Bottlenecks in planning and producing vocal, manual and foot responses

When subjects perform two sensorimotor tasks (T1 & T2) close together in time, T2 response selection is often delayed. Five experiments examined whether T2 response selection waits for both selection and execution of the response in T1. In the first two experiments, T1 required a sequence of vocal responses (R1). When the sequence length in R1 was increased, R1 took longer to complete (unsurprisingly); however, the (manual) second response (R2) was little affected, and R2 usually occurred while the R1 sequence was underway. Similar results were found when T1 involved a sequence of vocal responses and T2 required a foot movement (Experiment 3). However, when R1 was a manual sequence, and R2 involved either manual or foot movements, R2 was substantially delayed, and usually occurred after R1 was completed. When T1 required arm reaching, variability in reaction time (but not movement time) was associated with slowing of R2. The results argue for (1) a central bottleneck that does not include response production and (2) a separate responseproduction bottleneck specific to production of manual or foot responses (and likely to play no role in usual dual-task laboratory experiments). Dual-Task Interference August 7, 1994 3 Introduction When people try to perform more than one task at nearly the same time, the speed or accuracy of their performance is often impaired (dual-task interference). The simplest form of dual-task interference is observed when subjects are presented with two stimuli (S1 and S2), separated by a variable stimulus onset asynchrony (SOA), and attempt to produce a response to each stimulus (R1 and R2, respectively) as quickly and accurately as they can. Even when the tasks seem cognitively trivial, response time for the second task (RT2) increases as the SOA is shortened. This slowing of RT2 is usually referred to as the psychological refractory period or PRP effect (Vince, 1949; Welford, 1952; Bertelson, 1966; Smith, 1967a). The present research uses the PRP paradigm to explore dual-task interference in the production of response sequences, with the goal of learning more about the relationship between cognitive limitations and motor-control limitations. Recent evidence implies that a fundamental cause of dual-task interference (as observed in the PRP effect) is the fact that central processing (decision, response selection, and possibly response initiation) generally requires the exclusive involvement of a single mental mechanism (see Pashler, in press, for a review). The result is a bottleneck in the PRP task: while task 1 uses the mechanism, the operations that require the mechanism in task 2 are forced to wait (Welford, 1952; Pashler, 1984), thereby delaying task 2. One kind of evidence for this bottleneck comes from experiments that have manipulated the duration of particular stages of processing in the second task. Bottleneck Obviously, the term mechanism is used in the broad sense: it needn't be localized anatomically, and the bottleneck might be the result of inhibitory interactions among different sets of machinery. Dual-Task Interference August 7, 1994 4 models make distinctive predictions for the results of such experiments. When stages of processing in task 2 that occur before the bottleneck-dependent stage are slowed down and the SOA is short, R2 should still occur at approximately the same time (Figure 1 A), whereas at long SOAs R2 should be slowed. The bottleneck model therefore predicts an underadditive interaction between SOA and factors affecting the duration of pre-bottleneck stages. This interaction occurs because the effect of the factor is "absorbed" while processing in the second task waits for the bottleneck to be freed up. By contrast, any variable that slows the processing at the bottleneck stage itself adds a constant to RT2, regardless of SOA (Figure 1B). In short, pre-bottleneck manipulations in task 2 should interact underadditively with SOA, while bottleneck and post-bottleneck manipulations are additive with SOA. An analogy may help make these predictions more intuitive. If one walks into a bank immediately after another customer, and there is only a single teller on duty, the teller will represent a bottleneck: one will not leave the bank as soon as one would otherwise have (analogous to the PRP effect). However, since the teller is the bottleneck, there is no need to walk quickly small differences in walking time will be absorbed, leaving the total time one spends in the bank unaffected. On the other hand, once one reaches the teller, taking extra time will increase total time spent in the bank. *** INSERT FIGURE 1 A & 1B* * * Experiments using this sort of logic have manipulated factors that slow the perceptual processing in task 2, and observed the interactions predicted by a post-perceptual bottleneck (Pashler, 1984; Pashler & Johnston, 1989). On the other hand, variables slowing response selection in the second task (e.g., S-R compatibility) generally have the same effect on RT2 in the dual and single task conditions, and these effects do not change with SOA (Pashler, 1984; Pashler & Dual-Task Interference August 7, 1994 5 Dual-Task Interference August 7, 1994 6 Dual-Task Interference August 7, 1994 7 Johnston, 1989; McCann & Johnston, 1992; Fagot & Pashler, 1992). This pattern of results implies that response selection in task 2 must be part of the bottleneck. Is Response Execution Subject to Bottlenecking? The simplest model consistent with these results would postulate a bottleneck that encompasses response selection and claim that response execution plays no part in this bottleneck. Response selection in task 2 waits for the completion of response selection of task 1, on this view, while other stages (including execution) can overlap with each other or with response selection. (The assumption that perceptual processes in task 2 begin as soon as S2 is available is supported by additional observations beyond those mentioned, including the fact that accuracy in tasks requiring visual search of brief masked displays is almost unaffected by temporal overlap with another task; Pashler, 1989.) There are several alternatives to this simple view, however. At the other extreme, one could propose a single central bottleneck that includes both response selection and response execution. (Response execution is taken here to include whatever central processes are involved in taking the output of the response selection stage and generating appropriate commands to the muscles. Obviously, it is only the central response execution processes that could plausibly be part of the bottleneck, since it's hard to see how T2 could be delayed by peripheral activity in the motor system which operated under openloop control.) On this view, then, response selection in T2 would not commence until central motor-control machinery had issued commands to the muscles to produce R1. Another account would postulate two separate bottlenecks: one in selection of responses and another in production of responses. On this view, Dual-Task Interference August 7, 1994 8 selection in one task would have to wait for selection in another task, and production in one task would wait for production in another, but selection in one task could overlap production in another task. Possibilities like those just described -which would include response production in bottleneck(s) -are consistent with results of the laboratory PRP studies generally interpreted as favoring a response selection bottleneck. The findings described above involving manipulations of response selection duration imply that response selection (but not perceptual analysis) in task 2 is delayed by task 1. They also rule out the idea that producing the second response is the only operation in task 2 that is subject to delay (as proposed by Keele, 1973, and Norman and Shallice, 1985, among others). However, they do not provide any information at all about what stages of the first task are being waited for by the second task. Previous studies of dual-task interference directed specifically at uncovering interference with response execution have typically used one task as a "probe" measure of "capacity demands", and the relationship between performance of two responses has gone unexamined. For reasons to be described in the General Discussion, these studies cannot distinguish the different accounts of dual-task interference with response execution, either. The goal of the present studies is to explore this issue by looking at performance in two overlapping tasks while manipulating variables associated with response execution in one task. The specific strategy involved varying the duration of response execution in task 1 of a PRP situation, and determining how RT1 and RT2 are affected. The basic logic is as follows: If a central bottleneck encompasses response selection and execution in task 1, then increasing the time for this process will delay not only R1 but also task 2 response selection, at short SOAs. At long SOAs, on the other hand, task 2 will not be affected. Therefore, a bottleneck in selection and execution predicts an Dual-Task Interference August 7, 1994 9 overadditive interaction in RT2 between factors slowing response-execution and SOA. On the other hand, if the bottleneck does not encompass response execution, slowing task-1 response execution should obviously affect only task 1, regardless of SOA. Previous PRP Studies Manipulating Task-1 Variables In addition to the task-2 manipulations described earlier, the literature also contains several PRP studies that involve task-1 manipulations. These studies would seem to bear on the question at hand, but they too leave the main question unanswered. In several of these experiments, the difficulty of "cognitive" stages of task 1 was manipulated. For example, Smith (1969) varied the number of alternatives in a one-to-one choice first task. This factor number of alternatives -mostly affects