COGNITION AND EMOTION, 1996, 10 (4), 409± 423
Dimensions of Speech Perception: Semantic
Associations in the Affective Lexicon
Lee H . W urm and D ouglas A. V akoch
State U niversity of N ew Y ork at Stony Brook , U SA
The affe ctive lexicon has been explained in terms of three underlying
dimensions: Evaluation, Activity, and Potency. We assessed the importance
of these dimensions during online speech perception. Participants made
speeded lexical decisions about emotion w ords that w ere heard in a tone
of voice that w as either congruent or incongruent w ith the word’ s meaning.
The denotative semantic categ ory from w hich words w ere chosen w as signi® cantly related to lexical decision times (P < 0.001) . Tone of voice did not
in¯ uence decision times, nor did it interact w ith semantic category. Regression analyse s showed that lexic al decision times w ere signi® cantly predicted
by dimension w eights on Pote ncy, and by the three-w ay interaction betw een
dimension weights on Evaluation, Activity, and Pote ncy (both Ps < 0.001) .
The implications of this study for models of knowledge representation and
perception are discussed.
IN T R O D U C TIO N
The connotative m eaning s of w ords can be described in terms of a small
numbe r of unde rlying dimensions or semantic primitive s (e.g . Johns onLaird & Oatley, 19 89; Osgood & Suci, 19 55). The early w ork of Osgood
and colleague s is one im portant w ay in w hich this has been done (Osgood
& Suci, 195 5; Osgood, Suci, & Tanne nbaum , 195 7). From this w ork came
the dimensions of Evaluation, Potenc y, and Activity. Evaluation can be
thought of as a dimension rang ing from ``good’ ’ to ``bad’ ’ , activity as
ranging from ``active’ ’ to ``pas sive’ ’ , and Potency as ranging from ``strong’ ’
to ``weak’ ’ . These dim ensions (going by various names) have been studie d
extensiv ely and discov ered to play a role in a varie ty of contexts (e.g .
Requests for reprints should be sent to Lee H. Wurm, Department of Psychology, State
Unive rsity of New York at Stony Brook, Stony Brook, New York 11794± 2500, USA.
A portion of this work was presented at the 127th meeting of the Ac oustic al Society of
America, 9 June, 1994, in Cambridge, MA. We would like to thank Arthur Samuel for
allow ing this project to be completed in his laboratory. We would like to thank Annmarie
Cano, Arthur Aron, and David Cross for their help with statistical analy ses.
q
1996 Psychology Press, an imprint of Erlbaum (UK) Taylor & Francis Ltd
410
W U R M AND V AKO C H
Apple & Hecht, 1982; Daly, Lancee, & Polivy, 198 3; Green & Cliff, 1975;
Heise, 1965 ; Morgan & Heise, 1988 ). Evaluation, Potency, and Activity
have even be en propose d as cross-cultural unive rsals (Osgood, May, &
Miron, 19 75; but see also Russell, 1983, w ho believes the case is strong
only for Evaluation). All three dimensions are frequently found in judge ment tasks in w hich subje cts rate the similarity of a range of emotion states
(B ush, 1973; Daly e t al., 1983; Morgan & Heise, 1988 ).
R esearchers have made efforts to determine w hether the outputs of
cognitive processes are best unde rstood in categoric al or dimensional
terms. Often, although the best explanation for performance data may at
® rst seem to be categorical, on further study they can be explaine d dimensionally. This is true for such dive rse phenomena as obje ct classi® cation
(Ahn & Medin, 1992) and categoric al perception of phone mes (Massaro,
1987). One advantag e of using dimensional rather than categoric al information is that it is computationally cheape r to arrange inform ation dimensionally, w here all that must be stored for any give n item is a w eig ht on
each of a small numbe r of dimensions. Anderson (1991 ) points out that
each category maintaine d creates extra computational cost for the information-proc essing system (that is, the organism ). This is particularly important w hen inform ation processing is seen in an evolutionary context.
According to evolutionary theory, charac teristics that yield even a small
advantage in survival are more like ly to be passe d on to future generations
(Darw in, 185 9/1968). To the extent that organisms can process inform ation
about the external environme nt necessary for their survival in an ef® cient
manne r, their limite d resources are more readily available for other
ac tivitie s central to e volutionary success such as self-preservation and
reproduc tion.
Although some models of conceptual org anis ation and perception have
addre ssed the fact that seemingly categoric al perceptual effects can be
described in dimensional terms (e.g. Goldstone , 199 4; Masarro, 1987 ),
none has provide d much detail about w hat those dimensions are. The three
Osgoodian dimensions, which have been partic ularly useful for examining
the affective lexicon, may be candidate s. Each of them is associated with
psycholog ical and physiolog ical response s of the organism to emotioneliciting contexts (Scherer, 1986), and the same dimensions seem to be
® rmly established as e arly as age 8 (Russell & Ridge w ay, 1983 ). In addition, Osgood (1969) has propos ed a m odel in which these dim ensions are
important evolutionarily . It is advantage ous in terms of survival, ac cording
to this view , to be able to categorise items quic kly along the three dimensions: Is something good or bad? Fast or slow ? Strong or weak? To the
extent that this analysis holds, we w ould expe ct Evaluation, Activity, and
Potency (or dimensions like them) to be hardw ired into the perceptual
system .
D IM E N S IO N S O F S P E E C H P E R C E P T IO N
411
The present study inve stig ates w hether the dimensions of Evaluation,
Potency, and Activ ity are related dire ctly to perception, or w hether they
emerge only at highe r levels in the cognitiv e system. Before describing the
current study, how ever, it is necessary to clarify some distinc tions and
de® ne som e bas ic terms.
The lexic on is a special part of semantic memory. Traditionally, the
lexicon is view ed as dealing only w ith word s, w hich serve as labe ls for the
more comple x conc epts pre sum ed to be the main contents of semantic
memory. Lexical access is the process by which an acous tic signal makes
contac t w ith the representation in mem ory corresponding to the spoke n
word. It is a bas ic perceptual process that occurs autom atically w henever
ac oustic inform ation suf® cient for activation of such a representation
reache s an attending listener.
In the context of our research, perception is the involuntary, low -level
process w hereby a phys ical stim ulus makes initial contac t w ith a mental
representation. From the point of view of perception researchers, the question
of when variable s have an in¯ uence is as important as w hether they do. This
leads us to the important distinc tion betw een proce ssing that is truly
perceptual (calle d ``online’ ’ processing ) and that w hich occurs post-pe rceptually. In general, an effect is conside red online if the bas ic act of
perceiving a stim ulus (not interpreting it or making a judge ment about
it) is itself in¯ uenced. Word frequency is an example of such an effect:
High frequency w ords can be accessed more quic kly than low frequency
words (e.g . Forster & Cham bers, 1973; Old® eld, 196 6; Segui, Mehler,
Frauenfelder, & Morton, 1992 ).
V ariable s that in¯ uence later decision processes (i.e. those that occur
``highe r up’ ’ in the system) are conside red post- perceptual. Such variable s
have an effect only after the basic perceptual act has been comple ted. The
perceptual vs. post-pe rceptual issue plac es strong constraints on models of
perception, and cons equently the m atte r is being vigorousl y debate d.
The main purpose of the current study is to exam ine the relationship
betw een perceptual processing and the dimensions of Evaluation, Potency,
and Activity . The dimensional w ork described abov e has not done this, but
ins tead has involve d pos t-perceptual judge ments (e.g. deciding to w hich
category a partic ular stimulus belong s).
In addition to the question of dimensional relations hips in the lexicon,
we w ere interested in the possibil ity that an auditory Stroop effect could
in¯ uence speech perception. Although most familiar in the vis ual domain,
many researchers have demonstrated auditory Stroop effects as well (e.g .
Cohen & Martin, 197 5; Green & Barber, 1981, 1983; Pale f & Nickerson,
1978; Shor, 1975; Walke r & Smith, 1984, 19 85, 19 86). How ever, none of
these studie s has de® ned mismatching stimulus charac teristics in terms of
meaning and tone of voic e.
412
W U R M AND V AKO C H
We used emotion words for this inve stigation for tw o reasons. First, in
this domain we could easily de® ne matching and mismatching c onjunc tions
of words’ denotative semantic categorie s and the tones of voic e in w hich
they w ere uttered, which allow ed us to look for Stroop interference.
Secondly, in proposing that cognitive psy chologis ts and social cog nitive
psycholog ists do more to take notic e of each others’ work, B arsalou (19 90)
sugge sts that cog nitive psychologists begin to use categ ories for w hich
inferences are truly important. Surely em otions are such categorie s.
M E T HO D
S u bjects
Sixty-e ig ht unde rgraduate s from the psyc hology subje ct pool at the State
Unive rsity of New York at Stony Brook partic ipate d. All were native
speakers of English and all had normal hearing. They received course
credit for their partic ipation.
M a te rials
Forty-eight w ords were selec ted from Morg an and Heise’ s (198 8) study of
pure emotion w ords (em otion w ords that are relativ ely free of cog nitive
and behavioural connotations (Clore, Ortony, & Foss, 1987 ; Ortony , Clore,
& Foss, 1987). Hyphe nate d words w ere not allow ed. Four w ords w ere
chosen from each of three maxim ally distinc t denotative semantic categories (disgus ted, petri® ed, and happy ). The rem aining 36 w ere randomly
chosen from the full list of emotion w ords so that our pool of items w ould
not be concentrate d near the end-points of the dimensions.
Twenty-four of these 36 words w ere change d into nonw ords. We did this
by changing the phone me at the point in the acoustic sig nal w here the w ord
in question dive rges from all other w ords in English (the unique ness point;
see Marsle n-Wilson, 1984). The phone me at the unique ness point w as
change d to a different phonem e from the same broad class (i.e. fricative s
replac ed fricative s, vow els replac ed vow els, and so on). For example , the
word ``depressed’ ’ w as change d to the nonw ord ``deprussed’ ’ . The words
and nonw ords used in this study are listed in the Appendix.
Words from the Happy category had a hig her mean frequency (43.25 )
than thos e from the other categorie s (3.25 for Disgus ted, 21.0 0 for Petri® ed, and 10 .75 for Other; frequencies are from Francis & KucË era, 1982 ).
This poses a proble m, because frequency affe cts lexical access speed, as
we noted earlie r. How ever, it could not be av oide d in the c urrent study
because we did not have a large pool of pure emotion w ords from which to
choose a frequenc y-balanc ed subset.
D IM E N S IO N S O F S P E E C H P E R C E P T IO N
413
The location of each of the 24 stim uli (words) in three-dimensional
connotativ e lexical space w as speci® ed by the dimension w eights for
each w ord on Evaluation, Potency, and Activ ity, w hich were reported by
Morgan and Heise (1988 ), av erage d across male and female subje cts. Mean
weights w ere 2 0.76 for Evaluation (SD = 2.4), 2 0.51 for Potency (SD =
2.0), and 0.5 2 for Activity (SD = 1.4).
Eac h stimulus item w as read at the end of the carrier phrase : ``When that
happe ned, I felt Ð Ð Ð Ð .’ ’ The carrier phrase provide d a natural and
¯ uent pros odic context while keeping the potential effects of semantic
contex t to a m inim um. The carrier phrase w as read four times for each
stimulus item, once in each of the follow ing tones of voic e: happy, petri® ed, dis guste d, and neutral. Tone s of voic e w ere produc ed according to
speci® cations of tw o perceptual param eters found to differentiate the four
tones of voic e across num erous past studie s: spee ch rate and av erag e pitch
(Murray & Arnott, 19 93). In total, there were 19 2 stimulus sentences, each
of w hich w as digitise d at a sampling rate of 10kHz (low -pass ® ltered at 4.8
kHz) and stored in a dis c ® le.
M a nipu latio n C h eck
To determine whether the tone-of-voic e manipulation actually worked, w e
conduc ted a rating study. Fifteen unde rgraduate s at the State Unive rsity of
New York at Stony Brook listened to the 192 stim ulus sentences, presented
in a random order. The word or nonw ord that ended each sentence w as
dig itally strippe d from the carrier phrase , so that the tone-of-voic e rating s
would not be in¯ ue nced by w ord meanings. Subje cts were asked to give
each se ntence a rating from 1 to 7 on each of three dimensions, w ith anc hor
points correspo ndin gto the follow ing extrem es: slow vs. fast rate of presentation;lo w vs. high pitch; and good vs. bad. These dim ensions correspond
to the two perceptualpara m etersby which the stimuli w ere m odelled (speech
rate and averag e pitch) as well as the Evaluati ondim ension.
The results of this rating study are show n in Table 1. Consis tent with
Murray and Arnott’ s (1993 ) review of previous studie s charac terising
these four tones of voice, compare d to the neutral tone of voice: (1)
dis gust was slow er and lower pitched; (2) petri® ed w as very much faster
and very much higher pitched; and (3) happy was faster and highe r
pitched. Moreove r, petri® ed w as rated as being more negative on the
Evaluation dimension than either happy or neutral (P < 0.0 05 for all
comparisons ). We can thus be reasonably sure that the tones of voic e
were perceived as intended.
414
W U R M AND V AKO C H
TABLE 1
M ea n R atin g s o f S tim u lu s S e n te n c e s from M a n ip u la tio n C h e c k ( N = 15 )
Characteristic Rated
Speed
Tone of Voic e
Disgusted
Petri® ed
Happy
Neutral
Pitch
Valence
M
(SEM)
M
(SEM)
M
(SEM)
2.48
5.83
3.32
2.92
(0.22)
(0.15)
(0.17)
(0.20)
2.55
5.28
3.59
3.16
(0.22)
(0.22)
(0.13)
(0.17)
3.53
4.62
3.29
3.27
(0 .31)
(0 .19)
(0 .20)
(0 .22)
Note: Sente nc es were rated on 7-point scales. Anchor words on the scales were:
``slow ’’ and ``fast’’ , ``low pitch’’ and ``hig h pitch’’ , and ``good’’ and ``bad’’ (corresponding
to 1 and 7, respectively).
Proced u re
In order to avoid the poss ible effects of repetition priming, the 192 stimulus
sentences w ere div ided into four different lists of 48 sentences. Each w ord
was presented in exac tly one tone of voic e per list. For exam ple, the w ord
``scared’ ’ w as heard in a happy tone of voic e in one list, in a petri® ed tone
of voic e in anothe r list, and so on. Partic ipants w ere randomly assigne d to
hear one of the four lists of stim uli. Within each list the order of presentation w as randomise d. Each group of partic ipants got a diffe rent random
order.
Within each list, 25 % of the stimuli w ere from each of the four tones of
voic e. Across the entire experiment, each word w as he ard in each tone of
voic e an equal num ber of times. For each participant, half of the items w ere
words and half were nonw ords .
Partic ipants w ere directed to make a speeded lexical decision about the
item in sentence-® nal position. Partic ipants used their dominant hands to
make response s on a button board, pushing one button for w ords and
anothe r for nonw ords . Reaction times (RTs) w ere measured from the onset
of the phonem e at the unique ness point of each w ord. Partic ipants w ere
tested in groups of one to four in a sound-atte nuating c hambe r. Digitised
speech ® les w ere play ed for the partic ipants ove r headphone s at a comfortable listening level.
R E S U L TS
Subje cts w ho had error rates greater than 0.25 or mean R Ts greater than
1000m sec were exclude d from our analyse s. Four subje cts’ data w ere
exclude d by these criteria. For the remaining 64 subje cts, data were dis-
D IM E N S IO N S O F S P E E C H P E R C E P T IO N
415
carded for trials on which the w ord/nonw ord dec ision w as made incorrectly
1
(6.8% of all trials).
The RT data w ere not normally distribute d, so we performed a square
root transformation. The dependent variable in all reported analyse s w as
the transforme d RTs on correct trials (w ord targ ets only).
S tro o p A na lysis
Our ® rst analysis w as a repeated-measure s analy sis of varianc e (ANOVA),
with the denotativ e semantic category of the w ords and the tone-of-voic e
category serving as the factors.
For the Disgusted, Petri® ed, Happy, and Other semantic categories,
mean RTs w ere 55 1, 535 , 378 , and 484 msec, respectively (corresponding
SEMs w ere 17.84 , 17.5 7, 17.10, and 14.07) . A repeated-measures ANOVA
was performed, and revealed a signi® cant effect of semantic category
[F (3,54) = 62.7 3, MSE = 6.25, P < 0.001] . Subje cts responde d more
quic kly to w ords belong ing to the Happy semantic category than those in
the other three categories , by an ave rage of 145m sec. As w e hav e noted,
this could have been due to the highe r mean w ord frequency of those
words. Other effects in the analys is w ere not reliable (F < 1.0 for tone
of voic e and for the interaction) . Hearing a w ord spoke n in a tone of voic e
that is incong ruent w ith its denotativ e meaning did not produc e evide nce
2
for any Stroop-ty pe interference in RTs.
R eg re ss ion A nalysis
Because the main analysis did not reveal any evidence of a tone-of-v oice or
of Stroop-type interference, w e exclude d the tone-of-voic e factor from the
second analysis. (The pattern of results for this analy sis w as the same
1
Error rates did not differ across the four tones of voic e [ F(3,189) = 2.04, M SE = 0.009,
P > 0.10] , but there was a signi® cant effe ct of semantic category [ F(3 ,189) = 34.08, MSE =
0.009, P < 0.001 ] . Disgusted words had a higher error rate (16.8% ) than the other three
categories (1.6% , 3.9%, and 4.2% for Happy, Petri® ed, and Other, respectively).
2
One possible reason we did not observe a Stroop effect has to do with stimulus onset
asynchrony (SOA), whic h is the differe nce in time of onset betwe en the tw o characteristic s
of the stimulus under observation. Previous authors have noted that the basic Stroop
phenomenon seems most robust with SOAs in the 100msec range (Glaser & Glaser, 1982;
M cLeod, 1991). In our stimulus sentence s, the onset of the tone-of-voice information alw ays
preceded the onset of the semantic information by at le ast 800msec (and often by as much as
2500msec in the slow er sentences.)
416
W U R M AND V AKO C H
whether tone of voic e w as include d or not. Exclusion of the factor w as
done only in the interest of simpli® cation.)
The second analysis was a supple mentary hierarchic al regression analys is, in w hich the connotative semantic content of the words w as coded
continuously in terms of dimension weights on Evaluation, Potency, and
Activity (described earlier). This analysis provide d a detaile d inv estigation
of the roles of each of the three dimensions, as w ell as their possible
interactions. In addition, this analysis allow ed us to control statistic ally
for w ord frequency effects (described in Materials).
R egressors w ere entered in bloc ks, in the follow ing orde r: (1) log w ord
frequency and subje cts; (2) dimension w eights on Evaluation, Activity, and
Potency; (3) the 2-w ay dimension w eight interaction terms (i.e. Evaluation
3
Potency, Evaluation 3 Activity, and Activ ity 3 Potency); and (4) the
three-w ay dimension w eight interaction term. The results of this preliminary regression analys is revealed three suppre ssor variable s: dimension
weight on Evaluation; the interaction between dimension w eig hts on Evaluation and Potency; and the interac tion betw een dimension w eights on
Evaluation and Activ ity. Suppre ssor variable s are those for w hich the
direction (i.e. sig n) of the regression coef® cient is different from the sig n
of that variable ’ s bivariate correlation with the de pendent measure. For
exam ple, the biv ariate correlation between dimension weight on Evaluation
and RT w as signi® cant and negative , w hereas the regression coef® cient for
Evaluation was signi® cant and positive . B ecause the effects of suppre ssor
variable s cannot be meaningfully interpre ted, and because they enhanc e the
effects of variable s entered into the equation after them (Tabac hnic k &
Fidell, 1989), these three variable s w ere removed from the analysis. A
3
second regression analysis w as then run w ithout them.
The ® rst step w as to statistic ally remove (i.e. partial out) the effe cts of
log w ord frequency and individual subje cts’ varianc e [ for w ord frequency,
b = 2 0.19, F (1,13 67) = 72.42, MSE = 20.03, P < 0.001] . After this w as
2
done , w e found a signi® cant increase in R when w e adde d w ords ’ dimension w eights on Potency and Activity [ for these tw o regressors, F(2,13 65)
= 41 .29, MSE = 18.9 2, P < 0.001] . Potency had a signi® cant and negative
relationship w ith R Ts ( b = 2 0.22 , P < 0.001) : Words w ith highe r w eights
on Potency had faster lexical decision times. Dimension w eight on Activity
did not account for a signi® cant proportion of the varianc e (b = 0.03, P >
0.10 ).
3
There is no wholly satisfac tory way to de al with suppre ssor variable s. However, bec ause
suppressors result in better predictive pow er for the regression model (Darlington, 1990), we
opted for a conservativ e approach and exclude d them. The only major diffe rence be tween the
analy ses with and without suppressors was that the Activity 3 Potenc y interaction was
signi® cant with suppre ssors present, but not when they were removed.
D IM E N S IO N S O F S P E E C H P E R C E P T IO N
417
The next regressor entered w as the term for the two-w ay dimension w eight
interaction (Activ ity 3 Potency ). This interaction was not signi® cant [ b =
2 0.04 , F(1,13 64) = 1.54, MSE = 18.91, P > 0.10] .
Finally, w e found that the three-w ay interaction betw een dimension
weights on Evaluation, Activity, and Potency was signi® cant [ b = 0.56 ,
F(1,13 63) = 18 1.51 , MSE = 16.7 0, P < 0.0 01] . This inte raction can best be
unde rstood by exam ining Fig. 1. The ® gure show s mean RTs as a function
of weights on the three dimensions. High and low value s on each dim ension show n in the ® gure were determined by median splits. The resulting
dichotom ies are used for illustration and w ere no t used in the regression
analys is, but they illustrate the general pattern of results.
DIS C U S S IO N
This study show s that even at the very earlie st stage s of perception, the
basic units that allow entry into semantic memory are themselves coded
ac cording to continuous dimensions of m eaning . Furthermore, the current
study demons trates that the connotative semantic content of words allow s
us to predict the time course of online speech proce ssing m ore accurately .
Even after partialling out word frequency effects, w e found that dimension
weight on Potency w as a signi® cant predictor of RT. The hig her a word’ s
weight on Potency, the m ore quic kly it w as accessed. Of course, any
interpretation of the main effect of Potency must be quali® ed by a consideration of the signi® cant three-w ay interaction betw een dimension
weights, show n in Fig. 1. This interaction w as very strong, acc ounting
for more than twice as much variance as log word frequency, which w as
giv en ® rst access to the variance .
One interpretation of the interaction of the three dimensions, based on
an evolutionary m odel of the lexicon (Osgood, 1969 ), is offered here.
Cons ider ® rst the High Potenc y words (the top pane l in Fig. 1). For High
Evaluation w ords (i.e. those with a pos itive vale nc e), Activity is not of any
partic ular relevanc e. Althoug h the obje cts referred to by such w ords are
pow erful, they do not represent a dange r because they are pleasant. Thus ,
we w ould not expe ct their Activity level to m atter much. If Evaluation is
negative , how ever, the dis tinction between high and low Activ ity is absolutely critical. Survival itself could depend on the ability to make very fast
decisions about things that pose an imm ediate threat (i.e. that are very bad,
very strong , and very fast). Rapid processing is not require d for things that
are bad, strong , and slow .
An evolutionary interpretation of the bottom pane l in Fig. 1, w hich
show s the Low Potency words, is somewhat less clear. B ecause all of
the w ords show n in this pane l denote w eak or ineffectual things, Activity
should not play a partic ularly relevant role. We m ight expect R Ts for Low
FIG . 1 .
Mean RT as a function of dimension we ight on Evaluation, Activity, and Potency, in
msec. The top panel shows High Pote ncy words and the bottom panel shows Low Potency words.
Error bars show +/2 1 SEM.
418
D IM E N S IO N S O F S P E E C H P E R C E P T IO N
419
Evaluation w ords in this pane l to be faster than those for High Evaluation
words, and they are to a slight extent. How ever, because of the way it is
organise d, this area of connotative lexical space must be inte rpreted with
caution. It turns out that very few w ords have high w eights on Evaluation
and low w eights on Potency; ineffectual or w eak w ords are generally not
pleasant, w hereas strong words generally are. Overall, w e can see that
mean RTs for Low Potenc y w ords are slow er than for High Potency w ords ,
which again ® ts w ith an evolutionary perspective.
The relative pauc ity of w eak, good words is in accord w ith Morgan and
Heise’ s (1988) conclus ion that Potency is an important dimension primarily for distinguis hing betw een negative em otions . Morgan and Heise (19 88)
explain the importanc e of Potency in terms of its adaptive ness for the
organism. Thus, an organism must decide in a hedonic ally neg ative situation w hether it is being threatened, and if so, whether to respond with
``® ght’ ’ or ``¯ ight’ ’ .
Mode ls that stress the role of affectiv e-motiv ational states in cog nitive
developme nt prov ide an account that is consis tent w ith the evolutionary
ac count give n earlie r (Izard & Malates ta, 198 7). From the time w e are
childre n, w e hear certain w ords more often uttered in an emotion-lade n
contex t (e.g. ``poison’ ’ ). This covariation is a valuable source of information for listeners, and our data show that it is coded in the lexicon. Studie s
of childre n’ s e motion conce pts indic ate that Evaluation and Activity are
important dimensions for pos t-perceptual classi® cation of emotions as
early as third grade in elementary school (R ussell & Ridge w ay, 1983 ).
The time-scale w ithin which these dimensions become important for c hildren in organising their know ledge of the w orld must be inve stig ated more
thoroughly before any ® nal conclusions can be draw n about the relative
contributions of ontog enetic and phylog enetic models in accounting for the
orig ins of the dimensional structure of the lexicon.
B ow er (1987 ) c onclude d that perception is only w eakly affected (if at
all) by em otion, and only if the perceiver has a long-standi ng emotional
condition (e.g . depre ssion or an anxie ty disorde r). The current study
sugge sts a more pervasive impac t of the connotative meaning of emotion
words, even for a population of normal subje cts (i.e. subje cts not chosen
on the bas is of having long-s tanding emotional states of one kind or
anothe r).
The current study also has implic ations for models of speech recognition. Existing models have not addre ssed the pos sible role of connotative
meaning, but our w ork show s that connotative meaning is rele vant at the
earlie st stage s of perception. In ge neral, ne tw ork m odels of perce ption (e.g .
McClelland & Elman, 198 6) could easily be m odi® ed to accommodate
our results; presumably the resting activation levels of words could re¯ ect
the w ords’ Evaluation, Potenc y, and Activ ity dimension w eights , or,
420
W U R M AND V AKO C H
alternative ly, the c onne ction w eights betw een units in the netw ork could
be altered to re¯ e ct the dimension w eig hts.
This study reveals som e of the com plexity of lexic al represe ntations .
We found that lexical access time for any give n emotion w ord is dependent, in part, on the charac teristics of a larg er class of w ords know n to
the listener. We also found a previous ly unknow n link: Dimensions
derive d post-pe rceptually are also signi® cantly related to online perception.
The framew ork used here allow s us to locate easily any partic ular
word relative to other words on the basis of three dim ensions of connotative meaning. There is currently no sim ilarly com prehensive and yet
parsimonious system for categorising the de notative content of w ords .
The dimensions of Evaluation, Potency, and Activity have a fairly
obvious relationshi p to emotion w ords ; that, in fact, is why we chose to
study the partic ular w ords w e did. The extent to w hic h our results w ill
generalise beyond emotion w ords is not immediately obvious, but give n
that all w ords can be located within the connotative m eaning space that w e
exam ined (e.g. Osgood et al., 197 5), it seems plausible that access to words
in the general lexicon c an be described in terms of the same dimensions. If
this is the case, then the construc t of a spe ci® c ally affective lexicon could
be calle d into question. Future studie s w ill be require d to determine
whether these dimensions have comparable relevanc e w hen w e c onside r
words more representative of the e ntire lexicon.
The fact that lexical access can be in¯ uenced by w ord meanings has
been demonstrate d previously using various sorts of priming tasks, w here
ac cess time is facilitate d after prior presentation of a w ord related in
meaning (e.g. Warren, 1977 ). The present ® nding is partic ularly important
because w e hav e not used a priming paradigm ; the connotativ e semantic
effects w e identify are coded directly into the mental representations of
words them selves and exist inde pendently of priming . Thus , the present
study provide s a nove l approac h to uncove ring the structure of semantic
mental representations. The same approac h w ould be applic able to examining other dimensions that could in¯ uence lexical access.
Manuscript received 27 March 1995
Revised manuscript received 4 Oc tober 1995
R E FE R E N C E S
Ahn, W-K., & Medin, D.L. (19 92). A tw o-stage model of category construc tion. Cognitiv e
Science, 16, 81± 121.
Anderson, J.R. (1991). The adaptive nature of human categoriz ation. Psychologic al
Revie w, 98, 409± 429.
D IM E N S IO N S O F S P E E C H P E R C E P T IO N
421
Apple , W., & Hecht, K. (1982). Speaking emotionally : The relation betw een verbal and
vocal communication of affec t. Journal of Personality and Social Psychology, 42, 864±
875.
Barsalou, L.W. (1990). Access and infere nce in categorization. Bulletin of the Psychonomic
Society, 28, 268± 271.
Bower, G.H. (1987). Commentary on mood and memory. Behaviour Research and
Therapy, 25, 443± 455.
Bush, L.E., II (1973). Indiv idual differe nces in multidimens ional scaling of adje ctives
denoting feelings. Journal of Personality and Social Psychology, 25, 50± 57.
Clore, G.L., Ortony, A., & Foss, M.A. (19 87). The psychological foundations of the affe ctive le xicon. Journal of Personality and Social Psychology, 53, 751± 766.
Cohe n, G., & Martin, M. (1975 ). Hemisphere differe nces in an auditory Stroop test. Perception and Psychophysics, 17, 79± 83.
Daly, E.M ., Lancee, W.J., & Polivy, J. (1983 ). A conical mode l for the taxonomy of
emotional expe rience. Journal of Personality and Social Psychology, 45, 443± 457.
Darlington, R.B . (1990). Regression and lin ear models . New York: M cGraw-Hill.
Darwin, C. (1968). The origin of species, (J.W. Burrow, Ed.). New York: Penguin. (Original work publis hed 1859)
Forster, K.I., & Chambers, S.M. (1973). Lexical access and naming time. Journal of Verbal
Learning and Verbal Behavior, 12, 627± 635.
Francis, W.N., & KucË era, H. (1982). Frequency analysis of English usage: Lexicon and
grammar. Boston, M A: Houghton Mif¯ in.
Glaser, M .O., & Glaser, W.R. (1982). Time course analy sis of the Stroop phenomenon.
Journal of Experimental Psycholo gy: Human Perception and Performance, 8, 875± 894.
Golds tone, R. (1994). In¯ uence of categoriz ation on perceptual disc rimination. Journal of
Experimental Psychology: General, 123, 178± 200.
Green, E.J., & Barber, P.J. (1981). An auditory Stroop effect with judgments of speaker
gender. Perception and Psychophysics, 30, 459± 466.
Green, E.J., & Barber, P.J. (1983). Interference effec ts in an auditory Stroop task:
Congruenc e and correspondence. Acta Psychologic a, 53, 183± 194.
Green, R.S., & Cliff, N. (1975 ). M ultidimensiona l comparisons of structures of vocally and
facially expressed emotion. Perception and Psychophysics, 17, 429± 438.
Heise, D.R. (1965). Semantic differe ntial pro® le s for 1,000 most frequent Englis h words.
Psychological Monographs, 79, 1± 31.
Izard, C.E., & Malatesta, C.Z. (1987). Perspectives on emotional development I: Diffe rential emotions theory of early emotional development. In J.D. Osofsky (Ed.), Handbook of
in fant develo pme nt (2nd ed.). New York: Wile y-Interscience.
Johnson-Laird, P.N., & Oatley, K. (1989). The languag e of emotions: An analysis of a
semantic ® eld. Cognition and Emotion, 3, 81± 123.
M acLeod, C.M. (1991). Half a century of research on the Stroop effect: An integrativ e
review . Psychological Bulle tin, 109, 163± 203.
M arslen-Wil son, W.D . (1984). Func tion and process in spoken word recognition. In H.
Bouma & D.G. Bouw huis (Eds.), Attention and performance, X: Control of la nguage
processes. Hills dale, NJ: Lawrence Erlbaum Associates Inc.
M assaro, D.W. (1987). Categorical partition: A fuzzy-log ical model of categoriz ation
behaviour. In S. Harnad (Ed.), Categoric al perception (pp. 254± 283). Cambridge
Unive rsity Press.
M cCle lland, J.L., & Elman, J.L. (1986). The TRACE model of speech perception.
Cognitive Psycholo gy, 18, 1± 86.
M organ, R.L., & Heise, D. (1988). Structure of emotions. Social Psychology Quarterly, 51,
19± 31.
422
W U R M AND V AKO C H
M urray, I.R., & Arnott, J.L. (19 93). Toward the simulation of emotion in synthetic speech:
A review of the literature on human vocal emotion. Journal of the Acoustica l Society of
America, 93, 1097± 1108.
Old® eld, R.C. (1966). Things, words, and the brain. Quarterly Journal of Experimental
Psychology, 18, 340± 353.
Ortony, A ., Clore, G.L., & Foss, M.A. (1987). The referential structure of the affectiv e
le xic on. Cognitive Science, 11, 341± 364.
Osgood, C.E. (1969 ). On the whys and wherefore s of E, P, and A. Journal of Personality
and Social Psychology, 12, 194± 199.
Osgood, C.E., May, W.H., & Miron, M.S. (1975). Cross-cultural unive rsals of affectiv e
m eaning. Urbana, IL: Univ ersity of Illinois Press.
Osgood, C.E., & Suci, G.J. (1955). Factor analysis of meaning. Journal of Experimental
Psychology, 50, 325± 338.
Osgood, C.E., Suci, G.J., & Tannenbaum, P.H. (1957). The me asurement of meanin g.
Urbana, IL: Univ ersity of Illinois Press.
Palef, S.R., & Nicke rson, R.B . (1978). Re presenting auditory space. Perception and
Psychophysics, 23, 445± 450.
Russell, J.A. (1983). Pancultural aspects of the human conceptual organization of emotions.
Journal of Personality and Social Psychology, 45, 1281± 1288.
Russell, J.A., & Ridgeway, D. (1983). Dimensions unde rlying childre n’ s emotion concepts.
D evelo pmental Psychology, 19, 795± 804.
Sc he rer, K.R. (1986). Vocal affect expression: A review and a model for future research.
Psychological Bulletin , 99, 143± 165.
Se gui, J., Mehle r, J., Frauenfelder, U., & Morton, J. (1982). The word frequency effec t and
le xic al access. Neuropsychologia , 20, 615± 627.
Shor, R.E. (1975). An auditory analog of the Stroop test. Journal of General Psychology,
93, 281± 288.
Tabachnic k, B.G., & Fidell, L.S. (1989). Using multiv ariate statis tic s (2nd ed.). New York:
Harper Collins.
Walker, P., & Smith, S. (1984). Stroop interferenc e based on the synaesthetic qualities of
auditory pitch. Perceptio n, 13, 75± 81.
Walker, P., & Smith, S. (1985). Stroop interfe rence based on the multimodal correlates of
haptic size and auditory pitch. Perception, 14, 729± 736.
Walker, P., & Smith, S. (1986). The basis of Stroop interfere nce involving the multim odal
correlates of auditory pitch. Perception, 15, 491± 496.
Warren, R.E. (1977). Time and the spread of activation in memory. Journal of Experim ental Psychology: Human Learning and Memory, 3, 458± 466.
D IM E N S IO N S O F S P E E C H P E R C E P T IO N
423
A PP EN DIX
S tim u li
Words
Disgusted: disgusted, annoyed, dis pleased, irked
Petri® ed: petri® ed, terri® ed, afraid, scared
Happy : happy, ple ased, glad, contente d
Othe r: empty, shaken, outraged, delig hted, passionate, unhappy, apprehensive, resentful,
agitated, overjoyed, fearful, de¯ ated
Nonwords
melantoly, lonefome, deprussed, sickemed, eloted, emballassed, micherable, exsouted,
lonery, chee rlace, reliejed, downhorted, joyw ess, regletful, hurk, jealoos, frighkened,
lovezhic k, prud, mab, irete, charned, overyelmed, horrisie d