There is much food for thought in Garnham's (1998) commentary on our target article (Markman & Dietrich, 1998). Garnham objects to our endorsement of rule-governed mediating states. While he allows for using such states when describing cognitive capacities, he urges us, and all of cognitive science, to eschew rules as actual generators of behavior. Apparently, rules, for Garnham, are theoretical fictions like perfectly elastic balls and frictionless surfaces. We disagree with him on this point. Our interpretation of the relevant psychological data is that there are probably some rules that generate some behavior. Furthermore, given that there are rules that can be used to describe a certain kind of behavior, and given that other related processes are also in evidence, postulating that the rules are real and actually generate the behavior is indeed warranted.
2. Garnham's commentary reveals a deep disagreement between us over the conditions required to accept a theoretical construct as a real entity in the world rather than as a mere scientific expedient. This issue is important across the sciences. For example, in order to explain some physical phenomena, quarks have been proposed, with the weird property of partial charge. Does it follow that there really is such a property as partial charge? As another example from physics, in order to explain one set of properties of light (the diffraction properties), one has to view light as a wave, and in order to explain another set of properties (the photoelectric effect), one has to view light as a particle. Does it follow that there is something that is both a wave and a particle? In this reply, we do not intend to solve the problem of the status of theoretical properties in science, because it is one of the deepest problems in the philosophy of science. But we will suggest why moving from using rules to explain behavior to postulating them as the real generators of that behavior is sometimes warranted. We thereby suggest a methodological principle for a kind of psychological realism.
3. Garnham makes much of the distinction between describing or modeling certain behaviors with rules and actually imputing the exercise of those rules in the production of that behavior. His main argument against the existence of rules in the head is based on this distinction. He makes both an evidential and a logical claim, suggesting that all the positive data about generative rules is in fact just evidence that rules can be useful in describing or explaining that behavior. He adds that it does not follow from the fact that rules are useful in describing behavior that the rules actually generate or cause the behavior: "In classical physics, there is little temptation to see objects (e.g., thermostats) as having internal structure that corresponds to rules they follow. Why, then, has such a claim been made in the cognitive sciences? Why, indeed, has it so often been taken as an obvious truth?" (paragraph 4).
4. We agree that there is a difference between using rules to describe a certain behavior and claiming that rules cause or generate the behavior. But we never advocated viewing rules as generators of behavior merely on the basis that they were useful in describing that behavior. Indeed, in our target article we pointed out that Thelen and Smith (1994) provide an important demonstration of the case of infant stepping behavior, which was initially described as being rule-governed, but was later demonstrated not to be governed by rules.
5. Nonetheless, we do believe that there are cases in which the existing data warrant positing rules. These data are not simply findings that can be described as rule-governed; they also appear inconsistent with other explanations_. For example, as discussed in our target article, Marcus, Brinkmann, Clahsen, Wiese, and Pinker (1995) demonstrate that the productive form for plural nouns in German is relatively infrequent in the language. Thus, a statistical model will not account for this productive behavior. In this case, a rule-based model appears to be a better explanation than competing accounts.
6. More generally, we suggest that rules should be posited as psychologically real whenever the psychological data are better explained by positing a system of rules than by some alternative account. While it is always possible that rule-governed models will be proven wrong by subsequent experiments, it is not a necessary indictment of rule-governed mediating states that there are some systems that can be described as using rules that are not actually governed by those rules.
7. There are two other conditions that would, at least in principle, help determine whether rules are real generators of a certain kind of behavior. The first is that, as with any psychological mechanism, there is a nexus of other psychological capacities associated with rules that could in principle be detected before psychologists actually hypothesize that rules are the causes of observed behavior. For example, if rules produce behavior, then it is likely that they are stored in memory in some sort of rule-base. Hence, in order for the rules to cause the behavior in question, they must be retrieved from that rule-base. Hence, there must be evidence for memory access to a rule-base. The memory access could be tested for. The second condition is simply a use of Occam's Razor. Any psychological entity -- algorithms, rules, images, meanings, concepts, pronouns, a language of thought, etc. -- can be hypothesized to cause behavior on the basis of its explaining the behavior, provided that no other entity can explain the behavior in more simple way_. This use of Occam's Razor is a perfectly legitimate use of the principle of "inference to the best explanation." All sciences use this principle, and psychology should, too.
8. Garnham has a laudable goal in his commentary. By removing rules from the head, he saves cognitive science from those, like latter-day Wittgensteinians, who attack cognitive science by attacking the idea that all cognition is mediated by rules (paragraph 11). However, our solution to this problem is better than Garnham's, because he is throwing the baby out with the bath water. We save cognitive science from wholesale attacks on rule-governed systems while preserving the use of rules where they are in fact warranted using our DIVERSITY HYPOTHESIS that different levels and different kinds of cognition require different mediating states and different mediating principles (see Markman & Dietrich, paragraphs 15, 16, and 67).
9. We believe that there is sufficient evidence in a few domains of cognitive processing to suggest that rule-governed systems are psychologically real. Positing rules as explanations for some cognitive behaviors does not require that they be used for others. The beauty of our diversity hypothesis is that we need only posit the type of mediating state necessary for the cognitive processes being explained. There is no need to posit extremely complex mediating states for a wide range of cognitive processes simply because there are some cognitive processes that appear to be rule-governed.
Garnham, A. (1998) Rules in Whose Mind: The Modeler's or the Modeled's? PSYCOLOQUY 9(62) ftp://ftp.princeton.edu/pub/harnad/Psycoloquy/1998.volume.9/psycoloquy.98.9.62.representation-mediation.6.garnham
Marcus, G. F., Brinkmann, U., Clahsen, H., Wiese, R., & Pinker, S. (1995). German inflection: The exception that proves the rule. Cognitive Psychology, 29, 189-256.
Markman, A.B. & Dietrich, E. (1998) In Defense of Representation as Mediation. PSYCOLOQUY 9(48) ftp://ftp.princeton.edu/pub/harnad/Psycoloquy/1998.volume.9/psyc.98.9.48.representation-mediation.1.markman
Pinker, S., & Prince, A. (1988). On language and connectionism: Analysis of a parallel distributed processing model of language acquisition. Cognition, 28, 73-193.
Thelen, E., & Smith, L. B. (1994). A Dynamic Systems Approach to the Development of Cognition and Action. Cambridge, MA: The MIT Press.