Reasoning: Domain-General vs. Domain-Specific

One of the questions raised by the issues discussed in the first post on reasoning was: Is human reasoning domain-general or domain-specific? If you can answer this question definitively, you will be world famous, and I imagine someone will find a way to get you a Nobel Prize (they managed to get Daniel Kahneman one!). Obviously, I can't answer that question definitively, and the empirical research and theoretical arguments are a long way from being conclusive. In fact, the domain-general vs. domain-specific problem tends to rear its ugly head in all areas of reasoning, not just what traditionally falls under the rubric of "reasoning," and for the most part, it remains unsolved everywhere. However, since this is a series of posts on reasoning, I'm just going to focus on theories of domain-specific or domain-general reasoning mechanisms in reasoning research. While the domain-general view is the older view, I'm going to start with different domain-specific theories. After that, it's domain-general's turn. So, let's get started.

Domain Specific

The impetus for the development of domain-specific accounts of reasoning is not limited to errors and reasoning like those in the Wason selection task. The content of reasoning tasks can influence the way we reason in many different ways. For example, in one study, when asked to write addition and division word problems, the type of problem they wrote (either addition or division) depended on the content they were asked to include. If they were to use items from the same category, participants were more likely to write addition problems, while if the items were thematically related, they were more likely to write division problems. It's very difficult for traditional domain-general accounts of reasoning to explain thesephenomena. Thus, over the last couple decades, domain-specific accounts have begun to gain prominence.

There are a few different views of how and why reasoning is domain-specific. Perhaps the oldest (in cognitive science, at least), is the embodied cognition view, which is closely related to the phenomenology of philosophers like Edmund Husserl and Maurice Merleau-Ponty. A related view of domain-specific reasoning is the situated cognition perspective, which arose out of education theory and computer science. Finally, the most prominent domain-specific account of reasoning today comes from evolutionary psychology, which posits domain-specific modules that have evolved to solve specific types of problems. I'll briefly talk about these three different domain-specific perspectives individually.

I. Embodied Cognition

Embodied cognition is a pretty sweeping label for various views of cognition that see cognition as involving a tight coupling between an embodied organism and its environment. This view implies that environmental conditions, goals, and bodily makeup influence the structure of cognitive processes. Thus, there are few, if any domain-general reasoning mechanisms. For example, in J. J. Gibson's ecological approach to perception has been used by some theorists to explain how people learn and reason about objects in their environment. In particular, the concept of affordances is applied to reasoning situations, such that the way we reason about different objects in our environment depends on our goals and the actions that those objects afford. Another example, from empirical research, is the use of movement or the sensory-motor areas of the brain in solving what were previously seen as problems requiring abstract rules. For instance, when people complete a spatial reasoning task, movements consistent with the task facilitate performance. Thus, when performing tasks involving reasoning about routes between two points, people do better if they perform the movements involved in following the routes, and when performing tasks involving pointing to the location of an object after rotating away from the object, participants do better when they are blindfolded and allowed to actually rotate than when they are not blindfolded and mentally simulate the rotation¹.

There are more than a few problems with the embodied cognition approach as it is currently formulated. One problem is that it is currently formulated in several different ways², most of which are fairly vague. Another is that, while research like that on spatial reasoning described in the previous paragraph has shown that embodiment is important, and that we use our sensory-motor systems to aid in the solution of many types of reasoning problems, there has been very little research, and no non-linguistic research, attempting to show that all or even most cognition is "embodied" in the various senses that the embodied cognition approach posits. In fact, outside of spatial reasoning, the sort of reasoning likely to be most closely coupled with the sensory-motor system, there's almost no evidence outside of linguistics that reasoning is embodied.

II. Situated Cognition

The situated cognition approach is similar to the embodied cognition approach, but places more emphasis on the social environment. This perspective is heavily influenced by the work of Lev Vygotsky, and is similar to the pragmatism of people like Dewey. There are four primary features of the situated cognition view³:

(1) Action is grounded in the concrete situation in which it occurs.
(2) Knowledge does not transfer between tasks.
(3) Training by abstraction is of little use.
(4) Instruction must be done in complex, social environments.

Examples meant to demonstrate (1) often involve people being able to perform a certain task in the context in which it was learned, or is often used, but not in other (usually abstract) contexts. For instance, Carraher et al.⁴ reported that there were street children in Brazil who could perform complex calculations rapidly and accurately when selling items on the street, but were unable to perform the same types of calculations in the classroom environment. However, (1) is problematic because children are, in fact, able to use skills learned in school in non-school contexts. Thus, it does appear that learning can be tightly coupled with the learning context, but that this is not always the case, and it is possible to use knowledge in more than one type of situation. A similar criticism applies to (2) as well. While transfer between tasks is not always possible, there is a large body of research in which such transfer is demonstrated⁵.

Where (3) is concerned, it is true that in some cases, specific, task-oriented instruction is better than abstract instruction. However, abstract instruction is also valuable in many contexts, and task transfer is much more likely when instruction is abstract than when it is situation-specific. Finally, (4) claims that learning should be done with hard problems, in natural social situations. This is particularly true for tasks that are difficult and/or social to begin with, such as the distributed navigation tasks studied by Hutchins and his students⁶. However, this is not true for all situations. Often, particularly with children, overly complex problems cause frustration and a lack of motivation, which get in the way of learning. Similarly, there are benefits to both individual and social learning, and which type works better often depends on the type of information to be learned and the learner's goals.

III. Evolution and Modularity

The third, and currently most popular domain-specific view of cognition is the one that arose out of evolutionary psychology, and posits that different types of tasks activate different evolved modules. The most widely studied and discussed module is the cheater-detection module hypothesized in social exchange theory⁷. The primary evidence for the existence of a cheater-detection module comes from the Wason selection task research described in the first post. In addition to this empirical research, there are game-theoretic simulations and evolutionary stories that support the existence of cheater-detection modules. As Sugiyama, et al. wrote⁸:

Pressures favoring social exchange exist whenever one organism (the provisioner) can change the behavior of a target organism to the provisioner’s advantage by making the target’s receipt of a provisioned benefit conditional on the target acting in a required manner. This mutual provisioning of benefits, each conditional on the other’s compliance, is what is meant by social exchange or reciprocation.

Evolutionary biologists have shown through game-theoretic techniques that adaptations for social exchange can be favored and stably maintained by natural selection, provided they include design features that (i) enable them to detect cheaters (i.e., those who do not comply or reciprocate), and (ii) cause them to channel future benefits to reciprocators, not cheaters. (p. 11537)

There is also some neurological evidence, based on a patient with a severely damaged limbic system, for the neurological basis of a cheater-detection module⁹.

While both the embodied and situated cognition views seem to provide at least some insight into the nature of cognition, and indicate that in some cases, cognition is domain-specific, it's not clear that the social exchange theory can say the same. Almost all of the research on social exchange in human psychology involves either the Wason task or game-theory simulations. Given that Cosmides and her colleagues have been repeatedly criticized for completely misunderstanding the nature of the Wason task, and since it has been shown that relevance, rather than cheater-detection, maybe what creates improved performance in the task, it's hard to believe that the social exchange theory is still around, much less as popular as it is today. The evolutionary story described above is somewhat compelling, but it's just that, a story.

There is another view, which is not necessarily derived from evolutionary psychology, that agues for a modular, domain-specific view of cognition. This view, independent of the social exchange theorists, is championed by Sperber, who gives the following examples of domain-specific modules:

Avoidance of vertical drops: Human infants (and other baby animals also) perceive and avoid vertical drops in terrain, even if they have had no experience of falling before, as demonstrated by means of the well-known “visual cliff” experiments initiated by Gibson & Walk (1960). This is an obvious modular adaptation to a serious hazard facing animals moving on the ground. To be efficient, this particular module had better not depend on learning. It is as good an example of an innate cognitive module as one may ever hope to find.

The Garcia effect: Rats and other animals are innately equipped to develop an aversion to whatever type of food seems to have made them sick. This is a highly specialised one-pass-learning module. The outcome of such learning is a novel capacity, that of reacting with aversion to a specific kind of food. If the rat develops, say, three such aversions, then it has three distinct abilities. It could be that the learning process and each specific aversive reaction are all carried out by the same module: learning consisting in adding to the initially empty proprietary data-base of the module data about specific foods to be avoided. Or it could be that the learning process results each time in the setting up of a new module or sub-module dedicated to a specific aversive food. So, which is it: one general food-aversion module with a growing data-base, or a learning module producing as many micro-modules as there are aversions? This is an empirical issue that might be decided by answering questions such as the following: Do aversive reactions to different foods employ different detection procedures (as opposed to the same procedure using different data)? Does a new aversion recruit distinct brain tissues? Can the more general ability to generate new aversions and each of the more specific aversions be selectively impaired? Positive answers to such questions would suggest that to each new aversion corresponds a new mini-(sub)-module.

Face recognition: I assume that face recognition is modular (which is controversial, but see Kanwisher & Moscovitch 2000). If so, we are dealing, as in the case of the Garcia effect, with two types of modular abilities: a general learning ability to form specific abilities to detect specific faces. Is there a general face-recognition module that performs both functions or are individual-face-detectors developed as autonomous mini-(sub)-modules? This is an empirical question to which we do not have an answer. As in the case of the Garcia effect, these are nevertheless genuinely distinct possibilities involving subtle differences in the way these abilities may be carried out and impaired.

Language faculty and linguistic competences: The language faculty is a complex learning module that, given proper linguistic and contextual inputs, yields one or, in the case of plurilinguals, several mental grammars. Each of these grammars is itself a complex module subserving both verbal coding and decoding in a given language. Each mental grammar has a distinct developmental story, and can selectively decay or be impaired. It is plausible that, say, the two mental grammars of a bilingual individual are sub-modules of a more general mental universal grammar and, as such, share some resources (Dehaene et al. 1997, Kim & al. 1997).

Reading : Reading is too recent a cultural skill for a specialized innate module to have evolved. Yet reading systematically involves the same brain site located in the left occipito-temporal sulcus and sometimes described as the “visual word form area.” Dehaene speculates that “the human brain can learn to read because part of the primate visual ventral object recognition system spontaneously accomplishes operations closely similar to those required in word recognition, and possesses sufficient plasticity to adapt itself to new shapes, including those of letters and words. During the acquisition of reading, part of this system becomes highly specialized for the visual operations underlying location- and case-invariant word recognition. … Thus, reading acquisition proceeds by selection and local adaptation of a pre-existing neural region, rather than by de novo imposition of novel properties onto that region” (Dehaene, forthcoming). Reading skill can be viewed as resulting from a process of ad hoc modularisation of already specialised brain tissues.

As Sperber notes, some of the modules he posits are controversial. This is particularly true of certain views of the language module, and the face-recognition module, though neuroscientific research unequivocably indicates that face-recognition is occuring in brain regions that don't perform general object recognition (e.g., the right middle fusiform gyrus)¹⁰. Still, it's not clear from the research cited by Sperber and others just how modular the mind is (Sperber believes that it is massively so), or whether there are some domain-general rules for reasoning. It's also not clear that the evolutionary explanations offered for other modules are needed to understand the sorts of modules Sperber discusses.

Domain General

As the previous section makes clear, even if the current domain-specific theories of reasoning and cognition in general are lacking in many ways, theorists who believe that reasoning uses domain-general processes have a lot of data to explain. One way to do this is to posit that domain-specific differences that appear to be differences in processes are actually representational differences. For instance, in one of the most prominent computational models of reasoning, the ACT-R model¹¹, reasoning performance is partially determined by the structure of semantic memory. Furthermore, research on schematic memory has shown often that the types of inferences that we can make about things is largely determined by the structure of our representations of those things.

Another way to explain domain-specific effects in the literature is to refer to the distinction between active processing and automaticity¹². When people are learning how to perform a particular task, or to solve a particular type of problem, they tend to use domain-general processes. However, as they repeatedly encounter the same type of task, it becomes economical to store successful solutions as a whole. After these solutions are stored, people no longer need to use domain-general processes to solve those problems. Instead, they can simply retrieve the previously stored solution. There is a great deal of evidence for the the shift from active processing to automaticity, though it may not explain all of the instances of domain-specific processing. Furthermore, this shift isn't always helpful. Sometimes, this shift can be counterproductive, as when previous solutions make it difficult to discover new solutions to similar problems.

Thus, we're back where we started. There are cases that seem to entail domain-specific processing, but there may be domain-general explanations of those cases. Furthermore, domain-specific processes may actually be learned by using domain-general processes and storing specific solutions derived with them. Certainly there are some domain-specific modules in the brain, including some of the ones Sperber described in the passage quoted above. However, because there is no conclusive evidence for a general view of cognition from either perspective, and since both tend to be able to explain the bulk of the data available, we're just going to have to wait for anything like a definitive answer.

You may be thinking that I paid a lot more attention to domain-specific views than to domain-general ones. One reason for this is that I wanted to get in the criticisms of domain-specific views, which come from the domain-general camp. The other is that the next two posts, the one on categorical and analogical reasoning, and the one on mental models, will be largely from the domain-genera lperspective, and thus will in essence be lengthy descriptions of domain-general theories. So, if you waded through all that domain-specific stuff just to get to the domain-general theories, I'm sorry, but you're going to have to wait.

1 Glenberg, A. (1997). What memory is for. Behavioral and Brain Sciences, 20(1), 1–55.'
2 See this paper by Margaret Wilson for a description of 6 different senses of "embodied cognition."
3 Quoted from Anderson, Reder, & Simon(1996). Situated learning and education. Educational Researcher, 5-11.
4 Carraher, T. N., Carraher, D. W., & Schlieman, A. D. (1985). Mathematics in the streets and in the schools. British Journal of Developmental Psychology, 3, 21-29.
5 Anderson, et al. (1996); Perkins, D. N., & Solomon, G. (1989). Are cognitive skills context-bound? Educational Researcher, 18(1), 16-25.
6 See here for a short description of this research.
7 Cosmides, L. & Tooby, J. (1992). Cognitive adaptations for social exchange. In J. Barkow, L. Cosmides, & J. Tooby (Eds.), The adapted mind: Evolutionary psychology and the generation of culture. New York: Oxford University Press.
8 Sugiyama, L., Tooby, J. & Cosmides, L. (2002).Cross-c ultural evidence of cognitive adaptations for social exchange among the Shiwiar of Ecuadorian Amazonia. Proceedings of the National Academy of Sciences, 99(17), 11537-11542.
9 Stone, V., Cosmides, L., Tooby, J., Kroll, N. & Knight, R. (2002). Selective impairment of reasoning about social exchange in a patient with bilateral limbic system damage. Proceedings of the National Academy of Sciences, 99(17), 11531-11536.
10 Rossion, B.; Schiltz, C., Robaye, L., Pirenne, D., Crommelinck, M. (2001). How does the brain discriminate familiar and unfamiliar faces: a PET study of face categorical perception. Journal of Cognitive Neuroscience, 13, 1019-1034.
11 Anderson J. R.. 1993. Rules of the Mind. Hillsdale, NJ: Erlbaum; Anderson J. R.; & Lebiere C., eds. 1998. The Atomic Components of Thought. Mahwah, NJ: Erlbaum.
12 See e.g., Markman, A. B. & Gentner, D. (2001). Thinking. Annual Review of Psychology, 52, 223-247.