B.1.4.3. Discussion

First, this experiment replicated the results of the previous one relative to salience effects in the random condition. Again, salience effects appeared early, lasted long and discriminated the large, medium and small items, when salience was not harmful for the visual search task (i.e. in random and large-relevant conditions). However, cueing the target's size induced dramatic effects.

On the one hand, cueing the salient size increased the salience effects between the large and medium targets. RTs for the large target were shorter in the large-relevant than in random condition, even for the shortest quintiles, suggesting that these effects reflected a genuine processing speed up of the large targets. Salience and relevance seemed to have additive effects, and to combine to control attentional orienting. In this large-relevant condition, the medium item was still processed faster than the small one, even for the fastest RTs, despite the fact that both items were equally irrelevant (both had a 17 % probability to be the target). Salience effects were observed between all item sizes, at each RT distribution extremity (except for the small-medium differences at long RTs).

On the other hand, when the small item was made relevant, salience became detrimental, since salience effects would misguide attentional orienting away from the relevant item. Consistently, relevance seemed to completely override salience effects: the small item was processed before the two other items. Similarly to the large-relevant condition, RTs were shorter for the small target in the small-relevant than in random condition, even for the shortest quintiles, suggesting that these effects also reflected a genuine processing speed up of the small targets. Performances between the medium and large items did not differ. Moreover, only 21 subjects out of 38 presented shorter RTs for the medium target than for the large target. Thus, there remained no evidence of salience effects in the small-relevant condition.

The first conclusion was that endogenous featural attention could strongly modulate the observed salience effects, early in the RT distribution and strongly, either by enhancing them when they were relevant, or by overriding them when they were harmful for the task at hand. These results were in line with those of Ansorge, Horstmann, and Carbone (2005; Ansorge & Horstmann, 2007), who evidenced that top-down control settings could influence capture by colour even in fast correct responses in an RT distribution. This was also consistent with previous researches showing that cueing could favour even the processing of a salient item (Müller & Krummenacher, 2006), and that exogenous orienting driven by a salient item could be resisted (Lamy et al., 2004, Michael et al., 2006). On the contrary, these results seemed at odd with the two-stage model of Theeuwes (1993; 2010a), which claimed that endogenous featural attention could not influence "pre-attentive", but only "post-selective" processes. Indeed, in the discrimination task used, the relevant size (small or large) had no relation with the correct response (top or bottom). Moreover, these effects arose early in the RT distribution. All in all, cueing very likely influenced an attentional orienting level rather than a "response selection level", as in Theeuwes, Reimann, and Mortier (2006).

The second, most important, conclusion was that salience effects seemed to last long by default, that is, when they were not harmful for the task, while they could be overridden, or masked, when salience became detrimental for the task. This might explain why some authors (e.g. Donk & van Zoest, 2008) failed observing the salience effects predicted by many models (e.g. Itti & Koch, 2000; Theeuwes, 1993; Wolfe, 1994).

Contrary to our expectations, the two cueing conditions provided very similar results. No interaction involving the cue type approached statistical significance. Using a symbolic or a perceptual cue induced virtually identical performance patterns. This lack of statistical interaction suggested that the effects observed were not due to perceptual priming. Moreover, a not reported analysis was restricted to the symbolic cueing condition. Virtually identical results to the general ANOVA were obtained. Thus, the cueing effects observed in this experiment could confidently be attributed to endogenous influences. The induced priming effects were perhaps too weak, or the top-down relevance effects too strong to let appear a significant difference.