Perception & Psychophysics1975. Vol. 18 (2). 65-70

Reaction times for a same-different discriminationof vowel-consonant syllables

CHARLOTTE REEDBioacoustics Laboratory, Eye and Ear Hospital and School ofMedicine,

University of Pittsburgh, Pittsburgh, Pennsylvania 15213

Reaction times (RTs) were measured for decisions in a same-different discrimination of successivevowel-consonant nonsense syllables. Averaged data showed that "same" RTs were faster than "different"RTs and that the "different" RT decreased as the number of features (Wickelgren, 1966) by which a paircontrasted increased. For individual phonemic comparisons, two of the dependent variables, P(S I d), orprobability of responding "same" to a different trial and the mean correct "different" RT, were related inthat the RT increased as P(S , d) increased. The size of the difference between "same" and "different"RTs for a given phonemic contrast was directly related to P(S I d). The difficulty of a comparison, asdescribed by P(S I d) and by the difference between correct "same" and "different" RTs, was explainedthrough a markedness classification of phonemes.

Bindra , Donderi, and Nishisato (\968) haveproposed that the relative response latency of "same"and "different" decisions in a discrimination task is afunction of the codability of the test signals. Two ofthe experiments they reported involved stimuli thatwere readily codable. Verbal labels could be assignedeasily to these stimuli. which were colors in the visualmodality and clicks vs. pure tones in the auditorymodality. For these stimuli. "same" reaction times(RTs) were faster than or equal to "different" RTs.For tasks involving stimuli that were not as readilycod able (discrimination of pure tones. for example)the "different" RT was significantly faster than the"same" RT. Other investigators have reported similarresults. both for codable (Bamber. 1969. 1972;Burrows. 1972; Kreuger. 1970; Nickerson. 1965;Tversky, 1969) and noncodable signals (Bindra,Williams, & Wise. 1965; Kellogg, 1931; Nickerson,1969. 1971, 1972; Nishisato & Wise. 1967),

Mel nish and Tikofsky (\ 969) extended these resultsto auditorily presented speech. They measured RTsfor a same-different discrimination of consonant­vowel (CV) stimuli, with the vowel I al. Three lists of121 CVs were constructed, with each list having adifferent intersignal interval of 3, 6, or 12 sec.Different pairs contrasted one or two features (Miller& Nicely. 1955). Their 12 subjects were instructed torespond whether each syllable in the list was "same"

This research was supported by a grant from NINDS (NS 12501)and by an SRS training grant (44-P·1S072). The author wishes tothank the members of the advisory committee for her doctoraldissertation from which this paper was derived: Robert Bilger.Audrey Holland. Norma Hopkinson. Walt Jesteadt, CharlesPerfetti. and Marilyn Wang. In addition. the author wishes toexpress her appreciation to the following persons who read earlierversions of the paper: L. Braida, H. Cairns. R. Cole. N. Durlach, andC. Trahiotis.

65

or "different" from the preceding syllable. Thus, eachsou nd in the list (except the first and last) was used intwo separate judgments. Their results showed acorrelation between error rate and latency for theone-feature discriminations. thereby indicating thatlatency can be used as an index of discriminability.They also found that "same" comparisons had theshortest mean RT (397 msec) and one-featurecontrasts had the longest mean RT (466 msec), whiletwo-feature contrasts had a mean RT of 427 msec.

Cole and Scott (1972) presented pairs of phonemessimultaneously and dichotically for a same-differentcomparison. Of their 40 subjects, half served in aconsonant experiment and half in a vowel experiment.They used eight stimuli in each experiment. Theirresults indicated that the mean RTs for consonantswere significantly faster than for vowels. Within theconsonantal stimuli. their results showed an orderlydecrease in RT as the number of contrasting features(Halle. 1964) increased from one to six. The "same"RT was faster than the "different" RT for pairs ofconsonants contrasting one to four features.. Weiner and Singh (1974) measured RT for same­

different judgments of nine English fricativespresented in CV syllables. The RTs were analyzedusing multidimensional scaling, for which afour-dimensional solution was obtained. Thesedimensions were interpreted as the phonetic features ofvoicing, sibilant. front/back, and palatal. The timerequired for a "different" decision decreased as thenumber of features by which a pair contrastedincreased, both for a classification of phonemes withthe features derived from the scaling procedure and fora classification of phonemes using the features derivedby Miller and Nicely (1955).

None of the previous researchers have presenteddata for individual phonemic comparisons. Instead.

66 REED

RESULTS

o I 2 3 4 5

NO. CONTRASTING FEATURES

Figure 1. The mean correct RT across the four subjectli plotted asa function of the number of contrasting features (Wickelgren,1966). The "same" RT (0 contrasting features) is based on correctresponses for 6,000 trials. The data points for 1, 2, 3, 4, and 5contrasting features are based on correct responses for 4,400,1,200, 1,200, 400, and 800 trials, respectively.

I r

• •••

•

•

625

600I-a:I- 575

~ 550U

; 52S

SOO

~

pairs. were recorded. Each list contained an equal number of"same" and "different" trials. Each of the four "different"comparisons with each standard was presented twice in a list (40"different" comparisons per list). with the standard presented onceas the first member' of a pair and once as the second member. Eachof the 5 standard phonemes Ul:l/. Idi/. 131. Ig/. It I) and eachof the 10 comparison phonemes Uk/. Td/. If!. lsi. Iv I. /t:', IJ I.It! I. i e I. I IJ I) were presented 'in a "same" comparison twice ina list (resulting in 30 "same" comparisons). The 10 remaining"same" comparisons needed to balance the list were drawn fromthe phonemes 1011. /n.'. /p/. and /b/, which were neitherstandards nor comparisons.

Before analyzing individual phonemic comparisons.the data were examined in a manner comparable to theanalyses reported by Cole and Scott (1972), McInishand Tikofsky (1969). and Weiner and Singh (1974) todetermine the extent to which the present data

ProcedureThe same-different procedure was run under computer control.

Each trial was begun with a SOO-msec warning light. followed by theoccurrence of the first syllable of a pair. which was marked by anobserve light. A 75O-msec silent interval separated the first syllablefrom the second. and the presentation of the second syllablecoincided with a light labeled "answer." This answer lightremained on. for up to 4 sec until the subject responded bypressing one of two buttons labeled SAME and DlFF.Correct-answer feedback was provided. The RT was measured withl-rnsec resolution by timing out the interval from the onset of thesyllable to the depression of a button.

Six practice runs were conducted on the first day. three with eachleft-right positioning of the response buttons. On the 5 subsequentdays. subjects listened to 10 80-item lists per session. Each of the 25lists was presented twice. resulting in 100 observations percomparison per subject. The button order of SAME-DIFF for thefirst half of the experiment was reversed for the second half.Subjects were instructed to answer as quickly and as accurately aspossible and to respond using the index finger of the right hand.Subjects listened monaurally through the right ear.

they reported their data in terms of the means for all"same" trials and all "different" trials contrasting anequal number offeatures. The present study ofRTs forindividual phonemic comparisons was undertaken todetermine the differential speed of "same" and"different" RTs for specific phonemic contrasts.

SubjectsFour normally hearing young adults served as listeners. They

were tested simultaneously in six 2.S-h sessions. They were paidsz.».

METHOD

StimuliBecause the large number of English consonants discourages a

factorial study. selection of a subset of consonants was necessary.Selection of this subset was aided by the availability of confusionmatrices lor YCs for 22 subjects with both normal and impairedhearing. collected by Bilger. Wang. and Jesteadt (1972). Fivephonemes were selected as standards based on a tabulation of thefrequency with which the phonemes were misidentified by theselisteners. Three of the standards uf} I. I d 1 I. and 131) were thephonemes most frequently missed by tfie largest number oflisteners. The phoneme It I was chosen as a standard on the basis ofits high degree of recognition by all listeners. and I gl was chosen asa phoneme of intermediate ditficulty.

For each of these five standards. four comparison phonemes wereselected on the bases of confusion with the standard (as observedfrom the confusion matrices) and allowance for a range of featuredifferences. The standards and the phonemes chosen forcomparison with each standard are listed in Table 1. along with thenumber of contrasting features (Wickelgren. 19(6) for each pair.For the nonbinary place feature. each degree of difference betweenphonemes on this feature has been counted as a unit difference.Wickelgren's features have been used here because they werederived from consonantal confusions made in a short-term-memorytask. and the same-different task used here presumably involves'some use of short-term memory.

YC syllables (vowel I a.'). recorded by a male talker and stored ona Cognitronics Speechmaker (Wang & Bilger. 1973>. wereconverted to digital form using a to-bit analog-to-digital converterand a 10-kHz sampling rate. This sampling rate requires the use ofa S-kHz low-pass filter to eliminate distortion products that overlapthe speech frequencies. The digital values of the syllables werestored on a PDP-IS computer and subsequently plotted on an X-Yplotter by reconverting the digital values to analog form. Theduration of all syllables was determined by visual inspection of theseplots. The range of durations for syllables was 335 to 440 msec. Allsvllables were trimmed from the beginning and/or the end of thesyllable to produce a duration of 335 msec. ensuring a uniform,prescribed duration lor the same-different task. All of the syllablesexcept for /ct.', lag/. lat! I. lad~ I. /orn/, and lab I weretrimmed to some degree from the syllable onset. Only two syllables.laml and lag/. were not trimmed from the end. Two listenerscompared the trimmed and untrimmed versions of each syllable anddecided if there were any noticeable differences in quality orintelligibility between the two sounds. Each of the final. trimmedsyllables was judged to be an acceptable representation of thephoneme in question.

Tape-recording of the syllables for test presentation was effectedunder control of a PDP-IS computer in conjunction with an AmpexPR-IO tape recorder. The digitized values of the YCs werereconverted to analog form and passed through a S-kHz low-passfilter onto the tape recorder at a speed of 7.5 ips. The intersignalinterval was 750 msec. The time between pairs of syllables was6 sec. At playback the output of the tape recorder was passedthrough an attenuator into a four-way splitter which passed thesignal into calibrated earphones (TDH-49l in each of foursound-treated rooms (lAC 40IA). Syllables were presented at a levelof 60 dB SPL re a I-kHz calibration tone.

Twenty-live randomized lists. each consisting of 80 phoneme

REACTION TIMES FOR YCs 67

Note-Results are averaged across the four subjects and arebased on a total of 400 observations per phoneme. The finalcolumn of the table is the markedness value of each phoneme.

Table IA List of the 20 Phonemic Comparisons (Four ComparisonsWith Each of Five Standards) and the Feature Differences

(Wickelgren, 1966) Between Each Pair

Note-The four features in this system are voicing (V), open­ness (OJ. nasality (N), and place (P). Each degree of differencealong the place feature has been counted as a unit difference.Also provided are the P'[S ld), or probability of responding"same" to each "different" pair, and the key to Figures 2 and 3.

features (Wickelgren, 1966) by which a pair ofphonemes contrasted. was also related to the observedvariables. The largest values of P(S I d) were observedwith one- and two-feature contrasts. When the numberofcontrasting features was three or more, P(S I d) wasbelow 0.05. A wide range of P(S I d) values wasobserved. however, with the one-feature contrasts.While certain minimal pairs yielded high P(S I d)rates, others yielded rates as low as those obtained for

Table 2Mean Correct "Same" RTs and Mean Values of P(Sls), orProbability of a Correct "Same" Response, for the 15 Phonemes

that Were Either Standards or Comparisons

44333322222111o

Marked­ness

Value

.96

.96

.96

.95

.92

.96

.97

.95

.97

.94

.98

.98

.93

.97

.96

P(Sls)

562577575537542524542530488590452487560493550

Mean Correct"Same" RT

in Msec

1(11131Id31If I161IvlIzlIfI/tflIgIIqlIiiIdlIklItI

Feature Differences Key toPhonemic (Wickelgren,1966) Figures

Pair Number Features P(Sld) 2 and 3

13-vl 1 P .19 0

13-d1 1 0 .17 0

13-81 1 V .10 ()

lo-gl 4 O. P(3) .04 •/d"5-tfI 1 V .09 0

/d3-g1 1 P .02 G

/d3-d1 2 P(2) .02 [II

Id3-fl 5 v, 0, P(3) .03 •13-z1 1 P .20 013-fl 1 V .04 <:>13-g1 2 O,P .04 <J13-d31 1 0 .04 •Ig-dl 3 P(3) .05 /';

Ig-kl 1 V .06 L~

Ig~1 1 N .04 aIg I 5 V, 0, P(3) .06 •It-sl 2 O.P .09 "V

It-dl 1 V .05 "1}

[i-v] 3 V,O,P .02 ~

[t-z] 3 V,O,P .02 "If

Phoneme

replicate their results. The results of this analysis arepresented in Figure 1. where the mean correct RT isplotted as a function of the number of contrastingfeatures. The results are similar to those reported bythe earlier experimenters in that the mean correct"different" RT decreased as the number of contrastingfeatures increased. The mean correct "same" RT (0contrasting features) was shorter than any "different"RT.

For signal detection and RT analyses of individualphonemic comparisons, the trials of interest werethose for a given "different" comparison along withthe "same" trials for each of the two phonemes thatmade up the "different" comparion. A fourfoldstimulus-response matrix with dimensions of "same"and "different" was constructed for each comparison,with a total N of 300 trials per subject. Of these 300trials. 100 were the "different" trials for a givencomparison (SO in each order). The remaining 200trials resulted from grouping the 100 "same" trials foreach phoneme of the pair. For the comparison of It-d/,for example. the input "different" trials were the SOpresentations each of It-dl and Id-t/, The input"same" trials to the matrix were the 100 presentationseach of 11-11 and Id-d/. Although the ratio of "same"to "different" trials is 2: 1 in these matrices. the ratio of"same" to "different" trials in the experiment was 1:1.The cell entries in these matrices were either thenumber of each type of response for the signaldetection analysis or the mean RT for each type ofresponse.

"Same" trials with the two phonemes of a pair weregrouped since the value of P(S I S), lor probability of a"same" response to a same pair, did not vary greatlyfrom phoneme to phoneme. Both the sensitivity index.d ', and the criterion measure. (3, were determinedprimarily by the observed quantity P(S \ d), orprobability of a "same" response to a different pair.This finding can be explained by the fact that both d 'and (3 are based on the relation between P(S I s) andP(S I d). The quantity P(S I s) was a relatively stablevalue around 0.96 tor all comparisons, while thequantity P(S I d) varied from comparison tocomparison. Thus. the subjects were employing onlyone of the two degrees of freedom available in eachmatrix. For this reason. performance data will bepresented in terms of P(S I d).

The discriminability of a phoneme pair may beassessed through two of the variables studied in thisexperiment. a performance measure [P(S I d) in thiscase] and the time required to make a correct decisionof "different." The value of P(S I d) for each of the"different" comparisons is provided in the fourthcolumn of Table 1. The relation between P(S I d) andthe mean correct "different" RT is given. in Figure 2,tor phonemic pairs that covered the range of P(S Id)observed in the experiment. The size of the meancorrect "different" RT increased as P(S I d)increased.

An independent variable, that of the number of

68 REED

u TOO00::>w

III~

i?;660

.... e)0: 620 xru,

I!:>u,a

580....Juw

~ 540 ..au

Figure 2. The mean correct "different" RT for 10 phonemiccomparisons plotted as a function of P(S I d). Data points aremeans across subjects, and each point Is based on 400 observations.The key to this figure Is provided in Table 1.

were observed for certain of the one-feature contrasts.These comparisons also had the highest values ofP(S I d). A wide range of (0 - S) values was observedamong the one-feature contrasts.

Mean RTs for incorrect decisions, averaged acrossthe four subjects, are given in Table 3. The"different" stimuli have been grouped according tothe number of features by which the pair contrasted.The mean correct RTs, shown in Figure 1, have beenrepeated here so that RTs for correct and incorrectdecisions can be compared. An incorrect response of"same" to a different stimulus, (S I d), was the fastestof any of the RTs in all cases except for that of threecontrasting features. The next fastest RT was for acorrect "same" decision, (S I s). The RT for anincorrect response to a "same" stimulus, (0 I s). wasgenerally 40 msec longer than for the correct "same"response. The longest RTs generally were for correct"different" decisions, (0 I d).

DISCUSSION

Note-RTs are averaged across the four subjects. In the notationused to describe the stimulus-response condition, a capital letterdenotes a response and a lowercase letter denotes the stimulus.

Table 3RTs in Milliseconds for Correct and Incorrect Responses to"Same" Stimuli and to "Different" Stimuli Grouped Accordingto the Number of Features by Which a Pair Contrasts

"Same" "Different" StimuliStimuli Number of Contrasting Features

I 2 3 4&5

556493

567626

591416

611492

DidSid

534572

Mean RT results, presented in Figure 1, indicatethat "same" RTs were faster than "different" RTsand that the "different' RTs were a function of thenumber of contrasting features (Wickelgren, 1966)."Same" and "different" RTs were then examined forindividual phonemic comparisons in terms of adifference measure, (0 - S), between the two RTs.The results (Figure 3) showed a wide range ofvariability in this measure for comparisonscontrasting an equal number of features. particularlyfor the one-feature contrasts. Thus, the average dataare not generalizable to results obtained for individualphonemic comparisons.

Comparisons for which large values of (0 - S) wereobtained also had the largest values of P(S I d). Thesecomparisons were I~-v/, /~-d/, /~-e/, 13-z/,Id '3 -t J I, and It-s/, all one-feature contrasts exceptfor the two-featu re contrast of /t-s/. Three of the fourdistinctive features in the Wickelgren (1966) system(voicing. openness. and place) provided minimalcontrasts for these comparisons. The fourth feature.nasality. was contrasted minimally in only onecomparison, I g- ~ I, for which P(S I d) was low.Thus, the difficulty of a comparison. as described byP(S I d) and (0 - S). does not appear to be related to

SisDis

I 2 3NO. CONTRASTING

Figure 3. The difference between mean correct "different" and"same" RTs, (D - S), for individual phonemic comparisons plottedas a function of the number of contrasting features (Wlckelgren,1966). Data points are means across subjects and are based on1.200 observations per point. The key to this figure is provided InTable 1.

contrasts of more than two features. The one-featurecontrasts with the highest values of P(S I d) wereIJ-zi (.20). Iff-vi (.19), ItJ-dl (.17), 15ftl(.10), and Id3' -tJ I (.09). A P(S I d) of .09 was alsoobtained for the two-feature contrast of /t-s/.

Values ofP(S I s) and correct "same" RTs, averagedacross the four subjects, are given in Table 2. It can beseen that P(S I s) did not vary greatly from phoneme tophoneme. The phonemes are arranged by theirmarkedness values, a topic that will be raised in thediscussion of the results.

The difference between the mean correct"different" RT and the mean correct "same" RT,(0 - S), was computed for each of the 20 comparisonsfrom the matrices previously described. The size of(0 - S) was examined, i~ Figure 3, as a function ofthe number of features (Wickelgren, 1966) by whichthe "different" pair contrasted. The size of (0 - S)increased as the number of features, by which the paircontrasted decreased. The largest values of (0 - S)

minimal contrast on a particular feature.Previous reports in the literature for data averaged

across comparisons that contrast an equal number offeatures have led to the expectation that RTs and errorrates should be a function of the number of contrastingfeatures. The findings of the present study werecontrary to this expectation in that the total range ofexperimental results was observed among the 11comparisons that contrasted one feature. This resultindicated that the factors determining P(S Id) and(0 - S) involved something other than featuredifferences between phonemes. Thus, a search wasbegun for some other phonological property of soundsthat could account for the gradations in difficultyamong comparisons contrasting one feature. Theconcept of markedness (Cairns, Cairns, & Williams,1974; Chomsky & Halle, 1968) seemed to lend itself toan explanation of the variability of results obtainedamong the one-feature contrasts.

Although the concept of markedness is toocomplex to present fully in the present context,the basic percepts can be elucidated. Each of thefeatures that makes up a phoneme is assigned a valueof "marked" or "unmarked," determined by suchthings as "ease of articulation, perceptual saliency,and frequency of occurrence in languages of theworld" (Cairns, et al., 1974, p. 161). The number ofmarked features for a given phoneme is an indicationof its overall complexity, with a high markednessvalue corresponding to a high degree of complexity.The markedness values shown in Table 2 were takenfrom Cairns et al. (1974, Table I, p. 162). Thefeature system to which markedness values wereapplied was that of Chomsky and Halle (1968). Thesemarkedness values range from 0 to 4. Of thephonemes used as standards, I tj I and 13/ havemarkedness values of 4, Id 1 I a value of 3, Igl avalue of 2, and It! a value of O.

Three of the comparisons with high P(S I d) ratesinvolved I ~ I, a highly complex phoneme accordingto a markedness classification. The other morecomplex phonemes, 131 and Id 31, were involved inone comparison each when high P(S I d) rates wereobtained. Thus, most of the comparisons whichyielded high P(S I d) rates involved phonemes thatwere complex according to a markedness classifica­tion.

Difficulty or degree of complexity appears to havebeen a major factor in determining the size of thedifference between "different" and "same" RTs. Theeffect of difficulty on RTs in a same-different task hasbeen investigated by Bindra, Donderi, and Nishisato(1968) for discriminations of lengths of lines.Discriminability was varied by changing the similarityof the lengths of lines to be compared. For the low andmedium levels of difficulty, the "different" RT wasfaster than the "same" RT. For the high level ofdifficulty, however, this order was reversed, with afaster "same" than "different" RT. Overall latencies

REACTION TIMES FOR VCs 69

increased with the degree of similarity between lines,but the size of the "different" RT increased relative tothat of the "same" RT for a high level of difficulty. Asimilar effect was observed in the present experimentin that the "different" RT and (D - S) increased asP(S Id) increased.

Bamber (1969) has postulated a model ofsame-different decision making to explain hisobservation of faster observed than predicted "same"RTs. His model involves a rapid identity reporteroperating simultaneously with a serial, dimensionalprocessor. The rapid identity reporter processesstimuli as "gestalts" rather than on separatedimensions and can reach a decision of "same" fasterthan the serial processor. The decision-makingprocess is terminated by the rapid identity processoronly on the emission of a "same" response; otherwise,the serial process continues until either a "same" or a"different" decision is reached. Most "same"comparisons are assumed to be detected by the rapididentity reporter, thereby accounting for the observedfaster "same" decision time. The RT for a decisionmediated by the slower serial processor is an inversefunction of the number of dimensions on which thetwo stimuli differ.

This model accounts for the data shown inFigure 1. The RTs for incorrect decisions, given inTable 3, may give additional insight into how themodel can account for errors. Generally, an incorrectdecision of "same," (S Id), was faster than a correct"same" decision, (S I s). In terms of Bamber's model,a fast incorrect "same" response is the result of anincorrect decision by the rapid identity reporter. Anincorrect decision of "different," (D I s), generallytook longer than any correct or incorrect "same"response, indicating that incorrect "different"decisions were mediated by the serial processor.Incorrect "different" decisions tended to be fasterthan correct "different" responses. This findingsuggests that when the correct "same" response is nottriggered by the rapid identity reporter, then, if theserial dimensional process is not carried out to itscompletion (resulting ina correct decision and a long"same" RT), a premature decision of "different"based on false information is made. Thus, both typesof errors appeas to be the result of incompleteprocessing by either the rapid identity reporter or theserial processor.

Although Bamber's model accounts for the averageddata shown in Figure 1, it does not explain the largedifferences in RTs among the one-feature contrasts.This finding has been explained by a markednessclassification of phoneme complexity. When a highlymarked, that is. complex, phoneme is involved in aone-feature contrast, then both error rate and the timerequired for a "different" decision may increase. Asthe "difficulty" of a comparison increases, the size ofthe "different" RT with respect to the "same" RTincreases.

70 REED

REFERENCES

BAMBER. D. Reaction times and error rates for "same"­"different" judgments of multidimensional stimuli. Perception& Psychophysics. 1969. 6. 169-174.

BAMBER. D. Reaction times and error rates for judging nominalidentity of letter strings. Perception & Psychophysics. 1972,12. 321-326.

BILGER. R. C.. WANG. M. D .. & JESTEADT. W. Consonantconfusions in patients with hearing loss. Journal ofthe AcousticalSociety ofAmerica. 1972. 52. 113 (A).

BINDRA. D.. DONDER!. D. c.. & NISHI SATO. S. Decisionlatencies of "same" and "different" judgments. Perception &Psychophysics. 1968. 3. 121-130.

BINDRA. D., WILLIAMS. 1. A.• & WISE. J. S. Judgmentsof sameness and difference: Experiments on decision time.Science. 1965. 150.625-627.

BURROWS. D. Modality effects in retrieval of information fromshort term memory. Perception & Psychophysics. 1972. 11.365-372.

CAIRNS. H. S.. CAIRNS. C. E.. & WILLIAMS, F. Sometheoretical considerations of articulation substitution phenome­na. Language and Speech. 1974. 17. 160-173.

CHOMSKY. N.. & HALLE. M. The sound pattern of English.New York: Harper and Row, 1968.

COLE. R. A., & Scor'r, B. Distinctive feature control ofdecision time: Same-different judgments of simultaneously heardphonemes. Perception & Psychophysics, 1972. 12, 91-94.

GREEN. D. M.. & SWETS. J. A. Signal detection theory andPsychophysics. New York: Wiley. 1966.

HALLE. M. On the bases of phonology. In J. A. Fodor andJ. J. Katz (Eds.), The structure of language. Englewood Cliffs.N.J: Prentice-Hall. 1964.

KELLOGG. W. N. The time of judgment in psychometricmeasures. American Journal of Psychology, 1931. 43, 65-86.

K6l:UGER. L. E. Effect of bracketing lines on speed of "same"­"different" judgments of two adjacent letters. Journal ofExperimental Psychology, 1970. 84. 324-330.

McINISH. 1. R.. & TIKOFSKY. R. S. Distinctive features and

response latencies: A pilot study. Perception & Psychophysics.1969, 6. 267-268.

MILLER, G. A.. & NICELY, P. An analysis of perceptualconfusions among some English consonants. Journal of theAcoustical Society ofA merica, 1955. 27. 338-352.

NICKERSON. R. S. Response times for "same"-"different'·judgments. Perceptual and Motor Skills, 1965. 20, 15-18.

NICKERSON. R. S. 'Same'<'different' response times: A model and apreliminary test. Acta Psychologica, 1969, 30. 257-275.

NICKERSON. R. S. Same-different response times: A further testof a 'counter and clock' model. Acta Psychologica, 1971.35. 112-127.

NICKERSON. R. S. Auditory cod ability and the short-termretention of visual information. Journal of ExperimentalPsychology. 1972. 95.429-436.

Nrsmsxro. S.. & WISE. J. S. Relative probability. inter­stimulus interval. and speed of the same-different judgment.Psvchonomic Science. 1967. 7.59-60.

TVE~SKY. B. Pictorial and verbal encoding in a short-termmemory task. Perception & Psychophysics. 1969. 6. 225-233.

WANG. M. D.. & BILGER. R. C. Consonant confusions in noise:A study of perceptual features. Journal of the AcousticalSocietv ofAmerica, 1973. 54. 1248-1266.

WEINE~. ·F. F.. & SINGH, S. Multidimensional analysis ofchoice reaction time judgments on pairs of English fricatives.Journal of Experimental Psvchology, 1974, 102. 615-620.

WICKELGREN. W. A. Distinctive features and errors in short-termmemory for English consonants. Journal of the AcousticalSociety ofAmerica. 1966. 39, 388-398.

NOTE

I. The notational convention adopted by Green and Swets (1966)is used here: that is. lowercase letters represent a stimulus anduppercase letters. a response.

(Received for publication February 12. 1975;accepted April 14. 1975.)