Analogical apes and paleological monkeys revisited
Thompson and Oden (Reference Thompson and Oden2000) concluded that monkeys are paleologicians; their conceptual categories are based on shared predicates – absolute and relational features bound by perceptual and/or associative similarity, whereas symbol trained apes are analogical in the sense that they perceive abstract propositional similarities spontaneously. Moreover, symbol systems provide the representational scaffolding for manipulation and expression of propositional knowledge in relational matching-to-sample (RMTS) and related nonverbal analogy tasks (Gillan et al. Reference Gillan, Premack and Woodruff1981; Oden et al. Reference Oden, Thompson, Premack, Gentner, Holyoak and Kokinov2001). However, the authors of the target article dismiss this distinction, essentially arguing that only humans are truly analogical organisms, whereas all other animals are paleological in their conceptual abilities.
We suggest that the authors' necessary criteria as to what constitutes an analogy may be overly exclusive. As they note, when Sarah, a chimpanzee, under conditions of non-differential reinforcement, constructed both the base and target relations of an analogy from four of five elements (Oden et al. Reference Oden, Thompson, Premack, Gentner, Holyoak and Kokinov2001, condition 4), she did so by equating the number of within-pair featural differences, independently of the physical nature of those differences. We believe that Sarah's performance is evidence of her ability to construct a formal, if not material, analogy in which properties (i.e., number of featural changes) of one set of objects X are mapped onto those of another set Y according to the principles of mathematics (Rothbart Reference Rothbart2007).
In this regard Sarah is perhaps not alone in her formal analogical abilities, given the ability of other primates and birds to match entropy levels regardless of the physical icons used to instantiate this mathematical property (e.g., Fagot et al. Reference Fagot, Wasserman and Young2001; Wasserman et al. Reference Wasserman, Young and Fagot2001; Young & Wasserman Reference Young and Wasserman1997). However, Sarah's performance on functional analogy problems (Gillan et al. Reference Gillan, Premack and Woodruff1981) provides for now – the authors' concerns notwithstanding – the sole evidence of a nonhuman recognizing a material analogy by, “observing similarities between materials or types of phenomena” (Rothbart Reference Rothbart2007, p. 24).
Is the relational matching-to-sample task exemplary of analogical reasoning?
There is compelling evidence of relational conceptual capabilities in nonhuman animals (for reviews see, e.g., Wright & Katz Reference Wright and Katz2006, Zentall et al., in press). Macaque rhesus monkeys, for example, exposed to the same “symbol” training procedures as chimpanzees in a RMTS task (Thompson et al. Reference Thompson, Oden and Boysen1997) learned to generalize their responses to a circle whenever the two items in the discriminative cue matched (e.g., AA, etc.), and to a triangle whenever they did not (Washburn et al. Reference Washburn, Thompson and Oden1997).
Flemming et al. (Reference Flemming, Beran and Washburn2007) also found that rhesus monkeys (Macaca mulatta) not only correctly chose novel identical/nonidentical relational pairs in the presence of discriminative color cues, but they also correctly chose the color itself in the presence of the relational pairs. Importantly, however, unlike chimpanzees (Thompson et al. Reference Thompson, Oden and Boysen1997), none of the monkeys in these experiments (Flemming et al. Reference Flemming, Beran and Washburn2007; Washburn et al. Reference Washburn, Thompson and Oden1997) responded above chance on the RMTS task over literally thousands of trials.
The authors argue that, “a cognizer could pass a classic S/D [same-different] task by calculating an analog estimate of the variability between items in the sample display and then employ a simple conditional discrimination to select the appropriate behavioral response to this chunked result” (sect. 2.2, para. 4). But, given that monkeys can learn two-item conditional relational judgment tasks as described earlier, should they not then also acquire the relational matching task if judgments of sameness and difference may be reduced to the discrimination of between-item variability or entropy? Yet, clearly the monkeys do not.
Moreover, results obtained by Flemming et al. (Reference Flemming, Beran and Washburn2007) from rhesus macaque monkeys further suggests that categorical same/different judgments, although not necessarily prevalent in early stages of relational discriminations, can be learned and applied through the implementation of entropy-infused displays. Monkeys were not only successful in making a two-choice discrimination between sets of six identical or nonidentical stimuli, but also with pairs of novel stimuli. When the number of items in the display was systematically reduced to two, the monkeys' performance levels neither declined nor revealed asymmetric effects (on same vs. different trials), as would be expected if the animals' judgments were still under perceptual control of entropy. Flemming et al. (Reference Flemming, Beran and Washburn2007) argued that same/different judgments are not entirely based on entropy-infused displays, but rather that conceptual categorical judgments can emerge and overcome the initial dominance of perceptual-based responding.
A conditional cue is proto-symbolic
Nevertheless, as noted, these same monkeys still failed to acquire the RMTS task. Why might this be? As described by Thompson and Oden (Reference Thompson and Oden1996; Reference Thompson and Oden2000), conditional S/D tasks can be “solved” following application of a single matching operator, whereas for success in the RMTS task the animal must not only apply the matching operator to the sample and alternatives, but also to the abstract encoded outcomes.
The “profound disparity” in the performance of chimpanzees and monkeys in the RMTS task lies then in the chimpanzee's capacity – like that of children (Rattermann & Gentner Reference Rattermann, Gentner, Holyoak, Gentner and Kokinov1998b) – to symbolically recode abstract relations into iconically equivalent symbols, thereby reducing relational matching to a task that is functionally equivalent to physical/perceptual matching (Thompson & Oden Reference Thompson and Oden1996; Reference Thompson and Oden2000; Thompson et al. Reference Thompson, Oden and Boysen1997; Reference Thompson, Rattermann and Oden2001), a process, “akin to acquiring a new perceptual modality” (Clark Reference Clark, Carruthers and Boucher1998, p. 175).
Penn et al. suggest that, in part, the ability to label relational information is unique to the human mind and responsible for the discontinuity implicated by the relational reinterpretation (RR) hypothesis. In fact, we believe there is comparative evidence to suggest that similar symbolic systems also apply to our nearest primate relatives. In the case of other animals, like monkeys, however, no evidence as yet indicates that a conditional cue can acquire the full status of a symbolic label, although it would seem that symmetric treatment of a conditional cue lays the foundation for a recoding of relational information as set forth by the RR hypothesis.
