240 additional matched literal and metaphorical sentences for testing neural hypotheses abou

13 Pages·2016·0.5 MB·English

BehavRes(2017)49:471–483
DOI10.3758/s13428-016-0717-1
Stimulus needs are a moving target: 240 additional
matched literal and metaphorical sentences for testing
neural hypotheses about metaphor
EileenR.Cardillo1,3&ChristineWatson2&AnjanChatterjee1,3
Publishedonline:8March2016
#PsychonomicSociety,Inc.2016
Abstract As the cognitive neuroscience of metaphor has and auditory imagery). These itemsextend previously published
evolved, sotoo havethe theoretical questions ofgreatestin- stimuli,fillinganextantgapinmetaphorresearchandallowingfor
terest. To keep pace withthese developments, in the present testsofnewbehavioralandneuralhypothesesaboutmetaphor.
studywegeneratedalargesetofmetaphoricandliteralsen-
tence pairs ideally suitedtoaddressingthe current methodo- Keywords Figurativelanguagecomprehension .Nominal
logicalandconceptualneedsofmetaphorresearchers.Inpar- metaphors .Sentencenorms
ticular,theneedhasemergedtodistinguishmetaphorsalong
three dimensions: the grammatical class of their base terms,
thesensorimotorfeaturesoftheirbaseterms,andthesyntactic Despiteitspoeticassociations,metaphoriclanguageplaysavital
form in which the base terms appear. To meet this need, we andfrequentroleineverydaylanguage.Mostobviously,amet-
generatednominalmetaphors(andmatchedliteralsentences) aphormaybepreferredoveraliteralexpressioninordertofoster
using entity nouns as the base terms, with the intention that fresh insight or capture attention by virtue of its novelty. By
theybeusedinconcertwithalreadypublishedsetsofpredi- expressingfamiliarconceptsinnewways,subtletiesofmeaning
cate metaphors or nominal metaphors using event nouns. mayemerge.Metaphorsalsomakepowerfullearningtools.As
Using the results of three norming experiments, we provide theoft-citedlikeningoftheorbitofelectronsaroundanatomto
120pairsofcloselymatchedmetaphoricandliteralsentences theorbitofplanetsaroundthesunillustrates,wemorereadily
that are characterized along 14 dimensions: 11 at the sen- understandnewconceptsthroughfamiliar ones. Perhapsmost
tence level (length, frequency, concreteness, familiarity, nat- importantly,metaphorsmayallowustoconceptualizeandcom-
uralness, imageability, figurativeness, interpretability, ease municateaboutabstractconcepts(Jamrozik,McQuire,Cardillo,
of interpretation, valence, and valence judgment reaction & Chatterjee, in press; Lakoff & Johnson, 1980, 1999). As a
time), and three related to the base term (visual, motion, vehicle to abstraction, metaphor transcends the domain of
sonnetsandnovels,revealingitselftobeauniqueandessential
featureofhumancognition(Gentner,2003;Gibbs,1994).
Electronicsupplementarymaterial Theonlineversionofthisarticle
This consensus about the importance of metaphoric
(doi:10.3758/s13428-016-0717-1)containssupplementarymaterial,
thought and language contrasts with the disagreement
whichisavailabletoauthorizedusers.
concerning how we understand it. Several cognitive models
* EileenR.Cardillo of metaphor comprehension have attempted to explain the
[email protected] process.Accordingtooneview,comprehensioninitiallyrelies
onacomparisoninwhichtheconceptualsimilaritiesbetween
two superficially different domains (base and target) are
1 CenterforCognitiveNeuroscience,UniversityofPennsylvania, mappedandaligned(Gentneretal.,2001;Gentner&Wolff,
3720WalnutStreet,B51,Philadelphia,PA19104-6241,USA 1997).Inadifferentaccount,comprehensionisessentiallyan
2 MossRehabilitationHospital,Philadelphia,PA,USA actofcategorization,wherebyabasetermistakenasaproto-
3 DepartmentofNeurology,UniversityofPennsylvania, typicalmemberofanewlyconstructedcategory,ofwhichthe
Philadelphia,PA19104,USA target is also a member (Glucksberg, 2003; Glucksberg &
472 BehavRes(2017)49:471–483
Keysar, 1990). Alternatively, a process of sensorimotor 2009), but it may interact with other stimulus properties in
abstraction may drive comprehension. In this view, the theoreticallyandneurallysignificantways.
metaphoric sense of the base term is the conceptual core Previously, we highlighted the three conceptualquestions
remaining after irrelevant, literal sensorimotor features we believe are the most likely to contribute to differences
have been attenuated (Chatterjee, 2008; Chen, Widick, & between neural studies and the most useful to investigate in
Chatterjee, 2008). These proposals may not be mutually futuremetaphorresearchincognitiveneuroscience(Cardillo
exclusive. Different processes may predominate at different et al., 2010): (1) Might previous findings about hemispheric
stages as the metaphoric senses of base terms become lateralizationreflectdifferencesinnovelty,ratherthanfigura-
conventionalizedintheiruse(Bowdle&Gentner,2005),and/ tiveness? (2)Are metaphoricwordsunderstood byreference
ordifferentmetaphortypesmayrelyondifferentmechanisms totheirliteralsensoryandmotorproperties,therebyentailing
(Cardillo, Schmidt, Kranjec, & Chatterjee, 2010; Schmidt, reactivationofsensorimotorcorticesfortheircomprehension?
Kranjec,Cardillo,&Chatterjee,2010). (3)Dometaphorsofdifferentsyntacticforms(e.g.,predicate
These uncertainties regarding cognitive mechanisms are vs. nominal) rely on different cognitive and neural mecha-
paralleled by uncertainties regarding neural mechanisms. nisms for comprehension? To assist addressing these ques-
Early patient research suggested a right hemisphere (RH) tions,wepreviouslygenerated560matchedliteralandmeta-
specialization for figurative language, but it lacked the phoricalsentencesofdifferentsyntacticforms(nominal,pred-
methodological rigor common to current neuropsychologi- icate)and sensory modalities(auditory, motion)and normed
cal and neuroimaging investigations (Schmidt et al., 2010). them on ten sentence-level characteristics (frequency,
This traditional view is also not well supported by neuro- concreteness, length, familiarity, naturalness, imageability,
imaging studies, which outnumber patient studies at this figurativeness, interpretability, valence, and time required to
point. Functional magnetic resonance imaging (fMRI) and makeavalencejudgment;Cardilloetal.,2010).
positron emission tomography studies suggest a motley set Although progress has been made, these questions about
of critical areas, with poor consensus regarding hemispheric neuralorganizationlargelyremainoutstanding.Atthispoint,
lateralization, not to mention regional differences within the most headway has been made regarding the question of
hemispheres.Recentmeta-analyses,however,haveindicated howhemisphericity,figurativeness,andnoveltyinteract.Both
that metaphor comprehension is largely a left-hemisphere- thegradedsaliencehypothesis(GSH;Giora,Zaidel,Soroker,
mediated process, with the RH playing a weaker and less Batori, & Kasher, 2000) and the coarse coding hypothesis
consistent role (Rapp, Mutschler, & Erb, 2012). Rather (CCH;Jung-Beeman,2005)predictaspecialrolefortheRH
than resolve the question of hemispheric specialization, inmetaphorprocessing.TheCCHattributesthislateralization
the current challenge to the field is determining the func- totheRH’sspecializationforprocessingsemanticallydistant
tional roles of the many regions involved and the condi- relationships,whichare commoninmetaphoricexpressions.
tions that recruit them. In contrast, the GSH suggests that this lateralization reflects
Clarifying these uncertainties about neural substrates en- RHdominanceforprocessinglow-saliencemeanings,aprop-
tails stimulus optimization at three levels. The first level is ertyoftenconfoundedwithfigurativeness.Salientmeanings,
psycholinguistic:Lexicalandsententialpropertiesthatimpact bycontrast,arelexicalized,context-independent,andpromi-
processing difficulty must be controlled by careful stimulus nent senses of a word. According to the GSH, salience is a
design, selection, and/or statistical analyses. Comparisons compositeconstructbutisBdeterminedprimarilybyfrequen-
between metaphors and literal sentences, or familiar and cyofexposureandexperientialfamiliarity^withaparticular
novel metaphors, are common in neuroimaging studies, word sense (Giora, 2002, p. 491). Given the importance of
but standards for matching them vary widely. The second familiaritytosalience,theGSHpredictsthattheRHprocesses
is methodological: Task-related differences are especially novel metaphors but the LH is sufficient for understanding
important to better characterize in neuroimaging studies familiar ones (Giora et al., 2000). In partial support of these
and to distinguish from figurativeness-related processes. hypotheses, the emerging picture seems to be that both
Passive reading, imagery generation, and decisions about hemispheres, and especially bilateral prefrontal cortex, are
valence, semantic relatedness, or plausibility—all of which recruited when deriving a new figurative sense. In a novel
are tasks used in fMRI studies of metaphor—entail addi- familiarizationtask,Cardilloetal.(2012)parametricallyvar-
tional and different cognitive demands beyond metaphor ied participant experience with novel nominal and predicate
comprehension.Taskdifferenceslikelycontributesignificantly metaphors to test for a right–left shift in lateralization with
to the current poor consensus regarding the neural basis of increased familiarity. Instead, their results indicated that the
metaphorcomprehensionandthenecessityofRHengagement morefamiliarametaphorwas,thelessitengagedasubsetof
(e.g., Yang, Edens, Simpson, & Krawczyk, 2009). The third thesameregionsnecessaryforcomprehendingthemetaphor
levelisconceptual:Notonlyisfigurativenesssometimescon- for the first time [inferior frontal gyrus (IFG) bilaterally, left
founded withdifficulty(Schmidt&Seger,2009;Yangetal., posterior middletemporalgyrus (pMTG), and right postero-
BehavRes(2017)49:471–483 473
lateraloccipitalcortex].Studiesdirectlycomparingmetaphors such that sensorimotor areas are more strongly recruited for
ateitherend ofthe familiarity continuum havesimilarly im- understandingnovelfigurativesensesofwordsthanforcon-
plicated bilateral IFG, and more extensive RH activation, in ventional ones (Aziz-Zadeh & Damasio, 2008; Cardillo
the processing of novel metaphors (Desai, Conant, Binder, et al., 2010). The initial abstraction of the metaphoric sense
Park, & Seidenberg, 2013; Forgács, Lukács, & Pléh, 2014; of a base term may require activating its literal sense,
Mashal, Vishne, & Laor, 2014; Subramaniam, Beeman, via sensorimotor simulation, before selecting only those
Faust, & Mashal, 2013; Subramaniam, Faust, Beeman, & more abstract, conceptual features relevant to the figu-
Mashal, 2012). Meta-analyses have confirmed that RH in- rative extension. With repeated exposure to the meta-
volvement is more likely when metaphoric expressions are phoric sense, a simulation becomes less necessary, and
novel(Rappetal.,2012),buttwomajoruncertaintiesremain: the newly learned metaphorical sense can be accessed
WhatarethefunctionalrolesoftheseRHareas,andarethey directly without reliance on sensorimotor grounding.
necessaryforcomprehension,ordotheymerelyplayafacil- Observations of similar modality-specific activation
itatoryrolewhenLHprocessingdemandsarehigh?Studiesof graded by metaphor familiarity in domains other than
novel metaphor comprehension in patients with brain injury action/motion are needed to establish confidence in this
affectingnovelty-sensitiveareashaveyettobedone,butthis conclusion.
lineofresearchpromisesinsightintobothissues. Clarifying the current neuroimaging literature also re-
The answers to the other two conceptual questions—the quires model building and testing that incorporate both
importance of metaphor type and sensorimotor grounding noun- and verb-based metaphors orthogonal to object-
for metaphor comprehension—remain unclear. On the one and action-based semantics. Currently, action semantics
hand, a number of studies now support the hypothesis that and verb figurativeness are generally conflated, as are
words,evenwhenusedmetaphorically,areprocessedinpart object semantics and noun figurativeness. For instance,
by the same sensorimotor regions relevant for perceiving or Chen et al. (2008) concluded that the greater activity in
executing the sensory and motor features relevant to their pMTG for metaphorical and literal motion verb sentences
literal senses. Texture metaphors (She had a rough day) (The man fell under her spell; The child fell under the
engage somatosensory cortex (Lacey, Stilla, & Sathian, slide) than for abstract sentences (The merchant was
2012), taste metaphors (She received a sweet compliment) greedy and gluttonous) reflected the motion features of
engage primary and secondary gustatory cortex (Citron & the action verbs—but the conflation of base term gram-
Goldberg, 2014), arm action metaphors (The Congress is matical class and action semantics leaves open the possi-
grasping the new state of affairs) engage secondary motor bility that pMTG is recruited for verb or event processing
areas for action planning and coordination (Desai et al., more generally, rather than by motion features per se
2013), and motion verb metaphors (The man fell under (Bedny, Caramazza, Grossman, Pascual-Leone, & Saxe,
the spell; Chen et al., 2008) and fictive motion sentences 2008). To test this possibility, three kinds of contrasts
(Thepipegoesintothehouse;Saygin,McCullough,Alac,& are needed: (1) a comparison of noun-based metaphors
Emmorey,2010;Wallentin,Lund,Ostergaard,Ostergaard,& and verb-based metaphors, both generated from base terms
Roepstorff,2005;Wallentin,Østergaard,Lund,Østergaard,& with salient motion; (2) a comparison of verb-based and/or
Roepstorff, 2005) engage primary motion perception area noun-based metaphors with and without motion qualities
MT+/V5orthesecondarymotionassociationcortex,pMTG. (i.e., coming from different sensory modalities); and (3) a
Ontheotherhand,embodiedcognitionaccounts(e.g.,Gallese comparison of noun-based metaphors with motion qualities
&Lakoff,2005;Gibbs,2006)positthatfigurativeextensions that do and do not refer to events. The latter two contrasts
ofactionverbsshouldalsoengageprimaryandsupplementary havenot,toourknowledge,beentested.Adirectcomparison
motor cortex, a prediction that has not generally been sup- of predicate and nominal metaphors usingnominalized verbs
ported (Chen et al., 2008; Wallentin, Lund, et al., 2005; as the base terms in the nominal condition (Cardillo et al.,
Wallentin, Østergaard, et al., 2005). Rather, motor cortex 2012) is the closest approximation to the first proposed con-
engagement appears to be modulated by familiarity: trast.Theseresultsindicatednodifferenceinhowstronglythe
Highly conventional figurative uses of verbs, as are found twometaphortypesrecruitedpMTG,suggestingthatbaseterm
in idioms, do not engage M1/SMA (Aziz-Zadeh, Wilson, semanticsareamoreimportantdeterminantofneuralprocess-
Rizzolatti, & Iacoboni, 2006; Raposo, Moss, Stamatakis, & ing than syntactic structure or base term grammatical class.
Tyler,2009),butlessfamiliarmetaphoricsensesmay(Desai However,thebasetermsinthisstudywereprimarily,butnot
etal.,2011;Obertetal.,2014).Similarly,Cardilloetal.(2012) exclusively,motion-related,againleavingopenthepossibility
observed that pMTG activation was graded by familiarity that pMTG is responsive to event semantics more generally
with action verb metaphors. These studies provide prelimi- ratherthantomotionfeaturesinparticular.Toresolvethecur-
narysupport for previoushypothesesthatthe degree ofsen- rent ambiguity regarding the relative importances of syntax,
sorimotor engagement is determined by metaphor novelty, grammatical class, and base term semantics in determining
474 BehavRes(2017)49:471–483
theneuralbasisofmetaphor,existingstimulicanbeusedtotest Study1,weestablishedvaluesforninesentencecharacteris-
the first two contrasts, but new items are required in order to ticsthatimpactcomprehensiononthebasisoftheindividual
testthethirdcriticalcontrast. wordswithineachsentence.InNormingStudy2,weacquired
Thepurposeofthisstudywastoextendexistingmetaphor sixofflineratingsrelatedtotheoverallmeaningofeachsen-
stimulitomeetthisneed.Wegeneratedandnormednominal tence.InNormingStudy3,weacquiredtwoonlinemeasures
metaphors withentitysemantics sothat, inconjunction with relatedtotheoverallmeaningofeachsentence.
our previouslypublishedstimulusset (Cardillo etal.,2010),
we now provide a superset of nominal and predicate
metaphors that will allow us to fully teasing apart the Method
influences of semantics, grammatical class, and syntactic
structure.ThestimuliofCardilloetal.(2010)consistof240 Constructionofstimuli
nominal literal–metaphor sentence pairs and 240 predicate
literal–metaphorsentencepairs,using baseterms with either Aninitialpoolof312sentencesofnominalsyntacticformwas
salientauditoryormotionfeatures(forexamples,seeTable1). generated). Each sentence consisted of two noun phrases
Critically,thebaseterminthesenominalsentencesisalwaysa linked by a copula (i.e., BAn X is a Y^), with zero, one, or
nominalizedverb,tomaximizethesemanticsimilarityofthe twomodifyingadjectives.Theadjectiveswerenotdesignedas
basetermsinthetwometaphortypes.Bycloselymatchingthe partoftheexperimentalmanipulationbut,rather,tomakethe
sensorimotor features of nominal and predicate base terms, items comparable in terms of overall length, frequency, or
theseitemsareoptimizedfordetectingprocessingdifferences concreteness, and/or toclarify the metaphoricalextension of
related to grammatical class (noun vs. verb) and syntactic the base term, since most of the metaphors were unfamiliar.
construction (nominal vs. predicate), but not event se- For all four conditions, most of the items had one adjective
mantics. In the present study, we generated nominal (n=47–51).Halfofthesentencesexpressedaliteralmeaning,
literal–metaphor pairs using entity nouns with salient andhalfexpressedametaphoricalmeaning.
auditory or motion features as the base terms (for To generate these items, 78 concrete nouns with salient
examples, see Table 1). Entity nouns refer to static auditorypropertiesand78concretenounswithsalientmotion
and concrete persons, places, or things. Thus, a compar- propertieswerefirstselectedasbaseterms(i.e.,thewordtobe
ison of nominal-event and nominal-entity items is opti- extendedmetaphorically).Next,foreachnoun,bothaliteral
mized to detect processing differences related to senso- sentenceandametaphoricalsentencewerecreated,resulting
rimotor features (dynamic, action events vs. static, con- in 156 literal–metaphor sentence pairs. In this way, in each
crete entities), while holding grammatical class and syn- pairanidenticalnounimpliedaliteralorafigurativeinterpre-
tactic construction constant. The stimuli were generated tation, depending on its context. Critically, all of the meta-
using an extensive norming procedure identical to the one phorsinvolvingauditorybasetermsweredesignedsuchthat
established in Cardillo et al. (2010), in order to allow easy no sound was implied by the figurative interpretation of the
mixing of items from the two sets, if desired. In Norming sentence. Likewise, all metaphors involving motion base
Table1 Examplesofeachsentencetype
Modality SentenceType Literal Metaphorical
Auditory Predicate* Thelecturerdronedformanyhours. Thecontractdronedformanypages.
Theanxiousauthorshriekedatthemouse. Herpaleskinshriekedinthesun.
Nominal-Event* Herimmediateremarkwasasnigger. Thebookwasasexistsnigger.
Theconversationwasahushedwhisper. Hisglancewasafurtivewhisper.
Nominal-Entity Theprizewasanuprighthoover. Hismindwasahungryhoover.
Thesolemnsongwasananthem. Thesitcomwasanationalanthem.
Motion Predicate* Thegirltangoedwiththeinstructor. Thegarlictangoedwiththeginger.
Themodeltotteredinhighheels. Thecakeshoptotteredonbankruptcy.
Nominal-Event* Thesnake’smovewasaslither. Thedealwasagreedyslither.
Hisgaitwasaconfidentswagger. Hisyachtwasarichswagger.
Nominal-Entity Thepoliceevidencewasabullet. Thecoffeewasacaffeinebullet.
Thetouristattractionwasageyser. Histemperwasafaithfulgeyser.
*ExamplestimulifromCardilloetal.(2010),includedforcomparison
BehavRes(2017)49:471–483 475
terms were designed such that no physical or fictive motion used task in fMRI studies of metaphor (see Table 1 in Bohrn,
wasimpliedbythefigurativeinterpretationofthesentence. Altmann, & Jacobs 2012), valence RT provides an important
dimension for neuroimagers. Note that we added ease of
interpretationtothisnormingstudytoprovideresearchersasub-
Overviewofthenormingstudies
jective measure of interpretation difficulty to complement our
objectivemeasure(interpretabilityscore)andouronlinemeasure
For all norming tasks, we replicated exactly the procedures
(valenceRT).3Weanticipatethatwhichmeasure(s)willbemost
(tasks, sample sizes, selection criteria, and analyses) of
importanttocontrolwillvarywithstudydesignandquestion.
Cardilloetal.(2010),tofacilitatecombiningtheitemsintoa
singlesupersetofstimulitobesampled,ifdesired.Theinitial
Normingstudy1:words
poolofsentenceswasnormedbothofflineandonline,atboth
thewordandsentencelevels,inordertocharacterizeitspsy-
Participants Sixty native English speakers were recruited
cholinguistic and semantic properties and to highlight any
from the University of Pennsylvania community in compli-
problematicitems.Beforenorming,threemeasuresoflength
ance with the procedures established by the university’s
(number of characters, number of words, and number of
InstitutionalReviewBoardandwerecompensated$15orgiv-
content words) were calculated for each sentence, as well
en course credit for their participation. Forty of the partici-
as average frequency and concreteness scores based on the
pants(meanage=19.0years,SD=1.7;30females,tenmales;
values for the content words of the sentence (i.e., nouns,
mean education = 13.0 years, SD = 1.6) made concreteness
verbs, and adjectives). Frequency values were calculated
using the popular measure established by Kučera and ratingsandjudgedthebasetermsforauditoryandvisualim-
agery(20participantsratedonehalfoftheitems,theother20
Francis (1967), as well as using values from the more
participants rated the other half), and 20 participants (mean
recent and larger corpus, SUBTLEXus (Brysbaert &
age = 20.9 years, SD = 2.3; 16 females, four males; mean
New, 2009). Concreteness values were taken from the
education = 14.7 years, SD = 1.4) rated the base words for
MRC Psycholinguistic Database (Coltheart, 1981) and the
motionimagery.
University of South Florida Norms (Nelson, McEvoy, &
Schreiber 1998). For those words for which concreteness
Stimuli The initial pool of sentences contained 953 content
ratings were not found in either of these databases, we
collected our own (Norming Study 1).1 These participants words, 204 of which lacked published concreteness values
andthusrequirednorming(AppendixA).Allbasetermswere
also judged the strength of the auditory and visual
ratedfortheirassociatedauditory,visual,andmotionimagery
imagery associated with all of the base terms, to ensure
(n = 156; Appendix B). For participants who received the
a valid manipulation of sensory modality. A different set
motion imagery list, all base terms from Cardillo et al.
of participants rated the base terms from this candidate
(2010)werealsoincluded(n=226;AppendixC),sincethese
stimulus set and from Cardillo et al. (2010) for strength
had not been collected previously and doing so allows for
of motion imagery. A further set of individuals normed
future studies that combine sentences from the norming
the stimuli at the sentence level, interpreting them as well
as rating them in terms of familiarity,2 naturalness, studies.
imageability, figurativeness, and ease of interpretation
(Norming Study 2). Additionally, a valence judgment task Task An Excel workbook was generated with separate
wasadministeredtoafinalgroupofindividualstogeneratean worksheets corresponding to the four rating tasks (concrete-
online measure of comprehension difficulty for each item ness,auditoryimagery,visualimagery,andmotionimagery)
(NormingStudy3).GiventhesensitivityoffMRItoreaction and one line per worksheet corresponding to each item. For
time (RT) differences between conditions, coupled with the concreteness,participantswereinstructedtoratethewordsin
fact that valence judgment is currently the most frequently termsoftheiraccessibilitytooneormoreofthesenses,using
ascalefrom1(veryabstract)to7(veryconcrete).Foraudi-
1Wecollectedconcretenessvaluesontheassumptionthatconcreteness toryimagery,participantswereinstructedtoratethewordsin
judgmentsforcommonwordswouldnotdiffermuch,despiteaspanof terms of the speed and Bease or difficulty with which they
severaldecadesbetweenMRC,SouthFlorida,andournormingstudy. arouse a particular sound,^ using a scale from 1 (no sound)
Wedidnot,however,includeredundantitemstoconfirmthisconsistency
to 5 (clear sound). For visual imagery, participants were
acrosstime,sowecannotbecertainoftheirstability.
2We chose to norm items in terms offamiliarity rather than salience instructed to rate the words in terms of the speed and Bease
because(1)familiarityisastrongdeterminantofsalience,makingita ordifficultywithwhichtheyarouseamentalpictureorvisual
reasonable proxy; (2) familiarity is a simpler construct than salience, image,^ using a scale from1 (noimage) to 5 (clear image).
making its effects easier to interpret; and (3) familiarity is frequently
For motion imagery, participants were instructed to rate the
investigatedinindividualcognitiveandneuralstudies,makingitmore
usefulforcomparisonsacrossstudies.Forindividualsspecificallyinter-
estedinsalience,wedirectthemtoRonceroanddeAlmeida(2014). 3Wethankananonymousreviewerforthissuggestion.
476 BehavRes(2017)49:471–483
wordsintermsofthespeedandBeaseordifficultywithwhich naturalness, imageability, figurativeness, and interpretation),
theyarousevisualmotion,^usingascalefrom1(nomotion) withonelineperworksheetcorrespondingtoeachitem.Inthis
to5(clearmotion).Inallcases,theinstructionswerecoupled way,participantssawbothliteralandmetaphoricalitems,with
withseveralexamplesandexplanations(seethesupplemental 20responsescollectedforeachitem.
materialsformoredetail).4 Forthefamiliaritytask,participantswereinstructedtorate
theirfrequencyofexperiencewiththesentenceanditsmean-
DataanalysisForallwords, ratingswereaveragedoverthe ing,usingascalefrom1(veryunfamiliar)to7(veryfamiliar).
20participantsforeachofthejudgments.The204newcon- For the naturalness task, participants were instructed to rate
creteness values supplemented the previously published eachsentenceforhowBnaturalandnormal^itseemed,usinga
values for the other 749 content words in the stimuli set. scale from 1 (very unnatural) to 7 (very natural). For the
Theseindividualconcretenessratingswerethenusedtodeter- imageability task, participants were instructed to rate Bhow
mineanaverageconcretenessvalueforeachofthe312can- quickly and easily each sentence brings a visual image to
didatesentences(i.e.,thesumoftheconcretenessvaluesas- mind,^usingascalefrom1(noimage)to7(clear,immediate
sociated with each content word in any particular sentence, image). For the figurativeness task, participants were
dividedbythenumberofcontentwordsinthatsentence).The instructed to rate how literal an interpretation each sentence
imageryratingsofthebasetermsindicatedthreeproblematic suggested, using a scale from 1 (very literal) to 7 (very
base terms: One base term used in the auditory conditions figurative). For the interpretation task, participants were
(cab) elicited stronger motion than auditory imagery, and instructed to write the meaning of each sentence using their
twousedinthemotionconditions(sprinklerandzipper)elic- ownwords(fullinstructionscanbefoundinthesupplemental
ited stronger auditory than motion imagery. Overall, visual materials).Thefamiliarity,naturalness,imageability,andfig-
imagery was consistently strong in the motion base terms. urativeness ratings were collected for both literal and meta-
By contrast, its strength varied widely in the auditory base phoricalsentences.Giventhedifficultyofrestatingaconcrete,
terms,sincesomereferredtopalpableobjects(e.g.,instrument literal sentence innovel words and the absence of any theo-
names)whereasothersreferredtointangiblemusicalconcepts retical relevance for such descriptions, interpretations were
(e.g.,rhythm,beat,melody). onlycollectedforthemetaphors.
Normingstudy2a:sentences
DataanalysisTogeneratefamiliarity,naturalness,imageability,
and figurativeness ratings for each item, averages were cal-
Participants Forty participants were recruited from the
culated. Several steps were necessary to determine the inter-
University of Pennsylvania community in compliance with
pretability of each item. First, for each metaphor, two of the
the procedures established by the university’s Institutional
authors (E.C., C.W.) and a third researcher independently
Review Board, and were compensated $25 or given course
judged the number of interpretations that reflected a plausible
credit for their participation. All participants were native figurative construal of the sentence.5 In contrast, blank, non-
English speakers, and none had participated in Norming sensical, literal, or uninformative (e.g., BJust what it says^)
Task1.Becauseofthelargenumberofitemstobeevaluated
interpretations were not taken to indicate metaphoric compre-
and concerns about fatigue, 20 participants made judgments
hension.Toencourageconsistentevaluations,thejudgeswere
onhalfofthesentences(meanage=20.5years,SD=2.9;11
first trained on 100 interpretations from Cardillo et al. (2010).
females,ninemales;meaneducation=15.0years,SD=1.8)
Thethreejudgesevaluated3,120interpretations,resulting
and 20 participantsmade judgments on the other half of the
in9,360plausibilityjudgments.Theaveragepairwisepercent
sentences(meanage=22.1years,SD=3.2;tenfemales,ten
agreement between judges was 83.9 %, with the two judges
males;meaneducation=15.5years,SD=2.0).
beingingreateragreementwitheachother(88.2%)thanwith
the third researcher (82.2 % and 81.1 %, respectively). An
StimuliAll312candidatesentenceswereassessed.
interrater reliability analysis using the Kappa statistic was
used to further determine consistency among the judges. A
TaskTheitemswererandomlydividedinhalf,andforeachof FleissKappaformultipleratersof.53(SE=.01,95%CI=
thesesubsetsanExcelworkbookwasgeneratedwithseparate .51–.55,p < .00001) indicated moderateagreement between
worksheetscorrespondingtothefivenormingtasks(familiarity,
5Forsomesentences,allinterpretationsreflectedasinglemeaning;for
4TheinstructionswereslightlymodifieddirectionsfromPaivioandcol- manyothers,theresponsesindicatedmultipleoroverlappingmeanings.
leagues(1968),andalsowereverysimilartothoseusedinthetwoother Giventheplausibilityofmorethanoneinterpretationintheabsenceof
major sources of concreteness and imageability norms in the MRC contextandthesubjectivityofdeterminingwhereonemeaningendsand
Psycholinguistic Database (i.e., Gilhooly & Logie, 1980; Toglia & another begins, rather than tally the incidences of the most common
Battig, 1978). The exact instructionsfor all tasks can be found in the interpretation,anyplausiblefigurativeinterpretationwastakentoindicate
supplementalmaterials. thatthemetaphorhadbeenunderstood.
BehavRes(2017)49:471–483 477
thejudges.Thestronglevelofagreementbetweentwoofthe ofPennsylvaniaundergraduatecommunityincompliancewith
threejudges(Cohen’sKappa=.62)revealsthesensitivityof the procedures established by the university’s Institutional
theplausibilityjudgmenttocharacteristicsofthejudges(e.g., Review Board and were compensated $15 or given course
differencesinresearchbackgrounds).Thestrongerthree-way credit for their participation. All of the participants were
agreementinourpreviousstudy (Cardilloetal.,2010),with native English speakers, and none had participated in
nodifferenceintasktrainingbutmoresimilarresearchexpe- Norming Task 1 or 2.
rience,suggeststhattherichnessofinformationprovidedby
theinterpretationtaskmaybebestleveragedwhenthejudges StimuliAll312candidatesentenceswereassessed.
aregivenadditionaltraining.
Asinourpreviousstudy,aninterpretabilityscoreforeach Task Sentences were presented centrally in black 18-point
participantwascalculatedbydividingthenumberoftheirin- font on a white background, using E-Prime 1.1 software on
terpretationsthatweredeemedplausiblebyatleasttwoofthe aDellInspironlaptop.Sentencesweredisplayedfor3,000ms
judgesbythetotalnumberofitemsassessedbythatparticipant and separated by a 1,000-ms intertrial interval. Participants
(# plausible interpretations/all possible interpretations). This were instructed to read each sentence and then to judge its
assessmentrevealedpooroverallcomprehensionbyfourpar- emotional valence, using the ‘f’ key to indicate a positive
ticipants in one list and by two participants in the other list valence and the ‘j’ key to indicate a valence that was not
(>30 % of their interpretations were not considered plausible positive,definedaseitherneutralornegative.Theywerein-
and/orwereleftblank),sotheirdatawereexcludedfromsub- formed that there was no right or wrong answer and were
sequentanalyses.Togenerateaninterpretabilityscoreforeach encouragedtorespondasquicklyaspossible.Twelvepractice
item,thenumberofinterpretationsdeemedplausiblebyatleast trialsprecededfourblocksofexperimentaltrials.Eachpartic-
twoofthejudgeswasdividedbythetotalnumberofinterpre- ipantreceivedadifferentrandomorderofitemsandsaweach
tations for that item (# plausible interpretations/all possible sentenceonlyonce.
interpretations). These results indicated that 28 metaphors
failedtoreachtheminimumdesiredcomprehensibilitycriteria
DataanalysisForeverysentence,RTswereaveragedacross
(70 % plausible interpretations) established in our previous
participantsandtheproportionofpositivevalencejudgments
study(Cardilloetal.,2010).
wascalculated.
Normingstudy2b:sentences
Results
Participants Twenty participants (mean age = 22.5 years,
SD = 3.7; 16 females, four males; mean education =
To determine the reliability of the norms, we calculated
16.0 years, SD = 2.5) were recruited from the University
intraclasscorrelationcoefficientsforalldimensionsrequiring
of Pennsylvania community in compliance with the proce-
a subjective rating. All measures indicated high agreement
dures established by the university’s Institutional Review
acrossraters(Table2),whetherallratersratedallitems(i.e.,
Board and were compensated $10 for their participation.
two-wayrandomaveragemeasuresforeaseofinterpretation,
All participants were native English speakers, and none
concreteness,andbasetermimagery)ordifferentratersrated
had participated in the other norming tasks.
different items (i.e., one-way random average measures for
familiarity,naturalness,imageability,andfigurativeness).
StimuliAll312candidatesentenceswereassessed.
Task Items were presented in an Excel workbook, with one Table2 Intraclasscorrelationcoefficientsfortheratingtasks
linecorresponding to eachitem. Participantswereinstructed
Dimension IntraclassCorrelationCoefficient
to rate each item for its ease of interpretation, using a scale
from1(easytointerpret)to7(difficulttointerpret).Thefull Basetermauditoryimagery .969
instructionsarereportedinthesupplementalmaterials. Basetermvisualimagery .975
Basetermmotionimagery .962
DataanalysisTogenerateanease-of-interpretationscorefor Concreteness .969
eachitem,averageswerecalculated. Familiarity .857
Naturalness .886
Normingstudy3:onlinecomprehension
Imageability .872
Figurativeness .975
Participants Twenty participants (ages 18–22 years, 13
Easeofinterpretation .957
females, seven males) were recruited from the University
478 BehavRes(2017)49:471–483
Theresultsofthenormingstudieswereusedtoeliminate the visual nature of motion perception, the visual imagery
problematicstimulifromtheinitialpool.Theimageryratings provokedbythemotionbasetermswasalsogreaterthanthe
from Norming Study 1 and the interpretability scores from auditory imagery provoked by those terms, and was weaker
NormingStudy2indicatedthat31metaphor–literalsentence than the auditory imagery provoked by the auditory base
pairstobediscarded,resultingin60nominal-auditoryand66 terms.
nominal-motion items remaining possible sentence pairs. To Atthesentencelevel,theliteralandmetaphoricalsentences
generate a final stimulus set with equal numbers of items were similar in length and frequency. However, the literal
in each condition, the six items in the nominal-motion sentencesofbothmodalitieswerejudgednotonlytobeless
condition with the lowest interpretability values were also figurative (as they should be, by definition), but also more
discarded.Thelexicalandsententialcharacteristicsofthefinal concrete, familiar, natural, easy to interpret, and imageable
set of 240 sentences are summarized in Table 3 (see the thantheirmatchedmetaphoricalsentences.Thissamepattern
Electronicsupplementarymaterialforthefullsetofitemsand (with the exception of the concreteness difference) was ob-
theirnormingvalues). served in our previously normed stimuli (Cardillo et al.,
Wedidnotcalculatestatisticaldifferencesbetweenthesen- 2010). The greater familiarity and naturalness of the literals
tence types, because our aim is for the stimulus set to be highlightsthedifficultyofconstructingnovelliteralsentences
sampledinwaysthatcontrolforconditiondifferencesorfor withoutinadvertentlyevokingmetaphoricinterpretations.For
the norming data to be used to covary out the influences of example,considerthenovelliteralsentenceBHe’ssittingdeep
nuisance variables, rather than for the set to be used in its inthebubbles^inVanLanckerandKempler’s(1987)proverb
entirety. Nonetheless, the overall means confirmed desired study; it is not obvious that a participant wouldn’t interpret
differencesbetweentheconditions,aswellasindicatingsome thissentencefigurativelyifitwaspresentedinthecontextof
areasinwhichcontrolislikelytobenecessary.Asintended, many other metaphors. Rather than generate novel literal
the base terms consisting of nouns with motion properties sentencesthatmightunintentionallybeconstruedmetaphori-
were rated as having more salient motion imagery than cally, we erred on the side of creating more familiar literal
auditory imagery, and the base terms consisting of nouns items,knowingthattheirfamiliaritycouldbecovariedoutif
with auditory properties were rated as having more salient necessary.Indoingso,however,ourliteralitemsturnedoutto
auditoryimagerythanmotionimagery.Unsurprisingly,given beeasiertounderstand.Wepositthatthisisthelesseroftwo
Table3 Summaryoffinalstimuluscharacteristicsbysentencetypes
Literal Motion Metaphorical Motion
Auditory Auditory
M SD Range M SD Range M SD Range M SD Range
Baseauditoryimagery 4.1 0.6 2.2–5.0 2.0 0.6 1.1–3.4 4.1 0.6 2.2–5.0 2.0 0.6 1.1–3.4
Basevisualimagery 3.3 1.4 1.3–5.0 4.3 0.7 1.9–5.0 3.3 1.4 1.3–5.0 4.3 0.7 1.9–5.0
Basemotionimagery 2.2 0.6 1.3–3.9 3.3 1.0 1.4–4.7 2.2 0.6 1.3–3.9 3.3 1.0 1.4–4.7
Concreteness 465 66 310–596 492 78 201–609 444 89 45–588 461 59 317–583
Frequency1 90 120 4–589 73 101 1–577 66 80 2–438 86 118 0–549
Frequency2 85 121 1–653 54 70 2–334 67 106 0–484 67 92 2–405
#Characters 31.9 4.5 22–41 32.6 4.5 22–42 33.4 5.8 21–42 33.4 4.7 25–43
#Words 5.9 0.5 4–7 5.9 0.5 4–7 4.8 0.6 4–7 6.0 0.5 4–7
#Contentwords 3.0 0.4 2–4 3.0 0.5 2–4 3.0 0.5 2–4 3.1 0.5 2–4
Interpretability – – – – – – 0.91 0.07 .72–1.0 0.93 0.1 .75–1.0
Easeofinterpretation 1.3 0.5 1.0–7.0 1.2 0.3 1.0–7.0 3.5 0.8 1.0–7.0 3.3 0.8 1.0–7.0
Familiarity 5.6 0.6 3.6–6.7 5.6 0.6 4.2–6.8 4.1 1.0 2.4–6.4 4.3 0.9 2.8–6.3
Naturalness 5.8 0.7 3.6–7.0 5.9 0.5 4.8–6.9 4.2 0.9 2.4–5.9 4.4 0.9 2.7–6.4
Imageability 5.2 1.1 2.7–6.9 6.1 0.5 4.7–6.9 4.0 0.8 2.4–6.5 4.0 0.7 2.9–5.7
Figurativeness 1.9 0.7 1.0–3.9 1.8 0.4 1.0–3.0 6.0 0.5 4.9–6.7 6.1 0.5 4.3–6.7
ValenceRT(ms) 1,178 162 918–1,801 1,167 160 835–1,634 1,268 172 975–1,763 1,236 167 842–1,680
Valencepositiveratio 0.25 0.3 0.0–1.0 0.17 0.2 0.0–.70 0.35 0.3 0.0–.95 .26 0.3 0.0–90
Frequency1=valuesfromKučeraandFrancis(1967);Frequency2=SUBTL valuesfromBrysbaertandNew(2009)
WF
BehavRes(2017)49:471–483 479
evils, when faced with the possibility of literals being unin- stimulusset,itispossibletodisentanglecomprehensiondiffi-
tentionallyinterpretedmetaphorically.Wedonotbelievethis cultylevelfromfigurativeness.Despitetherelativenoveltyof
to be an inherent difference between metaphoric and literal the metaphors in this stimulus set, the correlations between
expressions,so much as a confound that will be important interpretabilityandfigurativeness,andbetweeneaseofinter-
to include as a covariate in any study with unmatched pretation and figurativeness, were both not significant. Also
items. notable,valenceRTdidnotsignificantlycorrelatewithanyof
By contrast, the consistent observation of reduced the sentence ratings or the interpretability scores. Balancing
imageabilityformetaphorsrelativetoliterals,acrossstimulus conditionsintermsoftime-on-taskiscriticalinfMRIstudies;
sets and metaphor types, suggests that this pattern may be thispatternsuggeststhatselectingstimulidifferingincharac-
inherent to metaphorical language rather than a confound. teristicsofinterestbutnotinprocessingtimeshouldnotbea
Abstract concepts, by their nature, are not imageable (e.g., challenge.'
justice,peace),andithasbeenarguedthatwerelyonfigura- By and large, the auditory and motion metaphor sets
tive expressions in order to talk about them (Lakoff & showedverysimilarrelationshipsbetweenthesentence-level
Johnson, 1980). However, not all low-imageability concepts factorswhencorrelationswerecalculatedseparatelyforeach
are abstract (e.g., auditory concepts like symphony, melody, (Table5).
song, etc.). For these items, a metaphorical sense may not
necessarily be any less imageable than its literal sense.
Alternatively,thispatternmaybeanaccidentalartifactofthe Discussion
particularwordschosenorofourcreatingauditoryandmotion
metaphors that did not imply any literal sound or motion, Thepurposeofthisstudywastoaugmentexistingpublished
respectively—a methodological necessity for testing sensori- metaphor stimuli with items that can address additional hy-
motor hypotheses that may have inadvertently biased us to pothesesorboostsamplesizeandstatisticalpowerwhencom-
generatemetaphorsencodingmoreabstract,low-imageability binedwithpreviouslypublisheditems.Tothisend,weoffer
meanings. Until future research clarifies the relationship of matched literal and metaphoric sentences characterized at
imagerytofigurativeness,ourdatasuggestthatmatchingmet- both the word and sentence levels on 14 variables of
aphors and literals on imageability may require anespecially methodological and theoretical relevance. Whether they
largesetofitemsfromwhichtosample. are used in combination with other stimuli or on their
Asinourpreviousstudy,whenconsideringthemetaphors own,ourhopeisthattheseitemswillfacilitatethemethodo-
separatelybymodality,interpretabilitywasveryhighforboth logicalandconceptualprecisionnecessarytoaddressexisting
auditoryandmotionmetaphors.Theonlydifferencebetween ambiguities and emerging questions in cognitive and neural
modalities regarded imageability in the literal condition: studiesofmetaphor.
Auditoryliteralitemswereratedasbeinglessimageablethan At240 items,the stimulussetjoinsthree others(Cardillo
motionliteralitems,anunsurprisingfindingconsideringthat etal.,2010[n=560];Katz,Paivio,Marschark,&Clark,1988
thebasetermsofauditoryitemswereratedashavingweaker [n=464];Roncero&deAlmeda,2014[n=168])asoneof
visualandmotionassociations. thelargestpoolsofextensivelynormedstimuliforstudiesof
Tofurtherexploretherelationshipsbetweenthesentence- metaphor. The earliest stimulus set, by Katz et al. (1988),
levelfactorsoftheoreticalinterest,thesevenvaluescollected provides204literaryand260nonliterarynominalmetaphors,
foreachsentence(familiarity,naturalness,imageability,figu- each rated on ten dimensions (comprehensibility, ease of
rativeness,interpretability,easeofinterpretation,andvalence interpretation, metaphoricity, metaphor goodness, metaphor
RT) were correlated with each other, both collapsed across imagery, subject imagery, predicate imagery, familiarity,
modalities(Table4)andseparately(Table5).Theresultsin- semanticrelatedness,andnumberofalternativeinterpretations).
dicated several expected relationships based on our prior More recently, Roncero and de Almeida (2015) provided 84
study.Aswepreviouslyobserved,familiarityandnaturalness nominal metaphors and matched similes, normed in terms of
werehighlycorrelated,againindicatingthattheseconstructs property associations, aptness, familiarity, conventionality,
areeitherconceptuallyindistinguishableoratleastsodifficult saliency, and connotativeness. Informative in their own
todisentanglethatwesuggestresearchersnotconcernthem- right, the Katz stimuli have also been a valuable resource
selveswithnaturalness,infavorofthemoretheoreticallyrel- in metaphor research since their publication, as well as the
evantconstructoffamiliarity.Sentencesratedhigherinfamil- only normed set of literary metaphors we are aware of.
iarityandnaturalnessalsotendedtoevokegreatervisualim- However, these items have not been normed on some of
agery,wereperceivedaslessfigurative,andweremoreeasily the characteristics germane to current debates concerning
understood.Yetthesepatternsarenotsufficientlystrongthat metaphor. By contrast, Roncero and de Almeida’s (2015)
they could not beorthogonalizedwithcareful itemselection carefully crafted metaphor–simile pairs are optimized for
(see the Discussion section). Critically, as with our previous addressing current, competing cognitive models of metaphor
480 BehavRes(2017)49:471–483
Table 4 Correlation coefficients between sentence scales, collapsed items should enable avoiding condition differences on pa-
acrossmodalities
rameters of noninterest and maximizing differences of
FAM NAT IMG FIG EASE INT RT theoretical relevance. We take the success of researchers
selecting items from the similarly normed Cardillo et al.
Familiarity(FAM) .90** .54** –.19* .79** .42** –.15 (2010) stimuli as a demonstration of the utility of our
Naturalness(NAT) .57** –.11 .83** .51** –.13 approach for a variety of study designs (e.g., behavioral:
Imageability(IMG) –.06 .43** .30** –.12 Bowes & Katz, 2015; Jalal & Ramachandran, 2014; fMRI:
Figurativeness(FIG) .17 .13 –.07 Cardillo et al., 2012; eyetracking: Columbus et al., 2015;
Easeofinterpretation .37** .06 event-related potentials [ERP]: Schmidt-Snoek, Drew,
(EASE)
Barlie,&Agauas,2015).Wesuggestthatcomputationalap-
Interpretability(INT) –.12
proaches to stimulus selection, such as SOS (BStimulus
ValenceRT(RT)
Optimization Software^; Armstrong, Watson, & Plaut,
**p<.01,*p<.05' 2012), be used to maximize the efficiency of this selection
process for complex designs. For instance, we used SOS to
select three sets each of nominal and predicate metaphors
comprehension (e.g., comparison vs. categorization mecha- fromCardilloetal.(2010)thatdidnotdifferintermsofsuch
nisms), and for testing the predictions of the GSH. criticalvariablesasfamiliarity,naturalness,imageability,fig-
However, without manipulations of the semantics, grammat- urativeness, and interpretability (Cardillo et al., 2012), a
ical category, or sensory associations of base terms, neither balancingactthatwouldbedifficultorimpossiblewithfewer
stimulus set can address the specific neural hypotheses itemsandtrial-and-errorselection.
outlined above regarding sensorimotor grounding or meta- The present items are also modifiable to suit a variety of
phor type. Nor do these sets include matched literal items, tasksandpopulations:Addingcontext,comprehensionques-
a critical comparison condition for both neuroimaging and tions,primes,semanticallyrelatedprobes,multiple-choicean-
patient studies. Our aim is to complement the strengths of swers,collectingadditionalratings,andsoforth,areallpos-
these other stimulus sets by filling the extant gap in re- sible elaborations. How researchers sample and/or augment
sources for testing neural hypotheses. thestimuliwilldependontheirquestionsofinterest,subject
The large number of items and normed properties in the populations,andmethodology.Forexample,werecentlyused
currentsetmaximizesitsflexibility.Whencombinedwithour SOStoselectfromboththepresentitemsandourpreviously
previousset,weofferanunprecedentedlevelofstandardiza- publishedsettogeneratematchedsetsofnominal-event,nom-
tion—800itemsthathavebeennormedinidenticalfashionon inal-entity, and predicate literal–metaphor pairs, combining
allthesame parameters.Wesuggestthatcarefulselectionof themwithnewlycraftedmultiple-choicequestionstogenerate
Table5 Correlationcoefficientsbetweensentencescales,separatedbymodality
FAM NAT IMG FIG EASE INT RT
AuditoryMetaphors
Familiarity(FAM) .90** .54** –.19* –.78** .31* –.19
Naturalness(NAT) .64** –.15 –.83** .37** –.13
Imageability(IMG) –.13 –.44** .29* –.03
Figurativeness(FIG) .17 .08 .09
Easeofinterpretation(EASE) –.28* .03
Interpretability(INT) –.20~
ValenceRT(RT)
MotionMetaphors
Familiarity(FAM) .89** .46** –.17 –.81** .54** –.07
Naturalness(NAT) .49** –.09 –.84** .64** –.10
Imageability(IMG) .02 –.43** .31* –.25
Figurativeness(FIG) .21 .15 –.21
Easeofinterpretation(EASE) –.46** .09
Interpretability(INT) –.01
ValenceRT(RT)
**p<.01,*p<.05,~p<.10