In contrast to a common assumption of dual-process models, Mitchell et al. argue that the formation of new associations in human memory is an exclusive product of controlled, propositional inferences, and that there is no empirical evidence for automatic processes of associative linking. In response to Mitchell et al.'s conclusion, we argue that their analysis conflates the distinction between the basic operating principles of a given process (i.e., associative linking vs. propositional reasoning) and secondary features pertaining to the conditions of its operation (i.e., automatic vs. controlled). If the conceptual independence of these dimensions is taken into account, the reviewed evidence regarding features of automaticity will be diagnostic about the operation of a particular type of process only to the degree that there is perfect overlap between the two dimensions (automatic=associative; controlled=propositional) – which seems debatable on both conceptual and empirical grounds.
Based on our own Associative-Propositional Evaluation (APE) Model (Gawronski & Bodenhausen Reference Gawronski and Bodenhausen2006; Reference Gawronski and Bodenhausen2007), we argue that the formation of a new association in memory should be understood as an effect that could be the result of two conceptually distinct mechanisms, associative linking and propositional reasoning. In our APE Model, we define associative linking as the creation of a new association between two concepts based on the mere co-occurrence of objects or events independent of the perceived validity of their relation. Propositional learning is defined as the creation of a new association as a result of syllogistic inferences about the validity of a given relation. The primary difference between the two processes is their dependency on subjective validity, in that only propositional learning, but not associative linking, takes the perceived validity of relations into account (see also Strack & Deutsch Reference Strack and Deutsch2004). As such, the two mechanisms should lead to the same outcome when the co-occurrence of two objects or events is interpreted as reflecting a valid relation. However, the two mechanisms may lead to different outcomes when the co-occurrence between two objects or events is regarded as non-diagnostic or invalid. This conceptualization incorporates Mitchell et al.'s emphasis of truth values as a core feature of propositional reasoning. However, it differs from Mitchell et al.'s approach, in that assumptions about automatic features represent empirical claims about the boundary conditions of the operation of the two processes, rather than defining characteristics that could be conversely used to identify their operation in a particular case.
To empirically distinguish between the two processes, we suggest that the actual operation of associative and propositional processes should be identified by means of their interactive effects on associations and beliefs. In the APE Model, we define associations as mental links between concepts independent of their subjective truth or falsity; beliefs are defined as the endorsed relations that are implied by validated or invalidated associations. This distinction has proven its usefulness in the social-cognitive literature, showing that activated associations can produce behaviors that are congruent with these associations, even when the relations implied by these associations are regarded as invalid (for a review, see Strack & Deutsch Reference Strack and Deutsch2004).
More importantly, there is suggestive evidence that such dissociations can sometimes be due to antagonistic effects of associative linking and propositional reasoning during the encoding of new information (e.g., Gawronski et al. Reference Gawronski, Deutsch, Mbirkou, Seibt and Strack2008; Rydell et al. Reference Rydell, McConnell, Mackie and Strain2006), supporting the usefulness of the proposed distinction in the formation of new associations. The basic notion of these studies is that the mere co-occurrence between two objects can create a mental association between these objects, even though the validity of the implied relation is rejected at the propositional level. Empirically, these differences are often reflected in dissociations between implicit and explicit measures (Fazio & Olson Reference Fazio and Olson2003), such that implicit measures (e.g., sequential priming tasks) reflect the mere co-occurrence between the two objects, whereas explicit measures (i.e., self-reported judgments) reflect the perceived validity of the implied relation.
Other evidence that is consistent with the notion of associative linking comes from research on spontaneous trait transference (e.g., Skowronski et al.Reference Skowronski, Carlston, Mae and Crawford1998), in which communicators have been shown to become associated with the traits they ascribe to others. In most cases, there is no logical basis to infer that a communicator has a particular trait (e.g., tidy) simply because he or she describes that trait in another person. Hence, it seems reasonable to assume that any such associations are the product of associative linking rather than propositional reasoning (Carlston & Skowronski Reference Carlston and Skowronski2005). To be sure, such associative linking processes may still depend on perceivers' attention, processing goals, or awareness of the co-occurrence. However, this by itself does not make the underlying learning process propositional, as defined in the proposed conceptualization.
Another important issue in this context is Mitchell et al.'s concern that proposing mutual interactions between associative and propositional processes would make the distinction between the two processes obsolete. Such interactions are a core assumption of our APE Model, which assumes that mutual interactions between the two processes are reflected in different mediation patterns of experimentally induced effects on activated associations and endorsed beliefs (Gawronski & Bodenhausen Reference Gawronski and Bodenhausen2006). Specifically, we argue that associative linking will often produce parallel effects on associations and beliefs, such that newly created associations provide the basis for explicitly endorsed beliefs. Conversely, newly created associations may be the product of propositional inferences, such that new beliefs generated in the course of validating currently accessible information may be stored in associative memory. Drawing on the abovementioned distinction between implicit and explicit measures, the first case is assumed to produce parallel effects on both kinds of measures, with effects on the explicit measure being fully mediated by the implicit measure. In contrast, the second case should produce parallel effects on both kinds of measures, with effects on the implicit measure being fully mediated by the explicit measure.
An illustrative demonstration of these diverging mediation patterns is a recent study by Whitfield and Jordan (submitted), who combined an implicit evaluative conditioning (EC) procedure (Olson & Fazio Reference Olson and Fazio2001) with a propositional impression formation task that used descriptive information about the conditioned stimulus. Their results showed that both the EC procedure and the impression formation task produced parallel effects on both explicit and implicit measures. However, in line with the predictions of the APE Model, EC effects on the explicit measure were fully mediated by the implicit measure, whereas impression formation effects on the implicit measure were fully mediated by the explicit measure (for related findings, see Gawronski & LeBel Reference Gawronski and LeBel2008; Gawronski & Strack Reference Gawronski and Strack2004; Gawronski & Walther Reference Gawronski and Walther2008).
Taken together, these results suggest that a conceptual distinction between associative and propositional processes in terms of their operating principles (rather than automatic vs. controlled features) has testable and empirically supported implications. More importantly, our analysis implies that the formation of new associations in memory can be the product of either associative or propositional processes, and that Mitchell et al.'s insightful review may speak only to the automatic versus controlled nature of these processes rather than to the general irrelevance of associative processes in human learning.
