Inner speech as a retrieval aid for task goals: the effects of cue type and articulatory suppression in the random task cuing paradigm

Transcription

1 Acta Psychologica 115 (2004) Inner speech as a retrieval aid for task goals: the effects of cue type and articulatory suppression in the random task cuing paradigm Akira Miyake a, *, Michael J. Emerson a, Francisca Padilla b, Jeung-chan Ahn a a Department of Psychology, University of Colorado at Boulder, 345 UCB, Boulder, CO , USA b Department of Psychology, University of Granada, Campus de Cartuja s/n 18071, Spain Abstract Articulatory suppression has been shown to increase switch costs in the list paradigm (e.g., [J. Exp. Psychol.: General 130 (2001) 641, J Memory Language 48 (2003) 148]). The present dual-task study examined whether this effect generalizes to the random task cuing paradigm. Participants performed color or shape judgments according to explicit word cues (COLOR or SHAPE) or less transparent letter cues (C for the color task and S for the shape task). In the word cue condition, the switch cost was equivalent for the articulatory suppression and the control (no dual-task) conditions, but, in the letter cue condition, the switch cost was significantly greater for the articulatory suppression condition than for the control condition. These results suggest that inner speech may be recruited as a tool for retrieving and activating the relevant task goal when the task cue is not transparent and hence imposes nonnegligible retrieval demand. Ó 2003 Elsevier B.V. All rights reserved. PsycINFO classification: 2340; 2346; 2720 Keywords: Inner speech; Articulatory suppression; Task switching; Goal retrieval; Verbal control * Corresponding author. Tel.: address: miyake@psych.colorado.edu (A. Miyake) /$ - see front matter Ó 2003 Elsevier B.V. All rights reserved. doi: /j.actpsy

2 124 A. Miyake et al. / Acta Psychologica 115 (2004) Introduction Language plays an important role in everyday living. Although not the only medium, the central role of language in human communication is incontrovertible. The utility of language is not restricted to interpersonal communication, however. It can also shape and support intrapersonal thinking processes in various ways (Carruthers, 2002; Gentner & Goldin-Meadow, 2003; Sokolov, 1972; Vygotsky, 1934/1986). One such intrapersonal cognitive function attributed to language is the regulation of thought and action (Barkley, 1997; Luria, 1961). In particular, at least among adults, inner (or subvocal) speech is considered to play some role in regulating one s behavior (Carlson, 1997; Sokolov, 1972; see also Gruber & Goschke, 2004). In this sense, inner speech (and, more generally, language) can be viewed as a contributing factor to efficient executive control. The current study examined the role of inner speech in one aspect of executive control processes, mental set shifting, and tested the hypothesis that inner speech supports task switching performance by serving as a goal retrieval aid. Although the regulatory role of inner speech in executive control has not been studied extensively from modern cognitive psychological perspectives, it seems clear that, at least under some circumstances, adults spontaneously recruit inner speech to facilitate their set shifting performance. For example, Dunbar and Sussman (1995) and, more recently, Cinan and Tan or (2002) showed that concurrent articulation of simple irrelevant words or phrases (i.e., articulatory suppression) significantly impaired healthy adults performance on the Wisconsin Card Sorting Test, a wellknown neuropsychological test of executive functioning related to set shifting ability (Miyake et al., 2000). In particular, the type of errors that increased under concurrent articulatory suppression was perseveration errors (i.e., failure to shift to a new task-relevant category), which patients with frontal lobe damage are known to predominantly make on this test. The negative impact of articulatory suppression on set shifting has also been demonstrated in a few recent studies that required participants to switch back and forth between two mental operations (Baddeley, Chincotta, & Adlam, 2001; Emerson & Miyake, 2003; Goschke, 2000; Saeki & Saito, in press; see also Kray, Eber, & Lindenberger, 2004). These studies used either the list paradigm, originally developed by Jersild (1927), or its variant, the precued task sequence paradigm, to examine the role of inner speech in task switching performance. In these paradigms, tasks are performed either in pure-task blocks (i.e., participants perform only one task) or in alternating-task blocks (i.e., participants alternate between two different tasks). In Baddeley et al. (2001) study, for example, participants performed various secondary tasks as they completed addition and subtraction problems in pure-task lists (i.e., addition only or subtraction only) or alternating lists (i.e., alternated between addition and subtraction). The key finding of the study was that the switch cost, defined as the difference between the completion time for the alternatingtask list and the average completion time for the pure-task lists, increased considerably under articulatory suppression (e.g., reciting the days of the week or saying the repeatedly). In contrast, other secondary tasks without a clear verbal compo-

3 A. Miyake et al. / Acta Psychologica 115 (2004) nent, such as spatial tapping (tapping four pegs of a square in clockwise direction) and a marble transfer task (moving marbles one at a time from one container to another), had a much smaller or negligible effect on switch costs. Similar dissociations have been reported in other studies using the list paradigm, regardless of whether the tasks to be switched between involved mental arithmetic or judgments related to numbers and letters (Emerson & Miyake, 2003; Saeki & Saito, in press). The negative effect of articulatory suppression on switch costs has also been shown with the precued task sequence paradigm (Goschke, 2000, Experiment 2). In this study, participants were presented with a colored letter (e.g., a green B) on each trial and judged either the color (red or green) or the identity (A or B) of the stimulus letter. Each experimental block consisted of trial pairs and, at the beginning of each block, participants were instructed to perform the same task on both stimuli of each trial pair (i.e., color only or letter only) or to switch from one task to the other within the pair (i.e., perform the letter task on the first item and the color task on the second item or vice versa). During the 1500 ms interval provided between the manual response to the first stimulus and the presentation of the second stimulus, half of the participants were instructed to say the task name ( color or letter ) aloud once, whereas the other half were instructed to say an irrelevant word ( Monday or Tuesday ) once instead. The results indicated that saying an irrelevant word caused a significant increase in switch cost relative to saying the task name. Moreover, the switch cost for the irrelevant word condition was comparable to the cost observed in a different experiment (Goschke, 2000, Experiment 1) when participants had only a short (14 ms) preparatory interval, whereas the switch cost for the task name condition was virtually identical to the cost observed when participants had a long (1500 ms) preparatory interval without any verbalization requirement. These results suggest that irrelevant articulation is particularly harmful if it is required precisely when the task goal information for the forthcoming trial must be retrieved (or reactivated). The evidence reviewed above suggests that inner speech may be involved in the retrieval or reactivation of the relevant task goal from long-term memory and may serve as an efficient internal self-cuing device that signals which task to perform next (Baddeley et al., 2001; Emerson & Miyake, 2003). Consistent with this interpretation, the presence of explicit external cues, such as the + and ) signs for addition and subtraction operations, considerably reduced the magnitude of the negative impact of articulatory suppression on switch costs (Baddeley et al., 2001; Emerson & Miyake, 2003; Saeki & Saito, in press). Similarly, underlining either the letter or the number in each letter number pair (e.g., B6) and thereby making obvious which stimulus to process also greatly reduced the magnitude of the articulatory suppression effect on switch costs (Saeki & Saito, in press). These manipulations are likely to have substantially reduced or even eliminated the need for internally retrieving the relevant task goal on each trial, hence minimizing the negative effect of articulatory suppression on switch costs. Further pointing to the role of inner speech in goal retrieval, manipulations that do not reduce the need

4 126 A. Miyake et al. / Acta Psychologica 115 (2004) for goal retrieval or internal self-cuing (e.g., adding or subtracting 1 vs. 3 vs. 6) did not have any influence on the magnitude of the articulatory suppression effect on switch costs, even though they strongly influenced the overall list completion times (Emerson & Miyake, 2003). Thus, a fair amount of existing evidence indicates that, as far as the list and precued task sequence paradigms are concerned, inner speech plays an important supporting role in task switching performance. A question remains, however, as to whether the results based on these paradigms will be obtained in other prevalently used paradigms in the task switching literature, such as the random task cuing paradigm (e.g., H ubner, Futterer, & Steinhauser, 2001; Mayr, 2001; Mayr & Keele, 2000; Meiran, 1996; Meiran, Chorev, & Sapir, 2000). In the random task cuing paradigm, a task cue indicating which task to perform is presented on each trial, typically before the target stimulus is displayed on the computer screen. As the name indicates, a task cue is chosen randomly for each trial and, thus, whether the forthcoming trial requires a task switch or not is unpredictable. Note that, because a task cue is presented on each trial, the random task cuing paradigm does not require participants to keep track of the task sequence to perform the correct task on a given trial, unlike the list and precued task sequence paradigms. Given these unique characteristics of the random task cuing paradigm, recruiting inner speech mechanisms for self-cuing purposes may no longer be needed and, if so, the effect of articulatory suppression on switch costs might be greatly reduced or even completely absent in the random task cuing paradigm. This would be particularly true if the role of inner speech in task switching performance in the list and precued task sequence paradigms is restricted to dealing with the sequencing requirement and has nothing to do with the process of goal retrieval and activation. The main purpose of the present study was to determine whether and to what extent inner speech serves as a goal retrieval aid by focusing on the random task cuing paradigm and minimizing the need for keeping track of the task sequence. The experiment reported below used two component tasks that were nonverbal in nature: making a color judgment (red or green) or a shape judgment (circle or triangle). Participants completed blocks of trials in which these tasks were presented in a random sequence, and a cue specifying which task to be performed was presented prior to the presentation of a bivalent stimulus (e.g., a triangle superimposed on top of a green color patch) on each trial. Two types of cues were used in the experiment to evaluate the involvement of inner speech in the process of goal retrieval and activation. One type of cue was a word that directly indicated the task name (i.e., COLOR or SHAPE), and the other type of cue was a letter (i.e., C for the color classification task, S for the shape classification task). The word cue type was chosen because word cues indicating the task name have been used in several previous studies of task switching (e.g., Mayr & Keele, 2000; Mayr & Liebscher, 2001) and because explicit, familiar word cues such as COLOR and SHAPE were considered likely to quickly activate a representation of the appropriate task goal. Moreover, according to Gathercole and Baddeley (1993), the role of inner speech (or, more specifically, the phonolog-

5 A. Miyake et al. / Acta Psychologica 115 (2004) ical loop subsystem of working memory) in the reading of highly familiar words is essentially none, except when complex judgments about phonological structure [is] required (p. 232). Thus, the activation of appropriate task representations was expected to take place automatically even under concurrent articulatory suppression. The letter cue type was chosen because it had some degree of association with the task name (i.e., the first letter of the task name) and hence was relatively easy to memorize, but was uncommon enough that it was unlikely to activate the task representations as efficiently as word cues would. Although this cue type manipulation may seem subtle, we chose letter cues because we wanted to avoid using intentionally misleading cues (e.g., the word COLOR for the shape task and the word SHAPE for the color task) or completely arbitrary cues that might be hard to remember (e.g., X for the color task, Y for the shape task). For the main part of the experiment, participants performed the color and shape tasks with both cue types in five secondary task conditions: concurrent foot tapping, concurrent articulatory suppression, cue-triggered verbalization of the task name (i.e., the say task name condition used by Goschke, 2000), and two conditions without any secondary task requirement (control-long and control-short). In four of the five conditions (foot tapping, articulatory suppression, say task name, and control-long), the preparation interval between the cue and the stimulus was 1200 ms. It has been well established that long preparation intervals allow participants to mentally prepare for the forthcoming task, which typically results in reduction of the switch cost (e.g., Meiran, 1996, Meiran et al., 2000; Rogers & Monsell, 1995). In the fifth condition (control-short), the preparation interval was only 200 ms. This condition was included for comparison to the control-long condition to ensure that participants did indeed try to mentally prepare in advance of the stimulus with a long preparation time. For each cue type, we compared the magnitude of the switch costs in the five secondary task conditions (control-short, control-long, say task name, foot tapping, and articulatory suppression). For the word cues, the switch cost in the control-short condition should be significantly larger than the costs in the four conditions with a long preparation time. However, because word cues provide the task name that participants would otherwise need to retrieve on their own, the switch costs in the control-long, say task name, foot tapping, and articulatory suppression conditions should not differ significantly from each other. For the letter cues, the control-short condition should again have a larger cost than most of the other four secondary task conditions. In contrast to the word cues, however, the letter cues should still impose some retrieval demand because of the relatively weak association between the letters (C or S) and the task name. For this reason, the switch costs under concurrent articulatory suppression should be significantly larger than the costs in the control-long, say task name, and foot tapping conditions. In fact, if articulatory suppression hinders or considerably slows down the retrieval of the task goal when letter cues are presented, the switch cost in this condition may be as large as that in the controlshort condition.

6 128 A. Miyake et al. / Acta Psychologica 115 (2004) Method 2.1. Participants The participants were 48 undergraduate students from the University of Colorado at Boulder who received monetary compensation or partial course credit for taking part in the experiment Design, materials, and procedure The experiment consisted of two parts administered in a single session that lasted approximately 90 min. In the first part, participants performed several blocks of pure-task trials in which only one task (either color or shape judgment) was performed in a given block of trials. In the second part, participants performed several blocks of mixed-task trials in which each trial was preceded by a cue indicating which task to perform. For both parts, participants first went through training or practice trials to familiarize themselves with the task requirements. They then performed target trials, with or without secondary tasks. The tasks for both parts were programmed with the PsyScope software (Cohen, MacWhinney, Flatt, & Provost, 1993) and were run on Power Macintosh 7200 computers. Button-press responses were recorded from stimulus onset to the nearest millisecond using a button box with an internal timer Pure-task trials Before starting the target trials, participants completed a total of 150 training trials, administered in five blocks of 30 trials each, to master the response-mapping rules. The responses for both color and shape tasks were mapped onto two buttons in a bivalent manner (e.g., the left button for red and circle and the right button for green and triangle ). Throughout this experiment (including both pure-task and mixed-task trials), the response-mapping rules were kept constant for each participant, although we counterbalanced the response mappings across participants by randomly assigning 12 participants to each of the four counterbalancing conditions. In each training trial, a stimulus was presented without any cue in the middle of the computer screen until a response was made. If the response was correct, the next trial commenced after a 600 ms intertrial interval (ITI). If it was incorrect, the computer beeped and an error message (i.e., incorrect ) was displayed for 1000 ms followed by an ITI of 600 ms. In the first three blocks of the training trials, the stimulus was univalent and was applicable to only one task. In the first block, a black line-drawing of either a circle or square shape was presented on each trial, whereas in the second block, a squareshaped color patch (red or green) was presented. In the third block, either a shape or a color patch was presented in a random sequence (e.g., a black line drawing of a square on one trial, followed by a red color patch on the next). The circle was 27 mm in diameter, the triangle was 25 mm wide and 28 mm tall, and the square-shaped color patch was 36 mm on each side for both colors (the viewing distance was

7 A. Miyake et al. / Acta Psychologica 115 (2004) approximately 60 cm). In the last two blocks of the training session, the stimulus was bivalent and consisted of a shape (either circle or triangle) superimposed on a square color patch (either red or green). Participants performed the shape task throughout the fourth block and the color task for the last, fifth block. The order of administering the five training blocks was fixed for all participants. After completing these response-mapping training trials, participants performed six blocks of target pure-task trials with 65 trials in each block. As in the last two blocks of training trials, the stimulus for each target trial was bivalent. The trial setup was identical to the training trials (including the absence of any task cue) except that participants no longer received an error message for incorrect responses but were informed of their accuracy (percentage correct) at the end of each block. The six blocks for the target pure-task trials were created by crossing the task type (color or shape) with the secondary task requirement (control, foot tapping, or articulatory suppression). The six blocks were organized in two sets of three so that participants performed one task in a fixed order of all three secondary task conditions and then repeated the procedure for the other task. For example, a participant might perform the color task with concurrent articulatory suppression in the first block, without any secondary task in the second block, and with concurrent foot tapping in the third block, and then repeat this secondary task order for the shape task in the remaining three blocks. The order of the three secondary task conditions and the order of the two task types were counterbalanced across participants. During pilot testing, it became clear that, despite explicit instructions, some participants failed to switch tasks between the third and fourth blocks and thus showed almost chance-level performance in the fourth block. To minimize the occurrence of this perseveration problem, we presented a visual reminder for 3 s at the beginning of each block that clearly indicated which task to perform for that block (e.g., Respond to Shape ). The secondary task procedure was similar to that used by Emerson and Miyake (2003). When the participants first encountered the articulatory suppression and foot tapping conditions, the experimenter described the relevant secondary task and then demonstrated how to perform it. A metronome was set to beat at the rate of 750 ms per beat for both secondary tasks. Participants were instructed to tap a foot once per beat in the foot tapping condition. They were allowed to choose whichever foot they preferred to use for tapping, but were asked not to switch in the middle of a block. In the articulatory suppression condition, participants were required to say either Tuesday or Thursday once per beat. Half of the participants were randomly assigned to say Tuesday for the color-task-only block and Thursday for the shape-task-only block, and the other half did the opposite. Participants first practiced the secondary task for 30 s to make sure that they could perform the task correctly. After this brief practice, the participant performed the appropriate secondary task concurrently with the color or shape task. The experimenter started the trial block after participants began the relevant secondary task. Performance on the foot tapping and articulatory suppression tasks was closely monitored by the experimenter. There was no secondary task requirement in the control condition, but the metronome was turned on to equate the level of external noise across conditions.

8 130 A. Miyake et al. / Acta Psychologica 115 (2004) Mixed-task trials After completing the pure-task trials and taking a short break, participants performed blocks of mixed-task trials. The sequence of events that took place during the mixed-task trials is schematically illustrated in Fig. 1. To familiarize themselves with the random task cuing procedure, participants first practiced mixed-task trials without any secondary task requirements for two blocks of 50 trials each. Each practice trial began with the onset of a cue, presented 9 mm above the stimulus and approximately 7 mm in height. The cue appeared 1200 ms before the stimulus and indicated which task was to be performed. The cue was a word printed in capital letters (COLOR or SHAPE) in one block of practice trials and was a capital letter (C for the color task and S for the shape task) in the other block of practice trials, with the order of these two blocks randomized for each participant. The stimulus was the same as in the target pure-task trials and consisted of a shape (circle or triangle) superimposed on a square-shaped color patch (red or green) presented in the middle of the screen. The cue and the stimulus then remained onscreen until a response was made. During this practice phase, immediate feedback was given for incorrect responses using the same procedure as in the pure-task trials. Word Cue Letter Cue SHAPE Cue: 200 ms or 1200 ms C SHAPE Stimulus: Until response C Time Intertrial Interval: 600 ms COLOR Cue: 200 ms or 1200 ms S Fig. 1. A schematic illustration of the sequence of events that happened in the word cue and letter cue conditions for the mixed-trial blocks. A cue was presented for either 200 ms (for the control-short condition) or1200 ms (for the control-long, say task name, foot tapping, and articulatory suppression conditions), indicating which of the two tasks should be performed (COLOR or C for color judgment and SHAPE or S for shape judgment). A bivalent stimulus (a color patch superimposed on top of a shape) appeared and remained on the display until a manual response was made. The intertrial interval was 600 ms.

9 After practicing the mixed-task trials with both cue types, participants performed 10 blocks of 110 target mixed-task trials each. Because performing concurrent articulatory suppression or foot tapping for an extended period of time may be tiring, each block was split into two runs of 55 trials apiece, with a brief break provided between runs. The trial setup for the target mixed-task trials was identical to the practice trials (see Fig. 1) except that feedback was no longer provided for incorrect responses. The 10 target blocks were created by crossing the two cue types (word cue or letter cue) with five secondary task conditions (control-short, control-long, articulatory suppression, foot tapping, and say task name). The order of the 10 target blocks was randomized for each participant, with the constraint that the same cue type not be repeated more than two blocks in a row. The control-short and control-long conditions differed only in terms of cue-to-stimulus interval (CSI). The CSI was 200 ms for the control-short condition and 1200 ms for the control-long condition. The CSI in the articulatory suppression, foot tapping, and say task name conditions was always 1200 ms. The secondary task requirements for the foot tapping and articulatory suppression conditions were the same as those in the pure-task trials. In the foot tapping condition, participants tapped a foot once per metronome beat (i.e., every 750 ms). In the articulatory suppression condition, participants said Tuesday throughout one block (e.g., the letter-cue block) and Thursday throughout the other block (e.g., the word-cue block), also at the same metronome pace. The days assigned to the word and letter cue blocks were counterbalanced across participants. As in the pure-task trials, participants were instructed to begin the relevant secondary task before the experimenter started the block, and the experimenter closely monitored performance on the secondary tasks. Having already practiced these secondary tasks, however, participants were not asked to practice the secondary tasks individually again. A new secondary task condition added for the mixed-task trials was the say task name condition, modeled after the Goschke (2000) study (Experiment 2). In this condition, participants were instructed to say the task name aloud ( color for the cues C or COLOR and shape for the cues S or SHAPE) immediately after the presentation of the task cue on each trial (i.e., during the CSI). As was the case with the first part of the experiment with pure-task trials, the metronome was set at the rate of 750 ms per beat for all target blocks, including the control and say task name conditions, to equate the noise level Preliminary data analysis A. Miyake et al. / Acta Psychologica 115 (2004) The first three trials in each run of the target pure-task trials and the target mixedtask trials were considered as warm-up trials and were excluded from analysis, leaving 62 maximum analyzable pure-task target trials per block and 104 maximum analyzable mixed-task target trials per block. The analyses reported below for the pure-task trials are based on reaction times (RTs) for correct trials, and the analyses for the mixed-task trials are based on RTs for correct trials in which the response to the previous trial was also correct. To prevent a small number of deviant data points

10 132 A. Miyake et al. / Acta Psychologica 115 (2004) from exerting undue influence on the mean RTs, we first excluded extreme RTs (RT < 100 ms or RT > 5000 ms) from analysis for both the pure-task trials and the mixed-task trials (0.15% and 0.03% of the total observed excluded, respectively) and then applied a logarithmic transformation to the remaining raw RTs. For ease of interpretation, the numbers mentioned throughout Section 3 and presented in the figures are backtransformed. For the RT data, we also conducted statistical analyses with trimmed RTs (specifically, any RTs that were longer than 3 standard deviations above a participant s mean RT in that condition were replaced with the value of the mean plus 3 standard deviations). The results with this trimmed RT method (without any transformation) were qualitatively the same as those based on log RTs. Thus, we only present the results from the log-transformed data. Regarding the error data, the raw proportions of errors tended to show a positive skew indicative of a floor effect. Thus, we subjected the error data to an arcsine transformation to achieve a satisfactory level of normality before performing ANOVAs. For ease of interpretation, the descriptive statistics presented below are untransformed. For all of the statistical analyses presented in this article, we used an alpha level of The effect size measure reported for the ANOVA results is partial eta squared ðg 2 Þ. 3. Results 3.1. Pure-task trials RT data The pure-task trial RTs were analyzed in a 2 3 repeated-measures ANOVA, with factors of primary task (color or shape judgment) and secondary task (control, articulatory suppression, or foot tapping). The RTs in the color (M ¼ 391 ms) and shape (M ¼ 406 ms) tasks were similar, but the 15-ms difference between these two primary tasks was reliable F ð1; 47Þ ¼7:14, MSE ¼ 0:003, p ¼ 0:010, g 2 ¼ 0:132. There was also a marginally significant interaction between the primary task and the secondary task, F ð2; 94Þ ¼2:84, MSE ¼ 0:002, p ¼ 0:064, g 2 ¼ 0:057, such that a small RT difference between the color and shape tasks was slightly magnified by the dual-task demands. The main focus for the pure-task trials, however, was on the effect of the secondary task requirement. This main effect, illustrated in Fig. 2, was statistically significant (M ¼ 354, 405, and 440 ms for the control, articulatory suppression, and foot tapping conditions, respectively), F ð2; 94Þ ¼ 48:58, MSE ¼ 0:005, p < 0:001, g 2 ¼ 0:508. A planned comparison of the control condition to the average of the foot tapping and articulatory suppression conditions indicated that there was a significant overall dual-task effect, F ð1; 47Þ ¼93:83, MSE ¼ 0:002, p < 0:001, g 2 ¼ 0:666, with longer RTs in the dual-task conditions (M ¼ 422 ms) than in the control condition (M ¼ 354 ms).

11 A. Miyake et al. / Acta Psychologica 115 (2004) Reaction Time (ms) Control Foot Tapping Articulatory Suppression Fig. 2. The mean reaction times (back-transformed from log-based means) for the color and shape puretask trials. Error bars indicate standard errors (also back-transformed). The reaction times in the two dual-task conditions were longer than those in the control condition, and concurrent foot tapping led to longer reaction times than concurrent articulatory suppression. Interesting to note, another planned comparison showed that the RTs were significantly longer with concurrent foot tapping (M ¼ 440 ms) than with concurrent articulatory suppression (M ¼ 405 ms), F ð1; 47Þ ¼12:19, MSE ¼ 0:003, p ¼ 0:001, g 2 ¼ 0:206. This result suggests that, although both secondary tasks significantly lengthened pure-task trial RTs, it was more difficult to combine the color and shape judgment tasks with foot tapping (a 24% increase in RT) than with articulatory suppression (a 14% increase). Thus, any negative effect of articulatory suppression on switch costs observed for the mixed-task trials reported below cannot be interpreted in terms of a greater attentional demand or difficulty associated with articulatory suppression than with foot tapping Error data Overall, the error rates for the pure-task trials were low, and there were no statistically significant effects. The main effect of primary task, F ð1; 47Þ ¼0:07, MSE ¼ 0:007, p ¼ 0:786, g 2 ¼ 0:002, was absent, with essentially equivalent proportions of errors for the color (M ¼ 0:048, SE ¼ 0:003) and shape (M ¼ 0:049, 1 Given that the primary task and the secondary task both required lateralized motor responses in the foot tapping condition, some sort of motor-level interference may have contributed to the greater dualtask effect found for the foot tapping condition. For example, pressing a left button with the left index finger might be somewhat difficult if the tapping is done with the left foot and the timing of the manual response does not synchronize with that of the tapping response. In this regard, it is possible that the nature of dual-task demands is somewhat different for concurrent foot tapping and concurrent articulatory suppression. However, we should note that both dual-task conditions required participants to keep up with the pace of metronome beats (i.e., tapping a foot once or saying an irrelevant word like Tuesday once every 750 ms) for the pure-task trials as well as for the mixed-task trials. Thus, any RT differences between the two conditions found in this study cannot be explained purely in terms of differences in the temporal characteristics of the two secondary task requirements.

12 134 A. Miyake et al. / Acta Psychologica 115 (2004) SE ¼ 0:004) tasks. There were also no significant differences among the control (M ¼ 0:044, SE ¼ 0:004), foot tapping (M ¼ 0:052, SE ¼ 0:004), and articulatory suppression (M ¼ 0:050, SE ¼ 0:004) conditions, F ð2; 94Þ ¼ 1:79, MSE ¼ 0:008, p ¼ 0:172, g 2 ¼ 0:037. The interaction between the two factors was also not significant, F ð2; 94Þ ¼1:47, MSE ¼ 0:005, p ¼ 0:234, g 2 ¼ 0: Mixed-task trials RT data According to the predictions outlined earlier, the RT difference between the repeat and switch trial types (i.e., the switch cost) in the five secondary task conditions should depend on the cue type. When word cues were available, the explicit nature of the cues should minimize retrieval demands and hence should yield small switch costs even under concurrent articulatory suppression. In contrast, when letter cues were presented, the mapping between the cue and the task is less transparent and may still impose some degree of retrieval demand. If so, articulatory suppression should have a greater negative impact on switch costs compared to the control condition. To test these predictions, we first conducted separate analyses for the word cue and the letter cue conditions. The RT data are graphically presented in Fig. 3A (for the word cue condition) and Fig. 3B (for the letter cue condition). For the word cue condition, an analysis of the interaction between secondary task and trial type (i.e., repeat vs. switch) indicated that there were significant differences in switch costs among the five secondary task conditions, F ð4; 188Þ ¼12:16, MSE ¼ 0:001, p < 0:001, g 2 ¼ 0:206. As illustrated in Fig. 3A, however, switch costs were uniformly low except in the control-short condition. A planned comparison of the switch costs in the control-short (M ¼ 110 ms) and control-long (M ¼ 42 ms) conditions indicated that the costs were significantly reduced with a long preparation time, F ð1; 47Þ ¼31:25, MSE ¼ 0:001, p < 0:001, g 2 ¼ 0:399, replicating the wellknown preparation effect (e.g., Meiran, 1996; Rogers & Monsell, 1995). When the control-short condition was excluded and the analysis focused on only the four secondary task conditions in which the CSI was 1200 ms (control-long, say task name, foot tapping, and articulatory suppression), the secondary task by trial type interaction was no longer significant, F ð3; 141Þ ¼0:86, MSE ¼ 0:001, p ¼ 0:462, g 2 ¼ 0:018, suggesting that the switch costs across these four conditions were essentially equivalent. Thus, as predicted, the word cues appear to have automatically activated a representation of the task goal, thereby eliminating any negative effect of articulatory suppression on switch costs. The RT data for the letter cue condition are displayed in Fig. 3B. As with the word cues, the RT differences between the task-repeat and task-switch trials (i.e., switch costs) significantly varied across the five secondary task conditions, F ð4; 188Þ ¼18:07, MSE ¼ 0:001, p < 0:001, g 2 ¼ 0:278. Again, long preparation times led to a significant reduction in switch cost (M ¼ 137 ms vs. 43 ms for the control-short and control-long conditions, respectively), F ð1; 47Þ ¼ 31:97, MSE ¼ 0:001, p < 0:001, g 2 ¼ 0:405. In contrast to the results of the word cue condition, however, when the control-short condition was excluded from analysis, a significant secondary

13 A. Miyake et al. / Acta Psychologica 115 (2004) (A) Reaction Time (ms) Word Cues Repeat Switch Control- Short Control- Long Say Task Name Foot Tapping Articulatory Suppression (B) 800 Reaction Time (ms) Letter Cues Control- Short Control- Long Say Task Name Foot Tapping Secondary Task Condition Articulatory Suppression Fig. 3. The mean reaction times (back-transformed from log-based means) for the word cues (A) and letter cues (B) for each secondary task condition in the mixed-task trials. Error bars indicate standard errors (also back-transformed) and switch cost values are printed on top of the bars for each secondary task condition. With word cues, the switch cost was larger in the control-short condition than in the articulatory suppression condition. With letter cues, the switch cost was larger in the control-short condition and in the articulatory suppression condition, and these two costs were not different from each other. task by trial type interaction was present, F ð3; 141Þ ¼ 10:03, MSE ¼ 0:001, p < 0:001, g 2 ¼ 0:176, suggesting there were still some differences in the switch cost magnitudes among the four conditions with a long preparation time. As Fig. 3B indicates, the only condition that showed a markedly different magnitude of switch cost was the articulatory suppression condition. In fact, the switch cost in the articulatory suppression condition (M ¼ 116 ms) was not significantly different from the cost in the control-short condition (M ¼ 137 ms), F ð1; 47Þ ¼2:17, MSE ¼ 0:001, p ¼ 0:148, g 2 ¼ 0:036, suggesting that, when letter cues were presented, articulatory suppression interfered with task name retrieval and thereby made advance preparation for the upcoming task as difficult as when participants had only a short time to prepare. As this comparison between the word cue and letter cue conditions indicates, the effects of articulatory suppression were dependent on the cue type, as predicted. A significant three-way interaction between trial type (repeat vs. switch), secondary

14 136 A. Miyake et al. / Acta Psychologica 115 (2004) task, and cue type (letter vs. word) further confirms that the data patterns for the word cue condition and for the letter cue condition were different, F ð4; 188Þ ¼4:52, MSE ¼ 0:001, p ¼ 0:002, g 2 ¼ 0:088. Moreover, when the articulatory suppression condition was excluded from analysis, the same three-way interaction was no longer significant, F ð3; 141Þ ¼1:90, MSE ¼ 0:001, p ¼ 0:132, g 2 ¼ 0:039. These results strongly suggest that the significant three-way interaction critically depended on the data from the articulatory suppression condition and that the effect of cue type was much stronger for the switch costs in the articulatory suppression condition than in the other four conditions Error data Table 1 presents a summary of error data for the mixed-task trials. In the word cue condition, more errors were made on task switch trials than on task repeat trials, F ð1; 47Þ ¼180:79, MSE ¼ 0:007, p < 0:001, g 2 ¼ 0:794. Error rates also differed among the five secondary task conditions, F ð4; 188Þ ¼ 6:78, MSE ¼ 0:008, p < 0:001, g 2 ¼ 0:126, primarily because of the relatively low error rate in the control-long condition. The interaction between trial type and secondary task condition was not significant, F ð4; 188Þ ¼1:37, MSE ¼ 0:007, p ¼ 0:247, g 2 ¼ 0:028, indicating that the error switch costs were equivalent among the five secondary task conditions. The results for the letter cues were identical. There was a significant difference between the repeat and switch trial types, F ð1; 47Þ ¼114:30, MSE ¼ 0:008, p < 0:001, g 2 ¼ 0:709. Error rates were also reliably different among the secondary task conditions, F ð4; 188Þ ¼5:59, MSE ¼ 0:007, p < 0:001, g 2 ¼ 0:106. There were no significant differences in error switch costs across the five secondary task conditions, F ð4; 188Þ ¼0:34, MSE ¼ 0:006, p ¼ 0:849, g 2 ¼ 0:008. Given this similarity of the error data patterns for the two cue types, it is not surprising that, when the data for the two conditions were combined and the cue type Table 1 Summary of descriptive statistics (proportion of errors) for the mixed-task errors and error switch costs Condition Repeat Switch Switch cost M SE M SE M Word cues Control-short Control-long Say task name Foot tapping Articulatory suppression Letter cues Control-short Control-long Say task name Foot tapping Articulatory suppression

15 A. Miyake et al. / Acta Psychologica 115 (2004) was treated as an additional factor, the three-way interaction was not significant, F ð4; 188Þ ¼0:46, MSE ¼ 0:006, p ¼ 0:767, g 2 ¼ 0:010. In this respect, the error data differ from the RT data, but the finding that the error switch costs were uniform across different secondary task conditions for both cue types rules out the possibility that the critical three-way interaction observed for the RT data was an artifact of a speed accuracy tradeoff. 4. Discussion The current study examined the role of inner speech in task switching performance using the random task cuing paradigm. The most important finding was the different patterns of results obtained for the word cue and letter cue conditions. Specifically, when participants were presented with letter cues (C or S), articulatory suppression caused an increase in switch cost compared to the other secondary task conditions with a long preparation time (CSI ¼ 1200 ms). In contrast, when the cues explicitly indicated the task names (COLOR or SHAPE) that participants would otherwise have had to generate or retrieve by themselves, concurrent articulatory suppression did not have any negative effect on switch costs. Given that the two cue types differed in the extent to which they automatically activate the task goal information, these results support the hypothesis that inner speech serves as a convenient and reliable medium for retrieving and activating the task goal for the forthcoming operation (Emerson & Miyake, 2003), at least among young adults (Kray et al., 2004). An additional finding from the control-short condition seems consistent with this interpretation. As Fig. 3 indicates, the switch cost in the control-short condition was 110 ms in the word cue condition and 137 ms in the letter cue condition. This 27 ms difference, tested in terms of the cue type by trial type interaction, was marginally significant, F ð1; 47Þ ¼2:89, MSE ¼ 0:001, p ¼ 0:096, g 2 ¼ 0:056. This somewhat smaller switch cost found for the word cue condition when the preparation time was short (CSI ¼ 200 ms) makes sense if, as we assumed, the explicit word cues activate the task goal information more rapidly or more directly (i.e., without going through the inner speech based retrieval process) than do the less transparent letter cues. Just as well-learned directional words (e.g., left ) seem capable of directing one s attention automatically toward those directions even when they are not predictive of the locations of subsequently presented targets (Hommel, Pratt, Colzato, & Godijn, 2001), it might be the case that such well-learned words as color or shape can automatically direct one s attention toward the designated stimulus dimension and thereby reduce the effect of perceptual interference from the other, task-irrelevant dimension. When the cues are less explicit or less well-learned, however, the process of translating those cues into more direct task instructions is needed at least on some proportion of trials, and this translation process seems to implicate the retrieval of task names via inner speech. As these considerations suggest, the main findings and conclusions from the current study are consistent with those from previous dual-task studies using the list and

16 138 A. Miyake et al. / Acta Psychologica 115 (2004) precued task sequence paradigms (Baddeley et al., 2001; Emerson & Miyake, 2003; Goschke, 2000; Saeki & Saito, in press): Although participants often recruit inner speech to support their task switching performance, they do so only when there are no strong or well-learned external cues available that automatically activate the appropriate task goal information or even automatically direct their attention toward the appropriate stimulus dimension. In this sense, inner speech may not be a necessary component of task switching performance, but it serves as an efficient and reliable tool or supporting mechanism that participants count on when the need for exerting endogenous control is strong (Emerson & Miyake, 2003). The current study goes beyond the previous studies by focusing on a paradigm that eliminates the need for actively keeping track of the task sequence (i.e., the random task cuing paradigm) and showing more clearly that one function of inner speech is to serve as a retrieval aid for task goals. Despite these overall commonalities between the current study and the previous studies, there are also some differences between the random task cuing paradigm and the list or precued task sequence paradigm concerning the effect of articulatory suppression. Perhaps the most obvious difference is that the absolute magnitude of the increase in switch cost caused by concurrent articulatory suppression (compared to the control condition) is much larger in the list paradigm than in the random task cuing paradigm. For example, in the Emerson and Miyake (2003) experiments (Experiments 1 4), the switch cost increases due to articulatory suppression were in the ms range, but, in the current study, the increase was only 21 ms in the word cue condition (42 ms cost in the control-long condition vs. 63 ms cost in the articulatory suppression condition) and 73 ms in the letter cue condition (43 ms cost vs. 116 ms cost, respectively). One possible reason for this marked difference in the absolute magnitude of the articulatory suppression effect between the two paradigms may be that the role of inner speech in the list paradigm goes significantly beyond the verbally mediated retrieval of task goals. As we emphasized earlier, one key characteristic of the list paradigm absent in the random task cuing paradigm is that participants are required to keep track of the task sequence for the alternating-task lists (e.g., alternating between addition and subtraction). Given that language is inherently linear and may be suitable for sequencing information (Carlson, 1997; Emerson & Miyake, 2003), it is likely that, in the list paradigm, inner speech was responsible not just for retrieving and activating task goals but also for actively keeping track of the task sequence for the alternating lists, although neither Baddeley et al. (2001) nor Emerson and Miyake (2003) clearly differentiated these potentially separable functions of inner speech in their discussion. We should note, however, that these increases in switch costs represent raw numbers and do not take into account the magnitudes of the switch costs in the control conditions, which were quite different across the two paradigms. In fact, when the negative effect of articulatory suppression on switch costs is expressed in terms of percentage increase using the switch cost in the control condition as the baseline, the results from the two paradigms are much more similar, with percentage increases of % in the Emerson and Miyake (2003) experiments and percentage increases

17 A. Miyake et al. / Acta Psychologica 115 (2004) of 50% and 170% in the current study for the word and letter cue conditions, respectively. Thus, although participants in the current study were not required to keep track of the task sequence, the percentage increase in cost due to articulatory suppression was substantial, particularly in the letter cues condition (170%). These observations suggest that, even though the absolute magnitude of the articulatory suppression effect may be different for the two paradigms, the contribution of inner speech in the random task cuing paradigm may be as substantial as that in the list paradigm, at least when the cues still impose some degree of retrieval demand. One interesting question raised by the results of the current study concerns which type of switch cost is affected by articulatory suppression. It is possible to calculate at least two types of switch cost, namely, (a) the cost associated with switching from one task to another (e.g., switch RT minus repeat RT for the mixed task trials), called local or specific switch costs, and (b) the cost of mixing two tasks in a trial sequence compared to always performing one task (e.g., repeat RT from the mixedtask trials minus RT for the pure-task trials), called global or mixing costs (Kray & Lindenberger, 2000; Mayr, 2001). These two types of switch costs are potentially dissociable, as indicated by the findings that reliable age-related changes have been observed for the latter type of switch cost (Kray & Lindenberger, 2000; Mayr, 2001; Meiran & Gotler, 2001), but not for the former type (Kray & Lindenberger, 2000; Mayr & Liebscher, 2001; Salthouse, Fristoe, McGuthry, & Hambrick, 1998; see Kray, Li, & Lindenberger, 2002, for an exception). The results of the current study have shown that, unlike aging, articulatory suppression significantly affects local switch costs (i.e., repeat vs. switch RTs in mixedtask trials), at least when the letter cues are presented (Fig. 3B). We interpreted this effect specific to the letter cue condition in terms of the involvement of inner speech in the retrieval of task goals when the cues are not direct or explicit enough. In addition, a comparison of the pure-task RTs (Fig. 2) with the repeat trial RTs from the mixed-task blocks (Fig. 3) suggests that global costs also significantly increase with concurrent articulatory suppression. In fact, it is interesting that, although the RTs were significantly longer in the foot tapping condition (M ¼ 440 ms) than in the articulatory suppression condition (M ¼ 405 ms) for the pure-task trials (as reported earlier), the RTs for the repeat trials of the mixed task blocks show the opposite pattern: The RTs were significantly longer under articulatory suppression than under foot tapping for both the word cue condition (M ¼ 560 ms vs. 509 ms, respectively), F ð1; 47Þ ¼10:59, MSE ¼ 0:004, p ¼ 0:002, g 2 ¼ 0:183 and the letter cue condition (M ¼ 582 ms vs. 521 ms), F ð1; 47Þ ¼14:16, MSE ¼ 0:004, p < 0:001, g 2 ¼ 0:231. One interpretation of this intriguing reversal of the dual-task effect is that, for the mixed-task trials, participants use inner speech to retrieve the task goal information even for task-repeat trials, perhaps because the task order is unpredictable in the random task cuing paradigm. Although this account seems plausible and deserves further investigation, one finding that seems somewhat inconsistent with this proposal is that the RT difference between the articulatory suppression and foot tapping conditions for the repeat trials of the mixed-task blocks was equivalent for the word cues and for the letter cues (51 ms vs. 61 ms, respectively), F ð1; 47Þ ¼0:18, MSE ¼ 0:003, p ¼ 0:674, g 2 ¼ 0:004.