Word Length and the Structure of Short-Term Memory

Home , Decay theory, Memory span, Modality effect

JOURNAL OF VERBAL LEARNING AND VERBAL BEHAVIOR 14, 575-589 (1975)

ALAN D. BADDELEY Medical Research Council, Applied Psychology Unit

AND NElL THOMSON AND MARY BUCHANAN University of Stifling, Scotland

A number of experiments explored the hypothesis that immediate memory span is not constant, but varies with the length of the words to be recalled. Results showed: (1) Memory span is inversely related to word length across a wide range of materials; (2) When number of syllables and number of phonemes are held constant, words of short temporal duration are better recalled than words of long duration; (3) Span could be predicted on the basis of the number of words which the subject can read in approximately 2 sec; (4) When articulation is suppressed by requiring the subject to articulate an irrelevant sound, the word length effect disappears with visual presentation, but remains when presentation is auditory. The results are interpreted in terms of a phonemically-based store of limited temporal capacity, which may function as an output buffer for speech production, and as a supplement to a more central working memory system.

Miller (1956) has suggested that the items, where items are defined experimental capacity of short-term memory is constant units. Words represent one commonly ac- when measured in terms of number of cepted type of item, and in this case, the chunks, a chunk being a subjectively meaning- chunking hypothesis would predict that the ful unit. Because of the subjective definition capacity of short-term memory, as measured of a chunk, this hypothesis is essentially in words, should be constant regardless of the irrefutable unless an independent measure of size or duration of the words used. the nature of a chunk is available. Typically A number of studies testing this hypothesis this problem has been avoided by making the have used the recency effect in free recall as simplifying assumption that such experi- an estimate of short-term memory capacity. menter-defined units as words, digits, and Craik (1968) found no reliable effect of word letters constitute chunks to the subject. length on performance in the free recall of Hence, although Miller's hypothesis is not separate groups of words comprising one to refutable in the absence of an independent five syllables. This invariance held true measure of a chunk, it is meaningful to test a whether performance was measured in terms weaker version, namely that the capacity of of either raw scores, or estimates of primary short-term memory is a constant number of memory and secondary memory components. We are grateful to the Social Science Research This result was replicated and extended by Council and the Medical Research Council for Glanzer and Razel (1974) who observed a financial support. We thank Ronald Bidgood for recency effect which was constant when running Experiment VII, Rob Baker for phonetic measured in number of items, even when an advice, and Graham Hitch for many useful discussions. Dr. Baddeley'saddress is: Medical Research Council, item comprised a whole proverb rather than a Applied Psychology Unit, Cambridge, CB2 2EF, single word, They concluded from their England. study that short-term or primary memory Copyright © 1975 by Academic Press, Inc. 575 All rights of reproduction in any form reserved. Printed in Great Britain 576 BADDELEY, THOMSON AND BUCHANAN has a capacity of two items regardless of item rate providing an indirect measure of word duration or complexity. length. The situation is, however, complicated Miller's generalization, however, was based by the fact that subjects in some cases were on the memory span paradigm, and it is asked to label pictures, and in others to read questionable whether recency and span de- words so that it is not clear whether the result pend on the same memory mechanisms. is due to articulation time or to difficulty in There is indeed a growing body of evidence retrieving the correct verbal label. Watkins suggesting that the recency effect in free recall and Watkins (1973) present the clearest is basically unrelated to short-term memory published evidence for an effect of word as measured by memory span. Such evidence length on memory span, in a study primarily includes: concerned with the modality effect. They (1) Craik's (1970) observation that a found evidence for a word length effect on subject's memory span correlates more highly earlier serial positions, but observed that the with the secondary memory than the primary modality effect (the enhanced recall of memory component of free recall. auditorily presented items) did not interact (2) Memory span shows clear evidence of with word length. They suggest that the word speech coding, being impaired by both length effect observed may have been due to phonemic similarity (Conrad, 1964; Baddeley, the greater difficulty of perceiving their 1966) and articulatory suppression (Levy, four-syllable words which were presented at a 1971). This is not the case for the recency 1/sec rate. effect in free recall which is unaffected by These studies do not support the hypothesis either phonemic similarity (Craik & Levy, that memory span capacity is a constant 1970; Glanzer, Koppenaal, & Nelson, 1972) number of items. However, it is always or articulatory suppression (Richardson & possible to save the item-based hypothesis by Baddeley, 1975). questioning the assumption that words con- (3) Baddeley and Hitch (1974) have shown stitute items. Given evidence that short-term unimpaired recency in free recall for subjects memory is a speech-based system, it could be performing a concurrent memory span task reasonably argued that its capacity should be involving the retention of a sequence of six measured in more basic speech units such as digits. Since the memory span task did not syllables or phonemes. The experiments that interfere with recency, it is difficult to main- follow aim first to study the influence of tain the view that the two tasks are based on word-length on memory span, secondly to the same limited-capacity system. explore the relative importance of number of Studies investigating the effect of word syllables and temporal duration of a word as length on memory span do not in general determinants of span, and thirdly to explore support the weak version of Miller's hypo- the implications of this for the question of thesis. Thus, unpublished work by Laughery, whether the underlying memory system is Lachman, and Dansereau (Note 1) and by time-based or item-based. Standing, Bond, and Smith (Note 2) have reported poorer performance in a memory EXPERIMENT I span task when longer words are used. Mackworth (1963) found a high correlation This study compared the memory span of between reading rate and memory span for a subjects for sets of long and short words of wide range of materials, including pictures, comparable frequency of occurrence in letters, digits, shapes, and colors. This result English. One set comprised eight mono- could be interpreted in terms of word length syllables, namely, sum, hate, harm, wit, bond, as a determinant of memory span, with reading yield, worst, and twice. The other set corn- WORD LENGTHAND MEMORYSPAN 577 prised eight five-syllable words, namely ance at each sequence length for the long and association, opportunity, representative, organ- the short words. There is a very clear advan- ization, considerable, immediately, university, tage to the short word set which occurs at all and individual. sequence lengths and is characteristic of all eight subjects tested. Method Five list lengths were used, comprising [] ONE SYLLABLE [] FIVE SYLLABLES sequences of four, five, six, seven, and eigh.t I00 " words. Eight sequences of each length were o made up from the pool of short words, and ~v BO - eight from the pool of long words. In both cases, sequences were generated by sampling 60" 8 at random without replacement from the appropriate pool of words. All subjects were c° 40 - tested on both long and short words, and all received the sequences in ascending order of u~ 20- list length, beginning with sequences of four o words and proce~..ing up to the point at 4 5 6 7 B which they failed on all eight sequences, Words per sequence whereupon testing on the pool of words in FIo. 1. Effect of word length on memory span. question was discontinued. Half the subjects Mean percentage recall of long and short words as began with the pool of long words, and half a functionof sequence length. with the short words. The words were read to the subject at a There is little doubt that the sample of short 1.5-sec rate, with each list being preceded words used results in better memory span by the spoken warning "Ready." Subjects performance than the sample of long words. were allowed 15 sec to recall the words However, it is arguable that polysyllabic verbally in the order presented. Subjects words tend to be linguistically different were allowed to familiarize themselves with from monosyllables. In particular, our poly- the two pools of words at the beginning of syllables tended to be of Latin origin, com- the experiment, and these two pools remained pared to the monosyllables which seemed to visible to the subjects on prompt cards comprise simpler words of Anglo-Saxon throughout the experiment. Several different origin. Experiment II attempted to avoid this prompt cards with the words in differing problem by using words from a single cate- orders were used in this and subsequent gory, country names, a sample of material experiments so as to prevent the subjects unlikely to come from any single language from using location on the card as a cue. The source. subjects were eight undergraduate or post- graduate students from the University of Stirling. EXPERIMENT II Method Results and Discussion Sequences of five words were constructed Performance was scored in terms of number by sampling without replacement from each of sequences recalled completely correctly of two pools. The pool of short words com- (i.e., all the items correct and in the correct prised the country names Chad, Chile, order). Figure 1 shows the level of perform- Greece, Tonga, Kenya, Burma, .Cuba, Malta, 578 BADDELEY, THOMSONAND BUCHANAN while the long names were Somaliland, temporal duration of the words determines Afghanistan, Venezuela, Czechoslovakia, Yugo- the size of memory span. The latter possibility slavia, Ethiopia, Nicaragua, and Australia. might be predicted by decay theory (Broad- The names were selected on the basis of their bent, t958) which assumes that forgetting probable familiarity to the subjects, and occurs as a function of time. Many studies because they had a similar frequency of have attempted to test the theory by measuring repetition of initial and final letters within performance as a function of presentation the pool. Subjects were tested on a total of rate, and while some studies report enhanced eight sequences of five short names and performance with rapid presentation as eight sequences of five long names. Eight predicted by decay theory (Conrad & Hille, undergraduate subjects were tested using the 1958), others have found the opposite (Sper- same presentation procedure as Experiment L ling & Speelman, 1970). However, in none of these studies was the subject prevented Results and Discussion from rehearsing, and this makes interpretation Table 1 shows the mean number of sequen- of the results difficult as the subject is effectively ces recalled completely correctly, and the re-presenting the list to himself at a rate of mean number of items recalled in the appropri- his own choosing. This problem can be ate serial position, for long and short names. avoided by allowing the subject to rehearse On both these scores all eight subjects showed while using lists of long- and short-duration a clear word length effect. Since the material words. As less long words than short words in this study was very different at a linguistic can be rehearsed in a given period of time, level from the material used in the previous a word duration effect will be predicted by study, and since the effect is very large in both decay theory (Sperling, 1963). On the other cases, it is clear that the word length effect is a hand, a simple displacement or interference robust phenomenon of some generality. model would predict an effect of number of items, but not duration. Thus, the hypothesis that short-term memory capacity is a constant TABLE 1 number of items, where the syllable is the MEAN NUMBER OF SEQUENCES AND ITEMS CORRECTLY item, predicts no word length effect for words RECALLED AS A FUNCTION OF WORD LENGTH IN EXPERIMENT II matched for syllable number, but differing in spoken duration. Decay theory, on the Short names Long names other hand, predicts that the amount recalled Mean SD Mean SD will be a function of word duration. The next experiment tests these predictions. Sequences correct Max = 8 4.50 2.00 .88 1.27 Items correct EXPERIMENT III Max = 5 4.17 .71 2.80 .24 Method Two pools of disyllabic words, matched for However, in these and all previous experi- frequency, were produced such that one set ments investigating the effect of word length, tended to have a longer duration when two major variables are confounded, namely spoken normally. The long word set com- a word's spoken duration and the number of prised: Friday, coerce, humane, harpoon, syllables it contains. The results could nitrate, cyclone, morphine, tycoon, voodoo, therefore indicate either that memory span and zygote, and the short words were bishop~ is limited in the number of items it can hold, pectin, ember, wicket, wiggle, pewter, tipple, with the item being the syllable, or that the hackle, decor, and phallic. The words were WORD LENGTH AND MEMORY SPAN 579 recorded by a female experimenter onto magnetic tape, which was then played through % eO an oscillograph. This plots the wave-form Correct "', Recall 6o "o SHORT DURATION of the signal against time, allowing the dura- "'. ~o tion of the utterance to be measured. The 4 NG DURATION mean duration of the long words was 0.77 o~ ,, ~ ~ ~ ~ tT see, and of the short words, 0.46 sec. Serial Position From each pool of words, 10 lists of five FIG. 2. Mean recall of disyllabic words of long and words were constructed by sampling at ran- short temporal duration. dom without replacement. The twenty lists were divided into four blocks of five, two differs from the Watkins and Watkins study comprising lists of short duration words and in using words which are matched for number two of long duration words. A Latin square of syllables, but differ in spoken duration. design was then used to present the blocks in As such, the results are consistent with decay counterbalanced order to each of the 12 theory, and are inconsistent with the hypo- subjects. Words were read at a 2-sec rate, and thesis that short-term memory holds a subjects were required to recall verbally at constant number of syllables. the same rate, paced by a metronome. The last version of Miller's weakened Recall was paced so as to ensure that the mean hypothesis to be investigated is that short- delay between input and recall was compar- term memory holds a constant number of able for long and short words (Conrad & phonemes. In the last experiment, there was a Hille, 1958). Subjects were familiarized with clear tendency for the long words to have more the set of words and with the procedure, constituent phonemes, thus the result is and were instructed to commence recall as open to the interpretation that the word soon as the last item in each list had been length effect represents a limit to the number of presented. Twelve undergraduates from the phonemes that can be held. Experiment IV University of Stifling served as subjects. compares performance on sets of words which are matched for number of constituent Results and Discussion phonemes, but which differ in duration. Figure 2 shows the mean number of words Decay theory again predicts a difference correctly recalled as a function of serial in performance in favor of the short duration position. A three-way analysis of variance words. involving subjects, word length, and serial position showed significant effects of word length, F(1, 11) = 11.33, p < .01, serial posi- EXPERIMENT IV tion, F(4, 44) = 36.82, p < .001, and a signifi- Two sets of words were generated with the cant interaction between word length and following constraints: They differed in spoken serial position, F(4, 44) = 3.28, p < .05. Analy- duration; they were equal in number of sis by t test showed that the word length syllables; they were matched for word effect was significant for serial positions 1, 2, frequency; and they were equal in number of and 3, but not for positions 4 and 5. phonemes (with Scottish pronunciation). These results are very similar to those of Given all these constraints, the previous sets Watkins and Watkins (1973) showing a of words reduced from 10 to five; details are word length effect only for the earlier serial given in Table 2. Sequences of five words positions; this could reflect the masking were produced, and the experiment performed of an underlying word length effect by the using a procedure identical to that used in modality effect. However, the experiment Experiment III, except that the presentation 580 BADDELEY, THOMSON AND BUCHANAN and paced recall rate was increased to 1 sec same experimenter read out the words and it per word. Eight Scottish undergraduates is possible that some incidental feature of served as subjects. her mode of delivery produced the observed effect. To avoid this possibility, the experiment TABLE 2 was repeated using visual presentation at a DETAILS OF WORDS USED IN EXPERIMENT IV 2-sec rate and to our dismay, a statistically reliable word length effect was not observed. Number However, a closer examination of the data of Duration revealed that most subjects did show the Words Frequency phonemes (sec) predicted effect, but that two out of eight Long Coerce 1 5 .80 did substantially better on the long words. Harpoon 1 6 .75 On testing a further set of subjects and Friday 40 5 .70 asking them how they remembered the Cyclone 3 6 .88 Zygote -- 5 .90 material, it was found that those who did best on the short words reported using a Short Wicket 1 5 .50 rehearsal strategy, whilst those who did Pectin 1 6 .60 better on the long words reported rising an Bishop 40 5 .28 imagery strategy. Use of this latter strategy Pewter 3 6 .40 Phallic -- 5 .42 was facilitated by the fact that the presentation rate had been reduced to 2 sec/word in order to obviate perceptual difficulties. As the Results and Discussion subject of investigation is the articulatory Subjects recalled a mean of 61.6 ~ of the short-term memory system, it is reasonable to long words and 72.2 ~ of the short. A three- instruct subjects to use a rehearsal strategy way analysis of variance showed a significant in order to avoid this difficulty. The next effect of word length, F(1, 7) = 18.9, p < .01, experiment, then, is a replication of the and of serial position, F(4, 28)=38.06, previous one, but using visual presentation p < .001, but no interaction between serial with an instruction to the subjects to rehearse. position and word length, F(4, 28)= 0.55, p > .05. It is clear then that word duration may EXPERIMENT V influence span when the number of both The same material and design were used as in syllables and phonemes is held constant. the previous experiment, except that the The absence of an interaction between word material was presented visually on flash length and serial position is puzzling in view cards at a 2-sec rate, and recall was unpaced. of the previous result, it may be due to either The duration of the words was also measured the change in material, or more likely, the in a different way. The duration of a word is change in presentation and recall rate. determined by two sets of variables, the However, despite this minor discrepancy acoustic nature of the word and the subjects' between experiments III and IV, both seem articulatory rate. The latter variable has been to concur in suggesting that the temporal shown to be very stable over a wide range of duration of items is a powerful determinant conditions within a subject, but to vary of memory span. Before finally dismissing the considerably between subjects (Goldman- hypothesis that short-term memory capacity Eisler, 1961), and, as decay theory assumes is a constant number of items, a procedural rehearsal rate determines performance, the point that could have distorted the results subject's rather than the experimenter's should be mentioned. In both experiments the pronounciation of the words was used. WORD LENGTH AND MEMORY SPAN 581

Two different estimates of rehearsal rate theory in this context empirically investigated. were made. In the first of these, subjects were Let us assume that the memory system timed for reading the I0, five-word lists in underlying the word length effect exhibits each condition, as quickly as they could out trace decay, but that rehearsal may revive loud, the 50 words being typed out in two a decaying trace. It then follows that the columns. This was done four times for each amount recalled will be a function of rehearsal word length after the memory task, times rate. Thus, if it can be assumed that reading being recorded by stopwatch. The times so rate (RR) and articulation rate (AR) are good obtained were transformed into reading rate estimates of rehearsal rate, then it should be (RR) scores in units of words per second. possible to use them as predictors of memory The second estimate of rehearsal rate involved span. Table 3 shows the ratio of memory span requiring the subject to repeat continuously to reading rate and to articulation rate across three of the words from one of the pools out conditions. A Wilcoxon matched pairs test loud. Subjects did this as quickly as they showed that there was no effect of conditions could, and were timed by stopwatch for 10 for either the memory span-reading rate repetitions of the three words. For each ratio, T = 9, N = 8,p > .05, or for the memory condition, they did this four times, always span-articulation rate ratio, T= 10, N= 6, with a different set of three words, and always p > .05. In short, Table 3 indicates that a after the memory task. These times were subject can recall as many words as he can transformed into articulatory rate scores read in 1.6 sec, or can articulate in 1.3 sec. (AR) in units of words per second. Half The next experiment explored this relationship the subjects did the reading rate test first, in more detail using five different word lengths and half the articulatory rate test first. rather than two. The subjects, who were instructed to remember the lists by repeating the words to themselves, TABLE 3 were eight members of the Applied Psychology RATIO O17 MEMORY SCORE TO READING RATE AND Unit subject panel who were paid for their TO ARTICULATION RATE FOR SUBJECTS IN EXPERIMENT services. V.

Results and Discussion Memory score Memory score Subjects recalled a mean of 53.4 ~ of the reading rate articulation rate Subject KL" Ks b KL ~' Ks b long words correctly and in the right order, and 71.7 ~ of the short words. Analysis of 1 1.78 t.70 1.48 1.43 variance showed that there was a significant 2 1.72 1.42 .93 .93 effect of word length, F(1, 7) = 15.14, p < .01, 3 1.55 1.80 1.15 1.14 4 1.43 1.83 1.34 1.48 indicating that the word duration effect is 5 1.30 1.46 1.14 1.32 not dependent on auditory presentation. 6 1.68 1.95 1.40 1.79 There was again a significant effect of serial 7 1.38 1.59 1.24 1.00 position, F(4, 28)= 14.79, p < .001, but the 8 2.15 1.63 1.95 1.36 interaction between word length and serial Mean 1.62 1.67 1.33 1.31 position failed to reach significance, F(4, 28) = "KL = Constant for long words. 2.43, .05

TABLE 4 POOLS OF WORDS~ MATCHED FOR CONCEPTUAL CLASS; USED IN EXPERIMENT VI

Number of syllables 1 2 3 4 5

Stoat Puma Gorilla Rhinoceros Hippopotamus Mumps Measles Leprosy Diphtheria Tuberculosis School College Nursery Academy University Greece Peru Mexico Australia Yugoslavia Crewe Blackpool Exeter Wolverhampton Weston-Super-Mare Switch Kettle Radio Television Refrigerator Maths Physics Botany Biology Physiology Maine Utah Wyoming Alabama Louisiana Scroll Essay Bulletin Dictionary Periodical Zinc Carbon Calcium Uranium Aluminium

Each pool comprised one word from each of correctly. Their reading times were measured 10 semantic categories, the items being by stopwatch. Subjects read each list a total matched as closely as possible for familiarity of four times, twice before beginning the to the subjects. The sets differed in comprising memory task and twice after completing it. words of either one, two, three, four, or five Half the subjects began both tests by reading syllables, as may be seen in Table 4. the one-syllable list and proceeding up to From each pool, 10 lists of five words were the five-syllable list, while the remainder of produced by sampling at random without the subjects were tested in the reverse order. replacement. The 50 lists were then presented The subjects, who were tested individually, visually on video tape in completely random comprised 14 members of the Applied order; hence subjects were unaware on any Psychology Unit's panel who were paid for given trial what set would be used and so were their services. unlikely to use a different strategy for words of different length. Half the subjects received Lists 1-25 first, and half Lists 26-50 first. Results Words were written on cards and presented Figure 3 shows the effect of word length at a 2-sec rate by a card changer which was on mean percentage of words correctly viewed by a video camera and recorded. recalled in the appropriate position, and mean A card containing a row of asterisks served reading rate. as a warning that the list was about to appear. Memory scores. Analysis of variance showed Twelve seconds were allowed for spoken a significant effect of conditions, F(4, 52)= recall. 36.70, p < .001, and of subjects, F(13,52) = Reading rate was measured in this experi- 11.84, p <.001. A Newman-Keuls test ment by requiring the subjects to read lists between conditions showed that words of of 50 words comprising five occurrences of one or two syllables were better recalled than each item in a given set. The words were typed words of three or four, which in turn were in uppercase in random order in two columns better than five-syllable words (p < .05 in on a sheet of paper. Subjects were instructed each case). to read the lists aloud as quickly as they Reading rate. Analysis of variance showed could, consistent with pronouncing each word a significant effect of conditions, F(4, 52)= WORD LENGTH AND MEMORY SPAN 583

IOO 2'5 where S is the memory span, R is reading rate,

9C 2.3 and k and c are constants. One final question of interest is whether 8C 2"1 such a relationship holds across subjects as 7C I.g well as across word samples, or in other [email protected]. -'x, words as to whether fast readers are also "--" 6C 1.7 ~ Rate " \--',_",. tJ good memorizers. This proved to be the so 1.5 case; there was a substantial correlation i) 3 between memory span and reading rate, 40 I-3 "2 Cr r(13) = .685, p < .005. aO I-I Discussion 20 0.9 The results show that the manipulations IO Or/ were effective in producing sets of words of 0.5 different spoken duration, and that memory 5 No. of Syllables score for these words was well predicted by their duration. It has also been shown that FIG. 3. Mean reading rate and percentage correct recall of sequences of five words as a function of word fast readers tend to be good memorizers. length. The relationship between reading rate and memory span thus appears to be remarkably 244.02, p<.001. A Newman-Keuls test straightforward. Again the ratio of reading between conditions showed that each condi- rate to span is approximately constant, tion was significantly different from every indicating in the present study that subjects other one (p < .01 in each case). are able to remember as much as they can The next set of analyses tested the predic- read out in 1.8 sec. At this stage of researcb, tion made by decay theory, that the ratio however, it is probably imprudent to general- memory span to reading rate is constant ize this result too widely. There are many across conditions. Figure 4 shows memory variables which change memory span, but span plotted as a function of reading rate, which are unlikely to change reading rate the line being fitted by the method of least (e.g., list length, word meaningfulness, inter- squares. The slope of the line is 1.87, and the polated delay). It would be of interest to know intercept on the ordinate 0.17. The standard whether these variables have an effect on the error of the estimate is 0.10. The value of the slope, as predicted by decay theory, or on the intercept differs significantly from zero, intercept of the function. Only if the intercept t(3)=3.71, p < .05. Thus the results are stays consistently near to zero for a variety well described by the function S = e + kR, of conditions can the simple form of decay

IOC

u 8c n- 6c

2c L o'.s i:o i:s ~o 2.s Fleadin9 Rate in words/sec

FIG. 4. The relationship between reading rate and recall observed in Experiment VI. 584 BADDELEY, THOMSON AND BUCHANAN theory under discussion be accepted. The the blocks. All subjects did all conditions. main result of the experiment however, when The lists were presented at a 1.5-sec rate on seen in conjunction with the previous studies, a memory drum, and subjects were instructed is that short-term memory capacity, as to recall the items in the order presented. In measured by memory span, is constant when the suppression conditions, subjects counted measured in units of time, not in units of repeatedly from one to eight, keeping rate of structure. articulation as constant as possible at about The time-based system, which presumably three digits per second. They began counting underlies the effects observed, is broadly before the list appeared and stopped to consistent with a decay theory component recall as soon as the last item had been of short-term forgetting. Decay theory ascribes presented. In the no-suppression condition, to rehearsal the role of reviving a decaying subjects were simply told to try to remember trace, and it is this function of rehearsal that the words. The subjects, 12 undergraduate requires the prediction of a word length students from the University of Stirling, effect. It follows that if rehearsal could be were familiarized with the pools of words prevented then, providing the presentation before being tested. rate was the same for both long and short words, no word length effect should occur. The next experiment was designed to test this Results and Discussion prediction. The technique used to stop Figure 5 shows the mean percentage of rehearsal was that of articulatory suppression words recalled in the correct serial position (Murray, 1968) in which the subject is required as a function of word length for the two to articulate an irrelevant item during pre- presentation conditions. Analysis of variance sentation of the list. showed a significant effect of word length, F(1, 1i) = 17.73, p < .005, of suppression, F(1, 11)=67.89, p< .001, and a significant EXPERIMENT VII interaction between word length and sup- In this experiment, the recall of visually pression, F(1, 11) = 16.30, p < .005. The presented lists of long and short words was data may be summarized by saying that the compared under two conditions: (1) with the IOO subject remaining silent during list input and being free to rehearse, and (2) with the subject required to articulate an irrelevant 8c 6"- sequence of items. The design thus involved t Words four conditions, comprising two word lengths 60 in each of two presentation conditions. 0 u 4C Method O N Two pools of 10 words each were produced, 2c one of one-syllable words and one of five- syllable words, matched for word frequency. • 1 r Control Articulation From each pool, 16 five-word lists were I-2- 3...... 8 constructed by sampling at random without FIO. 5. Effect of articulatory suppression on the replacement. Each set of 16 lists was divided word length effect. The influence of articulatory into two equal blocks and a Latin square suppression on the recall of auditorily presented long used to determine order of presentation of and short words. WORD LENGTH AND MEMORY SPAN 585 word length effect disappears under suppres- to the subject, whilst in the visual condition, sion. Thus, these results are consistent with the lists were presented on a memory drum. decay theory if it can be assumed that sup- In all conditions, recall was verbal. In all pression stops rehearsal. Unfortunately, this other respects the procedure was as for the latter assumption is open to dispute, since the previous experiment; the subjects were 16 effects of suppression seem to be dependent members of the Applied Psychology Unit on presentation modality (Levy, 1971; Peter- panel who were paid for their services. son & Johnson, 1971). In particular, suppression has been shown to have a large effect Results and Discussion on visually presented material, but little The mean percentage of words correctly effect on auditorily presented material. It recalled in the appropriate serial position is would seem unlikely that suppression stops shown in Fig. 6. Analysis of variance showed rehearsal with visual presentation, but not significant effects of word length, F(1, 15) = with auditory. An alternative explanation 14.02, p < .005, suppression, F(1, 15) = 85.68, might be to assume that suppression stops p < .001, and modality, F(1, 15) = 39.66, the transformation of a visual stimulus into p<.001. The interaction between word a phonemic code. Thus, given that the word length and modality was significant, F(1, 15) -- length effect is mediated by a system em- 8.81,p < .01, as was the suppression x modal- ploying a speech code, and that under sup- ity interaction, F(1, 15) = 33.13, p < .001. pression, visually presented material does not The remaining two-way interaction, word enter this system, we have an alternative length x suppression, just failed to reach explanation for the above results. Experiment significance F(1, 15) = 4.49, .05

Specifically, the word length effect is abolished subjects' memory span for letter sequences by suppression in the visual modality, but is to vary with order of approximation to unchanged in the auditory modality. English. McNulty (1966) has shown that The results demonstrate very clearly how higher orders of approximation to English the effects of suppression change with pre- lead to higher memory span performance sentation modality and provide support for when measured in terms of number of letters the view that suppression stops the visual to recalled, but not when measured in terms of auditory transformation. These results can number of adopted chunks. It seems highly still be fitted into the simple decay and probable that sequences which can be reduced rehearsal hypothesis, but only if the assump- to a relatively small number of chunks tion is made that articulatory suppression (e.g., THEMILEAKE) will be not only well does not prevent rehearsal, but simply remembered, but also spoken much more inhibits the translation of visual material rapidly than sequences which cannot be into a phonemic code. reduced in this way (e.g., YVSPCWUECR). Such a view has, of course, been explored in some detail by Glanzer and Clark (1962) GENERAL DISCUSSION in connection with the recall of verbally The experiments described have shown: recodable visual patterns. Both letter sequen- (1) That memory span is sensitive to word ces and the types of pattern used by Glanzer length across a range of verbal materials. and Clark are encodable into articulatory (2) That when number of syllables and number sequences which can be produced within the of phonemes are held constant, the word 1- to 2-sec limit implied by our results. It length effect remains. (3) A systematic would dearly be desirable to explore the relationship between articulation time and relationship between articulation time and memory span, such that memory span is memory for materials such as approximations equivalent to the number of words which to English prose, which are broadly consistent can be read out in approximately 2 sec. with Miller's chunking hypothesis (Tulving (4) That memory span is correlated with & Patkau, 1962), but which would seem reading rate across subjects. (5) That articu- likely to involve articulation times consider- latory suppression abolishes the word length ably in excess of 2 sec. effect when material is presented visually. A second general point arises from the We shall discuss the implications of these contrast between the recency effect in free results for existing empirical generalizations recall, which apparently shows no word about memory, and will then attempt to length effect (Craik, 1968; Glanzer & Razel, fit them into a conceptual framework. 1974), and the memory span task, which The most obvious implication of these clearly does show such an effect. This fits results is for Miller's (1956) suggestion that in with the general pattern of results men- memory span is limited in terms of number tioned in the Introduction, suggesting that of chunks of information, rather than their memory span relies on phonemic coding, duration. It suggests a limit to the generality whereas recency does not. This provides of the phenomenon Which Miller discusses, further evidence for the suggestion that the but does not, of course, completely negate two reflect quite different underlying memory it. The question remains as to how much processes. Since most current views of the of the data subsumed under Miller's original nature of short-term memory assumed a generalization can be accounted for in terms common primary memory system underlying of temporal rather than structural limitations. both, it is unclear how they would interpret Consider, for example, the tendency for the word length effect. WORD LENGTH AND MEMORY SPAN 587

One approach which does not have this role of short-term memory (Shallice & drawback is the framework suggested by Warrington, 1970), can readily be accounted Baddeley and Hitch (1974), who explicitly for within this framework if it is assumed that postulate a working memory system which is such patients are defective in the operation responsible for performance on memory span of the articulatory rehearsal system, while tasks, but is not responsible for the recency having the executive component of the work- effect in free recall. The formulation is based ing memory system intact. on a range of experiments in which subjects One of our results however does present were required to perform reasoning, prose a problem for such a view. This is raised by comprehension, or free recall learning tasks the observation that the word-length effect while simultaneously holding sequences of up does occur despite articulatory suppression, to six random digits in short-term memory. provided the material is presented auditorally. In general, the results suggested that subjects Levy (1971) has shown a similar pattern of could hold up to three items with virtually no results for phonemic similarity, with the effect on performance, but when required to similarity effect disappearing under suppres- remember six items, a decrement appeared. sion when visual presentation is used, but A tentative formulation was suggested in not when the presentation is auditory. terms of a working memory system which On the straightforward assumption of an acts as a central executive, and a supple- articulatory rehearsal loop which is entirely mentary articulatory rehearsal loop with synonymous with subvocalization, it should a capacity of about three items. follow that suppression effects could not be Most of the experiments in the present avoided by auditory presentation. Experi- series fit neatly into this broad framework, ment VIII, however, suggests that although on the assumption that the word-length articulatory suppression produces an overall effect is the result of the limited capacity of impairment in performance with auditory the rehearsal loop. Looked at from this presentation, it does not influence the word- viewpoint, our data suggest that the articula- length effect. This would therefore seem to tory loop system is time-based, and hence has point to articulation as being a means of a temporally limited capacity. When access converting the visual stimulus into a phonemic to the loop is prevented by articulatory code which may be accepted by some form suppression, memory depends entirely on the of storage system. With auditory presentation, capacity of the executive working memory the material is presumably already encoded system, which is not phonemically based, and in an appropriate form, and can be fed into does not have the same temporal limitation the supplementary system without the need as the articulatory loop. The tendency for for articulation. The fact that articulatory memory span to be impaired by phonemic suppression still impairs performance, even similarity among the items to be remembered with auditory presentation, may imply either can also be attributed to the operation of the that there is an additional advantage to be articulatory rehearsal system. As in the case gained by articulation, or simply that the of the word-length effect, the phonemic task of suppression provides a secondary similarity effect disappears when articulatory task which takes up some of the general suppression occurs with visually presented processing capacity which might otherwise material (Levy, 1971). Finally, the existence be devoted to remembering the items pre- of patients who have drastically impaired sented. digit span, and yet who appear to show none Suppose one tentatively assumes a supple- of the general cognitive impairments that mentary phonemically based store, what might be anticipated from most views of the might its other characteristics be? Could it, 588 BADDELEY, THOMSON AND BUCHANAN for example, be equivalent to the precateg- data in a way which is both internally con- orical acoustic store suggested by Crowder sistent and also likely to generate testable and Morton (1969)? This seems unlikely for hypotheses. two reasons: First because the word length effect occurs with visual presentation, provided suppression is avoided, whereas the REFERENCES precategorical acoustic store does not appear to be operative unless auditory presentation BADDELEY, A. D. Short-term memory for word is used. Secondly, Watkins and Watkins sequences as a function of acoustic, semantic and formal similarity. Quarterly Journal of Experi- (1973) have presented evidence suggesting mental Psychology, 1966, 18, 362-365. that the precategorical acoustic store is not BADDELEY, A. D., • HITCH, G. Working memory. sensitive to the effect of word length; if this In G. A. Bower (Ed.), The psychology of learning is so, it can clearly not be used to explain and motivation. New York: Academic Press, the word-length effect. An alternative is to 1968, Vol. 8. BROADBENT, D. E. Perception and communication. suggest that the system is an output buffer London: Pergamon Press, 1958. of some type: A limited-capacity store for CONRAD, R. Acoustic confusion in immediate memory. holding the motor program necessary for the British Journal of Psychology, 1964, 55, 75-84. verbal production of letter names has been CONRAD, R., & HILLS, B. A. The decay theory of suggested by Sperling (1963). It seems immediate memory and paced recall. Canadian Journal of Psychology, 1958, 12, 1-6. plausible to assume that some form of buffer CRAIK, F. I. M. Two components in free recall. store is necessary for the smooth production of Journal of Verbal Learning and Verbal Behavior, speech; and indeed the existence of the eye- 1968, 7, 996-1004. voice span in reading points to some such CRAIK, F. I. M. The fate of primary memory items temporary storage process (Morton, 1964), in free recall. Journal of Verbal Learning and Verbal Behavior, 1970, 9, 143-148. since what the reader is saying when reading CRAIK, F. I. M., & LEVY, B. A. Semantic and acoustic aloud lags consistently behind the point at information in primary memory. Journal of which he is fixating. Such a buffer system ExperimentaIPsychology, 1970, 86, 77-82. would need to be separate from the act of CROWDER, R. G., & MORTON, J. Precategorical articulation, since it is presumably necessary acoustic storage (PAS). Perception & Psycho- physics, 1969, 5, 365-373. to set up new articulatory programs while GEANZER, M., & CLARK, W. H. Accuracy of per- existing programs are operating. On this ceptual recall: An analysis of organization. interpretation, therefore, articulatory pro- Journal of Verbal Learning and Verbal Behavior, grams can be set up or at least primed, either 1962, 1, 289-299. by the act of overt or covert articulation, or GLANZER, M., KOPPENAAL, L., & NELSON, R. Effects of relations between words on short-term storage indirectly through auditory stimulation. It is and long-term storage. Journal of Verbal Learning tentatively suggested that such a system may and Verbal Behavior, 1972, 8, 435-447. be ,necessary for fluent speech, and may have GLANZER, M., & RAZEL, M. The size of the unit in the supplementary advantage of providing short-term storage. Journal of Verbal Learning an additional backup system for the im- and Verbal Behavior, 1974, 13, 114-131. GOLDMAN-EISLER, F. The significance of changes in mediate retention of phonemically codable rate of articulation. Language and Speech, 1961, material. Such a view is clearly very tentative 4, 171-174. and leaves unspecified the complex problem LEVY, B. A. The role of articulation in auditory and of how such a store might be interfaced with visual short-term memory. Journal of Verbal the other components of the system so as to Learning and Verbal Behavior, 1971, 10, 123-132. account for even the basic phenomena of the MACKWORTH, J. F. The relation between the visual image and post-perceptual immediate memory. memory span. It does, however, have the Journal of Verbal Learning and Verbal Behavior, advantage of linking together the existing 1963, 2, 75-85. WORD LENGTH AND MEMORY SPAN 589

McNULTY, J. A. The measurement of "adopted psychological study. Quarterly Journal of Experi- chunks" in free recall learning. Psychonomic mentalPsychology, 1970, 22, 261-273. Science, 1966, 4, 71-72. SPERLIN~, G. A model for visual memory tasks. MILLER, G. A. The magical number seven, plus or Human Factors, 1963, 5, 19-31. minus two: Some limits to our capacity for SPERLING, G., • SPEELMAN, R. G. Acoustic similarity processing information. Psychological Review, and auditory short-term memory: Experiments 1956, 63, 81-97. and a model. In D. A. Norman (Ed.), Models MORTON, J. The effects of context upon speed of of human memory. New York: Academic Press, reading, eye movements and eye-voice span. 1970. Quarterly Journal of Experimental Psychology, TULVING, E., & PATKAU, J. E. Concurrent effects of 1964, 16, 340-354. contextual constraint and word frequency on MURRAY, D. J. Articulation and Acoustic confus- immediate recall and learning of verbal material. ability in short-term memory. Journal of Experi- Canadian Journal of Psychology, 1962, 16, mental Psychology, 1968, 78, 679-684. 83-95. MURRAY,D. J. Articulation and Acoustic confusability WATKINS, M. J., & WATKINS,O. C. The postcategorical in short-term memory. Journal of Experimental status of the modality effect in serial recall. Psychology, 1968, 78, 679-684. Journal of Experimental Psychology, 1973, 99, PETERSON, L. R., & JOHNSON, S. F. Some effects of 226-230. minimising articulation on short-term retention. Journal of Verbal Learning and Verbal Behavior, REFERENCE NOTES 1971, 10, 346-354. 1. LAUGHERY, K. R., LACHMAN, m., ~ DANSEREAU, RICHARDSON,J. T. E., & BADDELEY,A. D. The effect D. D. Short-term memory: Effects of item- of articulatory suppression in free recall. Journal pronunciation time. (Unpublished.) of Verbal Learning and Verbal Behavior, 1975, 14, 2. STANDING,L., BOND, B., & SMITH, P. The memory 623-629. span. (Unpublished.) SHALLICE, T. & WARRINGTON, E. K. Independent functioning of verbal memory stores: A neuro- (Received July 1, 1975)