Auditory memory and the irrelevant sound effect: Further evidence for changing-state disruption

Transcription

1 MEMORY, 2002, 10 (3), Auditory memory and the irrelevant sound effect: Further evidence for changing-state disruption Tom Campbell, C. Philip Beaman, and Dianne C. Berry University of Reading, UK Four experiments investigate the hypothesis that irrelevant sound interferes with serial recall of auditory items in the same fashion as with visually presented items. In Experiment 1 an acoustically changing sequence of 30 irrelevant utterances was more disruptive than 30 repetitions of the same utterance (the changing-state effect; Jones, Madden, & Miles, 1992) whether the to-be-remembered items were visually or auditorily presented. Experiment 2 showed that two different utterances spoken once (a heterogeneous compound suffix; LeCompte & Watkins, 1995) produced less disruption to serial recall than 15 repetitions of the same sequence. Disruption thus depends on the number of sounds in the irrelevant sequence. In Experiments 3a and 3b the number of different sounds, the token-set size (Tremblay & Jones, 1998), in an irrelevant sequence also influenced the magnitude of disruption in both irrelevant sound and compound suffix conditions. The results support the view that the disruption of memory for auditory items, like memory for visually presented items, is dependent on the number of different irrelevant sounds presented and the size of the set from which these sounds are taken. Theoretical implications are discussed. Irrelevant sounds played either during visual presentation of to-be-recalled material or during a post-presentation retention interval disrupt subsequent serial recall of the to-be-recalled list. Irrelevant speech sounds have been shown to disrupt recall of lists of visually presented items across the entire list, even when instructions stress that the speech is to be ignored (e.g., Banbury & Berry, 1998; Beaman & Jones, 1998; Colle & Welsh, 1976; Jones & Macken, 1993; Neath, Surprenant, & LeCompte, 1998; Salamé & Baddeley, 1982, 1989). In addition to its practical relevance to human factors research (Banbury & Berry, 1997, 1998; Banbury & Jones, 1998), the irrelevant sound effect (Beaman & Jones, 1997) has also provided a probe into how pre-attentive perceptual, attentional, and memory mechanisms interact to support cognitive performance. An influential account of the irrelevant sound effect has been advanced within Baddeley s (1986) working memory (WM) model. This account states that only speech-like irrelevant sound disrupts phonological storage in immediate memory (Salamé & Baddeley, 1989). A problem for this view, however, is the finding that a changing sequence of non-speech tones, which arguably sounded nothing like speech, significantly disrupted serial recall (Jones & Macken, 1993). However, a single, repeated speech item does not necessarily produce a significant disruptive effect (Jones & Macken, 1993; Jones et al., 1992; but see also LeCompte, 1995). Speech is thus neither necessary nor sufficient to disrupt recall, a point emphasised by Beaman and Jones (1997). According to the changing-state hypothesis (Jones et al., 1992) the critical factor is whether or not the Requests for reprints should be sent to Philip Beaman, Department of Psychology, University of Reading, Earley Gate, Whiteknights, Reading RG6 6AL, UK. c.p.beaman@reading.ac.uk Tom Campbell is now at the School of Psychology, University of Leeds. Thanks are due to Dylan Jones, Alastair Nicholls, and Sébastien Tremblay for productive discussions. Nigel Holt deserves particular gratitude for his assistance with headphone calibration. Tom Campbell received support for his work from the UK s Engineering and Physical Science Research Council. # 2002 Psychology Press Ltd DOI: /

2 200 CAMPBELL, BEAMAN, BERRY sound contains changing-state information. For disruption to occur, there must be segmentation between each physical unit within the sound stream, and each unit must be different from the one that preceded it. The sound must change in state. 1 The assumption is that changing sounds contain serial order cues that interfere with memory processes which would otherwise maintain to-be-recalled material in sequence. Studies of the effects of irrelevant speech sound have shown that disruption of serial recall of visual items by changing-state speech is readily observable (Jones, 1994; Jones et al., 1992). Unchanging, steady-state speech in which a single word or utterance is repeated produced only a small and non-significant disruptive effect in these studies. When this small steady-state disruption has reached significance then changing speech sound accrues additional disruption over steady-state material (LeCompte, 1995; Tremblay & Jones, 1998). The changing-state hypothesis constitutes a core assumption of the Object-Oriented Episodic Record (O-OER) model of short-term memory (Jones, 1993; Jones, Beaman, & Macken, 1996). Two seriation processes are assumed to operate in immediate memory: The first, deliberate, serial rehearsal process retains attended items in the correct order. The second, pre-attentive, seriation process retains the order of the auditory items when there is a mismatch between successive items. A separate assumption of the O-OER model is termed the unitary hypothesis. Simply put, this states that these two processes of seriation operate on a common unitary resource an episodic surface that stores the order of presentation of both irrelevant and attended items regardless of the modality of presentation. Access to this resource for visual material is dependent on deliberate rehearsal. Auditory material is registered on this surface automatically. This unitary hypothesis is shared with Baddeley s WM model (Baddeley, 1986) which also assumes a common (phonological) level of representation for visualverbal and auditory-verbal stimuli with automatic representation of auditory-verbal stimuli and 1 As pointed out by an anonymous reviewer, no independent index of changing-state exists and with speech stimuli there are rapid spectral and intensity changes within each unit. However, following Jones, Macken, and Murray (1993) we suppose that for our stimuli to qualify as changing-state the units within a sequence must be physically segmented by pauses in presentation. subvocal recoding of visual-verbal stimuli. The locus of irrelevant sound disruption differs between the two models, however. In the WM model there is assumed to be confusion between two sets of phonological representations registered in a common phonological store. In the O- OER model conflict of the two seriation processes on the episodic surface permits irrelevant auditory material to disrupt the order in which the attended visual items are rehearsed and, eventually, recalled. One further model that addresses irrelevant sound effects is the feature model (FM) of Neath (2000). Currently, although a large number of mathematical/computational models of immediate memory exist, only the FM directly addresses the irrelevant sound effect (although others claim to be capable of simulating the effect in principle see Burgess & Hitch, 1999; Norris, Page, & Baddeley, 1995). Neath s (2000) FM is an extension of Nairne s original (1990) feature model (see also Neath & Nairne, 1995). Within this model irrelevant sound reduces levels of attention when encoding items and this reduction is greater with changing-state irrelevant sound than with steadystate irrelevant sound. If the irrelevant sound is speech then there is an additional effect. Speech is incorporated into the modality-independent (presumably phonological) features of the verbal to-be-recalled material at encoding, rendering all the list items more similar to one another and thus increasing the risk of recalling the items in the wrong serial order (Neath, 2000; for commentaries see Baddeley, 2000; Jones & Tremblay, 2000). These models are based on data from numerous studies examining the effects of changing-state irrelevant sound on visually presented to-berecalled material, whether orthographic (Jones & Macken, 1993; LeCompte, 1995), lip-read (Jones, 1994), or visuo-spatial (Jones, Farrand, Stuart, & Morris, 1995). Fewer studies have examined the effects of irrelevant sounds when the to-be-recalled and irrelevant material are both presented in the same modality, presumably because extant theories are concerned with interference at an amodal level. This scarcity of studies is surprising as one would expect a same-modality interference effect to be well-documented, as for example with the stimulus suffix effect. Same-modality interference is often assumed in dual-task situations (e.g., Wickens, 1992) so one might actually expect to see a larger irrelevant sound effect with auditory presentation of to-be-recalled material than with visual presentation. In fact, those few studies

3 AUDITORY CHANGING-STATE EFFECTS 201 that have examined the irrelevant sound effect within the auditory modality have presented the to-be-recalled and irrelevant material concurrently (Hanley & Broadbent, 1987; Surprenant, LeCompte, & Neath, 2000; Surprenant, Neath, & LeCompte, 1999). Concurrent presentation unfortunately leaves upon the possibility that the effects observed were either perceptual/discrimination or encoding difficulties rather than memory effects per se. For example, Martin-Loeches, Schweinberger, and Sommer (1997) found that with acoustic materials recorded at a sampling rate of 20 KHz, dichotic presentation of the to-be-recalled and irrelevant material rendered the to-be-recalled information unidentifiable when presentation was synchronous. Hanley and Broadbent (1987, Exp. 2) went some way to addressing this concern by ensuring that perceptual identification of the to-be-recalled items under conditions of irrelevant sound was near perfect before administering the memory test. However it is at least plausible that under conditions of existing memory-load, wherein attention is directed at maintaining those items already encoded, the attentional capacity for accurate registration of new items will be reduced (e.g., Lavie, 2000). Thus, the existence of an irrelevant sound effect within the auditory modality taking the same form as in the visual modality is open to some doubt. It remains questionable whether the effect is of the same magnitude and obeys the same laws as the more familiar irrelevant sound effect within the visual modality. Two studies have looked directly at whether a changing-state effect occurs with concurrent presentation of to-be-recalled and irrelevant auditory material. Experiment 3 of LeCompte (1996) directly examined irrelevant sound effects in the visual and auditory modalities and found no differences between the two in the appearance of the effect. However, LeCompte also reported equivalent disruptive effects of steady-state and changing-state material with auditory presentation (overall proportion correct in both cases was.54 compared with the control level performance of.60 correct). With visual presentation, however, the standard changing-state effect occurred and changing-state irrelevant material proved more disruptive than steady-state irrelevant material. In contrast, a study by Hamilton and Hockey (1974, Exp. 2) provides some support for a changingstate disruption of memory for auditory items. When auditory to-be-recalled digits were interleaved with a changing-state sequence of irrelevant letters, a larger disruption was found than when the interleaved irrelevant material was a repeated irrelevant letter. However, in Hamilton and Hockey s interleaved presentation procedure the closer temporal proximity of the to-be-recalled items to the irrelevant items also resulted in stronger disruption of performance (Hamilton & Hockey, 1974, Exp. 3) and use of interleaved digits produced greater disruption of recall of to-berecalled digits (Hamilton & Hockey, 1974, Exp. 2). Both of these findings are contrary to results obtained from related investigations of the irrelevant sound effect (Buchner, Irmen, & Erdfelder, 1996; Macken, Mosdell, & Jones, 1999). Arguably, the findings of Hamilton and Hockey s study, or that of LeCompte (1996), which also used interleaved presentation of irrelevant and to-be-recalled material, could have arisen from a perceptual masking phenomenon or a disruption of encoding, rather than of memory. Similar criticisms of other studies hold, as the precautions taken by Hanley and Broadbent (1987) in their study (outlined earlier) were not carried out. For example, perceptual disruption might also have occurred in the studies of Surprenant et al. (1999, 2000) because irrelevant and to-be-recalled items were presented to the same ear with no overt control over synchrony of presentation. Thus, although an irrelevant sound effect with auditory presentation of to-be-recalled items is well established (Hamilton & Hockey, 1974; Hanley & Broadbent, 1987; LeCompte, 1996; Surprenant et al., 1999, 2000) there is little evidence that the effect is a memory phenomenon. There is even less evidence to decide whether the effect is of the same type as that viewed with visual presentation. If the two effects are indeed the same, one might expect the predictions of the changing-state hypothesis to hold, suggesting greater disruption will be observed with changingthan steady-state irrelevant sound. Arguably, the WM model might predict the same result, as (presumably) the greater the number of different items entering the phonological store, the greater the potential for disruption. The FM, however, contains no obvious mechanism whereby post-list irrelevant material might interfere with representations in memory in this way. In the model either interference occurs at an amodal level of representation during encoding (Neath, 2000) or it occurs through overwriting of the immediately preceding item at both modality-specific and amodal levels of representation (Nairne, 1990). Thus, the FM can account for effects of irrelevant

4 202 CAMPBELL, BEAMAN, BERRY sound at encoding (concurrent effects) but cannot account for effects of irrelevant sound in memory (retrospective effects; Miles, Jones, & Madden, 1991; Norris, Page, & Baddeley, 2001) beyond the operation of a single item stimulus suffix. Although the evidence so far suggests a retrospective changing-state disruption of memory for auditory items, it is not conclusive. If no such changing-state effect is observed with auditory presentation of to-be-recalled items then the unitary assumption of the O-OER model, and possibly also the WM model, is seriously challenged. Experiment 1 provides a test of the unitary and changing-state hypotheses in a comparison of the relative disruptive effects of changing-state and steady-state irrelevant sound in both sensory modalities. The remaining experiments will then consider whether disruption by irrelevant material to recall of auditorily presented material is mediated by the number of utterances presented. The effects of the absolute number of irrelevant utterances presented will be tested (Experiment 2) and the number of different irrelevant utterances presented under conditions where the absolute number of irrelevant utterances is low (Experiment 3a) and high (Experiment 3b) will also be examined. EXPERIMENT 1 In Experiment 1, participants were given two different serial recall tasks. On one task, participants heard a list of items, and on the other they saw a list of items presented silently on a computer screen. The irrelevant sequence that followed was either a steady-state sequence a repeated syllable or a changing-sequence sequence where each utterance differed from that which preceded it. The unitary hypothesis of both the WM and O- OER models predicts that the same pattern of disruption would occur, regardless of the modality of the to-be-remembered items. The changingstyle hypothesis also predicts that this pattern of disruption would show a greater disruption with a changing-state than with a steady-state irrelevant sequence. Method Participants. A total of 24 students and staff from Reading University participated in exchange for course credit or a small honorarium. All reported intact hearing and normal or corrected to normal vision. All had English as their first language. Materials and design. A 3 (irrelevant speech: quiet, steady-state, changing-state) 6 9 (serial positions) 6 2 (modality: auditory, visual) repeated-measures design was used, with error probability as the dependent measure. In the quiet control condition, serial recall was performed under conditions of 52 db (+.5 SPL) of equipment hum, measured in situ, prior to attenuation by the cup of the headphone. Each list of to-be-recalled material comprised the digits 1 9 chosen randomly and without replacement. Easy-to-remember sequences were removed. There were 30 trials in each modality, presentation of which was blocked and counterbalanced. On visual trials, digits were presented in the centre of the screen. All visual materials were presented in white Times New Roman 72- point font on a black background. Each digit was presented for 800 ms with an inter-item interval of 200 ms when the screen was black. On auditory trials, presentation was synchronised with the presentation of a fixation cross (+) in the centre of the screen. Each heard digit lasted 500 ms followed by an inter-item silence of 500 ms ensuring that the total presentation time was equivalent across the two modalities. To-be-recalled auditory digits were spoken in a female voice, to-beignored irrelevant items were spoken in a male voice. All speech sound was recorded digitally to 16-bit resolution, with a sampling rate of 44 khz, using Goldwave 3.03 software. Use of Visual Basic 4.0 with Mabry Software Hi-Time and Wave Custom Controls upon a Pentium II processor gave millisecond accuracy in timing. Auditory material was presented as a sound level of 70 db (+ 0.5) SPL as calibrated with a 1 KHz pure tone, measured with KEMAR (Knowles electroacoustic mannequin for acoustical research) and a Bruel and Kjœr type 2425 microvolt meter. All sound was presented via C110- headphones. Before each list a row of five dots appeared, equidistant from one another, displaced just to the right of the centre of a VDU screen. These dots disappeared one-by-one, from left to right, at a rate of one dot per second. A warning tone, 3 seconds before the list, accompanied the disappearance of the third dot during this visual countdown. The screen was blank for 1 second following the disappearance of the last dot. A list of nine digits followed, pre-

5 AUDITORY CHANGING-STATE EFFECTS 203 sented one at a time. A 15-second retention interval followed list presentation, during which the word WAIT appeared in the centre of the screen. The first irrelevant sound coincided with the start of the retention interval, cued by the appearance of the word WAIT, the disappearance of which coincided with the offset of the final irrelevant sound, and 500 ms of silence followed this retention interval, when the screen was blank. The word RECALL appeared for 10 seconds, followed by the countdown that preceded the next list. The order of irrelevant speech conditions was randomised such that there were 10 lists followed by each irrelevant speech condition in each block. For changing-state trials, the changing sequence consisted of 30 sounds drawn randomly from the letters B (bee), I (eye), J (jay), N (enn), and Z (zed), with the constraint that each letter appeared at least once. Steady-state speech consisted of 30 repetitions of the same letter. The steady-state letter used was randomly chosen, such that each letter was used twice on each block. Each irrelevant letter lasted 300 ms followed by an inter-item silence of 50 ms. Procedure. Participants were tested in groups of up to five at a time. Participants were asked to attend to the digits and then silently rehearse them while the word WAIT was visible. They were asked to ignore any sound they might hear while the word WAIT was present on the screen. When the word RECALL appeared they were asked to write down the digits presented in a strictly left to right fashion, without correction, while attempting to preserve the correct serial order and position of those items, leaving a I when uncertain about an item. Participants initiated the first trial on each block with the click of a mouse; subsequent lists were then presented at regular intervals. Results The pattern of mean errors was consistent with the changing-style hypothesis. Whether the to-beremembered items were either heard or visually presented, a changing-state pattern of disruption occurred, as illustrated in the right hand panels of Figure 1. A three-way repeated-measure s analysis of variance (ANOVA) revealed this irrelevant sound effect to be significant: F(2, 46) = 32.33, MSE = 0.08, p <.001. The irrelevant sound effect did not interact significantly with the presentation modality of the to-be-recalled items, F(2, 46) = 0.24, MSE = 0.04, p =.79 consistent with the hypothesis that there is no overall modality difference in the way irrelevant speech disrupts serial recall. Analysis of error probabilities produced the expected changing-state effect. Steady-state irrelevant speech was significantly disruptive relative to quiet, F(1, 46) = 12.73, p <.05, while the changing-state disruptive advantage over steady-state material was also significant: F(1, 46) = 4.20, p <.05. The evidence thus stands as support for the changing-state and unitary hypotheses. Overall, changing-state irrelevant speech produced a significantly greater disruption than steady-state irrelevant speech. Although the main effect of presentation modality was not significant, F(1, 23) = 23.75, MSE = 0.11, p =.07, serial position effects were significant, F(8, 184) = 61.32, MSE = 0.09, p <.001, as was the modality by serial position interaction F(8, 184) = 16.59, MSE = 0.03, p <.001. The irrelevant sound effect also interacted significantly with serial position, F(16, 272) = 1.99, MSE = 0.02, p =.005, with larger irrelevant sound effects apparent towards the end of the list. Importantly, the three-way interaction between presentation modality, irrelevant speech condition, and serial position was unreliable, F(16, 368) = 1.45, MSE = 0.01, p =.12. It thus appears that the modality differences in the shape of the serial position curve operated in relative independence to the irrelevant sound effect. In other words, the final-item advantage for auditory presentation remained despite the presence of speech in a retention interval that might be expected to eliminate that advantage by acting as a post-list suffix (Crowder & Morton, 1969). This is consistent with data recently reported by Surprenant et al. (2000). A separate modality by irrelevant speech condition repeated-measure s ANOVA of error probability data from the final serial position corroborated this interpretation. There was a significant modality-specific recency effect such that performance was more accurate on the final serial position with auditory than with visual items, F(1, 23) = 25.48, MSE = 0.09, p <.001. However, this modality effect did not interact significantly with irrelevant speech condition, F(2, 46) = 1.13, MSE = 0.02, p =.34. Thus, regardless of the modality of presentation of the to-be-recalled material, disruption was greatest with changing-state irrelevant material.

6 204 CAMPBELL, BEAMAN, BERRY Figure 1. Experiment 1: Error-position functions (left-hand panels) and overall mean error probabilities collapsed across serial positions (right-hand panels) for each irrelevant speech condition when the to-be-remembered items were auditorily (upper panels) or visually presented (lower panels). Error bars represent standard error. Discussion The results demonstrate a changing-state disruption of memory for both visually and auditorily presented lists. This changing-state disruption showed no overall modality differences between auditory and visual presentation, consistent with unitary and changing-state hypotheses. Arguably, common mechanisms were responsible for the changing-state disruption of memory for visually and auditorily presented items. Some difference between the current experiment and that of LeCompte (1996, Exp. 3) must, however, have been pivotal to the demonstration of a changingstate disruption that did not occur in LeCompte s study. The most obvious methodological difference in this and later experiments is that the irrelevant material was presented after rather than during the presentation of the to-be-recalled items. There may be some form of disruption that

7 AUDITORY CHANGING-STATE EFFECTS 205 occurred at encoding in LeCompte s auditory task that obscured changing-state disruption of memory. One candidate form of encoding interference is the effort of keeping the irrelevant separate from the relevant items. Whatever the underlying reason for the past failure to observe a changing-state effect, Experiment 1 firmly establishes for the first time the importance of change in the irrelevant sound stream as a primary determinant of disruption across both visual and auditory modalities, a prediction derived from theories of the irrelevant sound effect. This is contrary to the more usual assumption of within-modality interference (e.g., Nairne, 1990; Wickens, 1992). A more commonly employed heuristic is that interference occurs primarily within sensory modalities. If the degree of overlap between irrelevant and to-be-recalled material were important, for example, one might expect a greater irrelevant sound effect with auditory than with visual presentations, as the irrelevant sound would interfere with modalityspecific features of the attended material as well as modality-independent features. However, Experiment 1 failed to show any interaction between irrelevant sound and modality of presentation supporting the assumption latterly and counterintuitively made, that irrelevant speech disrupts only modality-independent codes (Neath, 2000). It is possible, however, that a retrospective effect of irrelevant sound might occur with auditorily presented stimuli if the irrelevant sound items are perceptually grouped with the to-berecalled list. When an additional irrelevant item disrupts serial recall of the final item with auditorily presented to-be-recalled items, that disruption is termed a stimulus suffix effect (e.g., Morton, Crowder, & Prussin, 1971). The disruption is sometimes considered to result from grouping the suffix as part of an auditory-sensory code that is peculiarly sensitive to end-of-list positional codes (Frick, 1988). If the same suffix is presented three times at the end of a list, attenuation of the disruption has been observed (Crowder, 1978; Morton, 1976). This attenuation has been attributed to a tendency to group a repeated suffix away from the to-be-recalled list. In contrast, when a sequence of two different suffix items is appended to the list, it produces more disruption than either a repeated suffix or a single suffix (LeCompte & Watkins, 1995, Exps. 1 4). This disruption has been termed a heterogeneous compound suffix effect and it has been argued that the heterogeneous compound suffix effect cannot easily be accounted for by existing, modality-specific, explanations of the more traditional single-suffix effect (see LeCompte & Watkins, 1995, for details). In fact, there are ways in which accounts of the single-item suffix effect could be expanded to include the heterogeneous compound suffix effect. For example, in the feature model recall proceeds by comparing the representations of the memory items to a search set and finding the best match between a degraded memory representation and a pool of possible responses (the list items). The size of this search set is increased by the inclusion of the suffix (cf., Beaman & Morton, 2000) producing an effect akin to an increase in list-length. One could therefore account for the heterogeneous compound suffix effect by supposing that both suffixes were added to the pool of possible responses. However, the heterogeneous compound suffix effect also bears a strong resemblance to the changing-state effect described in Experiment 1. The compound suffix effect may then alternatively be viewed as an instance of the more general rules governing disruption by irrelevant sound. One major difference between the changingstate effect observed in Experiment 1 with auditorily presented to-be-recalled material and the heterogeneous compound suffix effect reported by LeCompte and Watkins (1995) is in the number of times each irrelevant utterance was presented. This is a condition referred to by Bridges and Jones (1996) as word-dose. It is, arguably, the most important difference between the two studies. In LeCompte and Watkins study, the dose was extremely low, comprising only two separate utterances, compared to the dosage used by Bridges and Jones. If the heterogeneous compound suffix effect is a special case of the changing-state effect comprising extremely low worddosage, then increasing the dose should increase the disruptive effect, mirroring the results obtained by Bridges and Jones using visual presentation of the to-be-recalled lists. If, however, the heterogenous compound suffix effect is best accounted for by assuming that both suffixes are added to the pool of possible responses, then increasing the dose should have no effect, as by increasing the dose no new irrelevant items are actually added to the pool of possible responses. As the increase in dose comprises only repetitions of earlier instances, the extra items added to the search set are identical to existing members of the

8 206 CAMPBELL, BEAMAN, BERRY search set. Consequently, the likelihood of erroneous match is not increased. In this manner the FM could account for the observation that no further disruption is observed with homogenous (repeated) compound suffixes beyond that observed using a single item suffix (LeCompte & Watkins, 1995; Morton, 1976). EXPERIMENT 2 Experiment 2 examines whether increasing the dose of the irrelevant sound does further disrupt serial recall of auditory lists in a manner analogous to that observed with visual lists (Bridges & Jones, 1996), or whether mere repetitions of the same irrelevant items do not add any new information to the search set at recall and so do not further disrupt serial recall. In this experiment, trials designated as low-dose changing-state trials were lists of to-be-remembered auditory items, followed by a heterogeneous compound suffix, irrelevant speech consisting of just two (different) utterances a dose of two. On trials designated as high-dose changing-state trials, these same two items were repeated 15 times each a dose of 30 also following the presentation of the list. The expectation is that the compound suffix effect reported by LeCompte and Watkins (1995) will be replicated but that increasing the dosage will increase the size of the effect in the same manner as occurs with irrelevant sound disruption of recall for visually presented items. However, this result would be contrary to the assumption that a repeated suffix adds no new information to a search set consulted at recall. Method Participants. A total of 24 staff and students from Reading participated either for course credit or a small honorarium. All participants were aged between 17 and 45 and were selected by the same criteria as in Experiment 1. None had participated in other experiments of this series. Materials and design. The memory task was identical to that used in Experiment 1, with the exception that there was just one block of 60 auditory lists. There were 20 trials in each of three conditions (quiet, low-dose changing-state, and high-dose changing-state) and presentation order of these trials was randomised. For each low-dose trial the changing sequence of two different sounds (e.g., I, Z) followed the list when the word WAIT was presented on the screen. The two sounds were drawn from the five letters used in Experiment 1. No irrelevant sound was presented in the remainder of the retention interval. High-dose sequences consisted of 15 repetitions of one ordered pairing of two different items (e.g., I, Z, I, Z, I, Z etc.). Onset and offset of the changing-state sequence coincided with that of the word WAIT. On both high- and low-dose trials, each of the possible ordered pairings a total of 20 orders of two different sounds that were drawn from a set of five tokens was used once. Procedure. The procedure was identical to Experiment 1, with the exception that only an auditory block of trials was presented. Results The pattern of mean overall error probabilities was consistent with the assumption that the extent of the disruption produced by a changing sequence was dose-dependent, as illustrated in the right hand panel of Figure 2 (Quiet: 0.47 < Low dose: 0.54 < High dose: 0.58). The overall disruption produced by irrelevant speech was found to be significant by a two-way repeated-measure s ANOVA, F(2, 46) = 26.56, MSE = 0.02, p <.001. Critical comparisons revealed that the disruption produced by a lowdose compound suffix relative to quiet was significant, F(1, 46) = 24.19, p <.05, and that the modest further disruption of high- over low-dose trials was also significant, F(1, 46) = 4.84, p <.05. Main effects of serial position were significant, F(8, 184) = 54.16, MSE = 0.05, p <.001, as was the interaction between serial position and the compound suffix condition, F(16, 368) = 4.27, MSE = 0.01, p <.001. These bowed error-position functions with marked recency characterise those of serial recall for auditory material. The significant interaction describes the numerically greater effect sizes that occurred on the later serial positions. This interaction is assumed to be a scalar effect whereby disruption was attenuated near floor on primacy positions (see Figure 2). Discussion The results from this experiment demonstrate that increasing the dose of a changing-state sequence following auditory presentation of the to-be-

9 AUDITORY CHANGING-STATE EFFECTS 207 Figure 2. Experiment 2: Error-position functions (left-hand panel) and mean error probabilities collapsed across serial positions (right-hand panel) for each irrelevant speech condition. Error bars represent standard error. recalled items increased the level of disruption observed. It is thus possible that compound suffix effect is a misnomer. The existence of change in the suffix increases the magnitude of the disruption caused by a suffix (LeCompte & Watkins, 1995) but, in the same manner, change also increases the disruption seen with auditory presentation of the to-be-recalled list and post-list irrelevant sound (Experiment 1). Equally, the larger the dose in the suffix, the larger the disruption with both auditory presentation (Experiment 2) and visual presentation (Bridges & Jones, 1996). Experiment 1 showed a changing-state disruption that occurred in recall of both auditorily and visually presented items. Experiment 2 showed that this changing-state effect was dosedependent. It was not simply the number of different irrelevant items that determined the extent of disruption, the number of repetitions of the irrelevant speech also contributed to the effect. Consequently, any explanation of the retrospective effect of irrelevant speech based on the addition of the post-list irrelevant item to the pool sampled at recall (Nairne, 1990; see also LeCompte, 1996) seems untenable. The results of the first two experiments seem to confirm the notion that irrelevant sound acts equally on representations derived from auditory or visual presentation, a finding consistent with the unitary assumption of the O-OER and WM models. These findings are, however, inconsistent with the idea that modality-dependen t features necessarily interfere with each other (e.g., Nairne, 1990) and further suggest that no special mechanism is necessary to account for the compound suffix effect beyond those already postulated to account for the effects of irrelevant sound. Furthermore, the idea that retrospective irrelevant sound disruption could occur within the auditory modality by adding the irrelevant items to the pool sampled at recall (Nairne, 1990) was tested. It is argued that the increase in disruption as dose increases is predicted from past studies of irrelevant sound effects with visual presentation (Bridges & Jones, 1996). If increasing the dose adds no new items to a pool sampled at recall but simply repeats existing (irrelevant) items, one would not expect to see an increase in disruption. A counter-argumen t is that even repetitions of the same item might be added to the pool sampled at recall. The probability of sampling a given item from the search set as the recall response for any one to-be-recalled item could then be calculated as a combined function of all items within the search set, even those that are only repetitions of earlier instances. However, such a counter-argument is difficult to sustain in the face of empirical evidence that a single irrelevant item repeated, a homogenous compound suffix, usually results in a reduction in disruption (Morton, 1976). A less complex situation arises when one considers the addition of different items to the irrelevant sound stimuli. It is non-controversial that these different items will add to the search set if the heterogeneous compound suffix effect is to be explained in the same manner as a single suffix

10 208 CAMPBELL, BEAMAN, BERRY effect (LeCompte & Watkins, 1995). Similarly, if one is to explain the compound suffix effect as an irrelevant sound phenonemon, then the changingstate hypothesis contains clear predictions that the addition of different items (new changes) to the irrelevant sound stream will not further increase disruption unless the total word dose also increases (Tremblay & Jones, 1998). It is, therefore, worth examing the possibility that adding different irrelevant items to the end of the to-be-recalled list would affect the extent of irrelevant sound disruption. EXPERIMENT 3A The explanation of the compound suffix effect given by the FM suggests that more disruption should result if more different irrelevant items are present in the search set (see also LeCompte, 1996). Increasing the utterances in the compound suffix from two to five should, therefore, increase the disruptive effect of the suffix. In contrast, the changing-state hypothesis of the O-OER model predicts that irrelevant sound disruption is caused by the number of changes, not by the nature of the changes. Thus increasing the set of utterances from which the irrelevant sound is drawn from one token (steady-state irrelevant sound) to two tokens (changing-state irrelevant sound) should produce an irrelevant sound effect, but beyond this there should be no further increase in disruption. A brief example may be of use. If compound suffix disruption is a result of grouping the compound suffix with the to-be-recalled list increasing the utterances in the compound suffix from two tokens (A-B) to five tokens (A-B-C-D- E) should increase the disruptive effect of the suffix as the number of possibilities within the search set will increase accordingly. From the point of view of the changing-state hypothesis, however, introducing new types of change to an irrelevant sound stream should have little effect: the occurrence of change is sufficient. The number of changes in the sequence A-B-A-B-A-B-A-B- A-B is identical to the number of changes in the sequence A-B-C-D-E-A-B-C-D- E (see Tremblay & Jones, 1998, for full details). Experiment 3a compares the effects of token set size on disruption of serial recall. One-token trials consisted of lists of auditory items, followed by a single suffix item. On two-token trials, lists were instead followed by a sequence of two different suffix items; on five-token trials by five different suffix items. On each trial, there were no repeats of tokens within the irrelevant sequence. The predictions are clear-cut, if change within a compound suffix operates by the principles described by the changing-state hypothesis there should be a changing-state effect (significant differences between token set sizes 1 and 2) but no further increase in disruption. If the compound suffix effect operates by perceptual grouping, a further increase in disruption from two to five tokens is anticipated as the number of irrelevant utterances grouped with the to-be-recalled list, and thus entering the search set, increases. Method Participants. A total of 24 students and staff from Reading University participated either for course credit or a small honorarium. All participants were aged between 18 and 45 and were selected by the same criteria as in Experiment 1. None had participated in other experiments of this series. Materials and design. The memory task was identical to that of Experiment 2. There were 20 lists in each token set-size condition. Presentation order of the lists was randomised. The five tokens used were identical to those used in Experiments 1 2. For each 1-token suffix trial, a single sound followed the list when the word WAIT was presented on the screen. There were four trials followed by one of the five possible tokens. On 2- token trials, one of 20 possible orderings of two tokens, chosen randomly without replacement, followed the list instead. On 5-token trials, a random ordering of the different five consonants followed the list, with no repetitions of each random order on subsequent trials. On all trials, no irrelevant sound was presented during the remainder of the retention interval. Materials and apparatus were otherwise identical to that used for Experiment 2. Procedure. The procedure was identical to that used in Experiment 2. Results The pattern of overall error probabilities was consistent with the view that the more tokens in the irrelevant speech, the more disruption occurs (1-token < 2-tokens < 5-tokens), as depicted in

11 AUDITORY CHANGING-STATE EFFECTS 209 Figure 3. Experiment 3a: Error-position functions (left-hand panel) and mean error probabilities collapsed across serial positions (right-hand panel) for each irrelevant speech condition. Error bars represent standard error. Figure 3. However, this pattern in the means was not unequivocally supported by the statistical analyses. Although a repeated-measure s ANOVA of error probability data revealed a significant main effect of token set size: F(2, 46) = 9.59, MSE = 0.03, p <.001, critical planned comparisons revealed that two tokens (changing-state compound suffix) produced reliably greater disruption than one token (steady-state single suffix item), F(1, 46) = 7.00, p <.01, but five tokens (changingstate 5-token suffix) did not produce a significantly greater disruption than two tokens, F(1, 46) = 2.97, p >.05. The effect of serial position was significant, F(8, 184) = 68.22, MSE = 0.04, p <.001, but did not interact significantly with the effect of set size, F(16, 368) = 1.15, MSE = 0.01, p =.308. Discussion The results from the overall pattern in the means and that on the final item show that despite modest numerical increments in disruption with enlargements of set size beyond two, only the increase from one to two tokens proved statistically reliable. At first blush it appears that the predictions of the changing-state hypothesis were supported the fact that the irrelevant sound was changing determined the extent of disruption rather than the number of tokens. Overall, only enlargements of set size from one to two produced a significant disruption. There are several causes for concern however. In the first place, the statistically unreliable increase in disruption beyond set size 2 mirrors similar statistically unreliable increases in disruption beyond set size 2 reported by Tremblay and Jones (1998), suggesting that a lack of statistical power might be an issue in both studies. It is possible that the difference between two- and five-token conditions missed statistical significance because of low statistical power coupled with a comparatively small effect size. In the second place, the increase in token set size from two to five items in Experiment 3a also entailed an increase in word-dose from two to five. Arguably, the impact of this increase is negligible because, in Experiment 2, a 15-fold increase in dose (from 2 to 30 items) produced only a further 3% increase in disruption. However this presumes that absolute, rather than proportional, differences in dose are the critical feature of the word-dose effect. This is obviously an empirical question, but it is worth noting that the difference in dose proportions in Bridges and Jones (1996) Experiment 1 (low:high) was approximately 2:5, the same proportion as appeared in Experiment 3a, although the absolute differences varied considerably between the two studies. Experiment 3b therefore replicated Experiment 3a using a more powerful design and higher word-dose but holding dose constant between the two token-set conditions (set sizes 1, 2, and 5).

12 210 CAMPBELL, BEAMAN, BERRY EXPERIMENT 3B The pattern of means from Experiment 3a indicated that the larger the set size, the larger the disruption. The fact that the numerical disruptive advantage of five over two tokens was not significant may have reflected the fact that higher doses were a pre-condition for this disruption to be reliable. Experiment 3b tested out this possibility by using high dose sequences of one-token, two-tokens, and five-tokens. Method Participants. A total of 24 staff and students from Reading University participated either for course credit or a small honorarium. All participants were aged between 18 and 45. None participated in other experiments of this series and all reported intact hearing and vision with English as their first language. Materials and design. The memory task was identical to that used in Experiments 2 3a. The tokens were identical to those used in Experiments 1 2. For each 1-token trial, 30 repetitions of a token followed the list, such that, for each of the five tokens, four trials were followed by 30 repeats of that token. With 2-token sequences, one of 20 possible orderings of two tokens, chosen randomly without replacement, was repeated 15 times. On 5-token trials, a random ordering of the five different consonants followed the list. This ordering was repeated six times on each trial with no repetitions of that random ordering on subsequent trials. Onset and offset of the changing-state sequence coincided with those of the word WAIT. The materials and apparatus were otherwise the identical to that used for Experiment 2. Procedure. The procedure was identical to that used in Experiments 2 and 3a. Results The pattern of overall error probabilities was consistent with the view that the number of tokens determined the extent of disruption, as depicted in Figure 4. A two-way repeated-measure s ANOVA of error probability data revealed a significant main effect of token set size, F(2, 46) = 23.23, MSE = 0.03, p <.001. Critical planned comparisons revealed that two tokens (changing-state) produced a greater disruption than one token (steadystate), F(1, 46) = 14.44, p <.01, and five tokens also produced a significantly greater disruption than two tokens, F(1, 46) = 9.01, p <.01. There was also a significant main effect of serial position, F(8, 184) = 53.24, MSE = 0.06, p <.0001, although there was no significant interaction of this serial position effect with token set size, F(16, 368) = 1.25, MSE = 0.01, p =.23. Figure 4. Experiment 3b: Error-position functions (left-hand panel) and mean error probabilities collapsed across serial positions (right-hand panel) for each irrelevant speech condition. Error bars represent standard error.

13 AUDITORY CHANGING-STATE EFFECTS 211 Discussion The results of this experiment show that the number of different tokens presented during a retention interval does influence the magnitude of the disruptive effect. A difference between one token and two tokens constitutes the welldocumented changing-state effect, but Experiment 3b further demonstrates that, with auditory presentation, an increase in the number of tokens beyond this two-token changing-state effect increases the irrelevant sound effect still further. This result confirms the trend towards a token set size effect in Experiment 3a, a trend also apparent in the results of Tremblay and Jones (1998) using visual presentation of the to-be-recalled material. Methodological differences between Tremblay and Jones s (1998) series and the current experiments may account for the slightly different findings with regard to the set-size effect. One possibility is that the modality of presentation of to-be-recalled material influenced the results. However, there was no difference between the changing-state effects shown whether the to-beremembered items were seen or heard (Experiment 1). This suggests that perhaps it may not be the presentation modality of the to-be-remembered items that underpins differences in set size effects. As here, Tremblay and Jones (1998) used high doses of phonologically distinct tokens, although their dose was spread across presentation and retention intervals. That this high dose was here confined to the retention interval seems an unlikely candidate explanation for the different pattern of findings. Elsewhere, equivalent levels of disruption have occurred when irrelevant speech was confined to the presentation or retention interval (Beaman & Jones, 1998; Macken et al., 1999; Miles et al., 1991). Whether the dose was presented during presentation or retention should thus have been of no functional significance. It may be that the experimental design was more sensitive to the effects of set size. Manipulation of set size within-participant, here, rather than between-participant s (Tremblay & Jones, 1998, Experiment 3), prevented individual differences in the susceptibility to disruption from obscuring the set size effect (Ellermeier & Zimmer, 1997). Most critically, fewer conditions were used such that, on repeated-measure s designs, observations for each set-size condition came from 20 trials rather than the 15 used by Tremblay and Jones (Experiments 1, 2, & 5). This increase in sensitivity was required to expose a reliable disruptive advantage of five over two tokens. GENERAL DISCUSSION The results of the present study show that changing-state effects occurred on delayed recall of auditorily presented material (Experiment 1), in a fashion that increased with both higher doses (Experiment 2) and enlargements of token set size (Experiments 3a and 3b). These set size effects were somewhat weaker at lower doses (Experiment 3a). The difference between two and five items observable in Experiment 3b appeared only as a non-significant trend at low dosage even though dose was higher with five items than with two items (Experiment 3a), suggesting that perhaps absolute rather than proportional differences in word-dose determine the appearance of the dose effect. The heterogeneous compound suffix effect shown with low-dose sequences (LeCompte & Watkins, 1995) thus shared many properties with the effects of irrelevant sound on visually presented lists, in that, like the irrelevant sound effect, it was influenced by both dose and set size (Bridges & Jones, 1996; Tremblay & Jones, 1998). The current series thus questions whether there is a distinction between the compound suffix effect and the irrelevant sound effect, and raises a number of issues for current models of irrelevant sound interference in memory. The first issue is how models should account for the retrospective effects of irrelevant sound found in Experiment 1. This is not a problem for either the WM model or the O-OER model, both of which predict that interference occurs in memory after the presentation of the to-be-recalled list. The O-OER model also specifically predicts the changing-state effect observed in Experiment 1. However, models such as the FM which assume that irrelevant sound captures attention at encoding do face difficulties in accounting for irrelevant sound effects that occur beyond this point. The second issue is how to account for the separate effects of dose and token set size. The effect of dose found in Experiment 2, an effect of repeatedly presenting the same irrelevant sound items, again does not seem very compatible with the FM unless it is assumed that presenting the same items will increase the number of distractor items in the search set. However, if this assumption is made, then the decrease in disruption