Withregardstonumeralswehaveagaintwocamps,withsometheoristsclaimingthattheyhavea
lower-boundoratleastmeaningwhiletheexactinterpretationissuppliedincontextviaimplicature
(e.g., Horn, 1972; Gazdar 1979; Levinson, 2000) and others claiming that numerals the exact
interpretationofnumeralsisnotanimplicaturebutpartoftheirtruthconditionalmeaning(Carston,
19982; Breheny 2008; Kennedy 2015). Both the study of Papafragou andMusolino (2003) and of
Huang,SpelkeandSnedeker(2013)provideconvincingevidencethatnumeralshaveanexacttruth-
conditionalinterpretation(i.e.threemeans‘exactlythree’)byshowingthatpre-schoolchildren,who
arenotoriously‘bad’atcalculatingscalarimplicatures(Noveck,2001;Chierchiaetal,2001;Gualmini,
Crain, Meroni, Chierchia, & Guasti, 2001; Hurewitz, Papafragou, Gleitman & Gelman, 2006;
Pouscoulous, Noveck, Politzer, Bastide, 2007) tend to give exact interpretations of numerals.
Furthermore, Huang and Snedeker (2009) found thatwhile processing the upper boundmeaning
scalartermslikesome isslowerforadultsthanprocessingtheliteralmeaningofthequantifierall,
2Carston(1998)actuallyarguesthatcardinalshaveanunderspecifiedmeaningandthatwhicheversensetheyassumeincontext(i.e.atleast,atmostorexactly)contributestothetruthconditionalmeaningoftheutterance.
8
processingtheexactmeaningofnumeralsisjustasfast;suggestingthattheformerinvolvesdrawing
apragmaticinferenceandtheseconddoesnot.
In this study we aim to explore how people use explicit and implicit communication in a non-
cooperativecontext.Weaskedparticipantstoplayasignallinggamesimilartotheoneemployedby
Ransomandcolleagues(2017)wheretheyhadtoproducehintsforareceivereitherinacooperative
orinacompetitivescenario.Ourgamehadanumberoffundamentaldifferencesfromthegameused
byRansomandcolleagues.Firstly,ourparticipantsweretoldthatthereceiverknewwhetherthegame
wascooperativeorcompetitive.Thisremovedthepossibilityforsignallerstopretendthattheywere
cooperativewhen theiractualgoalwas tomisinformthe receiver.This featureof thecompetitive
scenario eliminates the possibility of cooperation and, from a Gricean perspective, should push
signallers towards the strategy of opting out and being uninformative. Secondly, in our game
participantsgavelinguistichintsbycompletingshortdescriptions.Sincethesehintsareassertions,the
falsehintsusedinthegamesarelies, inthetraditionalsense,andthemisleadinghintsgiveriseto
falseimplicatures.Thirdly,thedescriptiontemplatesthatsignallerscompletedconstrainedtheirhints
intopre-determinedcategories.Halfofthedescriptiontemplatespushedparticipantstoconveythe
hint explicitly and half of them through an implicature – belonging to one of three types: ad hoc
quantityimplicatures,linkedtothescalarquantifiermostortotheuseofcardinals.
Methods
Materialsanddesign
Thegameusedinthisparadigmisasignallinggamewithtwoplayers.Eachroundofthegamehastwo
cardssuchasthecardsinFigure1:a‘winning’cardanda‘losing’card.Thesignallerknowswhichone
isthewinningcardandtheyhavetodescribeit.Thereceiverseesthesametwocardsbuttheydon’t
knowwhichoneisthewinningcard.Thereceiverhastodecidewhichoneisthewinningcardwith
thehelpofthedescriptionmadebythesender.
9
Figure1winningcardandlosingcardassociatedtothetemplatedescription“Onthewinningcardalloftheobjectsare___”
Participantsinourexperimentonlyplayedtheroleofthedescriber,whilethereceiverplayerwasa
virtual player. Participants were assigned to one of two conditions: a cooperative condition or a
competitivecondition.Inthecooperativeconditionparticipantswereaskedtohelpthereceiverfind
asmanywinningcardsaspossible(agameofpurecooperation,ingametheoreticterms),whileinthe
competitiveconditiontheirgoalwastomakethereceiverclickonasmanylosingcardsaspossible(a
so-calledzerosumgame).
Materialsincludedintotal36items(AppendixA),whichcorrespondedto36roundsofthegame:18
experimentalitemsand18controlitems.Eachitemconsistedofatemplatedescriptionandthetwo
cards:thewinningcard,markedbyagreenoutline,andthelosingcard,markedbyaredoutline(see
Figure1).Ratherthanwritethewholedescriptionofthewinningcard,participantswereaskedto
completeapre-madedescriptionwithonlyoneword(e.g.seetemplatedescriptionforFigure1).All
itemswereconstructedinsuchawaythattheyhadtwoobviouscompletions,referringeithertothe
shapeorthecolouroftheobjectsdisplayedinthecards(greenvs.pinkandrocketsvs.umbrellasin
theexampleinFigure1).Controlitemsuseddescriptiontemplatescontainingeitherthequantifiers
allornoneandtheirmostobviouscompletionswereatrueassertionandafalseassertionaboutthe
winning card. Experimental itemsuseddescriptions that couldgive rise to three typesofquantity
implicatures:exactinterpretationofnumerals,scalarimplicaturesassociatedwiththequantifiermost
orparticularisedadhocquantity implicatures.Experimental itemswereconstructed insuchaway
that oneof the twomost accessible completions resulted in a true assertion giving rise to a true
10
implicaturewhiletheothermostobviouscompletionproducedatruedescriptiongivingrisetoafalse
implicature.Withineachcategoryofitems,controlandexperimental,wecounterbalancedwhetherit
wasmentioningthecolourortheshapeoftheobjectthatgaverisetothefalseassertionforcontrol
itemsortothefalseimplicatureforexperimentalitems.Thefalseassertionsandfalseimplicaturesin
eachitemwerefalseofthewinningcardbuttrueofthelosingcard,sothattheycouldbeusedto
deceivetheguesser intothinkingthatthe losingcardwasactuallythewinningcard. InweTable1
provideexamplesforeachcategoryofitems.
Table1Examplesofeachcategoryofitems
ItemType Description Truecompletion
Falsecompletion
Cards
Control(All)
On the winning cardallof theobjectsare___
umbrellas rockets
Control(None)
On the winning cardnone of the objectsare___
blue green
Experimental(Numeral)
On the winning cardtwo of the objectsare___
blue mugs
Experimental(Most)
On the winning cardmost of the objectsare___
lamps yellow
Experimental(Adhoc)
On the winning cardthe objects in themiddleroware___
green apples
ParticipantsandProcedure
We recruited 103 native English speakers (66 females, Mean Age = 28.73) from the online
crowdsourcingwebsiteprolific.co.ukanddirectedthemtoaQualtricswebsitewhereourexperiment
washosted.Participantsweretoldthattheywouldplayagamewheretheywouldhavetocomplete
11
descriptionswithoneword.Sincethetaskrequirednormalcolourvisionwetestedparticipantsusing
twoplatesfromtheIshiharacolour-blindnesstest(Ishihara,1917)andparticipantswhofailedthetest
werepreventedfromcontinuingtheexperiment.Eachparticipantwasrandomlyassignedtoeithera
cooperativeoracompetitiveconditioninabetween-subjectsdesign.Participantsinbothconditions
weretoldthattheotherplayerscoredpointsbyclickingonwinningcardsandthatineachroundhe
orshewouldreadtheirdescriptionanduse it todecidewhichcardtoclickon.Participants in the
cooperativeconditionweretoldthattheythemselveswouldscorepointswhentheguesserclickedon
a winning card. Consequently, their goal was to help the guesser. Instead, participants in the
competitiveconditionweretoldthattheywouldscorepointswhentheguesserclickedonthelosing
cards.Consequently,theirgoalwastomaketheguesserlose.Participantsinthecompetitivecondition
wereexplicitlytoldthattheotherplayerknewthatthepersonwritingthedescriptionswasplaying
against themandvice versa forparticipants in the cooperative condition. Thiswas toensure that
participantsinthecompetitiveconditionknewthattheycouldnotpretendtobecooperativeasthe
otherplayerwouldexpectthemtobeuncooperative.
Participantsinbothconditionsweretoldthattheotherplayerdidnotknowthattheywerecompleting
thedescriptionsinsteadofwritingthemfreely.Thiswastopreventparticipantsfromanticipatingthat
theotherplayerwouldthinkthatthereasonwhytheyhadnotutteredamoreinformativedescription
in experimental itemswas because the game prevented them,whichwould effectively block the
derivationofquantityimplicatures.
Participantsinbothconditionsweretoldthattheymustcompletethedescriptionswithonlyoneword
andtheywereexplicitlytoldthattheywereallowedtolie.Beforeallowingparticipantstoperformthe
actual task of the experiment we asked them four multiple choice questions to check their
understandingofthegameandwepreventedparticipantswhoansweredincorrectlytoanyofthe
fourquestionsfromcontinuing.Instructionsforbothconditionsandcontrolquestionsarereportedin
AppendixB.
12
Eachparticipantsawallofthe36 itemsdividedintotworandomizedblocks.Participantswerenot
givenfeedbackonthechoicesofthereceiverastheybelievedthatthereceiverwouldplaythegame
in a second phase. After the last item participants were asked to predict their performance and
estimateonan11-pointscalerangingfrom0%to100%whichpercentageofroundstheguesserwho
wouldreadtheirdescriptionswouldclickonthewinningcard.
Results
DescriptionswereautomaticallycodedusinganRscriptwhichclassifiedeachentryaccordingtoa
predefinedlistofresponsetypesforeachitem.Thelistwasconstructedaprioriandthenadjusted
afterinspectionofthedatatoaccommodatespellingvariantsandunforeseenstrategies.Two-word
entries,whichwereexplicitlyforbiddenintheinstructions,wereexcludedfromtheanalysis.Because
ofacounterbalancingerrorweexcludedtwooutofthesixadHocitemsfromtheanalysis.
Wecategorisedeachdescriptioninoneoffourcategories:trueandfalse,whichcorrespondedtothe
twomostobviouscompletionsthatweanticipated(seeTable1),aswellasuninformativeandother.
Weclassifiedasuninformativethosedescriptionswhichcouldeitherapplytoboththewinningcard
andthelosingcardortoneither.Forexample,anuninformativeresponsefortheallcontrolitemin
Table1was“Onthewinningcardallof theobjectsarequaint”,andforthenonecontrol items in
whichbothcardscontainhelicopters“Onthewinningcardnoneoftheobjectsarehelicopters”.We
classifiedasotherdescriptions thatattempted todescribeonlyonecardbutnot through the two
obviouscompletionsweexpected.Thesedescriptionmainlyfellinoneoftwocategoriesofalternative
strategies.Onestrategyconsistedinmentioninganobjectassociatedonlywithonecardeventhough
thisresultedinastatementthatwasfalseofbothcards.Anexampleofthisstrategyforthemostitem
inTable1was“Onthewinningcardmostoftheobjectsaretelephones”,whichreferstothewinning
cardwheretelephonesarepresenteventhoughmostoftheobjectsarelamps.Theotheralternative
strategyconsistedinmakingreferencetowhethertheobjectssingledoutbythedescriptionwerethe
same ordifferent from the other objects in the card and relying on the preferred reading of the
statementasreferringtotheshapeoftheobjectsratherthanthecolour.Anexampleofthisstrategy
13
fortheadhociteminTable1was“Onthewinningcardtheobjectsinthemiddlerowareidentical”.
Becausetheamountofdatainthesetwocategorieswasrelativelysmallwedecidedtogroupthem
togetherunderthecategoryofotherstrategiesforthepurposeofouranalysis.
Inthecooperativecondition(Figure2)wefoundthatparticipantshadanoverwhelmingpreference
for true descriptions in both types of control items (all and none items) and in items containing
numerals.Inthemostandadhocitemsparticipantshadapreferencefortruedescriptionsbutthey
alsogaveaconsiderablenumberoffalsedescriptions.mostandparticularlyadhocitemsalsodiffered
fromtheotheritemcategoriesbecauseofthehighrateofotherresponsesgivenbyparticipants:12%
oftheresponsesformost itemsand30%foradhoc items. Inthepredictedperformancequestion
participantsinthecooperativeconditionestimatedthatthereceiverwouldclickonthewinningcard
78.15%ofthetime.
Figure2Proportionofresponsetypesinthecooperativecondition
In the competitive condition (Figure3) participants gavemostly equal numbers of true and false
descriptions for all types of items. Participants also gave a considerable amount of uninformative
descriptions incontrol itemsand itemswithnumerals.Therateofuninformativedescriptionswas
lowerformostandadhocItems.Mostandadhocitemsalsoexhibitedahigherrateofotherresponses
than other item categories but not as large as in the cooperative condition. In the predicted
performancequestionparticipants in thecompetitiveconditionestimated that the receiverwould
96%
93%
59%
46%
28%
22%
12%
30%
Control
Numerals
Most
AdHoc
True False Uninformative Other
14
clickonthewinningcard49.82%ofthetime,whichwasasignificantlylowerestimatethantheone
givenbyparticipantsinthecooperativecondition(t(98.81)=-8.351,p<0.001).
Figure3Proportionsofresponsetypesinthecompetitivecondition
Weanalysedthedatabyrunningthreebinomialregressionmodels.Eachmodelregressesadifferent
outcomevariableonthesamepredictors:condition,itemtypeandtheirinteraction.Itemtypeisa
four-leveldummycodedcategoricalpredictorwithcontrolitemsasthereferencelevel.Conditionisa
two-leveldummycodedcategoricalpredictorwiththecompetitiveconditionasthereferencelevel.
Therefore, the simple effect of the cooperative condition represents the difference between
conditions for control items; the simple effects of Numerals, Most and ad hoc represent their
differencefromcontrolitemsinthecompetitivecondition,andtheirinteractionswithconditionfactor
expresshowtheirdifferencefromthecontrolitemschangesinthecooperativecondition.
Model1addressesthequestionofwhatfactorsaffectthepreferenceforafalsedescriptionovera
truedescriptionandtheoutcomevariablewasabinaryvariablewheretruedescriptionswerecoded
as0andfalsedescriptionswerecodedas1.ThedetailsoftheanalysisaresummarisedinTable2.The
threetypesofexperimentalitemsarenotsignificantlydifferentfromcontrolitemsinthecompetitive
condition.Thenegativeeffectofthecooperativeconditionindicatesthatparticipantswerelesslikely
togivefalsedescriptionsforcontrolitemsinthecooperativeconditioncomparedtothecompetitive
condition.Thesignificantinteractionsofmostandadhocindicatethatthedifferencebetweenthese
37%
32%
40%
37%
41%
34%
47%
51%
22%
33%
6% 7%
10%
Control
Numerals
Most
AdHoc
True False Uninformative Other
15
items from the control items in the cooperative condition isdifferent from theirdifference in the
competitivecondition.Wedonotfindevidencethatthiswasthecasefornumerals.
Table2.Model1
Β S.E. Z p-value(intercept) 0.08 0.07 1.14 .254Numerals -0.03 0.15 -0.18 .854Most 0.07 0.14 0.50 .613AdHoc 0.24 0.16 1.48 .139Cooperative -4.22 0.29 -14.62 <.001Numerals*Cooperative 0.39 0.52 0.74 .459Most*Cooperative 3.32 0.34 9.77 <.001AdHoc*Cooperative 3.19 0.37 8.53 <.001Model2addressesthequestionofwhatfactorsaffectthepreferenceforanuninformativedescription
overallothertypesofdescriptions(true,falseandother)andtheoutcomevariableisabinaryvariable
where uninformative descriptionswere coded as 1 and all other responseswere coded as 0. The
details of the analysis are summarised in Table 3. The negative simple effect of the cooperative
condition indicates thatparticipantswere less likely togiveuninformativedescriptions for control
itemsinthecooperativeconditioncomparedtothecompetitivecondition.Thenegativesimpleeffects
ofmostandadhoc indicatethatparticipantswerelesslikelytogiveuninformativedescriptionsfor
these itemscomparedtocontrol items in thecompetitivecondition.Onthecontrary, thepositive
simple effect of numerals indicates that participants were more likely to give uninformative
descriptions for these items compared to controls in the competitive condition. The significant
interactionindicatesthatthedifferencebetweenadhoc itemsandcontrolitemsisdifferentinthe
cooperativeconditioncomparedtothecompetitivecondition.Infact,whiletheratesofuninformative
descriptionsforcontrolitemsandadhocitemsisroughlythesameinthecooperativecondition(i.e.
2%),theyareconsiderablydifferentinthecompetitivecondition.
Table3.Model2
Β S.E. Z p-value(intercept) -1.27 0.08 -16.68 <.001Numerals 0.55 0.14 3.96 <.001Most -1.49 0.24 -6.14 <.001
16
AdHoc -2.32 0.42 -5.52 <.001Cooperative -2.40 0.23 -10.27 <.001Numerals*Cooperative 0.17 0.38 0.45 0.655434Most*Cooperative 0.22 0.78 0.28 0.779232AdHoc*Cooperative 2.39 0.66 3.64 <.001Model3addressesthequestionofwhatfactorspushedparticipantstoresorttootherdescriptions
insteadofgivingtrue,falseoruninformativedescriptions.Theoutcomevariableforthismodelwasa
binaryvariablewhereotherdescriptionswerecodedas1andallotherresponseswerecodedas0.
The details of the analysis are summarised in Table 4. The only significant effects are the simple
positiveeffectsofmostandadhoc,whichindicatethatparticipantsweremorelikelytogiveother
descriptions to these items than to control items in the competitive condition. The fact that their
interactionsarenotsignificantmeansthatwehavenoevidencethatthistrendwasanydifferentin
thecooperativecondition.
Table4.Model3
β S.E. Z p-value(intercept) -6.915 1 -6.911 <.001Numerals 1.797 1.226 1.465 0.143Most 4.35 1.023 4.253 <.001AdHoc 4.747 1.024 4.634 <.001Cooperative -13.651 609.583 -0.022 0.982Numerals*Cooperative -1.797 1219.166 -0.001 0.999Most*Cooperative 14.262 609.58 0.02 0.981AdHoc*Cooperative 14.962 609.58 0.02 0.98Although theoverall proportionsof true and falsedescriptions in the competitive condition seem
roughlyequalforeachcategoryofitems(seeFigure3),wefoundthattheseoverallproportionswere
the result of different, sometimes opposed, individual strategies. For example, some participants
consistentlygavefalsedescriptionswhileothersconsistentlygavetruedescriptions,andwefound
thatthesestrategieswerereflectedintheperformancepredictionsthatparticipantsgaveaboutthe
receiverwhowouldreadtheirdescriptions.Forthisreasonweperformedaclusteringanalysisonthe
participantsinthecompetitiveconditioninordertofindoutwhethertheseindividualstrategiescould
be classified under a number of meaningful strategy profiles. We performed k-means clustering
17
analysisinRusingfourvariablesforeachparticipant:overallproportionoftruedescriptions,overall
proportion of false descriptions, overall proportion of uninformative descriptions and expected
performanceof thereceiver.Usingtheaveragesilhouettemethodology (Rousseeuw,1987),which
allows to visually compare the quality of different clustering solutions in terms of tightness and
separation of the clusters in each solution, we determined that the clustering solution that best
summarisedthedatawasathree-clustersolution.ThethreeclustersaresummarisedinTable5which
reports each cluster’s average values of the four variables we used in the analysis (i.e. expected
performance, overall proportions of true, false and uninformative descriptions; in the table as
clusteringvariables)togetherwitheachcluster’saverageproportionsofdescriptiontypesforeach
categoryofitems.
Table5.Clusteringvariablesandproportionsofresponsetypesforeachitemcategorybyclusteringgroups
Clusterandsize
Itemtype
Responsetype Performance
True False Uninformative Otherstrategies
Cluster1 Clusteringvar. 59% 17% 16% 69%N18 Control 74% 07% 19% 01% Numeral 55% 03% 37% 06% Most 48% 49% 02% 01% AdHoc 49% 42% 1% 8% Cluster2 Clusteringvar. 24% 57% 11% 38%N32 Control 21% 66% 13% 0% Numeral 24% 54% 22% 0% Most 37% 51% 3% 9% AdHoc 31% 55% 2% 12% Cluster3 Clusteringvar. 16% 15% 59% 55%N6 Control 16% 6% 79% 0% Numeral 6% 19% 75% 0% Most 31% 19% 33% 17% AdHoc 21% 50% 12% 17% Cluster 1 is characterised by a high rate of true descriptions and a high expected success rate
(performance)ofthereceiver.Participantsinthisclusterwerethereforemostlyplayingthegameas
18
iftheirgoalwastohelpthereceiverasahighsuccessrateofthereceiverinthecompetitivecondition
corresponds to a low performance of the signaller, who caused the signaller tomake only a few
mistkes.Cluster2,themostnumerous,ischaracterisedbyahighrateoffalsedescriptionsandbythe
lowestexpectedperformanceofthethreeclusters.Participantsinthisclusterweremostlylyingor
falsely implicating and they expected their strategy to cause the receiver to performworse than
chance.Inotherwords,thosewhobelievedtoperformwellasdeceptivesenders(andbetterthan
chance)areexactlythosewhousedmisleadingimplicatures.Cluster3ischaracterisedbythehighest
rateofuninformativedescriptionsandanexpectedperformancenearchance.Althoughtherateof
uninformativedescriptionsthattheseparticipantsgaveformostandadhoc Items isstill relatively
highcomparedtotheotherclustersitislowerthanforControlitemsandNumeralsasparticipants
seemtorelymoreonotherstrategiesandontrueandfalseresponses.
Discussion
Ourparticipantsplayedasignallinggameinwhichtheywereeitherhelpingorcompetingagainsta
receiver.Theirtaskwastocompletedescriptionsthatcouldhelpthereceiverchoosethewiningcard
outofeachpairofcards.Someitemspushedsignallerstoconveythehintviaassertionandothersvia
implicature (numerals,most,andadhoc).Wecategorised thehintsusedbyparticipants into four
types.Truehints,thatcouldbeeithertrueassertionsortrueimplicatures,falsehints,uninformative
hints and other, where participants used ways of referring to one of the cards that we did not
anticipate.Theitemswereconstructedinsuchawaythatthedescriptionsweanticipatedwerethe
obviouscompletionsforthedescriptiontemplates,thereforewefounditinterestingthatparticipants
resortedtootherstrategiesforcompletingthedescriptions.
Inthecooperativeconditionparticipantsoverwhelminglychosetruecompletionsforthecontrolitems
and the items containing numerals, with very few uninformative or other descriptions. This was
expected given that their aimwas to help the receiver find thewinning card. Participants gave a
considerablenumberoffalsedescriptionsformostandadhocitemsinthecooperativecondition.This
isincontrasttothegoalofhelpingthereceiverandthemostlikelyexplanationforthehighrateof
19
false descriptions is that in some cases the potential implicatures of the descriptions were not
available to participants and they randomly chose between the two most obvious completions.
Furthermore,mostandespeciallyadhocitemselicitedaconsiderablerateofotherdescriptions.As
wepointedout,this is interestingasthesealternativestrategieswerenotobviouslyavailable.One
possible explanation is that participants anticipated the potential implicatures in these items and
preferredtochooseotherstrategiesforcommunicatingthekeyinformationinthedescriptionrather
thantrusttherelativelyunreliablechannelofimplicitcommunication(Reboul,2017).Analternative
explanation,whichisalsoconsistentwiththehighrateoffalsedescriptions,isthatthisbehaviouris
also causedby participants not seeing the potential implicatures of the twoobvious descriptions,
whichwithouttheimplicaturesaresimplyuninformativeforthereceiver.Andinordertoavoidgiving
an uninformative hint participants may have preferred resorting to other strategies. Previous
productionstudieshaveinvestigatedsituationswherespeakersneededtocommunicateinformation
through an inference rather thanby asserting it. They found that speakersoften, but not always,
expressthemselves inawaythatallowthehearertodrawan informative inference. Inastudyby
DaviesandKatsos(2010)participantsneededtorefertoobjectsinsituationswhereusingabarenoun
wouldbeunderinformative(e.g.'passmetheapple'inasituationwheretherearetwoapples)and
theiradultparticipantsusedexpressionsthatallowedthehearertodrawacontrastiveinference(e.g.
pass me the red apple) almost 80% of the time. In a study by Degen, Franke and Jäger (2013),
participantsplayedasignallinggamewheretheycouldonlysendmessagesthatdidnotconveythe
key information unambiguously. Among the four messages they could choose from, only one
conveyedthekeyinformationthroughaninferencewhiletheotherswereambiguousor incorrect.
Theyfoundthatparticipantssentthetargetmessageonroughly80%ofthetrialsiftheinferencewas
simple and on 50% of the trials if the inference was complex. Our results from the cooperative
conditionarethereforetowardsthelowendofthespectrumasourparticipantsexpressedthemselves
inawaythatwouldallowareceivertoinferatrueinferenceon59%fortrialsformostitemsand46%
20
oftrialsforadhocitems.Theserateswereprobablyaffectedbythecharacteristicsofouritemsand
thecomplexityofthetask.
Thecompetitiveconditiondifferedfromthecooperativeconditionmainly intheratesof falseand
uninformativedescriptions.Participantsinthisconditionweremorelikelytogivefalsehints(i.e.to
lie)incontrolitems:theygavefalseandtruedescriptionsatroughlythesamerate.Wealsofoundno
evidencethattheratiooffalsetotruedescriptionswasdifferentforanyoftheotheritemtypesinthe
competitivecondition.Itemscontainingnumerals,likecontrolitems,showedalargeincreaseinthe
rateof falsedescription in the competitive condition compared to the cooperative condition. The
relativeincreaseintherateoffalsedescriptionwassignificantlysmallerformostandadhocitemsas
theseitemselicitedaconsiderableamountoffalsedescriptionsinthecooperativeconditionaswell.
Participantswerealsomorelikelytoproduceuninformativedescriptionsforcontrolitemscompared
tothecooperativecondition.Itemscontainingnumeralsalsoelicitedmoreuninformativedescriptions
inthecompetitivecondition,infacttheyelicitedevenmorethancontrolitems.Formostandadhoc
itemstheuninformativedescriptionswereveryfewandsignificantlylessthanorcontrolitems.The
factthatcontrolitemselicitedahigherrateoffalseanduninformativedescriptionsinthecompetitive
condition suggests that ourmanipulation had an effect as participants were aiming to cause the
receivertomakemistakeseitherbylyingorbybeinguninformative.Thefactthatparticipantsrelied
eitheronuninformativehintsandonequalratiosoftrueandfalsehintssuggeststhattheydidnot
expect to be able to cause the receiver to do worse than chance. This is also consistent with
participants in thecompetitivecondition indicating that theyexpectedthereceiver toclickonthe
winningcardroughly50%ofthetime.Inasimilarnon-verbalsignallinggame,Ransom,Voorspoels,
Perfors and Navarro (2017) found that when the signaller expected a distrustful receiver, their
participantsgaveuninformativehintsroughly75%ofthetimeandonlyafewmisleadingorhelpful
hints.Althoughwecanassumethatourparticipantswerealsoexpectingadistrustful receiver,we
foundamuchlowerrateofuninformativehintsthanRansomandcolleagues.Thisismostlikelydue
tothefactthatwhileparticipantsintheirstudycouldnotlie,ourparticipantswereallowedtolieand
21
thereforetheycouldtakeadvantageofthefactthatreceiverswouldnotknowif informativehints
weretrueorfalse.Twostudiesonimplicaturecomprehensioninnon-cooperativesettingfoundthat
participantstendedtoinferlessimplicaturesfromtheutterancesofacompetitivespeakercompared
to a cooperative speaker (Pryslopska, 2013; Dulcinati and Pouscoulous, 2017). Dulcinati and
Pouscoulous (2017) employed a competitive signalling game very similar to the one used in the
presentexperimentwhereparticipantsplayroleofreceiverandknewthatthesignallerwasallowed
to lie. They found that both assertions and implicatures communicated by the signaller were
interpretedasfalsehalfofthetimeandtruehalfofthetime.Therefore,theexpectationsoftheir
participantsseemtomatchthebehaviourofthesignallersinourstudy,whousedtrueandfalsehints
inroughlyequalmeasurebothinassertionsandimplicatures.
Oneinterestingaspectofthewaythethreetypesofexperimentalitemswereusedbyparticipantsis
thefactthatitemscontainingnumeralspatternedwithcontrolitemsratherthanwiththeothertwo
categoriesofimplicatureitems:adhocandmost.Wefoundnoevidencethatnumeralswereusedany
differentlythancontrolitemsintermsofpreferenceforfalsedescriptionsovertruedescriptionsorin
termsoftherateofotherdescriptionsineitherofthetwoconditions.Incontrasttomostandadhoc
items,itemscontainingnumeralsdidnotelicitmorefalsehintsthancontrolitemsinthecooperative
condition;whichsuggeststhatiftheexactinterpretationofnumeralsisaninferenceinourstudyit
wasasavailableasthesemanticmeaningofthequantifiersallandnone.Anotherdifferenceform
mostandadhocitemsisthatitemswithnumeralsdidnotelicitmoreotherresponsesthancontrol
itemsinthecompetitivecondition;whichmaybedueeithertotheavailabilityofotherstrategiesfor
numeral itemsor due to themotivation to seek alternative strategies for these items. Itemswith
numeralsdiddifferfromcontrolitemsinelicitingmorefalsedescriptionsinthecompetitivecondition.
However,thisdifferencewasintheoppositedirectionasmostandadhocitems,whichelicitedless
uninformative descriptions compared to control items in the competitive condition. Overall, we
interpretthispatternofresultsto indicatethatparticipantsusednumeral items inawaythatwas
closertothecontrolitemsthantotheimplicatureitems.Wefurthertakeittosuggestthattheexact
22
interpretationofnumeralsispartoftheirtruth-conditionalmeaningmeaning(Carston,1998;Breheny
2008)andnotanimplicature(Horn,1972;Gazdar,1979;Levinson,2000).
Aswementioned,mostandadhocitemswereuseddifferentlythancontrolandnumeralitems.They
elicitedahighernumberoffalsedescriptionsthancontrolsinthecooperativecondition,whichmay
beduetoaloweravailabilityoftheirupperboundinterpretation.Theseitemswerealsolesslikelyto
elicituninformativedescriptionscomparedtocontrol items inthecompetitivecondition.Thismay
also be attributed to a lower availability of the upper bound interpretation, as the lower bound
interpretationofmost andadhoc description resulted in a description thatwas as unhelpful and
indeedequivalenttoanuninformative. Inourview,thesedifferencescanbeattributedtothefact
thatthekeyinformationwasconveyedthroughassertionincontrolandnumeralitems,andthrough
animplicatureinmostandadhocitems.Thesimilarityinthewayparticipantsusedmostandadhoc
itemsinsteadsuggeststhattherewasnodifferenceinthewayparticipantscomputedthesetwotypes
of implicatures.Our interpretation is therefore in supportof theviewthat scalar implicaturesand
particularisedimplicaturesarecomputedinthesameway(Sperber&Wilson,1995;Carston,2002;
Breheny,Katsos&Williams,2006;Geurts,2010),and is incontrastwiththe ideathat implicatures
arising from lexicalisedscalesaredefaultmeaningscomputeddifferently from implicaturesarising
fromadhocscales(e.g.Levinson,2000;Chierchia,2004).
Although in the competitive condition the ratesof trueand false responsesareoverall equal,our
clusteranalysissuggeststhatthisisactuallytheresultofdifferentopposingstrategiesthatparticipants
tendedtowards.Onetendencywasforparticipantstogivemoretruehints,at least incontroland
numeral items, and expect the receiver to have a better performance as a result. The simplest
explanationforwhysomeparticipantschosethisstrategy,whichisincontrastwiththeirgoalinthe
competitivecondition,isthattheywerenotfollowingtheinstructionsinthisrespect.Incontrast,the
largestgroupofparticipantstendedtogivemorefalsedescriptionsandtheyexpectedthereceiverto
performworsethanchanceasaresult.Theseparticipantsgaveahigherrateoffalsedescriptionsfor
controlitemsaswellasforexperimentalitems,suggestingthattheywereexpectingthereceiverto
23
drawimplicaturesfromtheirutterances.Althoughthisisonlyanumericalobservationastherewere
notenoughdatatoperformmeaningfulstatisticaltestsonthissubgroupofparticipants,itwouldbe
interestingtoinvestigatethispreferencefurther.Ifparticipantsdidexpecttheirreceivertodrawfalse
implicaturestheymusthaveexpectedthereceivertoseethemnotasoptingoutbutascooperative
enough(inaGriceansense)tobecommunicatingimplicatures.Athirdsmallergroupofparticipants
tendedtogivemoreuninformativedescriptions.Aninterestingfeatureofthisstrategyisthatbygiving
anuninformativehintparticipantsmadetheunhelpfulnessoftheirdescriptionsmanifesttotheother
player.InGriceanterms,whilegivingfalseresponsesmightbeacaseofacovertviolationofthemaxim
ofquality,givinguninformativeresponsessignalsthatthespeaker isoptingoutofthecooperative
principle.Ononehand,thisstrategymightbeacalculatedwayofforcingthereceivertochooseat
random.Ontheotherhand,someparticipantsmightprefertobeseenasoptingoutbecausethey
haveanaversiontolying.Infact,multiplestudieshavefoundthatpeoplehaveanaversiontolying
even in economic games where they would benefit from deceiving their interlocutor (Lundquist,
Ellingsen,Gribbe&Johannesson,2009;Gneezy,Rockenbach&Serra-Garcia,2013).Inotherwords,
ourparticipantsmayhavegivenuninformativehintsinordertobehonestaboutthefactthatthey
werebeingunhelpful.
Inconclusion,wefoundthatuncooperativespeakerstendtobemoreuninformativeandtoliemore
thaniftheirgoalwastohelptheirinterlocutor,atleastinthekindofcompetitivesituationwesetup
inourstudy.Knowingthattheirinterlocutorcouldbecompletelydistrustfulseemstopushspeakers
towardsthestrategyoftellingasmanytruthsas lies,whichseemstomatchtheexpectationsthat
hearers have (Dulcinati & Pouscoulous, 2017). However, we found that our participants are not
uniform in their strategyanda largegroupofourparticipantsusedahigher rateof liesand false
implicatures,whichsuggeststhattheyexpectedtheir interlocutorstodrawimplicaturesfromtheir
utterances.Wehopethatfurtherresearchwillfindclearerindicationsastowhetherspeakersexpect
theirinterlocutorstoinferimplicaturesinuncooperativesituations.Ourresultsalsosuggestthatthe
exact interpretation of numeral expressions is part of their truth conditional meaning and that
24
particularised implicaturesand the implicaturesof scalar expressionsareusedand computed in a
similarway.Furtherexperimentalresearchontheuseofimplicaturesinnon-cooperativesituationsis
needed aswe believe that this could provide information on factors that influence the speakers’
decisiononwhethertocommunicatesomethingimplicitlyorexplicitlyaswellasaboutthenatureof
theimplicaturethemselvesandtheroleofcooperationintheirderivation.
25
ReferencesBale,A.C.,Brooks,N.,&Barner,D.(2010).Quantityimplicatureandaccesstoscalaralternativesin
languageacquisition.SemanticsandLinguisticTheory,20,525-543.Breheny,R.,Katsos,N.&Williams,J.(2006).Aregeneralisedscalarimplicaturesgeneratedbydefault?
An on-line investigation into the role of context in generating pragmatic inferences.Cognition,100(3),434-463.
Breheny, R. (2008). A new look at the semantics and pragmatics of numerically quantified nounphrases.JournalofSemantics,25(2),93-139.
Bull, P. (2016,October 24). TheresaMayhas a very special technique for avoidingquestions.TheConversation.Retrievedfrom:theconversation.com
Carston, R. (1998). Informativeness, relevance and scalar implicature. In: Carston, R. &Uchida, S.(eds.),RelevanceTheory:ApplicationsandImplications(pp179-236).Amsterdam:JohnBenjamins.
Carston, R. (2002). Thoughts and Utterances: The Pragmatics of Explicit Communication. Oxford:Blackwell.
Carston, R. (2009). The explicit/implicit distinction in pragmatics and the limits of explicitcommunication.InternationalReviewofPragmatics,1(1),35-62.
Chierchia,G.(2004).Scalarimplicatures,polarityphenomena,andthesyntax/pragmaticsinterface.InBelletti,A.(ed.),Structuresandbeyond.Oxford:OxfordUniversityPress.39-103.Chierchia,G.,Crain,S.,Guasti,M.T.,Gualmini,A.,&Meroni,L.(2001).Theacquisitionofdisjunction:
Evidenceforagrammaticalviewofscalarimplicatures.InA.H.-J.Do,L.Domingues,&A.Johansen,(Eds.),Proceedingsofthe25thBostonUniversityConferenceonLanguageDevelopment(pp.157–168).Somerville,MA:CascadillaPress.
Coleman,L.,&Kay,P.(1981).Prototypesemantics:TheEnglishwordlie.Language,26-44.Davies,C.,&Katsos,N.(2010).Over-informativechildren:Production/comprehensionasymmetryor
tolerancetopragmaticviolations?.Lingua,120(8),1956-1972.Degen, J., Franke, M., & Jäger, G. (2013). Cost-based pragmatic inference about referential
expressions.Proceedingsofthe35thannualmeetingoftheCognitiveScienceSociety,376–381.Doran,R.,Baker,R.E.,McNabb,Y.,Larson,M.,&Ward,G.(2009).Onthenon-unifiednatureofscalar
implicature:anempiricalinvestigation.InternationalReviewofPragmatics,1(2),211-248.Doran,R.,Ward,G.,Larson,M.,McNabb,Y.,&Baker,R.E.(2012).Anovelexperimentalparadigmfor
distinguishingbetweenwhatissaidandwhatisimplicated.Language,88(1),124-154.Dulcinati,G.,&Pouscoulous,N.(2017).Quantityimplicaturesinacompetitivegame.Talkpresented
atXPRAG,Cologne,Germany.Dynel,M. (2011).Awebofdeceit:Aneo-Griceanviewontypesofverbaldeception. International
ReviewofPragmatics,3(2),139-167.Dynel,M.(2015).Intentiontodeceive,bald-facedlies,anddeceptiveimplicature:InsightsintoLying
atthesemantics-pragmaticsinterface.InterculturalPragmatics,12(3),309-332.Gadzar,G.(1979).Pragmatics:Implicature,presuppositionandlogicalform.NewYork,NY:Academic
Press.Geurts,B.(2010).Quantityimplicatures.CambridgeUniversityPress.Gneezy,U.,Rockenbach,B.,&Serra-Garcia,M.(2013).Measuringlyingaversion.JournalofEconomic
Behavior&Organization,93,293-300.Grice,H.P.(1989).Studiesinthewayofwords.Cambridge:MA:HarvardUniversityPress.Gualmini,A.,Crain,S.,Meroni,L.,Chierchia,G.,&Guasti,M.T.(2001).Atthesemantics/pragmatics
interfaceinchildlanguage.InSemanticsandLinguisticTheory,11,231-247).Hardin, K. J. (2010). The Spanish notion of lie: revisiting Coleman and Kay. Journal of
Pragmatics,42(12),3199-3213.
26
Horn, L. (1972). On the Semantic Properties of Logical Operators in English. UCLA dissertation,distributedbyIndianaUniversityLinguisticsClub,1976.
Huang, Y., & Snedeker, J. (2009a). On-line interpretation of scalar quantifiers: Insight into thesemantics-pragmaticsinterface.CognitivePsychology,58,376–415.
Huang,Y.T.,Spelke,E.,&Snedeker,J.(2013).Whatexactlydonumbersmean?.LanguageLearningandDevelopment,9(2),105-129.
Hurewitz, F., Papafragou,A.,Gleitman, L.,&Gelman,R. (2006).Asymmetries in the acquisitionofnumbersandquantifiers.LanguageLearningandDevelopment,2,77-96.
Isenberg,A.(1973).DeontologyandtheEthicsofLying.InAestheticsandTheoryofCriticism:SelectedEssaysofArnoldIsenberg(pp.245–264).Chicago:UniversityofChicagoPress
IshiharaS.(1917).TestsforColor-Blindness.Handaya,Tokyo:HongoHarukichoKatsos, N. (2009). Evaluating under-informative utterances with context-dependent and
contextindependent scales: experimental and theoretical implications. In U. Sauerland & K.Yatsushiro (eds.), Experimental Semantics and Pragmatics. Basingstoke: Palgrave Studies inPragmatics,Language&Cognition,51-73.
Katsos,N.,&Cummins,C.(2010).Pragmatics:fromtheorytoexperimentandbackagain.LanguageandLinguisticsCompass,4(5),282-295.
Kennedy, C. (2015). A “de-Fregean” semantics (and neo-Gricean pragmatics) for modified andunmodifiednumerals.Semantics&Pragmatics,8(10),1-44.
Levinson, S. (2000). Presumptive meanings: The theory of generalized conversational implicature.Cambridge,MA:MITpress.
Lundquist, T., Ellingsen,T.,Gribbe,E.,& Johannesson,M. (2009). Theaversion to lying. JournalofEconomicBehavior&Organization,70(1),81-92.
Mahon,J.E.(2007).Adefinitionofdeceiving.InternationalJournalofAppliedPhilosophy,21(2),181-194.
Meibauer,J.(2005).Lyingandfalselyimplicating.JournalofPragmatics,37(9),1373-1399.Meibauer,J.(2014).Lyingatthesemantics-pragmaticsinterface.Berlin:MoutondeGruyter.Montague,R.,Navarro,D.,Perfors,A.,Warner,R.,&Shafto,P.(2011).Tocatchaliar:Theeffectsof
truthfulanddeceptivetestimonyoninferentiallearning.Proceedingsofthe33rdAnnualMeetingoftheCognitiveScienceSociety,1312-1317.
Morency, P., Oswald, S., & de Saussure, L. (2008). Explicitness, implicitness and commitmentattribution:Acognitivepragmaticapproach.Belgianjournaloflinguistics,22(1),197-219.
Nicolle, S.,&Clark,B. (1999). Experimentalpragmatics andwhat is said:A response toGibbsandMoise.Cognition,69(3),337-354.
Noveck,I.A.(2001).Whenchildrenaremorelogicalthanadults.Cognition,86,253-282Papafragou,A.&Musolino,J.(2003).Scalarimplicatures:experimentsatthesemantics/pragmatics
interface.Cognition,86,253-282.Pinker,S.,Nowak,M.A.,&Lee,J.J.(2008).Thelogicofindirectspeech.ProceedingsoftheNational
Academyofsciences,105(3),833-838.Pouscoulous, N., Noveck, I., Politzer, G., Bastide, A. (2007) Processing costs and implicature
development,LanguageAcquisition,14.4,347-376.Primoratz,I.(1984).Lyingandthe“MethodsofEthics”,InternationalStudiesinPhilosophy,16,35–57.Pryslopska,A.(2013).Implicaturesinuncooperativecontexts:Evidencefromavisualworldparadigm.
PosterpresentedatXPRAG,Utrecht,Netherlands.Ransom, K., Voorspoels, W., Perfors, A., & Navarro, D. (2017). A cognitive analysis of deception
withoutlying.Proceedingsofthe39thAnnualConferenceoftheCognitiveScienceSociety,992-997Reboul,A.(2017).Isimplicitcommunicationawaytoescapeepistemicvigilance?.InAssimakopoulos
S.(Ed.),PragmaticsatitsInterfaces,(pp.91-112).DeGruyter.
27
Recanati,F.(2004)‘‘Whatissaid’andthesemantics/pragmaticsdistinction’, inC.Bianchi(ed.)TheSemantics/PragmaticsDistinction,Stanford,CA:CSLIPublications.
Rousseeuw,P. J. (1987).Silhouettes:agraphicalaid to the interpretationandvalidationof clusteranalysis.Journalofcomputationalandappliedmathematics,20,53-65.
Russell, B. (2011). Topics in the computation of scalar implicatures (Doctoral dissertation, PhDdissertation,BrownUniversity).
Stiller,A.,Goodman,N.,&Frank,M.(2011,January).Ad-hocscalarimplicatureinadultsandchildren.Proceedingsofthe33rdAnnualMeetingoftheCognitiveScienceSociety,2134-2139.
Sperber,D.&Wilson,D.(1995).Relevance:CommunicationandCognition.Oxford:Blackwell.van Tiel, B. (2014).QuantityMatters: Implicatures, Typicality and Truth. PhD thesis, University of
Nijmegen.
28
AppendixANum_01 Num_02 Num_03 Num_04
Num_05 Num_06 Most_07 Most_08
Most_09 Most_10 Most_11 Most_12
Adhoc_13 Adhoc_14 Adhoc_15 Adhoc_16
Adhoc_17 Adhoc_18 None_19 None_20
None_21 None_22 None_23 None_24
None_25 None_26 None_27 All_28
All_29 All_30 All_31 All_32
All_33 All_34 All_35 All_36
29
Item Description:“Onthewinningcard…”
True False ItemType Description:“Onthewinningcard…”
True False
Num_01 twooftheobjectsare blue Mugs None_19 noneoftheobjectsare green pink
Num_02 twooftheobjectsare socks Pink None_20 noneoftheobjectsare cakes teapots
Num_03 twooftheobjectsare yellow saws None_21 noneoftheobjectsare pink yellow
Num_04 threeoftheobjectsare Boats green None_22 noneoftheobjectsare kettles bananas
Num_05 threeoftheobjectsare Buckets yellow None_23 noneoftheobjectsare Blue pink
Num_06 threeoftheobjectsare Pink candles None_24 noneoftheobjectsare Pans jars
Most_07 mostoftheobjectsare Pink keys None_25 noneoftheobjectsare Blue green
Most_08 mostoftheobjectsare lamps yellow None_26 noneoftheobjectsare cars sofas
Most_09 mostoftheobjectsare yellow flags None_27 noneoftheobjectsare Blue yellow
Most_10 mostoftheobjectsare beds blue All_28 alloftheobjectsare umbrellas Rockets
Most_11 mostoftheobjectsare pink books All_29 alloftheobjectsare green Yellow
Most_12 mostoftheobjectsare flowers blue All_30 alloftheobjectsare Shoes trophies
Adhoc_13 theobjectsinthetoproware blue kites All_31 alloftheobjectsare Pink green
Adhoc_14 theobjectsinthemiddleroware crowns blue All_32 Alloftheobjectsare Drums tents
Adhoc_15 theobjectsinthebottomroware yellow forks All_33 alloftheobjectsare yellow Pink
Adhoc_16 theobjectsinthetoproware vases pink All_34 alloftheobjectsare Trumpets Carrots
Adhoc_17 theobjectsinthemiddleroware green apples All_35 alloftheobjectsare blue Pink
Adhoc_18 theobjectsinthebottomroware pink bells All_36 alloftheobjectsare bottles Bikes
AppendixBInstructionsforthe[cooperative/competitive]condition:HOWTHEGAMEWORKS(pleasereadcarefully)Thisisa[cooperative/competitive]gamewithtwoplayers:adescriberandaguesser.Inthisgameyouarethedescriber.(theguesserwillplayinasecondphase)Ineachroundofthegameyou'llseeawinningcard(withagreenborder)andalosingcard(witharedborder)andyou'llhavetocompleteadescriptionofthewinningcard.Theguesserwillreadyourdescriptionandthey'llseebothcardsbuttheywon'tknowwhichoneisthewinningcard.[Inthisgamebothyouandtheguesserscorepointswhentheguesserclicksonwinningcards./Inthisgame both you and the guesser score pointswhen the guesser scores pointswhen they click onwinningcardswhereasyouscorepointswhentheguesserclicksonlosingcards.]Theguesserknowsthatthisisacompetitivegamebuttheydon'tknowthatyouarecompletingthedescriptionsinsteadofwritingthemfreely.PleasecompletetheinstructionwithONLYONEWORD.Youcantalkaboutacolouroratypeofobject.You canwrite false descriptions [and/but] remember that you are [helping / playing against] theguessersoyourgoalistomakethem[win/lose].Controlquestions[andanswers]forbothconditions:Beforeyouplay,let'scheckthatyouknowtherules:(Youwon'tbeabletoplayifyougetthesewrong)Thewinningcardis... [Theredone/Thegreenone]Yourdescriptionscanbe...[Onlytrue/Eithertrueorfalse]Inthisgameyouare... [Helpingtheguesserclickonwinningcards/Playingagainsttheguesser]Youcanwrite...[Maximum1word/Maximum3words]