In contradiction to stimulus-response- (S-R-) translation bottleneck models of dual-task control, stimulus processing in a primary task is affected by its compatibility with the response in a secondary, later performed task (Hommel, 1998a)- an indication of parallel S-R translation. Here we show that this backward-compatibility effect is independent of working-memory load, whether this is induced by an extra memory task (Experiment 1) or by increasing the number of S-R alternatives in the primary task (Experiment 2). However, backward effects occur even when the secondary task is no longer carried out (Experiment 3) and they are strongly affected by the inconsistency of previously used S-R mappings (Experiment 4). These findings suggest that S-R translation is (or can be) capacity-independent and automatic even under multipletask conditions, and that it is mediated by direct S-R associations that emerge after only little practice. The search for capacity limitations in human information processing has long been a central issue in cognitive psychology, for both theoretical and practical reasons. The assumption of two processing bottlenecks has received rather general agreement; one processing bottleneck associated with the selection of stimuli for further processing (selective attention) and the other associated with the translation of stimulus information into an appropriate response. Most evidence for this latter, response-related bottleneck, which our investigation aims at, comes from dual-task studies. In these studies, people are commonly asked to perform a primary response (R1) to a primary stimulus (S1) and secondary response (R2) to a secondary stimulus (S2). The stimulus-onset asynchrony (SOA) between S 1 and S2 often varies, so that primary and secondary tasks overlap temporally to varying degrees. With very few exceptions, overlap strongly impairs performance on the secondary task, and the greater the overlap, the more so (i.e., the shorter the SOA; for overviews see Meyer Kieras, 1997; Pashler, 1994). This has been taken to indicate a structural bottleneck in information processing (though not without exception: Meyer & Kieras, 1997). Peripheral stages do not seem to be responsible, as dual-task costs are observed under conditions where sensory or motor cross talk can be excluded (e.g., De Jong, 1993; Pashler & Johnston, 1989). However, dual-task costs do interact with S-R compatibility (McCann & Johnston, 1992), this pointing to the S-R translation stage. Indeed, several authors have suggested that only one stimulus can be translated into a response at a time, so that secondary-task translation cannot begin before primarytask translation is completed (De Jong, 1993; Fagot Pashler, 1992; McCann Johnston, 1992; Welford, 1952). Although most researchers agree with the idea that selecting a response provides some kind of bottleneck (the debated questions being, rather, whether this is the only bottleneck and whether it is structural or functional), its specific characteristics and possible functions are not well understood. Part of the problem has to do with the lack of conceptual clarity with respect to the terms S-R translation and response selection. Earlier approaches, like that of Smith (1967) or Welford (1952), assumed that responses are selected by translating a stimulus into a response (cf., Hommel, 2000a). Accordingly, the two terms were, and still are, treated as equivalent (Pashler, 1998, p. 277). Logically, however, they are not. The term "translation" refers to the act of transforming some code into some other code of a different "language" or reference system (obviously reflecting the widespread belief of a fundamental incommensurability of perceptual and action-related codes; Prinz, 1984, 1990) or, in neural-network terms, to activating one code as a function of the activation level of another.
