Action Planning and the Temporal Binding of Response Codes

Coding of Action Plans An important question about action planning concerns the level of abstraction on which the basic components of action plans are defined. According to Keele (1968), a motor program is a structured set of muscle commands—a relatively concrete and peripheral entity. Later approaches, such as Rosenbaum's (1985, 1987), have assumed that programs are built from abstract codes—from movement parameters that need not directly map on specific muscle commands. Although the present study did not aim to answer this question, its outcome supports the assumption that abstract codes are involved in action planning. In Experiment 2, we saw that planning a leftor right-hand action hampers foot responses on the same side of the body. That is, body side or egocentric location of the effectors must have been coded in the corresponding action plans independently of the effector's identity, which clearly rules out the idea that planning occurs in terms of muscle commands. Our findings fit well with other observations that also point to the abstract nature of action planning. For instance, Rosenbaum and colleagues have repeatedly shown that the time it takes to plan manual actions decreases with the amount of advance information the actor receives about abstract, outcome-related action features (see Rosenbaum, 1987, for an overview). That is, action plans seem to take abstract parameters. In the same vein, Ulrich, Moore, and Osman (1993) found that actions can be planned up to the level of a measurable lateralized readiness potential (LRP), even if the actual effector is not yet known. In their experiment, participants were given advance information about the hand of an upcoming manual action but not about the finger to be moved, so that the information was too abstract to allow for preparing a specific set of muscles. Yet, partial information about the hand significantly speeded up LRP onset, suggesting that the LRP reflects the implementation of abstract response parameters. A Two-Process Approach to Action Planning According to the common view, the gap between stimulus processing and action control is bridged by a single process, usually called stimulus-response translation or response 8 Here we presuppose that the same codes are used for coding the location of a stimulus and the location of an action—an assumption that finds ample support from studies on stimulus-response compatibility (see Hommel, 1997; Miisseler & Hommel, 1997a). 1638 STOET AND HOMMEL selection (e.g., Meyer & Kieras, 1997; Sanders, 1990; Teichner & Krebs, 1974; Theios, 1975). The basic assumption underlying most current models of human information processing is that after the stimulus is analyzed to at least some degree, some kind of selection process is called or some automatic translation device is activated, which then assures that the correct motor program is launched. However, this single-step translation view has been challenged by a study of Hommel (1998a), who was working with a standard dual-task design involving a primary manual key-pressing task and a secondary vocal color-naming task. On the one hand, secondary-task performance decreased as the time interval between primary and secondary stimulus shortened; hence, it decreased with increasing temporal task overlap. Of course, such an outcome is hardly surprising and merely replicates the wellknown finding that people are unable to select two responses at the same time (see Pashler, 1994, for a recent overview). On the other hand, however, the compatibility between the secondary response (the word red or green) and the primary stimulus (a red or green color patch) strongly affected primary task performance. This finding shows that the secondary response was activated before the primary task was completed and, thus, suggests that stimulus-response translation does not provide a bottleneck in dual-task performance. To account for this pattern of results, Hommel (1998a) suggested distinguishing between response activation, which typically results from stimulus-response translation, and eventual response selection, which involves recruiting those already activated response codes that belong to the same response. Our present findings fully support and extend this twoprocess view of response determination. For all conditions in which participants were encouraged to plan Response A, we found a substantial impairment of compatible B responses—an effect we attribute to the binding of response features in the process of planning Response A. If planning was neither encouraged nor likely, as in the no-plan condition of Experiment 3, we found a positive compatibility effect. This result is exactly what one would predict if stimulus-response translation and the resulting activation of response features were automatic, as Hommel (1998a) assumes. Therefore, we conclude that action planning consists of two processes or stages: an automatic one, responsible for activating those action-feature codes that are signaled by the stimulus, and another, more effortful one that binds the activated features. That is, constructing an action plan requires not just specifying the features of the intended action but also temporarily binding them to form an integrated whole. Of interest, the dissipation of action plans seems to reflect the way they emerge, with the disintegration or unbinding of feature codes being followed by their deactivation. In summary, the present findings provide substantial evidence for the idea that feature-binding mechanisms may be operative not only in perception but in action planning as well, suggesting that they play a major role in the cognitive coding of perceived and produced events. Future research is required to investigate the apparently different characteristics of pure action-related bindings and those involving stimulus codes. It would also be of importance to know whether and how the binding of stimulus features affects or interacts with the binding of action features. We have already begun to tackle these issues, and preliminary results suggest that the experimental design introduced in this article may be of some help in this enterprise. Further research is also necessary to find out whether binding effects can only be observed with spatial response features (which we focused on in this study) or whether other features (e.g., temporal) are associated with comparable phenomena.