A Dissociative Framework for Understanding Same-Different Conceptualization

Abstract

Cognitive, comparative, and developmental psychologists have long been interested in humans’ and animals’ ability to respond to abstract relations. Cross-species research has used relational matching-to-sample (RMTS) tasks in which participants try to find stimulus pairs that “match” because they express the same abstract relation (same or different). Researchers seek to understand the cognitive processes that underlie successful matching, and the cognitive constraints that create species differences in these tasks. Here we describe a dissociative framework drawn from cognitive neuroscience. It has strong potential to illuminate the area of same-different conceptualization. It has already influenced comparative research on categorization and metacognition. This dissociative framework also shows that species differences in same-different conceptualization have resonance with species differences in other comparative domains.

Introduction

Researchers have long explored humans’ and animals’ capacity to respond to abstract relations like sameness and difference [1]. James [2] described relational conceptualization as a backbone of thought. It may ground humans’ analogical reasoning [3]. It tracks cognitive maturation in children [4, 5]. And it may reveal discontinuities between human and animal cognition [6] and across animal species [7]. Thus, understanding the cognitive organization of same-different (SD) conceptualization is an important goal.

Here, we bring to this area a theoretical perspective drawn from cognitive neuroscience. We believe it has strong potential to illuminate SD conceptualization. It has already made substantive contributions to the comparative research areas on category learning and metacognition. Indeed, it may help explain a range of species differences in cognitive performance across multiple comparative domains.

A Theoretical Perspective

Neuroscientists exploring human learning generally accept that humans have multiple, dissociable learning systems. The implicit-procedural learning system is thought to be grounded in, and powered by, the primary reinforcement system. This system may underlie skill and habit learning [8] and performance in instrumental-conditioning, perceptual-categorization, and some discrimination-learning tasks [9, 10]. This system has a close connection to comparative psychology’s associative-learning construct [11]. Implicit-procedural learning occurs associatively through processes akin to conditioning—gradually, needing immediate reinforcement, not easily verbalizeable [11, 12, 13].

The explicit-declarative learning system depends on working memory [14] and executive attention [15]. It is useful in the present context to understand that these cognitive utilities would naturally support hypothesis testing and rule formation [16]. The explicit system learns by testing hypotheses and confirming one. In essence, participants self-construe the task, develop/discover their own performance rule, represent that rule in working consciousness, and can declare it to others. This self-construal process can even occur absent immediate reinforcement [17, 18], and rule discovery is often through a sudden insight [12, 17,19, 20]. These hypothesis-testing, rule-formation, and self-construal processes will be our working definition of explicit learning here. We propose that they play an important role in SD conceptualization—certainly for humans and perhaps for some other species as well. We understand that the conscious character, the verbal grounding, and the phenomenal feel of these processes could be different between humans and animals. It is an exciting prospect that new SD paradigms might give researchers the means to explore explicit cognition—and perhaps even working consciousness—in nonverbal animals.

Theoretical Contribution: SD Conceptualization

There are strong indications that this latter learning system—explicit-declarative learning—is often implicated in humans’ SD performances, especially their RMTS performance. In Smith, Flemming, Boomer, Beran, & Church [21], humans performed at high levels up to 99% correct, levels consistent with rule-based explicit cognition but not with implicit-procedural learning [22]. They also often discovered their RMTS task solution in a sudden insight. This qualitative learning leap is also diagnostic of explicit-declarative learning [19]. Smith, Jackson, and Church [23] asked participants to transition from a perceptually based RMTS task to a mandatorily conceptually based task. They found that this transition occurred seamlessly under control conditions, but that it was devastated under conditions of a working-memory load that drained away cognitive capacity. This targets the explicit processes of working memory as crucial to humans’ RMTS conceptualization. We believe there is sharp utility in exploring the possibility that humans’ RMTS performance is energized by a form of explicit-declarative cognition that involves particular neural structures, brain circuits, and levels of cognition that especially incorporate working memory.

Related evidence comes from analogical-reasoning tasks, within which working-memory interference also impairs performance [24,25]. For example, confirming analogies like noise:silence :: light:dark is impaired by manipulations that compete for working-memory resources. This convergence suggests a close theoretical relationship between RMTS and analogical-reasoning tasks [26].

The evidence from chimpanzees strengthens further the explicit-declarative hypothesis. It is particularly the language-trained or symbol-trained chimpanzees who succeed in RMTS tasks. Language and symbols would clearly offer a special cognitive affordance to an explicit-declarative rule encapsulating relational cognition. These data helped motivate the influential distinction between the paleological monkey and the analogical ape [7].

Regarding monkeys, the explicit-declarative hypothesis predicts they would falter in applying executive, prefrontal processes to RMTS tasks. They are weaker than humans in the cognitive capacities served by frontal brain systems. Their frontal cortices are small compared to those of apes and humans [27]. They are known to be compromised on frontal tasks that produce response competition or require response inhibition (Stroop tasks, flanker tasks, etc.,[28]).

Monkeys do falter. Fagot and Parron [29] used color patches as object pairs. Baboons matched successfully when the color pairs were grouped so closely as to appear a single stimulus. But, as spatial separation increased, requiring the matching of the relation between object pairs, performance collapsed. Wasserman, Fagot, and Young [30] studied baboons’ RMTS performance using multi-item Same or Different arrays. However, baboons matched arrays using a cue of visual entropy that might be first order and associative. When the visual entropy cue was removed by using object pairs instead of multi-item arrays, performance collapsed. Smith et al. [21] tried to help macaques transition between perceptual and conceptual forms of an RMTS task by gradually fading the first-order perceptual cues. However, macaques’ RMTS performance collapsed during this fading. None showed successful RMTS performance, despite sustained efforts over 260,000 trials. Remarkably, macaques collapsed during the fading near the point where humans make their successful transition to conceptual RMTS performance.

And yet Fagot and Thompson [31] found that baboons (closely related to macaques) have some cognitive capacity to perform RMTS tasks successfully, given dogged training with extensive trial repetition. Six baboons (of 29) met an 80% criterion after 15,000–30,000 trials on an RMTS task recycling 10 geometric shapes. Moreover, intriguingly, Mausgaard, Marzouki, & Fagot [32] found an analogue to the working-memory effect found in Smith et al. [23]. That is, a working-memory load compromised the baboons’ RMTS performance. So, it is not far-fetched to suppose that baboons bring to RMTS tasks a rudimentary, explicit-declarative cognitive capacity.

Finally, the explicit theoretical framework would predict that pigeons, having for evolutionary and brain-organization reasons the least explicit-declarative cognition, might show the least conceptual RMTS performance not grounded in an auxiliary cue like visual entropy. The data to date are consistent with this interpretation [33, 34].

Thus we believe a perspective from explicit cognition has great utility applied to SD conceptualization. A theoretically important form of relational cognition is allied to the cognitive utilities of working memory, hypothesis testing, and rule formation that are accepted to be resident in humans’ explicit-declarative learning system. Though other cues to SD performance exist, the explicit-declarative processes of relational cognition are distinctive and interesting from a comparative and evolutionary standpoint and deserve their own study. Our perspective also presents a provocative comparative map of the explicit cognitive capacity, consistent with many results from multiple species. It shows humans running on all explicit cylinders, children developing toward the mature explicit capacity with which to do so, apes hitting on some explicit cylinders (especially given language or symbol training), macaques showing a more rudimentary explicit capacity for responding relationally, and pigeons showing little sign of that capacity (to date!).

Theoretical Note on Wider Applications

The explicit theoretical perspective has the potential to apply more broadly in explaining comparative phenomena. We will note its extension to two additional research areas. Space permits, though, that we can only give these extensions a glancing blow.

In studies of categorization, humans have shown in many studies that rule-based category tasks foster in them the use of explicit-declarative cognitive processes. For example, humans learn rule-based tasks especially fast, to high levels, suddenly with the discovery of insightful realization, and with the need to have their working-memory capacity intact. In all these ways, the situation is like that described for the area of SD conceptualization. These category learners then hold their category rules in working consciousness, and they can declare their rules verbally to others [12, 17, 20]. Other studies have shown that monkeys have made some steps (not all steps!) toward learning category rules using their basic system for explicit cognitive processing [20, 35, 36, 37]. Monkeys are middling once more. And, in multiple studies, pigeons have shown almost no indication of having even the basic explicit cognitive system that macaques do [38, 39, 40, 41]. Rats have also failed to show explicit category learning [42].

It is the same comparative map as seen in the area of SD conceptualization: humans full-bore explicit-declarative, monkeys making beginning steps toward rule-based category learning, and pigeons showing extremely limited progress in this area. Pigeons may essentially not have started down the evolutionary pathway toward explicit, rule-based categorization—a pathway that the line of the primates clearly did start down.

The view from explicit cognition also informs the literature on animal metacognition. That literature was dominated for 20 years by a theoretical debate about whether animal metacognitive phenomena could be explained using only low-level associative learning processes [43, 44, 45, 46, 47, 48].

However, this idea has now collapsed, with a strong consensus emerging that some animal species have a basic form of metacognition that is more decisional and executive than associative. For example, apes have joined humans by showing highly sophisticated forms of uncertainty monitoring and metacognition [49]. Here, too, the crucial utility of working memory has loomed theoretically important. Smith, Coutinho, Church, & Beran [50] gave monkeys a task of metacognition, but on some trials also occupied monkeys’ working memory. The use of the two primary discrimination responses in the task were not affected by the working-memory load, consistent with the possibility that these were first-order behavioral responses. However, the metacognitive uncertainty response was sharply impaired by the memory load, consistent with the operation of an uncertainty-monitoring utility that uses working memory. Macaques’ URs seem to lie on a more executive plane of decision making. To still consider them associative would rob the important construct of associative responding of its disciplined meaning.

In this area, too, the comparative map completes in the expected manner. Pigeons have persistently not shown clear demonstrations of uncertainty monitoring or metacognition [51, 52] (compare monkeys in [53]). However, we do not foreclose pigeons’ capacity to show some aspects of metacognitive functioning, and promising studies continue in this area [54].

Thus, we believe that the present theoretical perspective, pointing to the important role of explicit forms of cognition in a variety of high-level cognitive domains, has an important role to play in explaining many task differences in cognitive performance, many species differences in cognition. Moreover, it may offer a unifying perspective on the nature of the cognitive systems emerging, or not emerging, during the course of cognitive evolution.

Discussion

Theoretical Questions

Our editors included in their precis probing target issues for contributors’ consideration. We do so now from the theoretical perspective of this article. That perspective—grounded in explicit-declarative cognition—of course is not binding on anyone. There are other possible perspectives [55]. Our comments are offered to promote dialog and to open new lines of research.

1. Our perspective describes an important dimension along which SD tasks vary—that is, in their associative-reactive or explicit-declarative character. For example, MTS tasks and RMTS tasks, respectively, likely lie toward the associative-reactive and explicit-declarative pole of cognition. This differentiating dimension may be broadly explanatory of multiple species performing multiple SD tasks.

2. For example, we would suggest that the nature of mental representations of same and different changes qualitatively across species. Some performances (e.g., MTS, 16-item SD performances by pigeons) likely occur associatively, reactively. Other performances (e.g., RMTS by humans and language trained chimpanzees) may make use of the explicit-declarative cognitive system. Our view would place monkeys (sometimes gravitating toward either pole) as intermediate. Premack [56], though not taking the neuroscience perspective of the present article, made similar species delineations.

3. SD representations probably are different even within a species during various SD tasks. Our theoretical perspective would suppose that pairwise RMTS tasks—demanding abstract conceptualization transcending concrete stimuli—recruit strongly explicit, rule-based cognition in the organisms that have that capacity. This explains why RMTS tasks produce a distinctive comparative data pattern.

4. Our perspective operationalizes the first-order—second-order distinction often made in the literature on SD conceptualization. First-order SD cues would be those that allow some tasks to be performed associatively, reactively. Second-order SD representations would be those framed as hypotheses and rules, resident within the explicit-declarative cognitive system. To illustrate, we believe that the representations required in two-item RMTS performances are often resident in working memory and in the explicit system. They would be considered second-order in that sense.

5. Our perspective also suggests how the representations of same and different change during cognitive development. The crucial mechanism could be that the explicit, rule-like system matures and strengthens and increasingly takes cognitive control in SD conceptualization tasks like RMTS tasks [5].

6. In this developmental transition, our perspective would target language and symbol labels as an important cognitive affordance available to humans. The labels (i.e., words) could concretize the abstraction needed for an RMTS performance. They could serve in generating, maintaining, testing, confirming, and declaring task hypotheses/rules. This may be why specifically the language-trained apes demonstrate RMTS competence. One might enliven rhesus macaques’ capacity for SD conceptualization by providing them with symbolic labels, and our present research concerns this problem.

7. There is ongoing interest in how SD conceptualization may grade into humans’ capacity for analogical reasoning. Our theoretical perspective suggests that the explicit cognitive processes typically energizing RMTS performance would be closely related to the capacity for analogical reasoning. In contrast, the capacity to equate 2 entropic displays would not have this close relation and would not provide a jumping-off place toward analogical reasoning.

8. Why do animals show glimmers of successful RMTS performance amidst abject failures to do so? We opine reluctantly, because we produced one of the abject failures [21]. The perspective from explicit-declarative cognition suggests that one will maximize an organism’s SD conceptualization by lifting its cognitive approach above the associative-reactive plane, and onto the explicit-declarative plane. The problem of accomplishing this opens an extremely rich domain from which many new lines of research could emerge.

9. Clearly, though, representations other than sameness and differentness may underlie SD responding (e.g., entropy). The question then arises whether those cues are deflationary with respect to claims of the capacity to represent same and different. We would frame this problem differently to avoid being critical. Alternative cues, like entropy, probably take information processing out of the realm of explicit-declarative processing and onto the plane of associative-reactive processing. So, the present theoretical perspective would suggest that the SD response to cues like entropy is quite different from explicit forms of SD conceptualization, because the former responses emerge from a different information processing system.

Conclusion

Through the decades, comparative theory self-organized around the unifying construct of associative learning. The area of SD conceptualization suggests the potential value added by a complementary perspective. That is, additional lines of research and theoretical progress may arise from fractionating the animal mind to study its distinct, dissociable learning and memory systems. A theoretical perspective from dissociative frameworks may also highlight new milestones with which to map the emergence of cognitive systems through evolution.