Arthur B. Markman (1998) Domain-independent Mediating States are Rare. Psycoloquy: 9(56) Representation Mediation (3)

Volume: 9 (next, prev) Issue: 56 (next, prev) Article: 3 (next prev first) Alternate versions: ASCII Summary
PSYCOLOQUY (ISSN 1055-0143) is sponsored by the American Psychological Association (APA).
Psycoloquy 9(56): Domain-independent Mediating States are Rare

Reply to Bringsjord on Representation-Mediation

Arthur B. Markman
Department of Psychology
University of Texas
Austin, TX 78712

Eric Dietrich
PACCS Program in Philosophy
Binghamton University
Binghamton, NY


Bringsjord (1998) suggests that Markman & Dietrich (1998) argued that there are no domain-independent abstract mediating states in cognitive systems. Bringsjord presents a small study using only logic experts, which he argues is a demonstration that there are abstract mediating states. His interpretation of Markman & Dietrich's claim is inaccurate. More important, his experiment does not provide evidence for the existence of domain-independent mediating states. Finally, Bringsjord suggests that an agent-based AI might be a good domain for studying the interactions among types of mediating states. This is an interesting proposal, worthy of further investigation, and is quite consistent with the arguments in our target article.


compositionality, computation, connectionism, discrete states, dynamic Systems, explanation, information, meaning, mediating states, representation, rules, semantic Content symbols
1. Bringsjord (1998) makes three points in his commentary. First, he suggests that Markman & Dietrich (1998) have made a weak argument that human cognitive systems have no abstract representations. Second, he presents data suggesting that people do in fact have abstract representations. Third, he suggests that an agent-based artificial intelligence may be a good way of examining the issues raised in our target article. We will address these issues in turn.

2. On the first point, our argument was not that there are no abstract, domain-independent mediating states in cognitive systems, but rather that "the balance [of evidence] seems to favor concreteness for many_ cognitive processes" (Markman & Dietrich, 1998, paragraph 54, emphasis added). For example, Bassok, Chase & Martin (1998) find that the content of arithmetic word problems written by college undergraduates has a strong influence. College students prefer to write addition problems using objects from the same taxonomic category (e.g., adding apples and oranges) and they prefer to write division problems using objects from different taxonomic categories that are thematically related (e.g., dividing apples among baskets). They prefer not to write addition problems using objects from different categories (adding apples and baskets) or division problems using objects from the same category (distributing apples among oranges). Arithmetic seems a likely domain for the use of abstract representations (particularly among college students), yet this evidence strongly suggests that people's reasoning is influenced by the content of the domain.

3. We took evidence like this (as well as evidence bearing on logical tasks like the Wason selection task) to mean that in many cases -- even those for which abstract representations would be useful -- people use mediating states tied to specific instances. Bringsjord does not discuss the implications of evidence like this. Instead, he focuses on the logic of our argument. We were quite clear that some cognitive processes may involve abstract mediating states, although they are rare. Demonstrations, such as the one Bringsjord presents, that some cognitive processes involve abstract representation only suggest that there are also some cases in which the cognitive system has highly abstract mediating states.

4. It is worth stressing that our conclusion that domain-independent mediating states are rare is based on a serious assessment of the available psychological data, as well as on computational work in areas like case-based reasoning (e.g., Kolodner, 1993), which suggests that reasoning often involves the use of specific cases. Domain-independent mediating states are still in vogue in some quarters of AI. Applying this work to human cognition (as Bringsjord does) requires an assumption that domain-independent abstract mediating states are essential to (and ubiquitous in) higher cognition, but this assumption has no empirical support. Applications of this type suggest that our conclusion (that domain-independent mediating states are rare) is not weak. Moreover, our point in this part of our target article was that some anti-representationalists, who made a serious assessment of the available data, have leapt to the conclusion that there are no representations, whereas this conclusion is likewise unwarranted.

5. Bringsjord's pilot study does not provide significant support for his position. He demonstrates that a difficult logical puzzle can be solved by people highly trained in logic. As he suggests, these logic experts may indeed have learned to reason with abstract representations (see also Rips, 1994). On the basis of these data, however, it is also possible that the experts have been exposed to so many cases of logical reasoning problems that they are able to retrieve similar problems they solved in the past and to use them to solve the new problem. In this case, the experts would be solving logic problems using other specific cases. It is also possible that people are recoding these problems to an intermediate level of abstraction in much the way physics experts seem to recode physics word problems into types such as "rate problems" or "incline-plane problems" (Chi, Feltovich, & Glaser, 1981). There is evidence that people often recode content-bound versions of the Wason selection task when they recognize them as instances of familiar social schemas such as permission (e.g., Cheng & Holyoak, 1985). The informal study provided by Bringsjord does not distinguish between these possibilities.

6. Finally, Bringsjord suggests that agent-based AI may be a fertile area for looking at these representational issues. We certainly agree, and our definition of mediating states is squarely in this camp. Computational approaches to cognition are a fruitful way to examine the role of representation in cognitive processing. Indeed, adopting a computational model of cognition requires positing some form of representation for the computation to act over. Any approach that is amenable to studying the interactions among types of representations will be useful for exploring the issues raised in the target article.


Bassok, M., Chase, V. M., & Martin, S. A. (1998). Adding apples and oranges: Semantic constraints on application of formal rules. Cognitive Psychology, 35(2), 99-134.

Bringsjord, S. (1998) Domain-Independent Abstract Mediating States and AI. PSYCOLOQUY 9(53)

Cheng, P. W. & Holyoak, K. J. (1985). Pragmatic reasoning schemas. Cognitive Psychology, 17, 391-416.

Chi, M. T., Feltovich, P. J., & Glaser, R. (1981). Categorization and representation of physics problems by experts and novices. Cognitive Science, 5, 121-152.

Kolodner, J. (1993). Case-based reasoning. San Mateo, CA: Morgan Kaufmann Publishers.

Markman, A.B. & Dietrich, E. (1998) In Defense of Representation as Mediation. PSYCOLOQUY 9(48)

Rips, L. J. (1994). The Psychology of Proof: Deductive reasoning in human thinking. Cambridge, MA: The MIT Press.

Volume: 9 (next, prev) Issue: 56 (next, prev) Article: 3 (next prev first) Alternate versions: ASCII Summary