In comparison with semantic network models, parallel distributed models of priming (see Masson, 1991, 1995; McRae et al., 1997) suggest a largely different structural organization of the semantic memory. Instead of allocating a single node to each semantic concept, the semantic knowledge is to be distributed over a multidimensional space of activation units displaying semantic features or micro-features. In order for this, each semantic concept is constituted by a specific pattern of activated units, while the number of shared activation units determines semantic relatedness between different concepts. Thus, semantic priming effects can be explained by a faster transition from the pattern corresponding to the prime concept to a semantically related than to an unrelated one (i.e., the pattern corresponding to the target) because the shared activation units are already in the appropriate mode of activation. Although this model was initially developed in order to interpret semantic priming effects (see McNamara, 2005, for a review), it is prima facie perfectly suited to account for S–S-based evaluative priming effects as well. It is sufficient to additionally assume that a considerable part of the activation pattern of a specific semantic concept corresponds to its evaluative connotation (see Spruyt et al., 2002; Wentura, 1999, 2000). Thereby, the activation patterns of evaluatively congruent concepts overlap in the activation units corresponding to their evaluative connotation. That means that in a sequential evaluative priming task, the transition from the prime pattern to the target pattern is facilitated in case of evaluative congruency, since the target pattern is partly pre-activated by the prime pattern. Thus, parallel distributed models provide an elegant mechanism for the explanation of S–S-based evaluative
priming effects as they allow for target-encoding facilitation by an evaluatively congruent prime (see, e.g., Wentura, 2000).
At this point, it is important to go back a step in order to emphasize a hidden inconsistency in the interpretation and the understanding of evaluative priming. S–R-based evaluative priming effects have typically been explained with response-based processes, assuming that an evaluatively congruent prime facilitates the target evaluation, while an evaluatively incongruent prime interferes with the target evaluation. Such response-related processes have been made responsible for the emergence of positive S–R-based evaluative priming effects (see Klauer et al., 1997; Wentura, 1999). An observable interaction between prime and target responses, however, requires a simultaneous activation of both concepts or―at least―both responses. This raises the following question: Does the parallel distributed structure of semantic memory provide a mechanism for response-related processes between two concepts as well as the concomitant parallel activation of these concepts?
In the distributed memory model―as it has been introduced by Masson (1991, 1995)―the activation of one semantic concept is necessarily accompanied by the activation of semantically related concepts because of shared activation units. Simultaneous activation of more than one concept is thereby, however, restricted to their overlapping parts and the completely activated pattern corresponding to one concept allows for only partial activation of related patterns. As previously mentioned, this aspect is yet crucial for the response-based explanation of S–R-based priming, since response facilitation and interference require the parallel activation of the activation units corresponding to prime and target responses, even if these activation units do not overlap in the response- incompatible condition. Thus, the distributed organization of semantic memory conflicts with the response-based account of S–R-based evaluative priming, since no simultaneous activation of the full pattern of two distinct concepts is allowed. In contrast, the distributed memory model (Masson, 1991, 1995) provides a conclusive implementation accounting for S–S-based evaluative priming: a currently activated concept (i.e., the prime) facilitates the encoding of an evaluatively congruent concept (i.e., the target) via pre-activation of the overlapping activation units.
Since the S–R-based and the S–S-based variants of evaluative priming differ in the required task only, while the whole task setting is comparable, it is quite dissatisfying to assume largely different representational structures of the semantic memory including the evaluative connotations of the semantic concepts. I concede that evaluative priming effects in both variants of the paradigm implicitly involve different explanatory mechanisms: The response-based explanation is simply not applicable to the S–S-based variant of evaluative priming, while it is the more plausible and more parsimonious explanation of S– R-based evaluative priming effects in comparison with the explanation when taking facilitated target encoding by evaluative congruency into account. Therefore, it is not necessary to create an overall explanation of evaluative priming effects, but it is crucial to search for a memory model that is compatible with evaluative priming in an S–S-based as well as in an S–R-based design. In Section 1.3, I will introduce the three-process model of evaluative priming that gives indications for the representation of the evaluative connotations in semantic memory and allows for both S–R-based and S–S-based evaluative priming.
1.3
The three-process model: A mutual facilitation account on
evaluative priming
The objective of this Section is to characterize the three-process model of evaluative priming and to discuss its suitability to account for the inconsistent findings in prior S–S-based evaluative priming studies. I will first characterize the theoretical conception and the main claims of the three-process model. Thereafter, I will post-hoc interpret previously reported effects in prominent S–S-based evaluative priming studies with the naming and the semantic categorization task, applying the logic of the three-process model. Based on the theoretical conception of the three-process model and the findings in prior studies, I will derive the hypotheses for my experiments on S–S-based evaluative priming.