Attentional control of the creation and retrieval of stimulus–response bindings

Research on feature integration in perception (commonly in vision) does not make much contact with feature integration in action planning or across perception and action, and researchers tend to think of the underlying processes as very different. However, not only are the representations of actions no less distributed as the representations of objects (Hommel & Elsner, 2009), thus raising the same kind of binding problems (Stoet & Hommel, 1999), but the functional and computational logic underlying feature integration in creating cognitive representations of objects, action plans, and sensorimotor coordination patterns may well be equivalent (Hommel, 2004).

Among other things, the relative benefit of complete repetitions depends on the question whether selecting a response presupposes the mere activation of the relevant feature codes (as for example considered by Jolicœur, Tombu, Oriet & Stevanovski, 2002) or the completed integration of these codes (as suggested by Kahneman et al., 1992). Specific predictions, especially regarding the comparison with alternations, are further complicated by the observation that complete stimulus repetitions seem to induce a tendency to shortcut response selection (Bertelson, 1963; we will come back to this issue in the discussion of Experiment 2) and that alternations may benefit from “integrated competition” (Duncan, 1996), in the sense that a previously created binding between competing features may make it easier to reject them as “one unit”. Given these complications, we neglect the precise relationship between complete repetitions and alternations (which empirically often produce comparable results) but focus on the observation that, and the explanation, why partial repetitions produce worse performance than both complete repetitions and alternations.

This reasoning assumes that attentional sets are implemented immediately after cue processing, so that the presentation of both S1 and S2 falls under the same set. One may object that, given that the set was unnecessary for processing S1, it may not have been implemented before S1 processing was completed, so that processing S2 but not S1 took place under the respective set. This consideration can be tested by comparing trials in which the set was the same as in the trial before (set repetition trials) with trials in which the sets differed (set alternation trials). Experiment 1 will show that set repetitions and alternations produced pretty much the same result pattern, which renders the objection unfounded.

One might argue that a task cue that appears 500 ms after S1 onset may still affect S1–R1 integration, which may blur the distinction between early and late cues. Even though it is impossible to rule that out entirely, it is rather unlikely. First, processing and interpreting the task cue must have taken some time, we would guess more than 200 ms. Second, processing the auditory cue must have taken even longer, which is evident from the increase in mean RT, and yet this did not interact with binding effects. Third, task switching studies have shown that shifting attention from one dimension to another is a rather capacity demanding and slow process, often taking several hundred milliseconds (e.g., Meiran, 1996; Rogers & Monsell, 1995). Considering all these complications, we doubt that the attentional control settings that were signaled by late cues were up and running much earlier than at S2 presentation—a time point at which stimulus–response bindings already start to decay (Hommel & Colzato, 2004).

Note that there was an anomaly regarding the way task modulated shape/response interactions. With late cues, the interaction of shape and response repetition was pronounced under the shape task and negligible under the color task. This was the expected pattern—a mirror image of the findings for color/response interactions. With early cues, however, the interactions were equally pronounced under both tasks, even though this did not render the related four-way interaction with task and time point significant. We have no explanation for this particular pattern to offer but note that similar asymmetries between perceptual dimensions have been reported before (e.g., Hommel, 1998).