Nick: *Jerry*
Oggetto: re:ESISTE SESSO SENZA AMORE 2
Data: 28/3/2004 20.6.51
Visite: 51
Beh innazitutto mi preme dire che uno sfruttamento reciproco sta alla base di tutti i processi di anodilatazione, ed è per questo che, in un'ottica tutta speculativa, non si dovrebbero inibire, ma anzi incentivare, per coloro i quali (come l'autore ed iniziatore del thread) favoriscono tale dilatazione onde agevolare più intimi scambi di vedute.
Mi pare sufficiente, in merito agli studi dell'Università di Berlino del 1997, citare soltanto alcune frasi degli eminenti scienziati autori dei test:
"il sesso senza amore è possibile dove la stimolazione dell'asse pituitario-gastronomico è indotta da precoci rapporti uomo-scimmia o anche uomo-pasilda"
(Prof. Ernest Faggot, Università di Berlino)
"la storia dell'olio di lino, tutta si evince dai cerchi concentrici individuabili nel tronco degli alberi di granoturco"
(Prof. Gordon Craig, Università di Berlino)
"il vino buono, come amava dire Michael Bubble, viene dalla distruzione del socialismo"
(Prof. Lugné Poe, Prof.ssa Eleonora Duse, Università di Brema)
Ora, stando ai pareri dei più autorevoli esponenti del dadaismo (nonché di alcune menti eccelse del positivismo) possiamo affermare che sì, un eccessivo aumento del GH endogeno può creare situazioni di squilibrio ormonale e feedbacks di varia natura.
Basandoci sullo studio che qui provvedo ad esporre, e cioé:
Executive Control of Cognitive Processes in Task Switching
Joshua S. Rubinstein
William J. Hughes Technical Center
Federal Aviation Administration
David E. Meyer
Department of Psychology
University of Michigan
Jeffrey E. Evans
Department of Rehabilitation Psychology and Neuropsychology
University of Michigan Medical Center
--------------------------------------------------------------------------------
ABSTRACT
In 4 experiments, participants alternated between different tasks or performed the same task repeatedly. The tasks for 2 of the experiments required responding to geometric objects in terms of alternative classification rules, and the tasks for the other 2 experiments required solving arithmetic problems in terms of alternative numerical operations. Performance was measured as a function of whether the tasks were familiar or unfamiliar, the rules were simple or complex, and visual cues were present or absent about which tasks should be performed. Task alternation yielded switching-time costs that increased with rule complexity but decreased with task cuing. These factor effects were additive, supporting a model of executive control that has goal-shifting and rule-activation stages for task switching. It appears that rule activation takes more time for switching from familiar to unfamiliar tasks than for switching in the opposite direction.
--------------------------------------------------------------------------------
Traditionally, experimental psychology has focused on studying repetitive performance of individual perceptual-motor and cognitive tasks. Nevertheless, daily life often requires performing multiple tasks either simultaneously or in rapid alternation, as when people prepare meals while tending children or drive automobiles while operating cellular telephones. To explain how such multiple-task performance is achieved, some theorists have proposed that executive control processes supervise the selection, initiation, execution, and termination of each task (e.g., Baddeley, 1986 ; Duncan, 1986 ; Logan, 1985 ; Meyer & Kieras, 1997a , 1997b ; Norman & Shallice, 1986 ; Shiffrin & Schneider, 1977 ). These proposals extend classical ideas about voluntary willed action ( James, 1890 ), which may be elaborated in terms of concepts from computer science and multitasking operating systems ( Kieras, Meyer, Ballas, & Lauber, 2000 ; Neisser, 1967 ).
Given this state of affairs, research on executive mental control requires asking various detailed analytical questions ( Monsell, 1996 ). Are executive control processes really separable from the basic perceptual-motor and cognitive processes used for performing individual tasks? How might executive control processes establish priorities among individual tasks and allocate resources to them during multiple-task performance? Of what functionally distinct subcomponents do executive control processes consist? The present article provides further answers to such questions through experiments with a successive-tasks procedure developed and used previously for studying executive control processes that enable task switching (e.g., Allport, Styles, & Hsieh, 1994 ; Botwinick, Brinley, & Robbin, 1958 ; Dark, 1990 ; Garcia-Ogueta, 1993 ; Jersild, 1927 ; Keele & Hawkins, 1982 ; Los, 1996 ; Meiran, 1996 ; Meiselman, 1974 ; Rogers & Monsell, 1995 ; Spector & Biederman, 1976 ; Weber, Burt, & Noll, 1986 ).
In subsequent sections of this article, we start by briefly reviewing available theories of executive control processes. Some past studies of task switching whose results bear on the veracity of these theories are summarized next. Then a new model of executive mental control in task switching is introduced. To test this model and to demonstrate its potential heuristic value, we report four experiments with the successive-tasks procedure. On the basis of data from them, we propose that task switching entails at least two functionally distinct stages of executive control, goal shifting and rule activation, which are separable from the basic perceptual-motor and cognitive processes used for performing individual tasks. Our proposed stage model provides coherent explanations of numerous previous findings about task switching and suggests promising directions for future research on executive mental control.
Theories of Executive Control Processes
For now, we focus on three representative theories: the attention-to-action (ATA) model ( Norman & Shallice, 1986 ), the frontal-lobe executive (FLE) model ( Duncan, 1986 ), and the strategic response-deferment (SRD) model ( Meyer & Kieras, 1997a , 1997b , 1999 ). These theories are especially relevant because they exemplify how task switching might be mediated by separable executive control processes that prepare systematically for transitions between successive tasks.
Attention-to-Action Model
The ATA model of Norman and Shallice (1986) has three subcomponents: action schemas, contention scheduling, and a supervisory attentional system (SAS).
Action schemas are specialized routines for performing individual tasks that involve well-learned perceptual-motor and cognitive skills. Each action schema has a current degree of activation that may be increased by either specific perceptual "trigger" stimuli or outputs from other related schemas. When its activation exceeds a preset threshold, an action schema may direct a person's behavior immediately and stereotypically toward performing some task. Moreover, on occasion, multiple schemas may be activated simultaneously by different trigger stimuli, creating error-prone conflicts if they entail mutually exclusive responses (e.g., typing on a keyboard and answering a telephone concurrently).
To help resolve such conflicts, the ATA model uses contention scheduling. It functions rapidly, automatically, and unconsciously through a network of lateral inhibitory connections among action schemas whose response outputs would interfere with each other (cf. Rumelhart & Norman, 1982 ). Through this network, an action schema (e.g., one for keyboard typing) that has relatively high current activation may suppress the activation of other potentially conflicting schemas (e.g., one for telephone answering). Contention scheduling allows task priorities and environmental cues to be assessed on a decentralized basis without explicit top-down executive control ( Shallice, 1988 ). However, this may not always suffice to handle conflicts when new tasks, unusual task combinations, or complex behaviors are involved.
Consequently, the ATA model also has an SAS. The SAS guides behavior slowly, flexibly, and consciously in a top-down manner. It helps organize complex actions and perform novel tasks by selectively activating or inhibiting particular action schemas, superseding the cruder bottom-up influences of contention scheduling and better accommodating a person's overall capacities and goals. For example, one might expect the SAS to play a crucial role during switches between unfamiliar incompatible tasks that are not ordinarily performed together.
Depending on conditions that prevail during multiple-task performance, the ATA model accounts qualitatively for a variety of empirical phenomena. In particular, slips of action that occur during daily activities (e.g., Reason, 1990 ) may stem from temporary failures of the SAS to regulate contention scheduling adequately. SAS failures may also explain behavioral abnormalities in patients with frontal-lobe brain damage ( Shallice, 1982 , 1988 , 1994 ).
Frontal-Lobe Executive Model
Assumptions similar to those of the ATA model have been embodied in the FLE model of Duncan (1986) . It has three main components: goal lists, means-ends analysis procedures, and action structures. Goal lists represent a person's current set of prioritized intentions. Means-ends analysis, somewhat like the SAS (cf. Norman & Shallice, 1986 ), updates the contents and order of goals in working memory, taking account of how well they are being achieved over time. Supplementing such functions, the action structures of the FLE model constitute a large store of procedural knowledge for goal-directed behaviors embodied as sets of condition-action production rules (cf. Allport, 1980 ; J. R. Anderson, 1983 , 1993 ; Hunt & Lansman, 1986 ; Logan, 1985 ; Newell, 1973 , 1990 ; Townsend, 1986 ). The conditions of these rules refer to goals and perceptual stimuli; the actions involve responses to achieve the goals (e.g., IF THE GOAL IS TO DO TASK A AND THE STIMULUS IS S, THEN PRODUCE RESPONSE R ). Action structures composed of such rules are functionally analogous to the ATA model's action schemas.
Furthermore, according to Duncan (1986) , goal lists and means-ends analysis are implemented primarily in the brain's frontal lobes. The FLE model implies that damage to particular frontal-lobe regions may disrupt people's ability to maintain and pursue their goals, reducing their effectiveness in planning and performing multiple tasks. This implication also agrees with claims of some other theorists (e.g., Kimberg & Farah, 1993 ; Shallice, 1994 ; Stuss & Benson, 1986 ).
Strategic Response-Deferment Model
Additional detailed ideas about how executive control contributes to multiple-task performance have been provided by Meyer and Kieras (1997a , 1997b , 1999) . Using a production-rule formalism, they constructed an executive-process interactive control (EPIC) architecture that combines various components of the human information-processing system in a unified theoretical framework (cf. J. R. Anderson, 1983 , 1993 ; Card, Moran, & Newell, 1983 ; Newell, 1990 ). EPIC includes perceptual, cognitive, and motor processors interfaced with working-memory stores whereby multiple-task performance can be described computationally. For example, on the basis of EPIC, Meyer and Kieras proposed an SRD model that simulates performance in a traditional dual-task paradigm, the psychological refractory-period (PRP) procedure ( Bertelson, 1966 ; Kantowitz, 1974 ; Pashler, 1994 ; Smith, 1967 ; Welford, 1952 , 1959 , 1967 ).
The PRP procedure exemplifies a simultaneous-tasks procedure. On each discrete trial of this procedure, a stimulus is presented for the first of two tasks that entail stages of processing such as stimulus identification, response selection, and movement production. In response to the Task 1 stimulus, a participant must react quickly and accurately. Soon after the Task 1 stimulus, another stimulus is presented for the second task, separated by a short (e.g., <=1 s) stimulus-onset asynchrony (SOA). In response to the Task 2 stimulus, the participant must again react quickly and accurately. However, instructions for the PRP procedure require that Task 1 receive higher priority than Task 2 (e.g., Pashler, 1984 ; Pashler & Johnston, 1989 ), and reaction times (RTs) are measured to assess the extent to which the two tasks interfere with each other.
To characterize this interference, the SRD model of Meyer and Kieras (1997a , 1997b , 1999) assumes that performance during the PRP procedure involves three sets of production rules. One rule set implements operations for Task 1 (e.g., selecting Task 1 responses). A second rule set implements operations for Task 2 (e.g., selecting Task 2 responses). A third executive-process rule set schedules these operations so that instructions about task priorities are obeyed and conflicts do not occur over the use of limited-capacity perceptual-motor components. By manipulating task goals and strategy notes in working memory, the executive process permits the Task 1 production rules to select and send Task 1 responses to an appropriate motor (e.g., manual or vocal) processor as soon as possible, regardless of the SOA. Task 2 production rules are also permitted to select Task 2 responses concurrently with Task 1 response selection (cf. Pashler, 1994 ; Welford, 1967 ). At short SOAs, however, the model's executive process defers movement production for Task 2 by storing selected Task 2 responses temporarily in working memory until Task 1 has been completed. This ensures that Task 2 responses do not inadvertently precede or interfere with Task 1 responses at peripheral levels. After completion of Task 1 on a trial, the executive process permits any previously selected and stored Task 2 response to be produced. Also, if Task 2 response selection has not yet started, then a subsequently selected Task 2 response is permitted to be produced immediately. Such temporal overlapping and interleaving of Task 1 and Task 2 processes accounts well for patterns of additive and interactive factor effects on empirical mean RTs from the PRP procedure (e.g., Hawkins, Rodriguez, & Reicher, 1979 ; Karlin & Kestenbaum, 1968 ; McCann & Johnston, 1992 ; Meyer et al., 1995 ; Pashler, 1990 ; Schumacher et al., 1999 , 2001 ). In essence, the SRD model demonstrates how some basic ideas from the ATA and FLE models can be formalized and tested successfully against quantitative data.
Tentative Theoretical Hypotheses
Given the success of these models in accounting qualitatively and quantitatively for major phenomena associated with multiple-task performance, some interesting theoretical hypotheses may be advanced. Perhaps executive control processes really do exist, and perhaps they incorporate multiple separable subcomponents that enable task switching. Thus, subsequent sections of this article consider these hypotheses and ways of testing them further.
Task Switching and the Successive-Tasks Procedure
Some evidence about the existence and separability of component executive control processes comes from a successive-tasks procedure for studying task switching. The successive-tasks procedure is similar in certain respects to the PRP simultaneous-tasks procedure mentioned earlier. However, there are also conceptually important differences between these two procedures. In what follows, we discuss the successive-tasks procedure more fully, and we summarize representative results that have been obtained with it.
Successive-Tasks Procedure
Several basic features characterize the successive-tasks procedure ( Monsell, 1996 ).
Assignment of task priorities. When the procedure is implemented, equal priorities are typically assigned to the individual tasks between which participants must switch. This assignment contrasts with that of the PRP simultaneous-tasks procedure, wherein one task is primary and the other secondary. Consequently, scheduling the stages of processing for the successive-tasks procedure may be relatively simple in certain respects, lessening the demands placed on executive mental control (cf. Kieras et al., 2000 ).
Temporal sequence of stimulus events. The assignment of equal task priorities is encouraged by the temporal sequence of stimulus events during the successive-tasks procedure. In this procedure, the stimulus for the next task is never presented until after a response to the current task stimulus has occurred. This constrains the response-stimulus interval (RSI) to be nonnegative and the SOAs to be all relatively long (i.e., equal to or greater than concomitant RTs). Thus, unlike the PRP simultaneous-tasks procedure, the successive-tasks procedure provides little, if any, opportunity to overlap the stages of processing for two or more tasks. Again, this may lessen the demands imposed on executive mental control.
Composition of stimulus-response mappings. Nevertheless, these demands can still be substantial because of the stimulus-response (S-R) mappings that are typically used during the successive-tasks procedure. Here the stimuli and responses are often the same for all of the tasks; one task's S-R mapping may differ from another's only in terms of which specific responses are associated with which specific stimuli. Consequently, under this procedure, task switching is potentially susceptible to proactive interference reminiscent of what occurs during verbal learning and memory ( Allport et al., 1994 ; cf. Crowder, 1976 ). To cope with such interference, executive control processes may need to incorporate response monitoring and inhibitory mechanisms. 1
Theoretical relevance. Because of its characteristic features, the successive-tasks procedure is especially relevant to addressing some important issues about the nature of multiple-task performance. It allows an investigator to examine how executive control processes enable task switching when task stimuli do not overlap temporally and responses to them need not be selected or produced in parallel, but alternative S-R mappings may induce considerable proactive interference between tasks. As discussed next, past studies conducted under such conditions have yielded many informative results (for other reviews of the literature, see Monsell, 1996 ; Monsell, Yeung, & Azuma, 2000 ).
Jersild's (1927) Study
An influential early study with a precursor of the successive-tasks procedure was conducted by Jersild (1927) . 2 It yielded substantial time costs of task switching whose magnitudes depended on the complexity of the operations that were performed during each task. This dependence could bear on the nature of underlying executive control processes.
In one experiment, Jersild gave participants columns of two-digit stimulus numbers. Proceeding down each column, the participants performed the same arithmetic task (e.g., adding 6 and reporting the sum verbally) with respect to each stimulus number in the column, or they alternated between two different tasks (e.g., adding 6 to the first stimulus number and reporting the sum verbally, subtracting 3 from the second stimulus number and reporting the difference, adding 6 to the third stimulus number and reporting the sum, etc.). The complexity of the required arithmetic operations was either relatively low (e.g., adding 6 and subtracting 3) or high (e.g., adding 17 and subtracting 13). Mean times taken to complete the columns of stimuli were measured as a function of operation complexity and task alternation versus repetition. High-complexity operations took longer on average. Task switching also increased the mean completion times. These two effects interacted reliably; the difference between completion times for task alternation and repetition was greater when the tasks required high-complexity operations.
According to the logic of Sternberg's (1969) additive-factor method, this interaction suggests that operation complexity and task switching influence at least one stage of processing in common. The affected stage may involve some type of executive control process. For example, it might serve to activate the rules used in performing each successive task.
Spector and Biederman's (1976) Study
Extending Jersild's (1927) research, a further study with various versions of the successive-tasks procedure was conducted by Spector and Biederman (1976) . It revealed that the sizes of switching-time costs depend on visual cues about what task should be performed next. This dependence suggests that there is an executive control process through which such cues are used along with other stored information to identify and prepare for impending tasks.
In one experiment, Spector and Biederman gave participants columns of two-digit stimulus numbers. For each column, the participants added 3 to every stimulus number and reported the sum verbally, subtracted 3 from every stimulus number and reported the difference, or alternated between adding and subtracting 3. No visible cues were presented to indicate which arithmetic operation should be performed next; instead, the relevant operations had to be recalled from memory. Under these conditions, task alternation took substantially more time than task repetition, as Jersild (1927) found.
In another experiment, Spector and Biederman modified their procedure, appending explicit visual cues (e.g., "+3" or "-3") to the stimuli that indicated which arithmetic operations should be performed. Task alternation still took extra time, but the switching-time cost was markedly lower than when participants were not explicitly cued about the next required operation. The reduction in the switching-time cost could stem from a contribution of task cuing to executive mental control. For example, there may be a control process that identifies what task should be performed next. This process may be facilitated by relevant external information, which helps forego time-consuming memory retrieval.
Yet not all investigators would attribute Spector and Biederman's (1976) or Jersild's (1927) results to anticipatory components of executive mental control. Instead, Allport et al. (1994) hypothesized that time costs of task switching stem from task-set inertia (TSI), a type of proactive interference between conflicting S-R mappings for successive tasks. Support for this hypothesis was provided by a study that Allport et al. conducted with the successive-tasks procedure.
Allport et al.'s (1994) Study
The study by Allport et al. (1994) yielded several sets of results (see Table 1 ). Some of them have been claimed to show that significant TSI occurs and that anticipatory executive control processes play little if any role in task switching. Nevertheless, other results of Allport et al. appear more consistent with executive mental control than with TSI. What led to this ambiguous state of affairs is discussed next in more detail.
Evidence against executive control processes. Some putative evidence against the importance of executive control processes for task switching is that switching-time costs may not depend on the scope of the switches (see Table 1 , Result A1). Allport et al. (1994 , Experiment 1) found this by presenting visual stimulus displays that contained multiple copies of a particular printed digit. For these displays, participants performed four alternative tasks with different S-R mappings defined by which stimulus attributes and response criteria were relevant. The tasks involved (a) saying whether the magnitude (i.e., absolute value) of the displayed digit was "odd" or "even," (b) saying whether the digit's magnitude was "more" or "less" than 5, (c) saying whether the numerosity of the digit's copies was "odd" or "even," or (d) saying whether this numerosity was "more" or "less" than 5. When participants alternated between two tasks that differed only in their relevant stimulus attributes (i.e., magnitude vs. numerosity), the mean RT was about 1,100 ms longer than for repetitive task performance. Approximately the same 1,100-ms switching-time cost occurred for alternations between two tasks that differed only in their response criteria (i.e., odd-even vs. more-less). Moreover, the switching-time cost was approximately the same for alternations between two tasks that differed in both their relevant stimulus attributes and response criteria. Widening the scope of the switches did not increase their time cost significantly.
According to Allport et al. (1994) , this invariance suggests that a "unitary central executive" does not mediate task switching. Their theoretical interpretation assumed that executive control processes have limited capacity and that more mental workload is imposed on these processes by switching between tasks whose relevant stimulus attributes and response criteria both differ. If so, then switching-time costs should increase with greater workload. However, such an increase did not occur empirically, which led Allport et al. to conclude that task switching involves time-consuming processes other than executive mental control per se.
More putative evidence against the importance of executive control processes for task switching is that persistent switching-time costs may occur as RSIs increase (see Table 1 , Result A2). Allport et al. (1994 , Experiment 5) found this by having participants perform four more tasks that involved alternative S-R mappings: vocally naming the colors of fonts in which different color words (e.g., the word red with blue ink) were printed (the standard Stroop task; MacLeod, 1991 ; Stroop, 1935 ), naming the colors of fonts in which rows of Xs were printed (the Stroop control task ), reading color words that were printed in fonts of different colors (the reverse Stroop task ), and reading color words that were printed in black font (the reverse Stroop control task ). For the reverse Stroop task, mean RTs were longer when it was performed in alternation with the standard Stroop task than when it was performed repetitively. This difference changed relatively little with the RSI. After RSIs of 20 and 1,100 ms, mean switching-time costs were about 180 ms and 135 ms, respectively.
From these results, Allport et al. (1994) again inferred that task switching entails little, if any, anticipatory executive mental control. They reasoned as follows. Suppose that executive control processes do mediate task switching and that these processes commence at the start of the RSI. Also, suppose that switching-time costs stem from the duration of these processes. Then after relatively short RSIs, some switching-time cost should occur. In contrast, after longer RSIs, there should be no switching-time cost; the executive control processes should finish before the next task's stimulus is presented, which would preclude them from contributing to RTs for the next task. However, for the reverse Stroop task, this expected pattern of results failed to occur ( Allport et al., 1994 , Experiment 5), casting doubt on whether executive control processes mediated switching to this task.
Also relevant is a further result from performance of the reverse Stroop and standard Stroop tasks ( Allport et al., 1994 , Experiment 5). Mean RTs for the standard Stroop task manifested almost no switching-time costs even when the RSI was very short (see Table 1 , Result A3). Allport et al. (1994 , Experiment 4) found other cases of null switching-time costs as well. This seems hard to reconcile with executive control processes that play a prominent anticipatory role in task switching. If task switching involves these processes, then after short RSIs, should they not generally yield substantial switching-time costs?
Evidence for task-set inertia. To explain why task switching sometimes but not always produces substantial switching-time costs, Allport et al. (1994) proposed the TSI hypothesis. It is based on two assumptions: (a) Performance of a prior task requires imposing a particular "task set" that increases the primacy of the task's S-R mapping and may also suppress other competing S-R mappings, and (b) the prior task's S-R mapping remains partially active even after long RSIs, potentially interfering with selection of responses for other subsequent tasks. According to Allport et al., this proactive interference is higher when the stimuli and responses for prior and subsequent tasks are similar and when a prior task involves a less dominant S-R mapping than does a subsequent task. 3
Some putative evidence for the TSI hypothesis is that switching-time costs may be very small when participants alternate between two tasks whose S-R mappings are dissimilar (see Table 1 , Result B1). For example, such a result occurred in the first phase of another experiment by Allport et al. (1994 , Experiment 4). Here a new group of participants performed only two tasks at the outset: reverse Stroop and digit-magnitude judgment. The stimuli, responses, and S-R mapping for each task differed from those of the other task. After participants had completed a few blocks of practice with the tasks, switching-time costs approached zero; they were much lower than when previous participants had alternated between the reverse Stroop and standard Stroop tasks (cf. Allport et al., 1994 , Experiment 1). Under the TSI hypothesis, this is what should have occurred given that proactive interference presumably influenced task switching in the former but not the latter experiment.
A second piece of putative evidence for the TSI hypothesis is that prior experience with related but currently irrelevant tasks may increase switching-time costs considerably (see Table 1 , Result B2). For example, this result occurred in a further phase of the experiment described earlier ( Allport et al., 1994 , Experiment 4). Here participants again performed the same two (reverse Stroop and magnitude judgment) tasks. In the interim, however, they performed two new tasks (Stroop color naming and digit-numerosity judgment) that were related to the preceding ones. After this additional experience, a large increase occurred in the switching-time costs for the reverse Stroop and magnitude-judgment tasks. Such an outcome follows naturally from the TSI hypothesis; residual proactive interference from the S-R mappings of the intervening new tasks may have impeded subsequent performance of the reverse Stroop and magnitude-judgment tasks on alternating-task blocks.
Evidence against task-set inertia. Nevertheless, there is also considerable evidence against the TSI hypothesis. Under some conditions, almost no switching-time cost may occur when TSI should be present (see Table 1 , Result C1). For example, let us again consider what happened when Allport et al. (1994 , Experiment 5) had participants alternate between the reverse and standard Stroop tasks. Here mean word reading RTs were significantly longer than those obtained when participants alternated between the corresponding control tasks. This result suggests that in alternating-task blocks, having to name ink colors for the standard Stroop task caused interference with subsequent word reading for the reverse Stroop task. Thus, under the TSI hypothesis, participants should have suppressed color naming and imposed a word reading task set for the reverse Stroop task in alternating-task blocks. In turn, such regulation should have caused a significant time cost for switching back to the standard Stroop task, which involves color naming rather than word reading. However, contrary to this prediction, the mean switching-time cost for the standard Stroop task was virtually nil.
Further evidence against the TSI hypothesis is that significant switching-time costs may occur when TSI should be absent (see Table 1 , Result C2). In particular, this occurred during performance of the control tasks for which participants named colored patches and read color words printed with black ink ( Allport et al., 1994 , Experiment 5). There, task alternation took longer than task repetition, even though the required S-R mappings should not have interfered with each other (i.e., their respective stimulus sets had no shared perceptual features). Such results have been obtained as well under other conditions in which TSI was presumably absent (e.g., Allport et al., 1994 , Experiment 3). It therefore appears that some source other than TSI contributes to the time cost of task switching.
What might this other source be? Of course, one possibility is executive mental control. Even if the S-R mappings for two different tasks are dissimilar, supervisory shifts of task set may be required to alternate between them.
Evidence for executive control processes. Additional evidence for the existence and separability of executive control processes is that switching-time costs, although substantial in size, may be unaffected by manipulations of within-task difficulty ( Table 1 , Result D1). For example, the Stroop and numerosity-judgment tasks of Allport et al. (1994 , Experiment 3) yielded considerably longer mean RTs than did the reverse Stroop and magnitude-judgment tasks. However, these RT differences were about the same on alternating-task and repetitive-task blocks; the difficulty of the individual tasks did not affect mean switching-time costs significantly. Allport et al. (1994 , Experiments 1-3) also reported several other cases of switching-time costs that were unaffected by task difficulty. Given the logic of Sternberg's (1969) additive-factor method, such data suggest that task switching and task difficulty may influence temporally separate, functionally independent stages of processing. Perhaps executive control processes mediate the effects of task switching, whereas other subordinate processes (e.g., stimulus identification, response selection, and movement production) mediate the effects of task difficulty.
Still, to maintain the latter theoretical interpretation, Allport et al.'s (1994) other results must be reconciled with it. For example, why did their long RSIs not eliminate the time cost of task switching? In answer, a study by Rogers and Monsell (1995) is relevant.
Rogers and Monsell's (1995) Study
Rogers and Monsell (1995) used a version of the successive-tasks procedure called the alternating-runs paradigm. During each trial block, runs of two or more successive trials for one task alternated with runs of two or more trials for another task. One task involved pressing keys to indicate whether printed digits were odd or even; the other task involved pressing keys to indicate whether printed letters were consonants or vowels. The stimulus display on each trial contained two characters, one relevant and the other irrelevant for the current task. Some of the irrelevant characters were either congruent or incongruent with impending responses; they came from the stimulus ensemble of the noncurrent task and corresponded respectively to keypresses that would be correct or incorrect for the current task. Other irrelevant characters were neutral (i.e., they did not come from the stimulus ensemble of either task). The spatial location of the stimulus display cued participants about which task should be performed next. RTs were measured as a function of the RSI and other stimulus factors. From these measurements, several instructive findings about the nature of executive control processes emerged.
Irrelevant-character effects. Incongruent irrelevant characters induced the largest switching-time costs ( Rogers & Monsell, 1995 , Experiment 1). This is consistent with the TSI hypothesis ( Allport et al., 1994 ). One would expect proactive interference from a previously applicable S-R mapping to be highest for such stimulus displays, thereby slowing responses especially on trials that require task switching.
However, the TSI hypothesis cannot explain other results of Rogers and Monsell (1995) so well. For example, a substantial switching-time cost also occurred in the context of neutral irrelevant characters, even though they presumably induced no proactive interference with the current task. What might the source of this particular cost be? A possible answer is that executive control processes are needed to switch between tasks regardless of which irrelevant characters appear in a stimulus display.
Response-stimulus interval effects. The latter possibility may be evaluated further from patterns of RSI effects. Under some conditions, Rogers and Monsell (1995 , Experiment 2)–like Allport et al. (1994 , Experiment 5)–found that switching-time costs did not vanish as RSIs increased. This occurred when the lengths of the RSIs varied within trial blocks. Nevertheless, when the RSIs all had the same length within a trial block, but their lengths varied between blocks, switching-time costs were substantially lower after longer RSIs ( Rogers & Monsell, 1995 , Experiment 3). The blocked-RSI effect was approximately additive with the irrelevant-character effect on switching-time costs.
On the basis of these results, three conclusions can be reached ( Rogers & Monsell, 1995 ). First, RSI and irrelevant-character effects on switching-time costs may occur during distinct substages of executive mental control. Second, if the RSI is predictable, then it may be used for completing some of the operations needed to switch between tasks. Third, if an RSI is unpredictable, then these operations may be postponed until after the next task's stimulus has appeared.
Still, like Allport et al. (1994) , Rogers and Monsell never found that the time costs of task switching entirely vanished after long RSIs. Even during trial blocks with constant 1,200-ms RSIs, which provided ample opportunity for executive control processes to complete their anticipatory operations, there were reliable switching-time costs. This persistence might be attributed either to residual TSI or to executive mental control that is postponed until after the RSI has ended.
Task serial-position effect. To test these possibilities further, Rogers and Monsell (1995 , Experiment 6) used trial blocks with alternating runs of four trials per task. They reasoned that if the TSI hypothesis were correct, then proactive interference from a prior task should decay gradually, slowing responses not only in the first but also in the second and perhaps even third serial positions of each four-trial run. However, no evidence of gradually decaying proactive interference was obtained. Mean RTs in the second, third, and fourth serial positions of the four-trial runs were virtually identical to each other and all reliably shorter than the mean RT in the first serial position. Thus, it appeared as though, on each four-trial run, the switching-time cost may have stemmed from executive control processes that completed their operations before the first trial of the run ended.
Theoretical interpretation. On the basis of their results, Rogers and Monsell (1995) proposed a model of task switching with two distinct types of executive control: endogenous and exogenous. According to this model, endogenous control takes place in a flexible top-down manner, executing anticipatory operations for impending tasks during predictable RSIs. These operations decrease switching-time costs as the RSIs increase, accounting for blocked-RSI effects. However, they leave the system in a partially unprepared state. Exogenous control, which completes final preparations for the next task, is triggered by the onset of the next task's stimulus. The occurrence of the exogenous control process after stimulus onset could yield irrelevant-character effects. The temporal separation of exogenous and endogenous control processes may also account for why irrelevant-character and RSI effects on switching-time costs are approximately additive ( Lauber, 1995 ). Although Rogers and Monsell did not specify exactly what these processes do, more conclusions about them may be reached through the new experiments that we report in this article.
Other Relevant Studies
Rogers and Monsell's (1995) theoretical ideas have also been reinforced by some other studies. For example, Meiran (1996) gave visual precues to participants during a modified version of the successive-tasks procedure, informing them explicitly about what their next task would be. On trials that required task switches, the precues reduced switching-time costs more when the RSIs were long than when they were short. This supports the assumption of endogenous executive control. More such support has been reported by other investigators (e.g., Biederman, 1973 ; LaBerge, Petersen, & Norden, 1977 ; Logan & Zbrodoff, 1982 ; Sudevan & Taylor, 1987 ).
In addition, complementary evidence of exogenous executive control has been reported by Gopher, Armony, and Greenshpan (2000) . Again using visual precues, they had participants make occasional unpredictable switches between tasks. Significant switching-time costs occurred even though the precues were presented at the start of long (1,200-ms) RSIs. However, these costs did not extend beyond the particular trials on which the switches took place; there appeared to be no residue of gradually decaying TSI. This result, reminiscent of Rogers and Monsell's (1995 , Experiment 6) all-or-none task serial-position effect, is what would be expected if an exogenous control process completes its task-set shifting immediately and fully after the onset of the next task's stimulus.
Interim Summary
Although the difficulty of task switching may be partly attributable to sources (e.g., TSI) other than executive mental control per se, our literature review suggests that both endogenous and exogenous control processes probably help supervise task switching and contribute significantly to observed switching-time costs. These contributions can account for patterns of effects by factors such as task cuing, operation complexity, and RSI. Thus, further efforts to formulate and test detailed models of executive control for task switching are presumably warranted.
A Stage Model of Executive Control for Task Switching
Given the preceding considerations, the purpose of the present article is to formulate and test a model of executive control that accounts more fully for task cuing, operation complexity, RSI, and other related factor effects on the time costs of task switching. In what follows next, our model's assumptions are outlined. After this, we discuss how they can explain various results in the task-switching literature. Then four new experiments are reported to evaluate some additional predictions of the model.
A schematic diagram of the model appears in Figure 1 . According to the model, performance during the successive-tasks procedure entails two complementary sets of stages: executive control processes and task processes (for more discussion about each type of process, see Lauber, 1995 , and Kieras et al., 2000 ).
Task Processes
We assume that task processes are used for performing individual perceptual-motor and cognitive tasks under both single-task and multiple-task conditions. In our model, these processes include three principal stages, stimulus identification, response selection, and movement production, which operate on the basis of information in declarative and procedural working memory (cf. Donders, 1868/1969 ; Meyer & Kieras, 1997a , 1997b ; Sanders, 1980 ; Sternberg, 1969 ). The stimulus-identification stage encodes perceptual features of stimuli and places them in declarative working memory for access during the response-selection stage. Through algorithms in procedural working memory, the response-selection stage converts the stimulus codes to abstract response codes. The movement-production stage converts the response codes to motor commands that generate overt physical action. Component operations in each stage are assumed to be tailored to the tasks' particular sensory modalities, response modalities, and S-R mappings.
Regarding the response-selection stage, we further assume that it uses production rules in procedural working memory, which specify actions to be executed whenever prerequisite conditions match the current contents of declarative working memory. For example, suppose that a task requires pressing finger keys in response to stimulus colors. Then a production rule for response selection might have the following form:
IF ((GOAL IS TO DO COLOR-DISCRIMINATION TASK) AND (STIMULUS COLOR IS RED)) THEN (PRESS RIGHT INDEX-FINGER KEY).
The numerosity and complexity of such rules depend on the task's S-R mapping, thereby affecting the duration of the response-selection stage ( Meyer & Kieras, 1997a , 1997b ).
When the same task is performed repetitively, response selection in the model starts immediately after stimulus identification on each trial. However, we assume that if a switch occurs from one task to another, there is a pause between the end of stimulus identification and the beginning of response selection for the current task (see Figure 1 ). This pause is used by an executive control process whose operations enable the subsequent response-selection stage to proceed correctly.
Executive Control Processes
To enable task switching, the model's executive control processes include two distinct stages, goal shifting and rule activation, which are accomplished through executive production rules. Together, goal shifting and rule activation respectively ensure that the contents of declarative and procedural working memory are appropriately configured for the task at hand, consistent with proposals of some previous theorists (e.g., Duncan, 1986 ; Kimberg & Farah, 1993 ; Logan, 1985 ; Meyer & Kieras, 1997a , 1997b ; Rogers & Monsell, 1995 ). 4
Goal shifting. The goal-shifting stage keeps track of current and future tasks, inserting and deleting their goals in declarative working memory as needed. Specific goal items in working memory let other components of the system "know" what the current task is. For example, in switching from a shape-discrimination to a color-discrimination task, goal shifting might involve updating the contents of working memory through the following production rule:
IF ((GOAL IS TO DO SHAPE-DISCRIMINATION TASK) AND (SHAPE-DISCRIMINATION TASK IS DONE) AND (NEXT TASK IS COLOR DISCRIMINATION)) THEN (((DELETE (GOAL IS TO DO SHAPE-DISCRIMINATION TASK)) AND (INSERT (GOAL IS TO DO COLOR-DISCRIMINATION TASK))).
By the application of such rules, various bits of information for initiating, executing, and terminating individual tasks can be maintained.
Furthermore, we assume that the time at which goal shifting takes place relative to concomitant task processes is flexible. Under some conditions, goal shifting may occur before stimulus identification starts for the next task (see Figure 1 ). For example, this might happen if the RSI is long and prior information is available about what the next task will be. Then the goal-shifting stage would be an endogenous control process of the sort Rogers and Monsell (1995) have proposed. However, our model also allows goal shifting to occur after the next task's stimulus has been identified. Such delayed goal shifting might occur if the RSI is short or the task stimulus is expected to provide an explicit cue about what task must be performed next. Then goal shifting would be an exogenous control process.
Rule activation. In the model, rule activation is another executive control process for task switching. Because of reasons explained subsequently, we assume that under at least some conditions, this stage is triggered exogenously and takes place during a pause between the end of stimulus identification and the beginning of response selection for the current task, after goal shifting has finished (see Figure 1 ). Two complementary functions are served by rule activation: enabling the rules for selecting the current task's response and disabling the rules for selecting the prior task's response. After these functions have been completed, the current task's response-selection stage can proceed.
How is rule activation accomplished? One possibility is that this stage involves "loading" the next task's rules into procedural working memory, just as a computer operating system reads new application programs from disk to core memory, overwriting old programs in preparation for executing the new ones. Such operations might be initiated by the following executive production rule:
IF ((GOAL IS TO DO COLOR-DISCRIMINATION TASK) AND (STIMULUS COLOR HAS BEEN IDENTIFIED)) THEN (((LOAD (COLOR-DISCRIMINATION TASK RULES)) AND (INSERT (WAIT FOR COMPLETION OF LOADING))).
On the basis of the loading-of-rules metaphor, it seems plausible that the numerosity and complexity of a task's production rules could influence the duration of the rule-activation stage.
Another complementary possibility is that this stage involves temporarily raising the activation levels of the current task's production rules in procedural long-term memory (cf. J. R. Anderson, 1983 , 1993 ). During such a process, the activation levels of the previous task's rules might be allowed to drop back toward baseline, or an intentional operation to suppress them might occur (cf. Goschke, 2000 ; Mayr & Keele, 2000 ). If either of these possibilities holds, then procedural working memory would constitute the part of procedural long-term memory that is currently activated. This could help account for putative effects of TSI on task switching (cf. Allport et al., 1994 ; Allport & Wylie, 2000 ).
Theoretical rationale. Of course, our assumptions about rule activation lead to other questions. Why is this stage necessary? Why are the production rules for multiple tasks not kept simultaneously enabled in procedural working memory during the successive-tasks procedure? Why might rule activation be an exogenous rather than endogenous control process?
One conceivable answer to some of these questions is that procedural working memory has only enough capacity for a single task's rules. However, this seems implausible. Results from previous studies with the PRP simultaneous-tasks procedure suggest that, under at least some conditions, sets of rules for two distinct tasks can be held in procedural working memory and used concurrently during multiple-task performance ( Meyer & Kieras, 1997a , 1997b ). Thus, there must be some other rationale for our proposed rule-activation stage.
In particular, perhaps rule activation is needed because of the successive-tasks procedure's special characteristics. As mentioned before, this procedure typically involves tasks that have the same stimuli but different S-R mappings. Under such conditions, it may be suboptimal to keep the production rules for all of the tasks enabled in procedural working memory. Doing so could create conflict, disruption, and errors during the response-selection stage for one task or another, because alternative rules whose stimulus conditions are satisfied simultaneously would yield inappropriate–and even mutually exclusive–actions (cf. Cohen, Dunbar, & McClelland, 1990 ; Kornblum, Hasbroucq, & Osman, 1990 ; MacLeod, 1991 ). One solution to these problems would be to enable the rules for only one task at a time, as our rule-activation stage does.
Given these considerations, proactive interference and TSI might play a significant role during rule activation (cf. Allport et al., 1994 ; Allport & Wylie, 2000 ). Suppose that some of the features of the next task's stimulus match those of the conditions in a production rule for the preceding task. Also, suppose that the matching features enter declarative working memory before rule activation has finished for the next task. Then the occurrence of such partial matches could make it more difficult to disable the preceding task's rules, thereby prolonging the rule-activation stage ( Mayr & Keele, 2000 ).
These considerations could also justify having the rule-activation stage be an exogenous (stimulus-triggered) control process. Perhaps attending to relevant features of the next task's stimulus helps determine which production rules should be enabled for dealing with it. If so, then on each trial that involves task switching, rule activation might benefit from waiting until the stimulus for the next task has been identified.
Ancillary Technical Assumptions
To derive explanations and predictions from the present model, we make some further technical assumptions that are commonly associated with discrete stage models of human information processing ( Sanders, 1980 ; Sternberg, 1969 ).
Strict successiveness. The model's component stages, including both executive control and task processes, are strictly successive. Each stage starts only after its predecessors have finished.
Summation of stage durations. Theoretical RTs are sums of component stage durations. On trials without task switching, the summed durations of stimulus identification, response selection, and movement production constitute the RTs. On trials with task switching, these summed durations in combination with those of goal shifting and rule activation constitute the RTs.
Selective influence of factors. Some factors may selectively influence the component stage durations of different executive control and task processes. However, other factors may influence multiple stages, and some stages may be influenced by multiple factors.
Constant output quality. The quality of the outputs produced by a component stage is constant regardless of the factor effects on its duration.
Additivity and interaction of factor effects. Factors that selectively influence the durations of different component stages have additive effects on mean RTs. In contrast, the effects of factors that influence the same stage may interact. 5
Justification of Assumptions
Although discrete stage models have enjoyed considerable popularity ( Donders, 1868/1969 ; Luce, 1986 ; Meyer, Osman, Irwin, & Yantis, 1988 ; Miller, 1988 ; Pachella, 1974 ; Sanders, 1980 ; Sternberg, 1969 ; Townsend & Ashby, 1983 ), their relevance is conceivably limited. For example, McClelland (1979) argued that human cognition and action are typically mediated by a cascade of contingent-concurrent operations whose outputs consist of continuous, gradually increasing activation. Similar arguments might be made by theorists who favor connectionist-network models (e.g., Cohen et al., 1990 ). If they were empirically correct, then our model's assumptions would not strictly hold. Nor would Sternberg's (1969) additive-factor method be entirely applicable here.
Nevertheless, we have strong grounds for initially adopting a discrete stage model and the additive-factor method. In other domains of cognitive psychology (e.g., studies of visual word recognition), such models have been especially useful even when later replaced by alternative theoretical frameworks ( Meyer, Schvaneveldt, & Ruddy, 1975 ). Their simplicity and rigor provide powerful heuristics for conceptualizing basic processes of human performance. Furthermore, discrete stage models account well for a wide range of RT data ( Roberts & Sternberg, 1993 ; Sternberg, 1969 , 1998 ). Thus, it seems likely that they may yield important insights about executive control processes and task switching as well.
Explanation of Past Findings About Task Switching
Our stage model of task switching explains a variety of findings from past studies with the successive-tasks procedure. Many reported differences in switching-time costs may be attributed to factor effects on either goal shifting or rule activation. Perhaps such effects have sometimes been additive because of their disparate loci in the hypothesized sequence of processing stages.
Effects on goal shifting. For example, one factor that probably affects the goal-shifting stage is task cuing. As mentioned earlier, Spector and Biederman (1976 , Experiments 3 and 4) found smaller switching-time costs when alternative types of arithmetic problems were accompanied by corresponding operation signs. This decrease may have occurred because the operation signs cued participants with useful information about which task goal should be placed in declarative working memory next, thereby shortening the number of steps taken to complete goal shifting.
A second factor whose effect probably occurs in goal shifting is the length of the RSI. Insofar as the RSI is relatively long or short, it would allow more or less of this stage to be completed before the onset of the stimulus for the next task. Consequently, goal shifting's contribution to RTs could be less when the RSI is long, reducing switching-time costs as Rogers and Monsell (1995 , Experiment 3) found with blocked RSIs. This explanation would also account for results of Meiran (1996) , who found that task cues reduced switching-time costs more after longer RSIs.
However, the model does not imply that, after long RSIs, switching-time costs should necessarily vanish. On the contrary, suppose that the length of the RSI varies randomly across trials. Then for each trial that requires a task switch, the goal-shifting stage may be postponed until the next task's stimulus is identified. Such optional postponement would preclude RSI effects on switching-time costs, as Rogers and Monsell (1995 , Experiment 2) found with mixed RSIs. Also, for each trial that requires a task switch, rule activation may occur after the RSI has ended. This stage could therefore yield residual switching-time costs even after long RSIs, as both Allport et al. (1994) and Rogers and Monsell (1995) found.
Effects on rule activation. The rule-activation stage is also a likely site of operation-complexity effects. Recall that Jersild (1927) found greater switching-time costs when participants alternated been complex rather than simple arithmetic operations. This may have occurred because more production rules are needed to perform complex operations, and it takes longer to activate them, just as larger "number crunching" programs take longer to be loaded in a digital computer's memory.
Irrelevant-character effects probably occur during rule activation as well. For example, recall that Rogers and Monsell (1995) found greater switching-time costs when stimulus displays contained incongruent rather than neutral irrelevant characters. This could occur because incongruent irrelevant characters make it harder to disable the production rules of prior tasks. Such difficulty would likewise explain why Allport et al. (1994 , Experiments 4 and 5) found greater switching-time costs when current stimuli contained perceptual features associated with previous familiar tasks.
Certain characteristics of the rule-activation stage might explain other results of Allport et al. (1994 , Experiment 1). As mentioned earlier, they found that under some conditions, mean switching-time costs were about the same regardless of the switches' scope (i.e., the costs did not depend on whether the relevant stimulus features, the response ensemble, or both changed between tasks). Perhaps this occurred because the scope of the switches did not affect the numerosity or complexity of the production rules that had to be enabled in task switching and so did not affect the duration of rule activation either.
In essence, rule activation presumably prepares the processing system so that response selection can proceed rapidly for the current task. After this has been accomplished, there need be no further switching-time cost until a subsequent task switch must take place. The immediate completion of the rule-activation stage accounts for why the time cost that Rogers and Monsell (1995 , Experiment 6) observed on the first trial after a task switch did not propagate in a gradually decreasing fashion throughout a run of successive trials with the same task.
Additive factor effects. If the preceding explanations are correct, then our model would account for why certain factor effects on switching-time costs have been essentially additive. For example, recall that Rogers and Monsell (1995 , Experiment 3) found such additivity in RSI and irrelevant-character effects. This may have occurred because these two factors respectively affect goal shifting and rule activation, which are successive stages whose durations jointly constitute the time cost of task switching.
Similarly, it is possible to account for some additive factor effects that Allport et al. (1994) found. Recall that their participants took more time respectively on the standard Stroop and digit-numerosity judgment tasks than on the reverse Stroop and digit-magnitude judgment tasks ( Allport et al., 1994 , Experiment 1). This effect of task difficulty was about the same during repetitive-task and alternating-task trial blocks, even though block type affected mean RTs reliably (i.e., the difficulty and block-type effects on mean RTs were additive). Our model explains such additivity because trial-block type may influence stages of executive control, whereas the difficulty of particular tasks may stem from within-task stages (i.e., task processes) such as stimulus identification and response selection. For analogous reasons, the model is also consistent with other additive factor effects found by Allport et al. (1994 , Experiments 2 and 3).
Overview of Experiments
The present article reports four experiments involving various versions of the successive-tasks procedure designed to further test several predictions based on our model. First, we show that as the model predicts, executive control and task processes can be empirically dissociated and affected separately by different factors (Experiment 1). Second, we show that executive control entails at least two component stages, goal shifting and rule activation, whose mean durations depend respectively–and additively–on task cuing and rule complexity (Experiment 2). Third, we show that because of how rule activation works, switching-time costs may be asymmetric in ways related to the familiarity of individual tasks between which participants must switch (Experiments 3 and 4). Taken together, the results of the four experiments strongly support the model's basic assumptions about the nature of executive mental control for task switching.
Methodological Approach
During each experiment, repetitive-task blocks of trials were completed for each of two tasks. There were also alternating-task blocks in which participants switched back and forth between the two tasks at hand. Short RSIs were used throughout the experiments, thereby helping to maximize observable contributions of goal shifting and rule activation to switching-time costs.
To show that these executive control processes are functionally distinct and separable from task processes, we manipulated several factors, including type of task, complexity of task rules, availability of task cues, and discriminability of task stimuli with which participants worked. In Experiments 1 and 4, participants rapidly classified visual patterns of geometric objects with respect to alternative perceptual categorization rules. For the pattern-classification tasks, rule complexity was manipulated by having participants apply unidimensional or bidimensional classification rules. In Experiments 2 and 3, participants solved arithmetic problems with alternative numerical combination rules. For the arithmetic tasks, rule complexity was manipulated by requiring addition and subtraction or multiplication and division operations. Visual task cues (i.e., arithmetic-operation signs) were presented during some trial blocks but not others. Our combined manipulations across the experiments allowed us to check for expected patterns of additive and interactive factor effects on mean RTs and switching-time costs, which provide diagnostic indicators of temporally separate processing stages ( Roberts & Sternberg, 1993 ; Sternberg, 1969 , 1998 ).
Estimation of Switching-Time Costs
Following Allport et al. (1994) , we estimated mean switching-time cost ( T S ) as follows:
where n - 1 is the number of task switch |
ESISTE SESSO SENZA AMORE 2 28/3/2004 19.18.18 (216 visite) pasilda
re:ESISTE SESSO SENZA AMORE 2 28/3/2004 19.18.50 (81 visite) Franti
