B.3.6.4.3. Salience and salience map

One might acknowledge that the effects of perceptual similarity on relevance-salience interaction might perhaps also be accounted for by salience map models. For that end, they would have, for instance, to take into account the endogenous influences on feature map weighting for the salience computation (e.g. Müller, Heller, & Ziegler, 1995; Navalpakkam & Itti, 2005; Wolfe, 1994). The efficiency of this kind of models to account for the present results would then probably have to be directly tested through computational modelling. However, it is still worth noting that this kind of models did not predict such effects a priori.

At first sight, the present results seem at odd with salience-map models. Indeed, one major characteristics of salience is that it is freed from any perceptual information (Koch & Ullman, 1985). Thus, salience effects should not differ depending on the feature that was mainly involved in inducing them. In the present experiments, however, salience and specific-feature relevance integrated differently, depending on their perceptual similarity. This suggested that salience effects could be dependent on the perceptual dimension from which they emerged. This seemed incongruent with traditional salience map models, in which a salience signal, independent of any featural information, would determine solely the attentional orienting. The present reasoning could be compared with the results of Koene and Zhaoping (2007), who evidenced feature-specific integration in salience computations. It seems that the notion of salience map need not be invoked for every exogenous attentional orienting. Li (2002) evidenced that one could consider salience effects within a frame that did not reject completely featural information from salience computing. Later, Koene and Zhaoping (2007) proposed an "augmented feature summation" process in V1. One could notice that the reasoning behind the salience map in V1 hypothesis is not very far from rejecting the necessity of passing by the salience map for orienting attention, since a single system could compute both salience and perceptual processes that were thought to be modulated by salience effects (Li, 2002). VanRullen (2003), for instance, argued that passing by a salience map before orienting attention and then processing stimuli was very unlikely for at least some extreme cases, such as rapid semantic categorization (Thorpe, Fize, & Marlot, 1996). Consistently with the biased competition hypothesis, he claimed that salience "must be extracted during object recognition itself" (VanRullen, 2003, p. 366), thus within the perceptual system. It is possible that a kind of feature combination in a salience map occurs and is perhaps necessary in some cases. However, it seems also that feature contrasts emerging within the perceptual system could more parcimoniously account for the salience effects in some conditions, such as those presented here.

In the continuation of the biased competition hypothesis (Desimone & Duncan, 1995; Duncan et al, 1997; 2006), we believe that the present results, along with the work of Zhaoping and colleagues (Koene & Zhaoping, 2007; Li, 2002) or VanRullen (2003), cast some doubt on the theoretical construct of a "salience map" as an independent and specialized cognitive module necessary for orienting selective attention. Obviously, they could certainly not invalidate a such venerable concept as "salience map". However, they might invite to examine more critically its theoretical necessity in the various conditions involving selective attention. Our understanding of attentional processes might perhaps gain to be thought as an emergent property of the perceptual and motor systems, in the vein of the works of, among others, Rizzolatti et al. (1987) and Desimone and Duncan (1995, Duncan, 2006). Indeed, this perspective already proved allowing powerful computational modelling (e.g. Deco & Rolls, 2005, Hamker, 2003; 2004). This approach might advantageously be led within the grounded cognition perspective (Barsalou, 2008), since the core idea would be to ground "attentional selection" on perceptual and motor processes.