Norms of valence, arousal, and dominance for 13,915 English lemmas Amy Beth Warriner, Victor Kuperman & Marc Brysbaert Behavior Research Methods e-ISSN 1554-3528 Behav Res DOI 10.3758/s13428-012-0314-x 1 23 Your article is protected by copyright and all rights are held exclusively by Psychonomic Society, Inc.. This e-offprint is for personal use only and shall not be self-archived in electronic repositories. If you wish to self- archive your work, please use the accepted author’s version for posting to your own website or your institution’s repository. You may further deposit the accepted author’s version on a funder’s repository at a funder’s request, provided it is not made publicly available until 12 months after publication. 1 23 Author's personal copy BehavRes DOI10.3758/s13428-012-0314-x Norms of valence, arousal, and dominance for 13,915 English lemmas Amy Beth Warriner& Victor Kuperman& MarcBrysbaert #PsychonomicSociety,Inc.2013 Abstract Informationabouttheaffectivemeaningsofwords inwhichtheyareproducedandperceived,theirinternalstruc- isusedbyresearchersworkingonemotionsandmoods,word ture,andtheconsequencesthattheyhaveforhumanbehavior. recognition and memory, and text-based sentiment analysis. For instance, Verona, Sprague, and Sadeh (2012) used emo- Three components of emotions are traditionally distin- tionallyneutralandnegativewordsinanexperimentcompar- guished: valence (the pleasantness of a stimulus), arousal ingtheresponsesofoffenderswithoutapersonalitydisorder (theintensityofemotionprovokedbyastimulus),anddom- to those of offenders with an antisocial personality disorder inance(thedegreeofcontrolexertedbyastimulus).Thusfar, whoeitherdidordidnothaveadditionalpsychopathictraits. nearly all research has been based on the ANEW norms The second line of research deals with the impact that collected by Bradley and Lang (1999) for 1,034 words. We emotional features have on the processing and memory of extended that database to nearly 14,000 English lemmas, words. Kousta, Vinson, and Vigliocco (2009) found that providing researchers witha muchrichersourceofinforma- participantsrespondedfastertopositiveandnegativewords tion, including gender, age, and educational differences in than to neutral words in a lexical-decision experiment, a emotion norms. As an example of the new possibilities, we finding later replicated by Scott, O’Donnell, and Sereno included stimuli from nearly all of the category norms (e.g., (2012) in sentence reading. According to Kousta, typesofdiseases,occupations,andtaboowords)collectedby Vigliocco, Vinson, Andrews, and Del Campo (2011), emo- Van Overschelde, Rawson, and Dunlosky (Journal of tionisparticularlyimportantinthesemanticrepresentations Memory and Language 50:289-335, 2004), making it possi- ofabstract words.Inother research, Fraga, Piñeiro,Acuña- bletoincludeaffectinstudiesofsemanticmemory. Fariña, Redondo, and García-Orza (2012) reported that emotional words are more likely to be used as attachment Keywords Emotion .Semantics .Genderdifferences .Age sitesforrelativeclausesinsentencessuchas“Someoneshot differences .Crowdsourcing theservant ofthe actress who. . . .” A third approach uses emotional ratings of words to estimate the sentiments expressed by entire messages or Emotionalratingsofwordsareinhighdemandbecausethey texts. Leveau, Jhean-Larose, Denhière, and Nguyen are used in at least four lines of research. The first of these (2012), for instance, wrote a computer program to estimate linesconcernsresearchontheemotionsthemselves:theways thevalenceandarousalevokedbytextsonthebasisofword measures (see also Liu, 2012). Electronicsupplementarymaterial Theonlineversionofthisarticle Finally, emotional ratings of words are used to automati- (doi:10.3758/s13428-012-0314-x)containssupplementarymaterial, cally estimate the emotional values of new words by com- whichisavailabletoauthorizedusers. : paringthemtothoseofvalidatedwords.BestgenandVincze A.B.Warriner V.Kuperman(*) (2012)gaugedtheaffectivevaluesof17,350wordsbyusing DepartmentofLinguisticsandLanguages,McMasterUniversity, TogoSalmonHall626,1280MainStreetWest, the rated valuesof words thatwere semantically related. Hamilton,OntarioL8S4M2,Canada Sofar,nearlyallstudieshavebeenbasedonBradleyand e-mail:[email protected] Lang’s(1999)AffectiveNormsforEnglishWords(ANEW) or on translated versions (for exceptions, see Kloumann, M.Brysbaert GhentUniversity,Ghent,Belgium Danforth, Harris, Bliss, & Dodds, 2012; Mohammad & Author's personal copy BehavRes Turney, 2010). These norms include ratings for 1,034 list of 30,000 lemmas for which Kuperman, Stadthagen- words. Three types of ratings were carried out, in line with Gonzalez, and Brysbaert (2012) collected age-of-acquisition Osgood, Suci, and Tannenbaum’s (1957) theory of emo- ratings. This list contains the content lemmas (nouns, verbs, tions.Thefirst,andmostimportant,typeofratingsconcerns and adjectives) from the 50-million-token SUBTLEX-US the valence (or pleasantness) of the emotions invoked by a subtitlecorpus.Weonlyselectedthehighest-frequencywords word, going from unhappy to happy. The second addresses knownby70%ormoreoftheparticipantsinKupermanetal., the degree of arousal evoked by a word, and the third giventhataffectiveratingsarelessvalid/usefulforwordsthat dimension refers to the dominance/power of the word—the are not known to most participants. Our final set included extent to which the word denotes something that is weak/- 13,915words,ofwhich22.5%aremostoftenusedasadjec- submissive orstrong/dominant. tives (Brysbaert, New, & Keuleers, 2012), 63.5 % asnouns, The number of words covered by the ANEW norms 12.6 % as verbs, and 1.4 % as other or unspecified parts of appeared sufficient for use in small-scale factorial experi- speech.Themeanwordfrequencyofthesetwas1,056(SD= ments. In these experiments, a limited number of stimuli 8,464,range=1to314,232,median=87)inthe50-million- would be selected that varied on one dimension (e.g., va- token SUBTLEX-US corpus; 152 words, or 1 %, had no lence) and were matched on other variables (e.g., arousal, frequencydata.Foreachwordinourset,wecollectedratings wordfrequency,andwordlength).However,thenumberof onthreedimensionsusinga9-pointscale. words in this set is prohibitively small for the large-scale Thestimuliweredistributedover43listscontaining346 megastudies that are currently emerging in psycholinguis- to 350 words each. Each list consisted of 10 calibrator tics.Inthesestudies(e.g.,Balotaetal.,2007;Ferrandetal., words, 40 control words from ANEW, and a randomized 2010; Keuleers, Brysbaert, & New, 2010; Keuleers, Lacey, selection of non-ANEW words. The calibrator words were Rastle, & Brysbaert, 2012), regression analyses of drawnfrom ANEWandwerechosen separatelyfor eachof thousands of words are used to disentangle the influences the three dimensions, with the goal of giving participants a onwordrecognition.TheANEWnormsarealso limited as sense of the entire range of the stimuli that they would inputforcomputeralgorithmsthatgaugethesentimentofa encounter.1 Participants always saw these calibrator words message/text or theemotional values ofnonrated words. first.TheremainingANEWwordsweredividedintosetsof Given the ease with which word norms can be collected 40 and served as controls for the estimation of correlations nowadays, we decided to collect affective ratings for a between our data and the ANEW norms. This meant that a majority of the well-known English content words (a total selectionofthesewordsappearedinmorethanonelistand of13,915).Becauseitwouldbeexpectedthattheemotional that the lists used for each of the three dimensions were valueswouldgeneralizetoinflectedforms(e.g.,sings,sang, mostly,butnotcompletely,identical.Thecontrolwordsand sung,andsingingfortheverblemmasing),weonlyinclud- the non-ANEW words were randomly mixed together in edlemmas(thebaseformsofwords—i.e.,theonesusedas each list. Once lists were created, the words in each one entriesindictionaries).Oursampleofwords(seebelowfor were always presented in a fixed order following the cali- theselectioncriteria)substantiallycoversthewordstockof bratorwords. the English language and forms a solid foundation from which to automatically derive the values of the remaining Data collection words (Bestgen &Vincze, 2012). ParticipantswererecruitedviatheAmazonMechanicalTurk crowdsourcing website. Responders were restricted to those Method who self-identified as being current residents of the US and whocompletedanygivenlistonlyonce.Thiscompletionofa Stimuli singlelistbyagivenparticipantwillhenceforthbereferredto asanassignment.Eachassignmentinvolvedratingwordsona The words included inour stimulus set were compiled from three sources: Bradley and Lang’s (1999) ANEW database, Van Overschelde, Rawson, and Dunlosky’s (2004) category 1Thecalibratorwordsfortherespectivedimensions wereasfollows (inincreasingorderofratings):Valence:“jail”(1.91),“invader”(2.23), norms, and the SUBTLEX-US corpus (Brysbaert & New, “insecure” (2.30), “industry” (5.07), “icebox” (5.67), “hat” (5.69), 2009).Ourfinalsetincluded1,029ofthe1,034wordsfrom “grin” (7.66), “kitten” (7.58), “joke” (7.88), and “free” (8.25). ANEW(fivewerelostduetoprogrammaticerror)and1,060 Arousal: “statue” (2.82), “rock” (3.14), “sad” (3.49), “cat” (4.50), oftheparticipant-generatedresponsesto60ofthe70category “curious” (5.74), “robber” (6.20), “shotgun” (6.55), “assault” (6.80), “thrill”(7.19),and“sex”(7.60).Dominance:“lightning”(4.00),“mil- names included in the category norm study (we did not dew”(4.19),“waterfall”(5.34),“wealthy”(6.11),“lighthouse”(6.24), include a few categories, such as units of time and distance “honey”(6.39),“treat”(6.66),“mighty”(6.85),“admired”(6.94),and ortypesoffish).Theremainingwordswereselectedfromthe “liberty”(7.04). Author's personal copy BehavRes single dimension only, in contrast to the ANEW study, for Results and discussion which participants rated each word on all three dimensions. The instructions given were minor variations on the instruc- Data trimming tions in the ANEW project, and are given below, with the respectivechangestothewordingfortheseparatedimensions Altogether,1,085,998ratingswerecollectedacrossallthree indicatedinsquarebrackets. dimensions. Around 3 % of the data were removed due to Youareinvitedtotakepartinthestudythatisinvestigating missing responses, lack of variability in responses (i.e., emotion,andconcernshowpeoplerespondtodifferenttypes providing the same rating for all words in the list), or the of words. You will use a scale to rate how you felt while completion of fewer than 100 ratings per assignment. The reading each word. There will be approximately 350 words. valenceandarousalratingswerereversedpost-hoctomain- The scale ranges from 1 (happy [excited; controlled]) to 9 tain a more intuitive low-to-high scale (e.g., sad to happy (unhappy [calm; in control]). At one extreme of this scale, ratherthanhappytosad)acrossallthreedimensions.Means youarehappy,pleased,satisfied,contented,hopeful[stimulat- and standard deviations were calculated for each word. ed, excited, frenzied, jittery, wide-awake, or aroused; con- Ratings in assignments with negative correlations between trolled, influenced, cared-for, awed, submissive, or guided]. agivenparticipant’sratingandthemeanforthatwordwere When you feel completely happy [aroused; controlled] you reversed(9%).Thiswasdoneonthebasisofbothempirical shouldindicatethisbychoosingrating1.Theotherendofthe evidence that higher numbers intuitively go with positive scaleiswhenyoufeelcompletelyunhappy,annoyed,unsatis- anchors(Rammstedt&Krebs,2007)andanexaminationof fied,melancholic,despaired,orbored[relaxed,calm,sluggish, these participants’ responses, which revealed unintuitive dull, sleepy, or unaroused; in control, influential, important, answers (e.g., indicating that negative words such as “jail” dominant,autonomous,orcontrolling].Youcanindicatefeel- made them very happy). Any remaining assignments with ingcompletelyunhappy[calm;incontrol]byselecting9.The ratingsthatcorrelatedwiththemeanratingsperitemsatless numbers also allow you to describe intermediate feelings of than .10 were removed, and the means and standard devia- pleasure[calmness/arousal;in/undercontrol],byselectingany tions were recalculated. The final data set consisted of of the other feelings. If you feel completely neutral, neither 303,539observationsforvalence(95%oftheoriginaldata happynorsad[notexcitednoratallcalm;neitherincontrolnor pool),339,323observationsforarousal(89%oftheoriginal controlled],selectthemiddleofthescale(rating5). data pool), and 281,735 observations for dominance (74 % Pleaseworkatarapidpaceanddon’tspendtoomuchtime of the original data pool). A total of 1,827 responders thinkingabouteachword.Rather,makeyourratingsbasedon contributedtothisfinaldataset,with362ofthemcomplet- yourfirstandimmediatereactionasyoureadeachword. ingassignmentsfortwoormoredimensions.Atotalof144 On average, assignments were completed in approximately participants completed two or more assignments within a 14 min. Participants received 75 cents per completed assign- single dimension. ment.Afterreadinganinformationalconsentstatementandthe For valence, 51 words received fewer than 18 (but more instructions,participantswereaskedtoindicatetheirage,gen- than 15) valid ratings. For arousal, 128 words had a total der,firstlanguage(s),country/stateresidedinmostbetweenbirth number of ratings inthat range. For dominance, 564 words and age 7, and educational level. Subsequently, they were hadatotalofeither16or17ratings,and17wordshad14or reminded of the scale anchors and presented with a scrollable 15ratingseach.Forallthreedimensions,morethan87%of pageinwhichallwordsinthelistwereshowntotheleftofnine the words had between 18 and 30 ratings per word. A total numbered radio buttons. Although we did not incorporate the of 50 words in each dimension received more than 70 Self-AssessmentManikins(SAM)thatwereusedintheANEW ratings each, due to the doubling up of ANEW words and study, we did anchor our scales in the same direction, with thererunningoflists.Toillustratehowourdataenrichesthe valence ranging from happy to unhappy, arousal from excited setofwordsavailableinANEW,Table1providesexamples to calm, and dominance from controlled to in control. In the of words that are not included in the ANEW list and that Results and Discussion section, we show that our numerical show very high or very low ratings in one of the three ratings correlated highly with the SAM ratings from ANEW, dimensions. demonstrating that the methods are roughly equivalent. Once finished,participantsclicked“Submit”tocompletethestudy. Demographics Lists were initially presented to 20 respondents each. However,missing valuesdue tosubsequentexclusion criteria Of the 1,827 valid responders, approximately 60 % were resulted in some words having fewer than 18 valid ratings. femaleinallthreecases(419valence,448arousal,and505 Severalofthelistswerereposteduntilthevastmajorityofthe dominance). Their ages ranged from 16 to 87 years, with wordshadreachedatleastthisthreshold.Datacollectionbegan 11 % being 20 years old or younger; 45 % from 21 to 30; onMarch14,2012,andwascompletedMay30,2012. 21%from31to40;11%from41to49;and12%age50or Author's personal copy BehavRes Table1 WordsattheextremesofeachdimensionthatwerenotincludedinANEW Valence Arousal Dominance Lowest pedophile 1.26 grain 1.60 dementia 1.68 rapist 1.30 dull 1.67 Alzheimer’s 2.00 AIDS 1.33 calm 1.67 lobotomy 2.00 leukemia 1.47 librarian 1.75 earthquake 2.14 molester 1.48 soothing 1.91 uncontrollable 2.18 murder 1.48 scene 1.95 rapist 2.21 Highest excited 8.11 motherfucker 7.33 rejoice 7.68 sunshine 8.14 erection 7.37 successful 7.71 relaxing 8.19 terrorism 7.42 smile 7.72 lovable 8.26 lover 7.45 completion 7.73 fantastic 8.36 rampage 7.57 self 7.74 happiness 8.48 insanity 7.79 incredible 7.74 older. Of the participants, 24 (3.3 %), 32 (4.3 %), and 23 distributions were not significantly different from rating 5, (2.7%)forthevalence,arousal,anddominancedimensions, which is the median of the scales (both ps > .1). The respectively, reported a native language other than English, tendency for more words to make people feel happy and while 10 (1.4 %), 12 (1.6 %), and 12 (1.4 %) participants, in control goes along with numerous former findings of respectively, reported more than one native language, in- positivity biases in English and other languages (see cludingEnglish.Table 2showsthenumbersofparticipants Augustine, Mehl, & Larsen, 2011, and Kloumann et al., at each of the seven possible education levels. Most had 2012 ). The positivity bias—or the prevalence of positive some college ora bachelor’sdegree. word types in English books, Twitter messages, music lyr- ics, and other genres of texts—is argued to reflect the Descriptivestatistics preference of humankind for pro-social and benevolent communication. Arousal, on the other hand, is positively Table 3 reports descriptive statistics for the three distribu- skewed (G = .47), meaning that only a relatively small 1 tionsofratings.Thedistributionsofbothvalenceanddom- proportionofwords(20%abovearatingof5)madepeople inance ratings are negatively skewed (G = −.28 and –.23, feel excited. 1 respectively), with 55 % of the words rated above the Ratings of valence were relatively consistent across par- median of the rating scale for both dimensions (see ticipants, while arousal and dominance were much more Fig. 1). The Mann–Whitney one-sample median test indi- variable. This is indicated by the difference between the cated that the medians of both the valence and dominance average standard deviations of the dimensions: 1.68 for valence, but 2.30 and 2.16 for arousal and dominance, respectively. In addition, the split-half reliabilities were Table2 Reportededucationlevelswithineachdimension .914 for valence, .689 for arousal, and .770 for dominance; EducationLevel NumberofParticipants see below for other examples of a higher variability of dominance and arousal ratings. Figure 2a–c show, for the Valence Arousal Dominance three emotional dimensions, the means of the ratings for (%) (%) (%) each word plotted against their standard deviations, with Somehighschool 28(4) 32(4) 28(3) each scatterplot’s smoother lowess line demonstrating the Highschoolgraduate 96(13) 98(13) 117(14) Somecollege–No 237(33) 252(34) 298(35) degree Table3 Descriptivestatisticsforthedistributionofeachdimensions, Associatesdegree 82(11) 79(11) 93(11) including the number of participants (N), number of observations, averagemean,andaverageSD Bachelorsdegree 212(29) 222(30) 218(26) Mastersdegree 55(8) 53(7) 78(9) N #ofObs Mean AvgSD Doctorate 13(2) 9(1) 13(2) Total 723 745 845 Valence 723 303,539 5.06 1.68 Arousal 745 339,323 4.21 2.30 Thenumbersacrossallthreecolumnsadduptomorethan1,827,as Dominance 845 281,735 5.18 2.16 somepeoplecontributedtomorethanonedimension Author's personal copy BehavRes variabilityforsuchwordsmaybeduetogenderdifferences inthe norms. For valence, the scatterplot in Fig. 2a (top left) is sym- metrical about the median, with relatively positive or nega- tive words associated with smaller variability in the ratings across participants, as compared to valence-neutral words (see Moors et al., in press, for a similar finding in Dutch). The same holds for the pattern observed in the dominance ratings, Fig. 2c (bottom left). The plot of valence strength (absolute difference between the valence rating and the medianofvalenceratings;Fig.2d)corroboratesthetenden- cy of more extreme (positive or negative) words to be less variable in their ratings than neutral ones. In contrast, for Fig.1 Distributionsofvalence(green),arousal(red),anddominance (blue) ratings. Dotted lines represent the medians of the respective arousal in Fig. 2b (top right), words that make people feel distributions calm generally elicit more consistent ratings than do those that make people feel excited. To sum up, in terms of the variabilityofratings,valenceanddominancepatterntogeth- overall trend in the data (red solid lines). For illustrative erandarebestconsideredintermsoftheirmagnitude(how purposes, each plot is supplied with selected examples of strong is the feeling) rather than their polarity (sad vs. wordsthataresubstantiallymoreorlessvariablethanother happy, or controlled by vs. in control); polarity, however, words with the given mean rating. Swear words, taboo determines variability inthe arousalratings. words, and sexual terms account for a disproportionally large number of words that elicit more variable ratings of Correlations between dimensions valence and arousal than would be expected given the words’ mean ratings (shown as words in blue above the WefoundthetypicalU-shapedrelationshipbetweenarousal red lowess line in Fig. 2a–c), in line with Kloumann et al. and valence (see Fig. 3a;Bradley & Lang, 1999;Redondo, (2012). Below we will demonstrate that the greater Fraga, Padrón, & Comesaña, 2007; Soares, Comesaña, Fig.2 Standarddeviationsof ratingsforvalence(a,topleft), arousal(b,topright), dominance(c,bottomleft),and valencestrength(d,bottom right)plottedagainstthe respectivemeanratings.Panels a–calsoprovideexamplesof wordswithdisproportionately largeandsmallstandard deviations,giventheirmeans Author's personal copy BehavRes Fig.3 Scatterplotsof dimensions(a,arousalvs. valence;b,arousalvs. dominance;c,dominancevs. valence),alongwithlowess lines(inred)showingthe functionalrelationships,and regressionlinesforarousalas predictedbyhigh(ingreen)and low(inpurple)valenceand dominance.Samplewordshave alsobeenincluded Pinheiro, Simões, & Frade, 2012): Words that are very study (Bradley & Lang, 1999), stemming from original positive or very negative are more arousing than those that factor analyses done by Osgood, Suci, and Tannenbaum are neutral. This is corroborated by the positive correlation (1957). Future research will have to demonstrate that dom- between valence and arousal for positive words (mean va- inance explains unique variance over and above valence in lence rating > 6; r = .273, p < .001) and the negative language-processing behavior. The fact that extreme values correlationbetweenvalenceandarousalfornegative words ofvalence anddominancearemore arousingpoint again at (mean valence rating < 4; r = −.293, p < .001). The rela- the utility of considering valence/dominance strength (i.e., tionship between arousal and dominance is also U-shaped howdifferentawordisfromneutral)ratherthanpolarityas (see Fig. 3b), as corroborated by the positive correlation theexplanatory variable. We return tothis point below. between dominance and arousal for high-rated dominance words(meanrating>6;r=.139,p<.001)andthenegative Reliability correlation between dominance and arousal for low-rated dominance words (mean rating < 4; r = −.193, p < .001). Wecomparedourratingswithseveralsmallersetsofratings The relationship between valence and dominance is linear, that had been collected previously by other researchers, withwordsthatmakepeoplefeelhappieralsomakingthem feel more in control (see Fig. 3c). Table 4 shows that a quadratic relationship between arousal and valence and be- Table 4 Pearson’s correlations, linear and quadratic coefficients and tweenarousalanddominanceexplainsmoreofthevariance thequadraticR2foreachdimension than does a linear relationship. However, this does not rule r Linear Quadratic R2 out the possibility that the high and low levels of these Coefficient Coefficient associations might be explained better by a regression with Arousaland –.185 −0.130 34.883 .143 a break point at the median of the scale (see Fig. 3). The valence relationship between dominance and valence, however, is Arousaland –.180 −0.172 21.842 .075 fitted better by alinearmodel. dominance The strength of the correlation between dominance and Dominanceand .717 0.974 – .518 valence valence casts doubt on the claim that the three dimensions underconsiderationherearegenuinelyorthogonalaffective For both arousal/valence and arousal/dominance, the quadratic rela- states.ThisassumptionwasthebasisoftheoriginalANEW tionshipexplainsmorevariancethanthelinearfunction Author's personal copy BehavRes Table5 Correlationsofpresentratingswithsimilarstudiesacrosslanguages DataSet Correlations Source Language N(source) N(overlap) Valence Arousal Dominance a English 1,040 1,029 .953 .759 .795 b Dutch 4,299 3,701 .847 .575 N/A c Spanish 1,034 1,023 .924 .692 .833 d Portuguese 1,040 1,023 .924 .635 .774 e Finnish 213 203 .956 N/A N/A f English 10,222 4,504 .919 N/A N/A Sources:a,Bradley&Lang(1999);b,Moorsetal.(inpress)—Englishglosses;c,Redondo,Fraga,Padrón,&Comesaña,(2007)—Englishglosses; d,Soares,Comesaña,Pinheiro,Simões,&Frade(2012)—Englishglosses;e,Eilola&Havelka(2010)—Englishglosses;f,Kloumann,Danforth, Harris,Bliss,&Dodds(2012).AllstudiesexceptMoorsetal.(inpress)utilizedanine-pointscaleinacquiringtheirratings.Moorsetal.(inpress) usedaseven-pointscale Table6 Correlationsbetweenemotionaldimensionsandsemanticvariablesreportedinpriorstudies[degreesoffreedomarebasedonthenumbers ofdatapointsreportedasN(Overlap)] Source Measure N(Source) N(Overlap) Valence Arousal Dominance a Imageability 5,988 5,125 .161 –.012 .031 b Imageability 326 318 –.037 .099 –.160 Concreteness 326 318 .109 –.244 –.019 ContextAvail. 326 318 .196 –.147 .044 c Concreteness 1,944 1,567 .105 –.258 .009 d Imageability 3,394 2,906 .152 –.045 .006 Familiarity 3,394 2,906 .206 –.028 .215 e AoA1 30,121 13,709 –.233 –.062 –.187 %Known2 30,121 13,709 .094 .078 .103 f SensoryExp. 5,857 5,007 .067 .228 –.044 g Body–Object 1,618 1,398 .203 –.143 .172 h Familiarity 559 503 .272 –.193 .329 Pain 559 503 –.456 .579 –.343 Smell 559 503 .139 .052 –.043 Color 559 503 .401 .052 .081 Taste 559 503 .309 –.102 .084 Sound 559 503 –.176 .407 –.286 Grasp 559 503 .024 –.121 .252 Motion 559 503 –.113 .328 –.328 i Sound 1,402 1,283 –.04 .311 –.121 Color 1,402 1,283 .322 –.072 .100 Manipulation 1,402 1,283 .070 .026 .255 Motion 1,402 1,283 .011 .335 –.140 Emotion 1,402 1,283 .902 –.206 .658 j LogFrequency3 74,286 13,763 .182 –.033 .167 1AoA,ageofacquisition.2Theoverlappingwordsinthisstudyrepresentabiasedsample,duetothefactthatwordsinthepresentstudywere restrictedtoonlyincludewordsthatwereknownby70%ormoreparticipantsinthestudiescitedhere.3Sincewechosewordstofillourquotathat werehigherinfrequency,theoverlaphereisalsobiasedtowardtheupperrange Sources:a,Cortese&Fugett(2004)andSchock,Cortese,&Khanna(2012);b,Altarriba,Bauer,&Benvenuto(1999);c,Gilhooly&Logie(1980); d,Stadthagen-Gonzalez&Davis(2006);e,Kuperman,Stadthagen-Gonzalez,&Brysbaert(2012);f,extendeddatasetofJuhasz&Yap(inpress) and Juhasz, Yap, Dicke, Taylor, & Gullick (2011); g, Tillotson, Siakaluk, & Pexman (2008); h, Amsel, Urbach, & Kutas (2012); i, Medler, Arnoldussen,Binder,&Seidenberg(2005);j,Brysbaert&New(2009) Author's personal copy BehavRes including the ANEW set from which we drew our control emotionalimpactofawordarelikelytobeaffectedbyother words. The correlations arelistedin Table 5. aspects of the word’s meaning. Table 6 reports correlations Valence appears to generalize very well across studies ofvalence,arousal,anddominancewitharangeofavailable and languages, as evidenced by high correlations. Both semantic variables. In the remainder of the article, words, arousal and dominance show more variability across lan- rather then the trial-level data, were chosen as units of the guages and studies, as reflected in the lower correlations. correlational analyses. Notethatthesestudiesthemselves(thosethathavereported Most of the correlations that the emotional ratings show theinformation—i.e., c,d,ande)alsofound alowercorre- with other semantic properties are weak to moderate lation between their arousal and dominance ratings and the (Cohen, 1992), with the exception of correlations with var- arousal and dominance ratings reported in other studies iables that directly tap into emotional states (h and i in (arousal range = .65 to .75; dominance range = .72 to .73). Table 6). Specifically, words that make people happy are Importantly, however, cross-linguistic correlations were easier to picture [r(5123) = .161, p < .001] and more stronger (the range of Pearson’s r for arousal was concrete [r(1565) = .105, p < .001], familiar [r(2904) = .575–.759) than those between gender, age, and education .206, p < .001], context rich [r(316) = .196, p < .001], and groups within our study (the range of Pearson’s r was easytointeract with[r(1396) =.203,p<.001],areofhigh .467–.516), see Table 8 below. This observation clearly frequency[r(13763)=.182,p<.001],andarelearnedatan indicates the validity of using emotional ratings to English early age [r(13707) = −.233, p < .001]. They are also glossesofwordsinalanguagethatdoesnothaveanexten- associatedwithlowpain[r(501)=−.456,p<.001],intense sivesetofratingsattheresearcher’sdisposal.Thisseemsto smell [r(501) = .139, p < .01], vivid color [r(1281) = .322, be more the case for valence and dominance than for p < .001], pleasant taste [r(501) = .309, p < .001], quiet arousal. sounds [r(501) = −.176, p < .001], and stillness [r(501) = −.113, p < .05]. Virtually all of these properties are also Correlations with lexical properties associatedwithwordsthatmakepeoplefeelincontrol;that is, they correlate inthe same way withdominanceratings. Asisknownforothersubjectiveratingsoflexicalproperties Wordsthatmakepeoplefeelexcitedaremoreambiguous[r (cf.Baayen,Feldman,&Schreuder,2006),judgmentsofthe (1565)=−.258,p<.001],unfamiliar[r(501)=−.193,p<.001], Fig.4 Relationshipsbetween thethreedimensionsandageof acquisition,wordfrequency, imageability,andsensory experienceratings,presentedas scatterplotsmootherlowess trendlines