Whether advanced cognitive competencies can be grounded in more elementary perceptual or associative processes is a matter of considerable current interest. Goldstone and Barsalou (Reference Goldstone and Barsalou1998) suggested that the distinction between low-level perception and high-level cognition is inapt; there is no chasm, but rather a continuum, from perception to conception. Similarly, Leech et al. (Reference Leech, Mareschal and Cooper2008) have proposed that relational priming is a basic building block for analogical reasoning. And, in the realm of language development, sequential learning and domain-general mechanisms may pave the way for language (Christiansen et al. Reference Christiansen, Dale, Ellefson, Conway, Cangelosi and Parisi2002). These examples illustrate how complex mental abilities could emerge from simpler behavioral mechanisms, thereby opening the door to understanding the ontogeny and phylogeny of higher-order cognition.
Mitchell et al. proceed in the opposite direction. They hypothesize that “associative learning depends on high-level cognitive processes that give rise to propositional knowledge” (target article, Abstract). For Mitchell et al., a basic and general ability – learning the relations between environmental events – is not due to the formation of links between representations of the events (associative account), but to the formation of mental propositions about how the events are related (propositional account).
On this view, propositions predicate some property of a subject, are generally held to be true or false, and are combined by laws of logical inference (Braine Reference Braine1978). Propositional reasoning thus entails processing and storing verbal premises and assertions. For example, “blocking” results from applying the following rule (De Houwer & Beckers Reference De Houwer and Beckers2003): If cue A alone causes the outcome with a particular intensity and probability, and if cues A and B together cause the outcome with the same intensity and probability, then cue B does not cause the outcome.
How organisms such as infants and nonhuman animals, who do not have language, deploy such propositions to infer relations about events is unclear; but Mitchell et al. believe that possibility and cite two studies (Beckers et al. Reference Beckers, Miller, De Houwer and Urushihara2006; Blaisdell et al. Reference Blaisdell, Sawa, Leising and Waldmann2006) which hint at rats' engaging in inferential propositional-based reasoning.
Beckers et al. (Reference Beckers, Miller, De Houwer and Urushihara2006) found that, after presenting rats with A+ trials in Phase 1 and AX+ trials in Phase 2, blocking (e.g., low responding to X) did not take place if the rats had earlier experienced C+, D+, and CD+ trials. According to Beckers et al., rats possess (as do humans) prior knowledge that, when two potential causes are presented together, a larger effect should occur than when only one cause is presented – additivity. Mitchell et al. agree, but they never say if this prior knowledge is innate or acquired; and if it is acquired, then how all organisms have come to the same understanding.
Nevertheless, armed with the additivity assumption, rats infer that X is not the cause of the outcome, when AX+ trials are presented after A+ trials and the outcome remains the same. But, if rats are pretrained with C+, D+, and CD+ trials, then they can tell that the additivity assumption is now false. The rats consequently reassess their beliefs; when they are later presented with A+ trials followed by AX+ trials, they deduce that X, as well as A, is a cause of the outcome.
But, perhaps something simpler is happening here. Haselgrove (under review) has shown that Rescorla and Wagner's (Reference Rescorla, Wagner, Black and Prokasy1972) associative model can readily explain the results of Beckers et al. (Reference Beckers, Miller, De Houwer and Urushihara2006). Haselgrove noted that five out of the six experimental cues in Beckers et al. were from the same modality (audition); all of them were of the same duration; and all of them were trained in the same context. Under those conditions, it is conceivable that the cues used for pretraining and the cues used for blocking entailed a common element. The Rescorla and Wagner model predicts that the conditioned properties acquired by the pretraining cues can transfer to the blocking cues via this common element. Thus, generalization of the pretraining contingencies to the experimental contingencies can explain Beckers et al.'s findings – reasonably and elegantly.
Mitchell et al. also consider the Blaisdell et al. (Reference Blaisdell, Sawa, Leising and Waldmann2006) project as support for propositional reasoning in animals. Here, rats could distinguish between a situation in which the outcome had been observed and a situation in which the outcome had been produced by the rats. This distinction actually reflects the dichotomy between classical and instrumental conditioning. In classical conditioning organisms learn that events in the environment are related to one another, whereas in instrumental conditioning organisms learn that their own actions change environmental events. The fact that animals can distinguish between observation (classical conditioning) and intervention (instrumental conditioning) is not new. Killeen (Reference Killeen, Commons and Nevins1981) showed that pigeons can discriminate whether their own behavior or “something else” caused changes in a light. How associative models can accommodate the specific data of Blaisdell et al. is a challenge, but it is hardly compelling evidence that rats are propositional reasoners.
Mitchell et al. are curiously silent about the emergence of associative learning. Must we also assume that human infants engage in complex propositional reasoning? If so, then this proposal is difficult to reconcile with studies of human development. Clancy et al. (Reference Clancy, Jacobsen and Silva1976) found that use of the if-clause (the meaning, without containing the connective) emerges at 2 or 3 years of age. And, “If p, then q” propositions (the type involved in blocking) are not understood until children are 6 years old (Braine & Rumain Reference Braine, Rumain, Flavell and Markman1983). Importantly, there seems to be consensus that children's early-developing inferences are likely to be acquired as part of learning their language (Braine & Rumain Reference Braine, Rumain, Flavell and Markman1983; Falmagne Reference Falmagne and Falmagne1975). Does that mean that human infants who have not yet acquired language cannot learn about regularities between environmental events? Doubtful. Sobel and Kirkham (Reference Sobel and Kirkham2006) found backward blocking in 8-month-old children; so, infants exhibit the same associative learning phenomena as do animals and human adults.
The great advantage of associative accounts is that their mechanisms seem to be available to all species across all developmental stages. Without explaining its origin and its developmental trajectory, we cannot fully comprehend any psychological process, particularly one as essential as associative learning.
