The target article offers an exciting new perspective on how the cognitive system samples visual input for relevant information. Although we are in agreement with this account with regard to detection tasks, our commentary focuses on a potentially crucial limitation: generality across task settings. Despite Hulleman & Olivers' (H&O's) endeavour to highlight similarities between different search tasks, we argue that this issue deserves more attention. In particular, we contend that item-based selection is likely to be mandatory in tasks that require precise, focal-attentional stimulus analysis. Here we discuss three experimental findings that challenge the notion that the search process is equivalent in detection and compound tasks.

As noted by H&O, some of our findings indeed suggest that attentional selection is similar across tasks (Exp. 1; Töllner et al. Reference Töllner, Rangelov and Müller2012b). This view is supported by the activation profile of the Posterior Contralateral Negativity (PCN, or N2pc) – an ERP wave generally agreed to reflect attentional selection (Eimer Reference Eimer1996; Luck & Hillyard Reference Luck and Hillyard1994; Töllner et al. Reference Töllner, Müller and Zehetleitner2012a) – which was elicited equally for search tasks in which observers had to detect, localize, or identify physically identical targets. However, this pattern was restricted to conditions with relatively high target prevalence (between 66.6% and 100%). In the same study (Exp. 2), we observed a strong influence of target prevalence on attentional selection: Using a detection task, PCN amplitudes increased gradually with decreasing target frequency (80% < 50% < 20%). Following one proposal according to which the PCN may represent the target's saliency/priority signal on the attention-guiding “master map” (Töllner et al. Reference Töllner, Zehetleitner, Gramann and Müller2011; Reference Töllner, Conci and Müller2015a), this amplitude modulation may result from a cortical amplification mechanism (Egner & Hirsch Reference Egner and Hirsch2005) that comes into play when searching for low-prevalence targets. In particular, the visual system may adapt to reduced target probabilities by boosting incoming relevant information through a context-sensitive amplifier that keeps track of environmental statistics. This biasing of target signals may automatically translate into enhanced activations at the next processing stage – the attention-guiding master map (likely represented by the PCN) – thereby raising detectability of rare targets to the level of frequent targets. Another factor that may have contributed to this amplitude gain is refractoriness (Woods et al. Reference Woods, Courchesne, Hillyard and Galambos1980). It is conceivable that neural populations associated with the coding of attentionally selected targets are more prone to refractoriness in conditions of 80% target prevalence in which the average inter-target interval was 2.5 s, as compared with 20% prevalence with, on average, a 10-second interval. Whatever the exact interplay of mechanisms, findings of attentional selection being frequency-dependent have implications for H&O's framework. Because a target is present on every trial in the typical compound task but only on half of the trials in the standard detection task, we surmise that attentional selection is not directly comparable among different search tasks using variable target frequencies.

Furthermore, H&O propose that compound-search tasks involve a cascade of processing steps, with sequential reductions of the functional visual field (FVF) to meet the requirements of focal-attentional analysis: the higher the requirements, the smaller the FVF. Critically, H&O assume that the first stage of these processes is similar across tasks, and that the initial size of the FVF depends exclusively on target discriminability. Although we agree with the general idea, we doubt that the initial stage of target selection is entirely detached from the subsequent target identification stage. Our scepticism derives from recent EEG evidence (Mazza et al. Reference Mazza, Turatto, Umilta and Eimer2007; Rangelov et al. Reference Rangelov, Töllner, Müller and Zehetleitner2013b; Töllner et al. Reference Töllner, Conci, Rusch and Müller2013) demonstrating that target identification involves a post-selective process that extracts detailed object information from working memory (WM). Specifically, we observed the CDA (or SPCN) wave – a well-established ERP marker of the maintenance of (Vogel & Machizawa Reference Vogel and Machizawa2004; Wiegand et al. Reference Wiegand, Töllner, Habekost, Dyrholm, Müller and Finke2014) but also the access to (Töllner et al. Reference Töllner, Eschmann, Rusch and Müller2014; Reference Töllner, Mink and Müller2015b) WM representations – being elicited exclusively in target identification tasks, but not in tasks requiring simple target detection or localization. That target identification is carried out in WM may consequently place further constraints on the number of items participants may select in parallel in the first place. Given that WM is limited in capacity to about three to four items (Cowan Reference Cowan2001), it would appear plausible that participants strategically adjust the size of their FVF in relation to their available WM capacity. Note that this consideration is strikingly supported by the fact that three to four objects match exactly the target number at which both the PCN (Mazza et al. Reference Mazza, Pagano and Caramazza2013) and the CDA amplitudes (Vogel & Machizawa Reference Vogel and Machizawa2004) reach asymptotic level. It is as if the stage of attentional selection (PCN) were naturally tuned to take in only an amount of information that can be fairly handled at subsequent WM stages (CDA). Moreover, no such set size modulations of the PCN are evident if the task requires simple detection, as opposed to exact enumeration (in WM), of multiple targets (Mazza & Caramazza Reference Mazza and Caramazza2011). In the light of these findings it appears unlikely that the initial stage is identical across tasks. Instead, we argue that, while in standard detection tasks the unit of selection may well be set according to target discriminability, in compound tasks the unit is additionally constrained by individuals' capacity to evaluate target information in WM. The unit of selection may even scale down to a single item; for example, when the task requires precise focal-attentional analysis in the face of highly similar distractors.

Finally, investigations of search slopes indicate qualitative differences between detection and compound tasks. Specifically, in detection tasks reaction times to pop-out targets remain constant across different display densities (Young & Hulleman Reference Young and Hulleman2013), whereas reaction times to the same targets decrease as more distractors are added in compound tasks (Bravo & Nakayama Reference Bravo and Nakayama1992; Rangelov et al. Reference Rangelov, Müller and Zehetleitner2013a). In H&O's framework, search slopes are determined by the FVF size, which, in turn, depends on target discriminability. Consequently, FVF size should be the same for similar targets. As negative search slopes indicate, however, this is not the case.