Psychological ReviewI960, Vol. 67, No. S, 332-342

WORD VALUES, WORD FREQUENCY, AND VISUALDURATION THRESHOLDS

RONALD C. JOHNSON, CALVIN W. THOMSON, AND GERALD FRINCKESan Jose State College

A number of studies have shown arelation to exist between word fre-quency, word value, and visual durationthresholds. Certain issues have devel-oped concerning the interpretation ofthe observed relations. Two types ofinterpretation of results can be dis-tinguished :

1. Those interpretations that claimword frequency to be the major deter-minant of visual duration threshold.This point of view contends that theresponses made to the tachistoscopicpresentation of words are learned in thesame manner that other responses arelearned. Differences in the visual dura-tion thresholds of words can then beaccounted for in terms of word fre-quency so that the introduction of suchtenuous and "unanchored" variables asperceptual selectivity and perceptual de-fense is a violation of the law ofparsimony. Thus, the problems thatexist with regard to the differentialvisual duration thresholds of words areproblems in learning, not problems inperception.

2. Those interpretations that attemptto show that differences in the visualduration thresholds of words are due,all or in part, to differences in the affec-tive qualities (values, goodness-bad-ness, affective tone, emotional valence)attached to words. These affectivequalities affect visual duration thresh-olds directly, through perceptual selec-tivity and/or perceptual defense. Vis-ual duration thresholds of words aredetermined by variables that can becalled perceptual variables instead of oralong with those variables that usuallybear upon learning.

We wish to center our introductorydiscussion around those few studieswhich, for us, best exemplify the twopositions described above. Solomonand Howes (1951) take the positionthat differences between words can beaccounted for on the basis of wordfrequency. The perceptual process doesnot differ in any fundamental way fromthe learning process. They say:

Emotional factors undoubtedly operate toan important extent in the building of wordfrequencies in a given life history. In thisway they would be related to word frequencyand, indirectly, to the duration thresholds.. . . But to date we can find no evidence tosuggest that emotional factors operate in thetachistoscopic situation independently of theireffect on word frequency (p. 267).

According to Solomon and Howes(1951, p. 258) visual duration thresh-old can be accurately predicted from aknowledge of population-wide word fre-quency as given in the Thorndike-Lorge tables (1944). There is no needto bring in concepts such as perceptualselectivity or perceptual defense inorder to explain differences in visualduration thresholds of words. To Sol-omon and Howes emotional factors op-erate only in producing idiosyncraticvariation in word frequency. Thesedifferences in frequency account for dif-ferences in the visual duration thresh-olds of words such as those found be-tween the visual duration threshold ofa value-oriented word for subjects (5s)who score high as opposed to low onspecific scales of the Allport-VernonStudy of Values. They do not haveany other influence on visual durationthreshold.

332

WORD VALUES, WORD FREQUENCY, VISUAL DURATION THRESHOLDS 333

The second position concerning therelation between word frequency, wordvalue and visual duration thresholdis clearly expressed by Postman andSchneider (1951). Faced with theproblem of salvaging value as a variableinfluencing visual duration threshold,they say that

It may . . . be more profitable theoret-ically to regard both frequency of wordusage and duration thresholds as dependentvariables, both manifestations of more funda-mental psychological properties attributed tothe organism, such as "habits," hypotheses,or even, perhaps, "personal values" (p. 277).

While not denying that word fre-quency can account for a considerableproportion of the variance in visualduration thresholds of words, theyclaim that values (called directive fac-tors in this paper) directly influence thevisual duration thresholds of infrequentwords.

Postman, in a later article (1953),suggests that the frequency of usage ofa specific word is determined by itsreinforcing qualities. He suggests thatfrequent as opposed to infrequent wordsdiffer systematically in affective tone,He presents evidence to show that thesheer number of pleasant words is muchgreater than that of unpleasant words.Postman says, "Social control over en-vironmental stimuli will, then, tend tocreate a correlation between frequencyand positive value" (p. 68). Whethercalled values, directive states, or affec-tive qualities, certain emotive or affec-tive aspects of words influence theirgeneral (e.g., Thorndike-Lorge) fre-quency. Affective quality determinesfrequency, not merely in an idiosyn-cratic way as Solomon and Howes(1951) suggest, but also in a verygeneral sense for all words used by allindividuals.

It has been hypothesized from thissecond position that values can influ-ence visual duration threshold in two

ways. The affective qualities of wordsinfluence visual duration threshold di-rectly as in the case of an S's Allport-Vernon score influencing his thresholdsfor infrequent value oriented words(Postman & Schneider, 1951). (It hasbeen shown that the 5" who values anarea such as aesthetics highly reportsinfrequent "aesthetic" words at lowerthresholds than an individual who doesnot value aesthetics highly. However,as Solomon and Howes point out, thismay be due to "idiosyncratic" differ-ences in frequency of exposure.) Theaffective qualities of words influencevisual duration threshold indirectly,since the frequency of general usage ofany word is determined to a consider-able extent by the affective tone of thatword, with this frequency then actingas a determiner of visual durationthreshold (Postman, 1953).

A third position might also be takenconcerning the relations between wordfrequency, word value, and visualduration threshold. Postman (1953)touches upon this last possible interpre-tation. He says, "Finally, we cannotentirely discount the possibility thatfamiliarity resulting from frequencymay be in itself a source of positivevalue" (p. 68).

It appears, historically, as if the firstpoint of view—that emphasizing wordfrequency, playing down the influenceof value except as it produces idiosyn-cratic differences in word frequency-has won out. None of the three posi-tions described above have been provento be untenable, however. Indeed, wehave all of the original questions left,plus a few new ones as well. The ques-tions that we see as being testableand requiring further investigation arethese:

1. Is word value related, in a generalsense, to word frequency? All of the threepositions described above would admit theexistence of a relation between value and

334 RONALD C. JOHNSON, CALVIN W. THOMSON, AND GERALD FRINCKE

frequency—they would diverge in opinionwith regard to the generality of the relation-ship. It is true that a knowledge of wordfrequency allows us to predict visual durationthreshold. But why is the frequent wordfrequent? Perhaps we should go one stepbeyond frequency to determine whether otherattributes of words vary systematically withfrequency.

2. If a general relation between value andfrequency exists, does it exist because morepositively valued words are used more fre-quently or because one increases positiveaffect merely by increasing frequency? Ordoes the dependent-independent relationshipdepend on the experimental procedure used?

3. If value and frequency are related in ageneral, not merely an idiosyncratic sense,then one would have to separate out theinfluences of each on visual duration thresh-old. Are there significant differences in vis-ual duration threshold between "good" and"bad" words of equal population-wide fre-quency? Are there significant differencesbetween frequent and infrequent words equalin affective tone?

These are the three problems thatwe have attempted to deal with in theseries of experiments reported below.

Our first concern was to determinewhether a general relation exists be-tween word value and word frequency.As we noted above, there is reason tosuspect that emotional factors do op-erate in the building of word frequency.In a large sample of words for whichwe have actual measures of goodness,the words in the semantic atlas pre-pared by Jenkins, Russell, and Suci(1958), one finds better than a two toone preponderance of good over badwords. While the words in this atlasare not a random selection of words inthe English language, the sources fromwhich the words were selected seemunlikely to be systematically biased infrequency of pleasant as opposed to un-pleasant words. One is led to believethat this ratio is characteristic of theentire English language when examin-ing proportions of pleasant to unpleas-ant words in a sample of ISO wordsselected at random from the Thorndike-

Lorge tables for use in the experimentsin Series I, discussed below. Of theseISO words, 95 were on the good end(4.00 or less) of the semantic differen-tial, a ratio of approximately 2:1. Itseems highly probable that the sheernumber of pleasant words in the Eng-lish language is far greater than thenumber of unpleasant words.

It seems equally likely that the fre-quency with which any specific wordis used also depends to a considerableextent on its goodness or badness:hence on its reinforcing qualities. Thisquestion is amenable to determination.As Solomon and Howes (1951) state,"In the absence of further data it seemsbest merely to define word frequencyfor English words as the frequency ofwords in the Thorndike-Lorge tables"(pp. 264-265). If affective factors op-erate in the building of word frequency,then pleasantly toned words shouldhave higher frequencies in the Thorn-dike-Lorge word count than unpleas-antly toned words. Experiment I wasdesigned to determine whether this wasthe case.

EXPERIMENT IIn the first of this series of ex-

periments we attempted to determinewhether the pleasantness of words—inthis case measured by the good-badscale of the semantic differential—wasrelated to word frequency as measuredby the Thorndike-Lorge tables.

We selected a sample of words fromthe Thorndike-Lorge tables. This sam-ple of words consisted of two wordsselected at random from within everyalphabetic category (except X) in thetables. These words were rated on thegood-bad dimension of the semanticdifferential by 24 5s, all freshmen tak-ing an introductory psychology course.The rank order correlation between thegoodness and the L count frequency ofthese words was +.63. Another ran-

WORD VALUES, WORD FREQUENCY, VISUAL DURATION THRESHOLDS 335

dom sample of 50 words was takenfrom the Thorndike-Lorge word list inthe manner described above. Thesewords were rated by 28 5"s, a differentgroup than those who made the first setof ratings. The rank order correlationbetween goodness and L count fre-quency in this list was +.40, again sig-nificant to the .01 level of confidence.A third random sample of 50 wordswas drawn from the Thorndike-Lorgetables. These words were rated by 245"s, none of whom had taken part inprevious experiments. The correlationbetween L count frequency and good-ness for this set of words was +.38.All of these three correlations are sig-nificant to the .01 level. The results ofthese experiments, regardless of differ-ences in the magnitude of the correla-tions, indicate that there is a positiveand significant correlation betweenword frequency and goodness. Valueattributes of words are related to wordfrequency—and not merely in an idio-syncratic way, but in a far more gen-eral sense as well.

EXPERIMENT IISolomon and Howes (1951) list 30

frequent and 30 infrequent words thatthey used in a study of the relationof word frequency to visual durationthreshold. Five frequent and 5 infre-quent words had to do with each of thesix value areas of the Allport-VernonStudy of Values. Frequent and infre-quent words were sometimes, but notalways, synonymous. We paired thesewords, matching the first frequent withthe first infrequent word, and so on,through the two lists. Position of pres-entation within pairs was determinedby coin flip. These 30 pairs of wordswere presented to 34 freshman intro-ductory psychology students, none ofwhom had taken part in any previouspsychological experiments. 6"s weretold to "encircle the most pleasantly

toned word of each pair." hi 26 of theword pairs, the more frequent wordwas chosen by the majority of the 5s asthe most pleasantly toned. (The 4exceptions, in which the least frequentword was chosen as the more pleasantlytoned, were the word pairs limousine-automobile, economics-assets, orchestra-ensemble, and celestial-heavenly.) Asign test (Guilford, 1956, p. 248) indi-cates that this preference for the morefrequent word in each pair could occurby chance less than one time in a hun-dred. Again, the more frequent wordis rated as the better word.

In the experiment discussed belowwe attempted to determine whether fre-quency and goodness were related whennonsense syllables were used as stim-ulus materials.

EXPERIMENT IIITwenty-two 5"s rated two separate

visually represented lists of nonsensesyllables for goodness, as measured bythe semantic differential. An intervalof one week separated the ratings of thetwo lists. Each list consisted of eight100%, eight 47-53%, and eight 0%association value (Glaze, 1928) non-sense syllables in random order. Syl-lables were selected at random fromwithin each block of ten Glaze syllablesas listed in the Handbook of Experi-mental Psychology (Stevens, 1951, pp.540-546) at each of the associationvalues used. Then the eight syllablesused in each list at each associationlevel were drawn from this pool ofrandomly selected syllables. We be-lieved that if frequency of exposure wasrelated to goodness, then the higher theassociation value of the syllable, thebetter the syllable should be rated. Thisseemed likely, since the associationvalue of nonsense syllables appears tobe determined largely by the frequencyof occurrence of the letter combina-tions in meaningful words (Underwood.

336 RONALD C. JOHNSON, CALVIN W. THOMSON, AND GERALD FRINCKE

1959). Table 1, showing two sets ofsyllables and their mean ratings ofgoodness, indicates this to be the case.

A Mann-Whitney (1947) test of thesignificance of differences between syl-lables of 100% and 47-53% associationvalues in List 1 indicates that the 100%list is rated as somewhat better (P< .10). The difference in goodnessbetween 47-53% and 0% syllables issignificant (P < .01); Mann-Whitneytests of List 2 show the 100% and the47-53% list to differ in goodness (P< .01); and the 47-53% list and the0% list to differ significantly (P< .05) as well.

Just as the goodness and the fre-quency of meaningful words in theEnglish language are related to oneanother, so also is goodness related tothe association value of nonsense syl-lables ; the higher the association value,the better the syllable.

Our data indicate that value and fre-quency are related. We have moved at

TABLE 1RATINGS OF NONSENSE SYLLABLES OF DIF-

FERING ASSOCIATION VALUES ON THEGOOD-BAD DIMENSION OF THE

SEMANTIC DIFFERENTIAL

100%Syllables

List 1BULFESHONJINRACROVSURWIL

List 2DULBALLOVSARLIKHAVWATROV

Rating

4.723.412.112.983.983.742.722.46

4.272.351.443.232.502.353.463.54

47-53%Syllables

WOBKAWDUXBOZVOKGEYTIZVOZ

BEKPIJWEPRIOYAFLUBGOWRUV

Rating

4.594.334.334.003.462.933.524.30

3.465.174.714.463.964.153.753.83

0%Syllables

ZEQWUHXIWWUQQTHGUVVECXUR

YUFQIFZtlKZIFXADXUCQITJGUQ

Rating

4.914.485.365.264.874.393.915.28

4.464.445.044.064.855.355.005.98

least somewhat beyond the empiricalfact that word frequency is related tovisual duration threshold. We knowsomething else about the stimulus qual-ities of frequent as opposed to infre-quent words; frequent words are notonly more frequent but are better aswell. Our second question then be-comes relevant. Does a manipulationof frequency cause systematic variationin value or do differences in value pro-duce differences in word frequency, ordoes the independent-dependent statusof each of these variables vary depend-ing on the experimental proceduresused? Experiment IV bears upon thisquestion.

EXPERIMENT IVFourteen 5"s rated the goodness of

twenty nonsense words. These wordswere from the list of words used bySolomon and Postman (1952) in theirstudy of the effect of "built in" fre-quency on visual duration threshold.The words used were: JANDARA, AF-WORBU, BIWOJNI, NANSOMA, ENANWAL,IKTITAK, SARICIK, ZABULON, CIVADRA,LOKANTA, KADIRGA, ADAFNAW, BOR-ULCE, NIJARON, ENSHIMI, INDULAM,TAVHANE, UDIBNON, DILIKLI, and MEC-BURI. Mean goodness scores were ob-tained. One week later each 5" wastested individually. When the 5" en-tered the experimental room he wasgiven the following instructions (fromSolomon & Postman, 1952):

This is an experiment concerning the effec-tiveness of repetition in learning to pro-nounce strange words correctly. It has adirect bearing on the problem of readingwords in a foreign language, as compared tohearing the words spoken. In addition, weare interested in knowing whether the rela-tive effectiveness of the two kinds of learningmethods depends on general reading ability.

We are going to give you a deck of cardsOn each card is printed a strange word. Wewould like you to look at each card carefullyand then pronounce the word in the way itwould be pronounced if it were a word in the

WORD VALUES, WORD FREQUENCY, VISUAL DURATION THRESHOLDS 337

English language. Proceed steadily fromcard to card, turning over each one after youhave finished with it. Go right through thedeck and then stop unless you have seriousdoubts about your pronunciations.

The experimenter (£) handed the 5 apack of 90 3" X 5" cards. In this packwere the words listed above. The firstfive words above were presented 10times apiece, the next five, 5 timesapiece, the next five, 2 times apiece,and the last five, once each. The cardswere shuffled thoroughly after each £completed this task, so that they werein a relatively random order.

All responses were taped in order tolend realism to the expressed purposeof this new situation. Since Ss hadparticipated in another experiment onthe pronbunceability of words, it seemslikely that (although the 5s had seenthe words before while rating them onthe semantic differential) they weremisled as to the purpose of this ex-periment. Each 6" rated the nonsensewords a second time following this pro-nounceability test.

Figure 1 indicates the mean pre-exposure and postexposture ratings forwords of each frequency of exposure.While the differences between ratingsprior to exposure were not significant,a Kruskal and Wallis H test (1952)shows that words of differing frequen-cies of exposure do differ significantlyin the postexposure test (P < .01, H= 7.83).

In this experimental situation a ma-nipulation of word frequency alters 5Vevaluations of word goodness. Onecan, however, think of instances wherethe reverse might well occur. The mostobvious example of this is in verbalconditioning, where the differential re-inforcement of a class of words pro-duces systematic variation in the fre-quency with which this class of wordsis emitted. We would suspect that inthis case frequency of emission (and

en§£BU

ii1

Good1.0 ,

3.0 '

3.2

3.4

3.6

3.8

4.0

4.2

4.4

4.6

4.8

5.0

j

/'

X '/

**y '

\ y "\l ** ^^^***~\. - ' ' ̂^^^

" ' N̂ ^̂

prr B^

» - _ _ ~ a Post Exposure

1 , 2 . 5 . 10

NUMBER OF EXPOSURES

FIG. 1. Ratings of nonsense words beforeand after word frequency was manipulated.

hence of reinforcement) alters the af-fective quality of the words. Reinforce-ment might then produce further varia-tion of frequency.

No matter whether frequency of ex-posure alters value or vice versa, orboth, one thing is evident—there is arelation between word frequency and5V ratings of word goodness. Sincethis is so, our third question also be-comes relevant. Since value and fre-quency are generally related, it becamenecessary to experiment with samplesof words matched in frequency, varyingin goodness, plus samples of wordsvarying in frequency, matched in good-ness, in order to separate out the influ-ence of value and of frequency onvisual duration threshold. This wasattempted in Experiment V.

EXPERIMENT VWe had various groups of 5"s rate

the goodness of a large number ofwords on the semantic differential.These groups rated 60 words of aThorndike-Lorge G count of 12, 13, or14, three random samples of 50 wordseach (described in Experiment I), and39 words, one sample of 30 words plus

338 RONALD C. JOHNSON, CALVIN W. THOMSON, AND GERALD FRINCKE

TABLE 2WORDS USED IN THE TACHISTOSCOPIC EXPERIMENT

Words of Matched Goodness

Frequent

Word

Sample 1ThinkButterDoctorHealthSleepSweetMotherHouseFlower

Sample 2MountainWaterWindowGreenSmoothFatherPrettyStreet

SDRating

2.341.971.601.972.221.931.732.461.87

2.732.432.682.932.572.362.132.77

Thresh-old

.0373

.0692

.0369

.0492

.0362

.0631

.0315

.0323

.0369

.0687

.0217

.0400

.0296

.0522

.0800

.0661

.0496

Infrequent

Word

IncomeCaressIvoryPianoRadiantPatriotVisionAgileBible

BodiceDoughGradualElegantZenithJellyRefreshLenient

SDRating

2.521.971.681.972.081.901.572.741.87

2.742.472.633.002.622.492.092.77

Thresh-old

.0500

.1085

.0262

.0546

.0662

.0992

.0546

.0623

.0462

.0965

.0757

.0965

.0852

.0330

.0800

.0661

.1522

Words of Matched Frequency

Good

Word

RallyGlistenProsperIvoryGingerAlertDeputyVehicleCaress

BodiceCandyEngineBraveStatueCushionGarmentChurch

GCount

121313131314131312

232AA

322140AA

Thresh-old

.1092

.0777

.1046.0262.0446.0392.0446.0492.1085

.0965

.0409

.0565

.0739

.0748

.0900

.1043

.0530

Bad

Word

StammerCorruptDefectHinderWretchSneerScandalDismalPenalty

SlimeBeggarBitterAngerDivorceInferiorGriefDanger

GCount

121214131212131313

429AA

291945AA

Thresh-old

.1538

.0985

.0815

.0969

.1285

.0785

.1123

.1162

.0469

.0713

.1087

.1704

.0417

.0948

.1948

.0687

.0583

9 words which happened to appear onthe other lists, from Jenkins' semanticatlas. From these words, and fromJenkins' atlas, we selected 17 pairs ofwords matched in frequency, varyingin goodness, and 17 pairs of wordsmatched in goodness, varying in fre-quency. (Some words among thosematched in goodness were rated byJenkins' 5s and by our own. Althoughthe rank order correlation of goodnessratings between our own and Jenkins'sample of 5s was +.98, for a group of30 of these words, there was somevariation in ratings between the twosamples. Whenever this occurred, weused the mean goodness rating assignedby our 5s.) The first 9 words from eachlist were presented to the first group of26 5s; the last 8 to a second group of23 5s. The four lists of words, dividedas they were presented to two samplepopulations, are presented in Table 2.

The words were put on slides in theform of bold faced capital letters. Thewords were presented to 5s by meansof a Revere 888 tachistoscope. Each 5

sat 18 feet from the screen on whichthe words were projected. The testingroom was fully illuminated while thetachistoscope was used at the smallestpossible lens opening. The ratio' ofroom illumination to screen illumina-tion was 1:1.08. The words were pre-sented in random order with the orderremaining constant in all presentationsto any specific 5 but 'with the ordervarying between 5s. A modifiedmethod of limits was used in whicheach of the 36 words was presentedonce at each of the following speeds:1/100, 1/50, 1/25, 1/5, 1/2, and 1 sec-ond. Under the conditions of presen-tation used in this experiment, only oneword was recognized at 1/100 of asecond while all words were recognizedat 1 second.

Each 5's mean visual duration thresh-old for frequent as opposed to infre-quent and good as opposed to badwords was obtained. The 5s in thefirst sample (N — 26), exposed to thefirst group of words, differed signifi-cantly in mean thresholds for good vs.

WORD VALUES, WORD FREQUENCY, VISUAL DURATION THRESHOLDS 339

bad words (t = 3.56, P < .01, usingthe formula for correlated data), withgood words reported at lower thresh-olds. These 5s also differed in meanthresholds for frequent vs. infrequentwords (t = 2.33, P < .05), with fre-quent words reported at lower thresh-olds. The 5s in the second sample (N= 23), exposed to the second group ofwords, differed significantly in meanthresholds for good vs. bad words (t= 3.25, P < .01), with good words re-ported at lower thresholds, and inthreshold for frequent vs. infrequentwords 0 = 3.82, P < .01), with fre-quent words reported at lower thresh-olds.

DISCUSSIONWe attempted to obtain data relevant

to three specific questions. Let us lookat these questions, along with the rele-vant data.

1. Is value related, in a general sense, toword frequency? Our data is unequivocallyaffirmative. Whether dealing with samplesof words in the English language, with fre-quent vs. infrequent words in various All-port-Vernon value areas, or with nonsensesyllables, frequency, as measured by theThorndike-Lorge tables and by associationvalue, and positive value, as measured by thegood-bad scale of the semantic differential,are significantly related in a general sense.

2. If this general relation exists, does itexist because more positively valued wordsare used more frequently or because one in-creases positive value merely by increasingfrequency? We found that a manipulationof frequency produces systematic variation in5V ratings of the goodness of words. Itseems highly probable, however, that thereverse may also be true—that by increasingthe goodness of words, one also increasesfrequency. At this point it seems likely thatthe independent-dependent relation betweengoodness and frequency would depend on theexperimental situation. We would not haz-ard a guess as to the effect of each on theother in general speech, although a relationbetween frequency and goodness has beenshown to exist.

The data gathered in the first threeexperiments suggest that other attrib-

utes of the stimulus vary systematicallyas one varies frequency. Perhaps thisis true for other forms of learning also.The fourth experiment would suggestthe possibility that a manipulation ofthe frequency of occurrence of a stim-ulus and a contiguous response, in it-self and without any regard to drivereduction, changes the reinforcing qual-ities of the stimulus. Perhaps weshould look beyond the influence of fre-quency on response probability and askwhy frequency changes response prob-ability. Is it the difference in associa-tion value (and hence, in frequency ofoccurrence) that causes 100% associa-tion value syllables to be learned morerapidly than 0% syllables ? Or is it thevariation in affective quality accom-panying differing frequencies of occur-rence? Since the relation betweengoodness and frequency is not perfect,one can obtain syllables that vary from0 to 100% in association value, yet areequal in semantic rating—this shouldprovide a test for our question.

These results seem to have consider-able social consequence. If word fre-quency and word goodness are related,and if a manipulation of the frequencyof occurrence of a word produces sys-tematic variation in word value, thenone wonders whether ideas and per-sonal values are manipulable in thesame fashion. If so, then perhaps thoseelements of society that believe in cen-sorship would find support in these ex-periments for one of their major as-sumptions, viz., that frequent exposureto an idea or a set of behaviors—vio-lence, drinking alcoholic beverages, orwhat have you—lessens the "evil" asso-ciated with the idea of behavior. Thevalidity of their second major assump-tion—that if we make a thing evilenough, people won't do this thing—is,of course, a very different problem.

If the more frequent becomes dem-onstrably better merely because it is

340 RONALD C. JOHNSON, CALVIN W. THOMSON, AND GERALD FKINCKE

more frequent, one can well understandthe values of advertising. If it is gen-erally true that the most frequent wordsin the English language are also themost good, the logical conclusion thatthis would lead to is that ads would bemore effective if written at an evenlower vocabulary level than is currentlyused. "Oh, brave new world. . . ."

3. If value and frequency are related, whatare the influences of each on visual durationthreshold when the other variable is heldconstant? Words equal in frequency, vary-ing in value, are correctly reported by 6"s atdifferent mean thresholds, with good wordsbeing correctly identified at significantlylower thresholds than are bad words. Wordsequal in goodness, varying in frequency, arereported by ^s at different mean thresholds,with frequent words being correctly identi-fied at significantly lower thresholds.

While it has been generally acceptedthat values have some secondary effecton visual duration thresholds, it is alsogenerally accepted that the influence ofvalues on threshold is small, producingidiosyncratic variations in frequencyand hence influencing thresholds in thisway. Our results indicate that valuesinfluence threshold even when fre-quency of exposure is held constant,and that this influence is approximatelyequal to that. of frequency with valueheld constant, using these lists ofwords.

Since word goodness, independentfrom frequency, does significantly influ-ence visual duration threshold, we areleft with the problem of "Why?" Theterms "perceptual selectivity" and "per-ceptual defense" are descriptive, notexplanatory. We have no answershere, although we would like to discussseveral possible explanations.

One interpretation of the fact thatgood words are perceived at lowerthreshold than bad words of the samefrequency might be called the "simple"guess theory. Good words appear tobe about twice as common as bad words

in the English language. Specific goodwords are likely to be more frequentlyused than specific bad words. It seems,then, that good words occur betweenthree and four times as frequently asbad words in terms of usage. Eventhough the frequency of occurrence ofthe stimuli are equaled, by matchinggood and bad words for Thorndike-Lorge frequency, response probabilitiesprobably are still not equal. If an 6"sees a word in a tachistoscope but iden-tifies only a "g d " he is prob-ably three or four times more likely, onthe basis of responses made in the past,to fill in the gaps with a good word,such as "gradual," rather than a badword, such as "goudge." This simpleguess explanation may be partially cor-rect but is not too satisfying to us, sincean examination of our Ss' responses towords on the good-bad lists indicatesthat there seems to be a rather stronglikelihood that when 5"s guess at butmiss a bad word, they respond with an-other bad word—incorrect but still bad,while when they respond incorrectly tothe tachistoscope presentation of a goodword, the incorrect response is usuallyanother good word. This finding runscontrary to the idea that perceptual de-fense is a significant variable but alsosuggests that something more system-atic than mere guessing is involved inerrors and in correct responses to tach-istoscopically presented words.

A second explanation for the dif-ferences in ,Ss' mean visual durationthresholds might be called the "com-plex" guess theory. This explanationis derived from Bruner and Postman's(1949) hypothesis theory of perceptionand cognition. According to Postman(1951) hypotheses are, "in the mostgeneral sense, expectancies or predis-positions of the organism which serveto select, organize, and transform thestimulus information that comes fromthe environment" (p. 249). The

WORD VALUES, WORD FREQUENCY, VISUAL DURATION THRESHOLDS 341

strength or availability of certain hy-potheses is stronger than that of others.This strength is a function of the fol-lowing factors: frequency of past con-firmation of the hypothesis, number ofalternative hypotheses, motivational rel-evance, and cognitive support. Thefirst two factors—frequency of pastconfirmation and number of alternativehypotheses—fit well into the simpleguess theory, while the last factor, thatof cognitive support, may not be rel-evant to this discussion. The factor ofmotivational relevance might possibly,however, produce differential predic-tions between the two explanatory de-vices that we have labeled simpleand complex guessing. As Postman(1951) puts it: "The stronger the mo-tivational support for a given hypothe-sis, the less of appropriate informationis required to confirm it" (p. 255).We would limit this statement some-what by tying it only to the effect ofreward or reinforcement on the numberof cues required for the confirmation ofa hypothesis. It may be that the dif-ferential reinforcement of various stim-uli causes those stimuli (as well as thoseresponses) that have most frequentlyled to reinforcement in the past to bemore readily observable or available inthe present. The availability of a spe-cific response may well depend on thefrequency that this response has beenrewarded in the past—-but perhaps be-cause that aspect of the stimulus situa-tion that has called forth the rewardedresponse in the past has become moresalient—has become more of a "figure"as opposed to the other stimuli becom-ing more like the "ground." To para-phrase our quotation from Postmangiven above: The more reinforcing orrewarding a word is, the less cues areneeded for the recognition of this word.We are advancing the proposition thatreinforcement not only influences theprobability that a given response will be

emitted, but also that reinforcementcauses that aspect of the stimulus situa-tion that is salient at the time that therewarded response is made to becomemore likely to be salient at the nextpresentation of the stimulus. If this istrue, then even if word frequencies ofgood and bad words are equal and evenif habitual modes of response (such asguessing good words) are controlled,the stimulus qualities of rewardingor reinforcing words would be suchthat less information is needed beforethe word is correctly recognized thanwould be the case for words whichwere presented with equal frequencybut were not equally reinforcing. Some-thing that might be called perceptualsensitivity might then be said to exist.This sensitivity would not, however, bea unique perceptual process, but wouldinstead depend on those old standbys inall forms of learning: frequency andreinforcement, a little intermingled tobe sure, but still observably present.

We have no way of testing out the"why" question with these data, al-though the types of wrong guessesmade in our study incline us away fromthe simple guess theory. We believethat the data presented in this study arevaluable chiefly in reviving the conceptof perceptual selectivity, not in solvingthe problem of why what is called per-ceptual selectivity exists. By system-atically varying frequency of exposureand frequency of reinforcement andnonreinforcement of nonsense syllablesin future experiments, we hope to even-tually be able to deal with the questionof why good words are perceived atlower mean thresholds than are badwords.

SUMMARYThree positions may be taken con-

cerning the relation between word valueand word frequency, and the relation ofeach of these to visual duration thresh-

342 RONALD C. JOHNSON, CALVIN W. THOMSON, AND GERALD FEINCKE

old. Two of these points of view wouldsuggest a general relation to exist be-tween word frequency and word value.Three experiments demonstrate the ex-istence of this relationship. The morefrequent a word or a nonsense syllableoccurs in the English language, thebetter it is likely to be rated on thegood-bad scale of the semantic differen-tial. Since this relation does exist, weattempted to test out, in ExperimentIV, one possible explanation of why therelation exists. A manipulation of thefrequency of nonsense words producessystematic variation in the rated good-ness of these nonsense words. Wordsthat are frequent are also very likely tobe rated as good. Hence, to separateout the influence of word value andof word frequency on visual durationthreshold, we tachistoscopically pre-sented lists of words matched in fre-quency, varying in goodness; matchedin goodness, varying in frequency.Subjects reported the good words atsignificantly lower thresholds than thematched bad words; the frequent wordsat significantly lower thresholds thanmatched infrequent words. Both fre-quency and value appear to operate inproducing differential visual durationthresholds for words. Possible ex-planations and certain implications ofthese findings were discussed.

REFERENCESBRUNER, J. S., & POSTMAN, L. Perception,

cognition, and behavior. /. Pers., 1949,18, 12-31.

GLAZE, J. A. The association value of non-sense syllables. /. genet. Psycho!., 1928,35, 255-267.

GuiLFORn, J. P. Fundamental statistics inpsychology and education. (3rd cd.) NewYork: McGraw-Hill, 1956.

JENKINS, J. J., RUSSELL, W. A., & Suci,G. J. An atlas of semantic profiles of 360words. Amer. J. Psychol., 1958, 71, 688-699.

KRUSKAL, W. H., & WALLIS, W. A. Use ofranks in one criterion variance analysis./. Amer. Statist. Ass., 1952, 47, 583-621.

MANN, H. B., & WHITNEY, D. R. On atest of whether one of two random vari-ables is stochastically larger than theother. Ann. math. Statist., 1947, 18, 50-60.

POSTMAN, L. Towards a general theory ofcognition. In J. H. Rohrer & M. Sherif(Eds.), Social psychology at the cross-roads. New York: Harper, 1951. Pp.242-272.

POSTMAN, L. The experimental analyses ofmotivational factors in perception. In J. S.Brown (Ed.), Current theory and researchin motivation. Lincoln: Univer. NebraskaPress, 1953. Pp. 59-108.

POSTMAN, L., & SCHNEIDER, B. H. Personalvalues, visual recognition, and recall.Psychol. Rev., 1951, 58, 271-284.

SOLOMON, R. L., & HOWES, D. H. Wordfrequency, word value, and visual durationthresholds. Psychol. Rev., 1951, 58, 256-270.

SOLOMON, R. L., & POSTMAN, L. Usage asa determinant of visual duration thresholdsof words. /. exp. Psychol., 1952, 43, 195-201.

STEVENS, S. S. (Ed.) Handbook of experi-mental psychology. New York: Wiley,1951.

THORNDIKE, E. L., & LORGE, I. The teach-er's word book of 30,000 words. NewYork: Bureau of Publications, TeachersCollege, 1944.

UNDERWOOD, B. J. Verbal learning and theeducative process. Harvard educ. Rev.,1959, 29, 107-117.

(Received September 15, 1959)