Perception and immediate recall of normal and "compressed" auditory sequences

DORISAARONSON, NANCY MARKOWITZ, and HOLLISSHAPIRONew York University,New York, New York 10003

Perception and immediate recall ofnormal and "compressed" auditory sequences*

For a fixed presentation rate, the ratio of speech-to-pause time was varied in threeexperiments. Ss recalled seven-digit sequencesor monitored for item or order informationin addition to recall. Removing 33% of the speech and substituting pause time improvedrecall accuracy and monitoring reaction times. The data suggest that loss of orderinformation in recall may result from cumulative perceptual delays when adequate pausetime is unavailable.

The relationship between presentationrate and accuracy of immediate recall iscritical for any theory of short-termmemory. However, published experimentson this issue are in conflict (Posner, 1963;Aaronson, 1967). One difficulty with thoseexperiments using auditory material maybe that the duration of the items has beenconfounded with the rate at which they arepresented. For example, when presentationrate is increased naturally, by speakingfaster, the durations of both the words andthe intervals between them are decreased(Hutton, 1954). Also, when rate isincreased by "speech compression,"temporal segments are usually deleted in auniform pattern from the stimulusmaterial, from both the words and theintervals between words (Fairbanks,1957a, b, c; Foulke, 1969). Variations ineach of these temporal parameters mayhave different effects on the S's processingin an immediate recall task. Hence, it is oftheoretical importance to obtaininformation on their separate effects.

Several lines of research suggest that (atleast) two stages of perceptual processingare performed in series on the stimulusrepresentation to increase its refinement,abstraction, or permanence. These stagesmay be differentially affected by thestimulus duration and the pause timebetween stimuli. At an early stage, whichwe will call "sensing," the stimulusduration (within limits) might determinethe amount of stimulus information takeninto a perceptual storage. However, thesilent interval between stimuli might becritical for a higher level process, which wewill call "identification." Pastinvestigations are consistent with thisapproach. For example, Broadbent (1957,1958) has postulated a large-capacity rapid

*This study was supported in part by USPHSFellowship MH-14,589 and ARPA ContractSD-187 to Harvard University, and by USPHSGrant MH 16,496 to New York University. Wethank Lloyd Kaufman and Joseph Markowitz fortheir helpful comments.

"decay" sensory storage, the S-system,followed by a limited-capacity perceptualstorage, the P-system, which processes itsinput for later recall. Sternberg (1967)considered stages of perceptual processingranging from a "raw image," or "directcopy," of the physical stimulus features toan identification based on a name or verballabel. Pollack (1959) provided evidencethat auditory stimuli can be maintained asa "direct representation, much like that ofa sound recording," but a more durable"categorized" representation can also beformed.

To study the effects of temporalparameters on immediate recall, Yntemaet al (1964) and Aaronson (1967) variedpresentation rate by varyingonly the silentintervals between words while holdingword duration constant. Both studiesfound that increasing the silent intervalsincreased recall accuracy.I The presentstudies were conducted in order to assessfurther the effects of stimulus duration andpause time on perception and immediate

.recall of auditory sequences. In three serialrecall experiments, the presentation rate ofseven-digit sequences was held constant,while the ratio of stimulus duration topause time was varied by "com pressing"the speech and compensating withadditional pause time between words.

Within the above theoretical framework,what general trends might one expect forthe present experiments? For a low-levelsensory stage of processing, the duration ofthe physical stimulus should be important.Within limits, the longer the physicalstimulus is available for reference and themore information the stimulus contains,the more accurately and completely it canbe represented. According to Brown(1958), coding of stimulus redundancy isimportant for later recall, when decay timeand other items intervene, Hence, at anearly stage of processing, representations ofhigher quality, which yield higher recallaccuracy, may result from normal digitsequences, with their longer stimulusdurations, than from compressed digit

sequencespresented at the same. rate.On the other hand, for higher levels of

processing, such as identlfying2 orencoding item and order information, theamount of processing time availablebetween items may be critical. Forexample, the internal processing may notbe carried out efficiently while the externalauditory stimulus is present. Hence, abetter identification or encoding, leadingto higher recall accuracy, should resultfrom the longer silent intervals betweencompressed digits than from the shorterintervals between normal digits.

EXPERIMENT 1: SERIAL RECALLMethod

Stimulus materials. The method used torecord stimuli guarantees that a particularitem is acoustically the same, regardless ofits position in the sequence. (Naturalspeech often contains changes in stress andtemporal grouping that dependsystematically on serial position.) First, amaster tape was recorded, containing oneexample of each digit from 0 (pronounced"oh") to 9, spoken by a male voice. Thisoriginal master tape of 10 digits, equatedfor loudness, had a mean digit duration of225 msec. A second master tape of 10"compressed" digits was made by deletingseveral sections of about 17 msec(approximately uniformly) from eachoriginal digit and combining the remainderto form a continuous acoustic signal,usingGarvey's (l953) chop-slice method. In thisway about 33%of the speech was removed,resulting in a "compressed" master tapewith a mean digit duration of 150 msec.

With the use of two Ampex taperecorders and electronic timing and controlcircuitry, seven-digit sequences of thesenumbers were rerecorded, one digit at atime, from the master tapes to form thestimulus tapes. For both normal (N) andcompressed (C) seven-digit sequences, a3-digit/sec presentation rate was used, Forthe fixed 3/sec rate, there was more speechtime and less silent time for N than for esequences.

Sequences were determined by takingthe first seven digits from randompermutations of 0 to 9, excludingconsecutive forward or backward runs ofLength 3 or greater. Different sequenceswere used for Nand e tapes. Intelligibilitytests were performed on these digits bypresenting them, one at a time (S-paced),at 35 dB SPL, the intensity used in thethree experiments. Six Ss heard 10practice, 88 normal, and 88 compresseddigits in the order NC (three Ss) or eN(three Ss). Ss named the digits correctly on99.5%of both Nand e trials.

Procedure. The sequence of events on asingle trial was as follows. At the start of

338 Copyright 1971. Psychonomic Journals, Inc., Austin, Texas Perception & Psychophysics, 1971, Vol. 9 (4)

Fig. 2. A model of the temporal course of perceptual processing.

12

time

34

expected for sequences of normal digitswith their shorter pause times.

Further, if two assumptions are made,such accumulations could yield abow-shaped serial position curve of ordererrors (Fig. I C) whose degree ofasymmetry is determined by temporalparameters. (1) Perceptual processing isserial in nature; items are handled one at atime, and unprocessed items accumulate ina buffer storage while early items areprocessed. Broadbent (1957, 1958) andothers provide evidence supporting this.(2) The order of identifying (and recalling)items is not determined strictly by thepresentation order but, instead, by aprobability distribution over theaccumulated unidentified items in thebuffer.

Figure 2 illustrates how a bow-shapedcurve might be obtained. The top linerepresents stimulus items arriving at aconstant rate. The second line indicates thetime at which each of seven (possiblydegraded) items are taken from the buffer(not necessarily in their presented order,i.e., Assumption 2). The third line indicatesthe number of unidentified items in thebuffer at the time of each identification.As the stimulus sequence progresses, for aserial identifier that lags behind thepresentation, the buffer accumulationbuilds up. However, when the stimulussequence terminates, the buffer size beginsdecreasing. When items are selected foridentification, the chances of selectingthem in the wrong order would be higherwith larger buffer accumulations, as in themiddle of the series." If identificationoccurs primarily during the silent intervals,greater accumulations and 1110re ordererrors should result for normal than forcompressed digits.

The shape difference between the NandC curves of item errors (Fig. 1B) isconsistent with the hypothesis ofidentification delays. The presence of more

32

l-1st L-

2nd L-3rd L-

4th L.5th L

6th t7th

/:.__.-_.\.• 'C '\

' '., ., \. ~,,,,, 0,,

2 C.ORDER

STIMULUS

IDENTI­FICATION

BUFFERSIZE

DiscussionConsider the data in relation to the

two-stage processor discussed earlier. Ifstimulus duration for both normal andcompressed digits is about adequate for"sensing" physical features of the items,then item errors should be few and thereshould be little overall difference in itemerrors between conditions. However, ahigher level verbal "identification" maydetermine order information primarilyduring the silent intervals. If the necessaryidentification time exceeds the silentinterval, a backlog of items mayaccumulate and be identified in the wrongorder. Hence. more order errors would be

Although frequency of item errors didnot differ between conditions, the shape ofthe serial curves did. As an index of shape,least-squares lines for Positions I to 6(Fig. IB) were determined for each S. Thedifference in slopes between Nand Csequences was significant (p < .05,one-railed t test). Relatively more of theerrors occurred toward the end of thesequence for N than for C.

5 2

POSITION

5 B.ITEM

SERIAL

A. TOTAL

·N°"'O\! -_.-C..... 0,. -.. , \, \, \\\\\

Fig. l. Experiment 1. Serial position curves of total, item, and order errors in recall.

en0::.o0::0::W

ResultsEven though 33% of each item was

removed, immediate recall accuracy wassignificantly higher for C than for N. Theordinate of the serial curve in Fig. I A is thepercentage of stimuli recalled in error.vFor 13 of the 16 Ss, the percentage oferrors was smaller for the sequences ofshorter digits and longer pauses (p < .02,binomial test).

Recall errors were separated into itemand order errors. An item error was scoredfor a given stimulus position when thatitem was not recalled anywhere in theresponse sequence. An order error wasscored when a stimulus item was recalled,but not at its original stimulus position. Nooverall difference in item errors occurredbetween Nand C (Fig. IB). Rather, thedifference in accuracy can be attributed todifferences in order errors, as shown inFig. IC (p < .01, t test).

each trial a yellow "ready" light signaledSthat he could start by pressing a pedalwhen he was ready to attend. He thenheard a I,OOO-Hz tone in his earphones(Permoflux PDR.8), warning him that aseven-digit sequence would start in .67 sec.Four seconds after the last digit, a 500·Hztone signaled S to recall aloud the digits inthe presented order, making the best guesshe could when uncertain. If S recalledcorrectly and in the correct order, a greenlight was flashed; otherwise a red light wasflashed.

The experimental session was dividedinto four blocks of 32 trials each, adifferent seven-digitsequence on each trial.Half of the Ss received blocks in the orderNCCN and halfCNNC. At the beginningofthe session, Ss had 10 practice trials of thesame condition as their first block ofexperimental trials. Sixteen undergraduateSs were paid for their services, and the top25% received an extra monetary bonus asmotivation for high accuracy.

Perception & Psychophysics, 1971, Vol. 9 (4) 339

times) at each of the six possible stimuluspositions, and the left and right responsekeys were each designated to be correct on50% of the trials. Conditional on theserestrictions, the critical pairs were chosenrandomly from the seven-digit sequences.Ss received 12 practice trials of the samecondition as their first block ofexperimental trials. A new group of 24undergraduate Ss was paid for theirservices, and the top 25% received amonetary bonus.;2 3 4 5 6

SEQUENCE LENGTH

f ' . , . .( . .1; i J; 1 7SP OF MONITORED DIGIT

.7 A. /. (/)100

a:.6

• N ~~ 90

o Q.

w \/~._. Q(/) .5 W 80

~. ~._- .. >" . , a:

I- .4 " ,·-_··'C W 70

a: ' , (/)··-_e' UJ.3 It 60

Fig. 3. Experiment 2. Monitoring reaction times and recall accuracy. N =normal; C =compressed; SP =serial position.

•30C.ORDER D.ITEM 1\

7 I \

., ): 6

I \25 I \: X·20 I '. 5 elj \(/) f-a: I \

0 4 . \15 , .a: l·Na:

3w f/10 N .,~ I!e •

1 .//

o .......Ll'o ........,..2 3 4 5 6 7 2 3 4 5 6 7

SEQUENCE LENGTH SEQUENCE LENGTH

stimulus features for normal digitsproduces better "sensing" and item recallearly in the sequence. However,identification delays (and consequentdegradation of "memory traces") wouldaccumulate over the series more rapidly fornormal sequences with their short pausetimes, reducing recall accuracy toward theend. For both normal and compresseddigits, item errors increase over the series,except for the last position. Recall for thelast digit may improve partly because nosubsequent digits are available to delay itsidentification. Similarly shaped item errorcurves have been obtained in previousstudies (Aaronson, 1968; Ryan, 1967;Glanze r, 1966).

EXPERIMENT 2: MONITORING FORAN ORDERED PAIR PLUS RECALLTo test the hypothesis of perceptual lag,

it would be valuable to obtain furtherevidence about the time at which orderedidentification occurs. One possibility is toask the S to respond as soon as he perceivesthe order of a pair of items in thesequence. The greater the delay betweenthe presentation of the pair and the S'sperception of its order, the longer theresponse time (RT) should be. Therefore,we conducted Experiment 2, in which Sswere required to monitor the sequence foran ordered pair, as well as to recall in serialorder.

MethodThe same equipment and stimulus tapes

were used as in Experiment I. One secondafter S started each trial he was shown a"critical pair" of digits that wouldsu b sequentIy occur in the auditorysequence in adjacent positions. The pairappeared in one order on the left side ofthe visual display and in the reverse orderon the right, for example "36" and "63."(An element of the pair could not occuralone in the sequence.) The display lasted2 sec, and I sec elapsed before theI,OOO-Hz tone signaled the auditory digits.Ss were told to press a response key assoon as they heard the critical pair in theauditory sequence: the left-hand key if thepair was spoken in the order on the leftside of the visual display and the right-handkey otherwise. The RT from the onset- ofthe first digit of the critical pair wasrecorded. Four seconds after the sequenceterminated, Ss recalled aloud the items upto and including the critical pair, makingtheir best guesses when uncertain. At theend of each trial, Ss saw two sets offeedback lights: green lights for correctrecalls and correct keypresses, and redlights for errors.

The experimental session was dividedinto four blocks of 30 trials each, half theSs receiving the order NCCN and halfCNNC. In each block of 30 trials, thecritical pair occurred equally often (five

ResultsReaction times. Figure 3A shows that

for correct monitoring responses, RTs tocritical pairs were longer for N than for C(p < .01, F test). The graphed data are forthe first half-session. During the secondhalf, reaction times decreased by a greateramount for N than for C, reducing thedifferences (see section on practice effectsbelow). Figure 3A shows that RTdifferences increased with position of thecritical pair in the sequence. Least-squarelines were determined for each S, and theslope of RT vs serial position was steeperfor N than for C for 22 out of the 24 Ss(Serial Position by Item-durationinteraction: p < .01, F test).

Recall performance. Figures 3C and 3Dshow the percentages of order and itemerrors. The sequence length wasdetermined by the location of the criticaldigits. "Preserved pairs" (Fig. 3B) providesa second index of information about order:the percentage of recalled stimulus pairs(for positions up to and including thecritical pair) that occurred in successiveadjacent response positions. As expected,error rates increased with sequence lengthfor both groups (p < .01, F tests). Also,the data show that there were more itemand order errors, and fewer preserved pairsfor N than for C. Although these durationeffects did not quite reach statisticalsignificance, the interaction betweenduration and sequence length did (p < .01,F tests): the greatest duration effectsoccurred at intermediate sequence lengths.

In Experiment I, there were many moreorder than item errors (Figs. I B and IC)and the ratio of order-to-item errors wasgreater for N than for C (5.2:1 and 4.1:1,respectively). Similarly, in Experiment 2there were more order than item errors(Figs. 3C and 3D) and, again, the ratio oforder-to-item errors was greater for N thanfor C (5.3: I and 4.6: I, respectively).

DiscussionThe Ss in Experiment 2 were required to

respond as soon as they perceived the orderof the "critical pair" of items in eachsequence. Hence, we have considered RT asan index of perceptual delay; the greaterthe delay between the presentation of the

340 Perception & Psychophysics, 1971, Vol. 9 (4)

Fig. 4. Experiment 3. Monitoring reaction times and recall accuracy.

2 5 6

SEQUENCE LENGTH

3 6

SEQUENCE LENGTH

fcc,

(/) 100

a:~ 90

cUJ 80

>a:UJ 70(/)UJg: 60

7 D.ITEMi\

I \I \

I \

")(C!'. \I \. \I •

,'.N;/

./f./,'o •.--....__ .•

possible stimulus positions (i.e., 21 criticaltrials), and 9 trials were "catch trials."Conditional on these restrictions, thecritical digits were chosen randomly fromthe seven-digit sequences. A new group of26 undergraduate Ss was paid for theirservices, and the top 6 received a monetarybonus.Results

Reaction times. As shown in Fig. 4A,RTs for correct monitoring responses werelonger for N than for C (p < .0 I, F test),and this RT difference increased with serialposition (Duration by Position interaction:p < .05, F test). For both Nand C, RTsge ne rally increased over the series.However, for both groups, RTs arerelatively long at the first serial position .This has been found in earlier studies(Aaronson, 1968) and seems to indicatethat Ss are less prepared to respond at thestart of the series than at subsequentpositions.

Recall performance. Figures 4D, 4C, and4B show that item and order errorsincrease and preserved pairs decrease withsequence length for both Nand C (p < .01,F tests). There were slightly more ordererrors and fewer preserved pairs for N thanfor C. However, slightly more item errorsoccurred for C than for N. As previouslyfound, order errors exceeded item errors,and the ratio of order-to-item errors wasgreater for N than for C (5.1: I and 3.9: I,

20

25

SP OF MONITORED DIGIT

5

SEQUENCE LENGTH

( ,1; s

JO C.ORDER

10

(/)a:o 15a:a:UJ

MethodThis experiment was identical to

Experiment 2, with the followingexceptions. On each trial, Ss first saw asingle "critical digit" that would occur inthe auditory sequence with a probability of.70 (3CYfo of the trials were "catch trials"on which the critical digit was not spoken).Ss were told to press a YES key as soon asthey heard the critical digit in the auditorysequence. When they did not hear thecritical digit, Ss pressed a NO keyimmediately after the sequence ended.Four seconds after the last digit, a SOD-Hztone signaled S to recall aloud the items upto and including the critical digit, or theentire sequence on catch trials.

The session was divided into 12 practicetrials and four blocks of 30 experimentaltrials, half of the Ss receiving the orderNCCN and half CNNC. In each block of 30trials the critical digit occurred equallyoften (three times) at each of the seven

unidentified items accumulated inperceptual storage, the greater theopportunity to select them in the wrongorder for identification. To test thishypothesis, an experiment was conductedon the perception of item informationduring a recall task using the sameexperimental paradigm and stimulussequences used in the aboveorder-perception experiment.

critical pair in the auditory sequence andthe S's perception of its order, the longerthe response time should be. With thisassumption, the data of Fig. 3A provideevidence that in a serial recall task Ss lagbehind in their identification of orderinformation to a greater extent for normalthan for compressed digits. By the lastposition, the RT difference is 200 msec,even though the difference in durationbetween normal and compressed items isonly 75 msec. Hence, the increased RT fornormal items does not arise simply becauseSs wait until the digit of longer duration iscompleted before responding. Rather,inadequate processing time betweennormal digits causes perceptual delays toaccumulate over the series. Thesemonitoring data support the hypothesisthat differences in order errors for our firstexperiment were due in part to differencesin identification delays between normaland compressed sequences.

The data for order errors and preservedpairs are also consistent with those ofExperiment I. The available time betweenitems to identify them in proper order isless for N than for C; consequently moreorder information is lost. Also, as inExperiment I, there is more difficultymaintaining order than item information,and this difference is greatest for thenormal sequences. In Experiment 2, thedifference in shape for serial curves of itemerrors was again maintained. Averaged overall sequence lengths, the mean position ofitem errors was later in the series for Nthan it was in the series for C, SerialPositions 2.82 and 2.61, respectively. ForExperiment I, it was hypothesized thatthis shape difference between Nand CC'!P'~S resulted from a trade-off betweendigit redundancy (improving norn.c. Jl!;IlS

early in the series) and identification delays(degrading normal digits late in the series).In Experiment 2, even though the shapedifference was maintained, slightly moreitem errors occurred for normal digitsregardless of serial position. Perhaps theadditional requirement to monitor for acritical pair increased iden tification delays(even at early positions) enough to offsetthe advantages of the longer stimulusduration.

EXPERIMENT 3: MONITORING FORA SINGLE DIGIT PLUS RECALL

We hypothesized that the difficulties oforder perception in the above experimentswere not due to any special processing oforder information per se. Rather, ordererrors result from delays in theidentification of the items themselves.6

Our perception of speech lags slightlybehind the presentation. The greater thelag, and the greater the number of

Perception & Psychophysics, 1971, Vol. 9 (4) 341

respectively). Also, as in Experiments 1and 2, a greater percentage of the itemerrors occurred toward the end of theseries for N than for C. Averaged over allsequence lengths, the mean serial positionof item errors was 3.4 for Nand 3.2 for C.

Catch trials. For catch trials (when thecritical digit did not occur during theauditory sequence) recall trends weresimilar to those of Experiment 1 and to therecall data for "critical" trials ofExperiments 2 and 3. There were moreitem errors and order errors (p < .05,F test) and fewer preserved pairs for Nthan for C. Also, the ratio of order-to-itemerrors was greater for N than for C, 4.4: 1and 4.0: 1, respectively.

Practice effects. As in Experiment 2,RTs decreased with practice to a greaterextent for N than for C. RT differencesbetween the first and second half-sessionswere 173 msec for Nand 30 msec for C.Perhaps Ss decreased their identificationtime by decreasing the amount ofinformation they extracted, especially fornormal digits. Indeed, with practice, itemerrors and monitoring errors increasedslightly for N but decreased slightly for C.Also, the percent of preserved pairsincreased less with practice for N than forC.

DiscussionSs in Experiment 3 were required to

respond as soon as they perceived the"critical" digit in the auditory sequence.Their RT was taken as an index ofperceptual delay; the greater the delaybetween the occurrence of the critical digitin the sequence and its perception, thelonger the response time should be. Withthis assumption, the data of Fig. 4A areconsistent with the notion of alimited-capacity processor whoseidentification rate cannot keep up with thestimulus rate. The positive slopes of RT vsserial position suggest perceptual lags forboth Nand C. The steeper slope acrossserial positions for normal digits wouldresult from greater accumulations ofunidentified items. Again, as we foundearlier, such accumulations in a temporarybuffer should increase order errors anddecrease preserved pairs for normal digitson "critical" and "catch" trials.

In Experiments 2 and 3, we aimed tostudy the temporal course of perception oforder and item information during ashort-term memory task. Hence, correctserial recall was emphasized for thesemonitoring tasks. The similarities in recalltrends in all three experiments support thehypothesis that Ss were processing theitems in similar ways in all three tasks. Thesimilarity between recall data for "catch"trials of the monitoring task and recall-only

342

trials of Experiment I provides a controlcondition in this light.

Now compare the RT data, when Ssmust monitor for item or orderinformation. The absolute times are greaterin Fig. 3A than in Fig. 4A, as Ss inExperiment 2 responded after the seconddigit in each pair (anticipatory responsesfor both experiments are low and showlittle difference). Note that RTs, as afunction of serial position and digitduration, show similar trends when Ss mustidentify the items themselves (Fig. 4A) orthe order in which they occur (Fig. 3A).These similarities between Experiments Iand 2 are consistent with the hypothesisthat the listener uses the pauses betweenitems for identifying them in a particularorder. When this pause time is reduced, asfor the normal digit sequences, theresultant identification delays lead to ordererrors in recall. Aaronson (1968) drewsimilar conclusions from monitoring andrecall experiments that varied pause timebut kept speech duration constant.

Our experimental results are directlyrelevant to problems in using compressedspeech to enable blind people to receiveverbal information faster than the normalspeech rate. Fairbanks et al (1957a, b, c),Foulke (1969), and others have shown thatrecall and comprehension decrease whenspeech is compressed uniformly by deletingsegments from the words and from theintervals between words. The presentexperiments suggest that for maximumrecall and comprehension, for a fixedamount of compression, it may be optimalto delete more from the words than fromthe intervals between words. English is soredundant that much of the word can beeliminated without decreasingintelligibility, but the interword intervalsare needed for perceptual processing.

ALTERNATIVETHEORIESMasking

Forward and backward masking(Licklider, 1951) might increase perceptiontime for the more closely spaced normaldigits. Such intelligibility differencesshould also produce more item errors. ForExperiments I, 2, and 3, the differences initem errors between Nand C were,respectively, zero, slightly negative, andslightly positive, contradicting thisalternative.

Psychological Momentsand Prior EntryPsychological time may consist of

discrete "moments" of attention, eachabout 100 msec, during which we cannotsimultaneously attend to, or discriminateorder between, two events (Schmidt &Kristofferson, 1963; Stroud, 1955; Harter,1967). Hence. differences in order

perception between Nand C may result ifthe end of one digit and the start of thenext occur within the same "moment" forN (I08-msec pauses), but not for C(l80-msec pauses). This hypothesis isinsufficient, as it does not predict the serialposition effects for RTs and errors.

ExpectancyFor variable foreperiods, simple RT

varies inversely with time between thewarning and stimulus, perhaps due todecreased expectancy or readiness toperceive (Snodgrass, 1969). Analogously,the pauses may provide shorter foreperiodsfor N than for C. However, expectanciesregarding (1) the event time of each digitand (2) the occurrence of the particularcritical digit or pair should increase overthe series, predicting decreased RTs withserial position, in contradiction to ourdata.

Digit DurationShorter digits may require less

identification time, reducing RTs anderrors for C. First, this doesn't account fordifferences between C and N in shape ofthe serial curves of item errors and in ratioof order to item errors. Second, accordingto Brown (1958) and Crossman (1961) ashorter and less redundant code for Cshould lead to more item and order errors.To compare the digit-duration andpause-time theories, one might keep pausetime fixed and vary digit duration. Thedigit-duration theory then predicts betterperformance for shorter digits, andconsequently faster presentation rates. Wecould not locate such experiments forauditory stimuli, but sequential immediaterecall studies using visual items contradictthis prediction (Mackworth, 1962; Sitterly,1968; Haber & Nathanson, 1969).Attention

Because compressed speech differs fromour usual acoustic input, Ss might attendmore closely to C than to N, resulting infaster RTs and fewer order errors. One testof this hypothesis (as in the aboveparagraph) is to vary digit duration bycompression but to hold pause time fixed.Secondly, for an attention theory, onewould expect habituation and consequentdegradation of performance over thesession. On the contrary, performanceimproved with practice, and recall accuracyimproved more for C than for N.

In summary, theories based on masking,psychological moments, expectancy, digitduration, and attention can each accountfor some aspects of our data. Buthypotheses based on identification delaysmore completely account for the presentresults and for many related short-termmemory studies.

Perception & Psychophysics, 1971, Vol. 9 (4)

PARAMETRIC EFFECTSIn reviewing short-term memory studies

that vary the ratio of on-to-off time, wefound that none shows better performancefor longer on-time. The advantage forlonger off-time depends on other stimulusparameters: stimulus modality, numberand complexity of stimuli, andpresentation rate.

Thor and Spitz (1968) and Fraisse(1966) required Ss to report (and henceremember) which of two visual displaysoccurred first. Accuracy was determinedprimarily by total presentation time (on +off + on) and increased to loem by200 msec. However, Fraisse's graphs doindicate better performance with longer offtime for fixed values of total time, inagreement with our results.

In several immediate recall tasks, on-offratio was varied for visual stimuli. Fordisplays of one letter at a time, Haber andNathanson (1969) report that recallincreased with total time, regardless ofon-to-off ratio. However, their data matrixindicates that the proportion of cellsfavoring longer off-times increases withoverall presentation rate (35 to160 msec/letter) and with sequence length(4 to 8). Sitterly (1968) used a greaterrange of on- and off-times, each varyingfrom 100 to 1,000 msec per digit (4 to 14digits). Again, the critical variable was totaltime, although longer off-times had a slightadvantage in four out of six possible timecomparisons, and interactions betweensequence length, duration, and off intervalwere significant. For pairs of displays,containing one, three, or five letters each,Greenberg et al (1968) varied on + off + ontime from II 10 225 msec. They report astrong advantage of longer off-tline forfixed values of total timc, and thisadvantage increases slightly for longersequences.

Taub, Monty, and Laughery (1967),with visual displays, and Glucksberg andLaughery (1965), with auditory recordings.did parallel experiments in which Ssremembered the number of times Q, R, S,and T occurred in sequences of ~ to 20letters. Taub found longer off-times highlysuperior for overall rates of 2 and3 sec/letter, no difference for I sec/letter.and a reversal for 4 sec/letter. Glucksberg.using 2 sec/letter, obtained significantlyfewer errors for long than for shortoff-time. In both studies the advantage oflonger off-times increased significantlywith sequence length, suggesting thatinsufficient processing time was morecritical for longer sequences.

For nine-digit (.3 sec/digit) auditory orvisualsequences, Corballis ( 1966) graduallyincreased (I) or decreased (D) off-time overthe sequence from .2 to 1.6 sec. Recallaccuracy was slightly worse in the middle

of the sequence for I than for D butsignificantly better toward the end. This isconsistent with the hypothesis thatidentification delays may cumulate overthe sequence when insufficient processingtime is available.

In summary, experimental parametersinfluence the advantage of longer off-timein short-term memory tasks. (I) Auditorytasks (CorbaIlis, Glucksberg) are helped bylonger off-time, but this is not always thecase for visual stimuli: Thor found noadvantage for longer off time; Fraissc,Haber, and Sitterly found a slightadvantage; while Greenberg and Taubfound stronger effects. (2) Theimprovement with longer off-timegenerally increases with number of stimuli(Glucksberg, Greenberg, Taub , Sitterly)and with stimulus complexity. Thor foundno improvement using a singlesimple form(upright or inverted triangle), whereas taskshelped by off-time used verbal material.Greenberg found more improvement whenthe displayed letters formed words than forrandom letters. Indeed, the determiningfactor may be whether Ss code the materialverbally or simply rely on a sensory image.(3) The effect iveness of longer off-timeinteracts with rate (Haber, Taub]. Off-timemay be more effective at presentation ratesnearer the normal speech rate (Corballis,the slowest rates of Haber, intermediaterates of Taub) than for extremely fast(Thor) or slow (Taub) rates. The crrticalfactor may he the presentation rate relativeto the complexity of coding performed bySs. Our present experiments are consistentwith these trends: the stimuli are auditory.the sequences are moderately long, and thespeech rate IS close to normal.

REFERENCESAARONSON, D.. & STERNBER(;, S. Eflecls of

presentation rute and signul-to-no isc ratio oni m m cd ia t c recall. Eastern PsycholoaicalAssociation. Philadelphia. 1964.

AARONSON. D. Percept ion and unmcdiate rcrullof auditory sequences. Doctoral dissertation.University orPennsylvania, 1966.

AARONSON. D. Temporal f",'lors in perceptionand sh o r t-t c rm IIf1'mOr}'. PsychologicalBulletin. 1967.67. 130-144.

AARONSON, D. Temporal course of perceptionin an innncdi.ue recall task. Journal ufExperimental Psychology. 1961-:.76. 129-141l.

BERGSTROM. 1. A. Effects of changes in timevariables in ll1l'1nOriling. together with .\O!l1C

discussion or the techniques of memoryc x pc r i mc n t at io n. American Journal ofPsychology, 1907, t8, 206·231-:.

BERTELSON, P.. & TISSEYRE. L Apparentorder and st iniulu, unccr tuinty, PsychonomicScience, 1969. 15,65-66.

BROADBENT. D. E. A mechanical model forhuman at tent ion and immediate memory.Psychological Review, 1957.64,205-215.

BROADBENT. D. E. I'cr cep t ion andcommunication, New York: Pcruamon Press.1958.

BROWN. 1. lntor ma tion , redundancy. and decay

of the memory trace. Symposium on themechanization of thought processes. Vol. II.London: Her Majesty's Stationery Office,1958. Pl'. 729-752.

BUSCHKE, H., & LENNON, R. Perceptualidentification and encoding. Perception &Psychophysics, 1969,5,338-340.

CORBALLlS, M. C. Rehearsal and decay inimmediate recall of visually and aurallypresented items. Canadian Journal ofPsychology, t 966,20,43-51.

CROSSMAN, E. R. r. W. Information and serialorder in human immediate memory. In C.Cherry (Ed.), Information theory. London:Butterworths, 196 J.

CUMMING, G., & COLTHEART, M. Short-termmemory: Are item and position informationstored independently') Psychonornic Science,1969, 15,216-218.

FAIRBANKS. G., GUTTMAN, N., & MIRON, M.Effects of time compression upon thecomprehension of connected speech. Journalof Speech & Hearing Disorders, 1957a, 22,10·19.

FAIRBANKS, G., GUTTMAN, N., & MIRON, M.Auditory comprehension of repeated highspeed messages. Journal of Speech & HearingDisorders, 1957b, 22, 20-22.

FAIRBANKS. G., GUTTMAN, N., & MIRON, M,A u d itory comprehension in relation tolistening rate and selective verbal redundancy.Journal of Speech & Hearing Disorders, 1957c,22.23-32.

FOULKE, Eo, & STiCHT, T. G. Review ofresearch on the intelligibility andcomprehension of accelerated speech.Psychological Bulletin, 1969,72,50·62.

FRAISSE, P. Visual perceptive simultaneity andmasking of letters successively presented.Perception & Psychophysics, 1966, I,285-287.

GARVEY. W. D. The intelligibility of speededspeech. Journal of Experimental Psychology,t953,45,102-107.

GLANZER. M. Encoding in the perceptual(visual) serial position effect. Journal of VerbalLearning & Verbal Behavior, 1966,5,92-97.

GLUCKSBERG, S.. & LAUGHERY, K. R.Sequential memory as a function of total timeof intormation exposure and availability ofinforma tion-processing strategies. Proceedingsof the American Psychological Association,1965.79-1-:11.

GRLENBERG, M.. HELFER, M. S.. &MAYZNER, M. S. Information processing ofkiter and word pairs as a function of on andoil limes. Perception & Psychophysics, 1968.4,357-3611.

HAllER, R. N.. & NATHANSON, L. S.Processing or seq ucn tiaJly presen ted letters.Perception & Psychophysics, 1969. 5,359-361.

HARTER. M. R. Excitability cycles and corticalscanning: A review of two hypotheses ofcentral intermittency in perception.Psychological Bulletin, 1967, 68, 47-58.

IIUTTON. C. L. A psychophysical study ofspeech rate. Unpublished doctoral dissertation,University or Illinois, 1954.

LICKLIDER, J. C. R. Auditory stimulus. In S. S.Sleven; (l-d.), Handbook oj experimentalpsychologv, New York' Wiley. 1951.

MACKWORTII, 1. L Presentation rate andimmediate memory. Canadian Journal ofPsychology. 1962.11>,42-47.

MONTY, R. A.. FISHER, D. I .. & KARSH, R.Stimulus characteristics and spatial encoding insequential short-term memory. Journal ofPsychology. 1967.65, 109-116.

POLLACK. I. Mcssage uncertainty and messagereception. Journal of the Acoustical Society of

Perception & Psychophysics, 1971. VoL 9 (4) 343

America, 1959, 31,1500-1508.POSNER, M. I. Immediate memory in sequential

tasks. Psychological Bulletin, 1963, 60,333-349.

RYAN, J. Grouping and short-term memory.Unpublished doctoral dissertation, Universityof Cambridge, 1967.

SCHMIDT, M. W.. & KRISTOFFERSON, A. B.Discrimination of successiveness: A test of amodel of attention. Science, 1963, 139,112-113.

SITTERLY, T. E. Short-term retention ofsequentially presented digits as a function ofinterdigit interval, digit duration, and serieslength. Journal of Experimental Psychology,1968,78,174-178.

SNODGRASS, J. G. Foreperiod effects in simplereaction time: Anticipation or expectancy?Journal of Experimental PsychologyMonograph, 1969,79, H9.

STERNBERG, S. Two operations in characterrecognition: Some evidence from reaction timemeasurements. Perception & psychophysics,1967,2,45-53.

STROUD, J. M. The fine structure ofpsychological time. In H. Quastler (Ed.),Information theory in psychology. Glencoe,11l: The Free Press, 1955.

344

TAUB, H. A., MONTY, R. A., & LAUGHERY,K. R. Keeping track of sequential events:Effects of stimulus on-time and interstimulusoff-time. Perceptual & Motor Skills, 1967,24,159-166.

THOR, D. H., & SPITZ, H. H. Temporaldiscrimination as a function of totalpresentation time. Psycho nomic Science,1968,13,291-292.

YNTEMA, D. B.. & TRASK, F. P. Recall as asearch process. Journal of Verbal Learning &Verbal Behavior, 1963,2,65·74.

YNTEMA, D. B., WOZENCRAFT, F. T., &KLEM, 1. Immediate serial recall of digitspresented at very high rates. PsychonomicSociety, Niagara Falls, Ontario, Canada, 1964.

NOTES1. Similar trends were obtained for the visual

presentations of Bergstrom (1907) and of Monty.Fisher. and Karsh (I967).

2. Some psychologists would question theinclusion of identification as a perceptual ratherthan a mnemonic process. However, data to datedo not even justify a clear dichotomy between

thc two levels. We arc concerned here simplywith two stages of processing that increase thepermanence and usability of the internalrepresentation.

3. Ss were instructed to recall seven digits, andto guess if uncertain. Responses of an incorrectlength amounted to less than 1%of the data andwere omitted from SUbsequent analyses.

4. A strictly random selection from the bufferproduces a highly nonsymmetric curve with fartoo many errors at the end of the series.However, if selection probabilities are weighted,based on imperfect information about arrivalorder or "trace strength" of the items, bowedcurves can be generated (Aaronson, 1966).

5. Digits on the original master tape wereslightly clipped at the beginning and end toprovide sharp transients for temporal reliabilityin activating electronic components,

6. Other short-term memory studies indicatingnonindependence between item and orderinformation have been done by Bertelson andTisseyre (\ 969), Buschke and Lenon (\969),Cumming and Coltheart (\ 969), Aaronson(I968), Aaronson and Sternberg (1964).

(Accepted for publication October 17. /970.)

Perception & Psychophysics, 1971, Vol. 9 (4)