The Mental Models Theory of Reasoning: Refinements and Extensions.
Author: Morris, Bradley J.
Pub Date: 09/22/2010
Publication: Name: The Psychological Record Publisher: The Psychological Record Audience: Academic Format: Magazine/Journal Subject: Psychology and mental health Copyright: COPYRIGHT 2010 The Psychological Record ISSN: 0033-2933
Issue: Date: Fall, 2010 Source Volume: 60 Source Issue: 4
Accession Number: 242454427

The Mental Models Theory of Reasoning: Refinements and Extensions

Mahwah, NJ: Lawrence Erlbaum

pp. vii-224, ISBN 978-0-8058-4183-1

The mental model theory (MMT) is one of the most influential theories of cognition and reasoning. MMT suggests that reasoning occurs via semantic-situational models that represent possible states (Johnson-Laird, 1983). Mental models have been used to explain performance in a variety of areas such as language comprehension (MacWhinney, 2008), analogical reasoning (Gentner, 2002), and deductive reasoning (Johnson-Laird, 2005). For example, MMT has been used extensively in the investigation of deductive reasoning as an alternative to rule-based theories. Taking the title of The Mental Models Theory of Reasoning: Refinements and Extensions as a starting point for discussion, the following sections outline what I take as both the refinements and extensions/modifications of MMT. Any change to a theory must increase its explanatory power and scope yet must not change the basic axioms of the theory or create novel axioms that modify its essential tenets (see Kuhn, 1969; Laudan, 1977, for a discussion). In his seminal book, Mental Models, Johnson-Laird (1983) outlined an extremely ambitious idea in cognitive science. After such a "revolution," the next step was the process of "normal science" in which scientists accumulate empirical results (Kuhn, 1969), which by their very nature cause modifications to the theory. The Mental Model Theory of Reasoning: Refinements and Extensions describes at least four refinements to the original theory.

1. The role of working memory and presentation format. MMT suggests that working memory capacity limits the number of models created and searched and the amount of information that can be conveyed within these models. Barrouillet and Grosset (Chapter 1) examine the influence of working memory on model creation in a developmental investigation of conditional reasoning. The authors conducted two experiments that demonstrate that fleshing out models (i.e., creating veridical models) requires sufficient space to integrate statements with meanings stored in long-term memory. Schaeken, Van der Henst, and Schroyens (Chapter 7) demonstrated that participants in their study attended more to relevant than irrelevant information and tended to avoid redundant information. These results suggest that meaning provides critical links to existing information and between concepts (e.g., causal or temporal relations). Further, these links may be aided by strategies in which efficient representation of information (e.g., chunking) reduces working memory burden. Dierckx and Vandierendonck (Chapter 6) investigate how the temporal ordering of stimuli influences reasoning. Their results indicate that inconsistent orders (i.e., in which the presentation sequence was different than the temporal sequence) were more difficult to encode and process than consistent orders because inconsistent orders required the construction of alternative models, a process that is costly for working memory. These results are similar to previous research demonstrating that presentation formats (e.g., diagrams) reduce working memory costs by making relations more transparent in model-based representation than propositional representations (Stenning & Oberlander, 1995). Overall, these results suggest that model creation may be linked to available processing resourses, current knowledge, and reasoning context (more on the latter two below) and do not occur in a content-independent manner. The results also suggest variation in how models are created (i.e., strategies) and extend recent research into the role of model creation and search strategies (see Van der Henst, Yang, & Johnson-Laird, 2002).

2 The role of content. Because models are created via meaning, a second critical issue is how meaning and context influence representation. Because of this focus on meaning (rather than underlying syntactic structure), MMT often better accounts for performance than theories outlining propositional representations (e.g., mental logic theory; Braine & O'Brien, 1998). Byrne (Chapter 3) suggests that the use of whether indicates a more restrictive set of conditions under which relations are evaluated compared to if (suggesting that the antecedent may not be necessary). Although provides slightly different restrictive conditions than whether in that although seems to eliminate many presuppositions from consideration. The results suggest that the contexts in which people hear and use conditionals in language exchanges may be very different than the contexts used in experimental investigations. Roberts (Chapter 5) demonstrates that task construal (defined as how individuals interpret information, e.g., logical terms) influences task solutions. For example, the term some is often misinterpreted, and this interpretation may change the nature of the task itself.

3 The number of models created. Early MMT research demonstrated that increases in problem complexity were associated with increases in the number of models necessary for a solution (Johnson-Laird, Byrne, & Schaeken, 1992). Although this certainly holds true under some conditions, it is possible that model creation itself is "costly" and that the production of a single model may allow reasoners to satisfice rather than reason through a problem exhaustively. As noted previously, working memory limitations restrict model creation and search; thus it is possible that reasoners may use heuristics in the construction and search of models (Gigerenzer, 2000; see also Schaeken, Van der Henst, & Schroyens, Chapter 7). The results from chapters by Handley and Feeney (Chapter 2) and Barrouillet and Grosset (Chapter 1) suggest that reasoners often create a single model during reasoning, which strongly influences the types of inferences drawn. This finding is consistent with earlier research that demonstrated that young children typically create a single model (Sloutsky & Goldvarg, 2004) when solving a variety of logical propositions (Morris & Sloutsky, 2002).

(4.) How reasoners establish truth-falsity. How do reasoners determine whether statements are true or false? Evans, Over, and Handley (Chapter 4) suggest that the Ramsey test (i.e., Probability[q|p]) is a better approximation of human conditional performance than an implicit truth table. The authors suggest that reasoners do not use unweighted possibilities (i.e., binary truth values) but that possible states are always linked to their probability. This probability is likely derived from background knowledge of objects and their relations. Evans et al. offer the most trenchant suggestions for extension and revision in the volume and challenge several foundational axioms of MMT. They suggest that models may encode not only semantic information but also the probability that this information occurs with other information. Such information may provide the foundations for relational events (including causal relations) as well as truth conditions for propositions and may point to a possible convergence with researchers in computational modeling in both the connectionist (McClelland, 2009) and Bayesian traditions (Tenenbaum, Griffiths, & Kemp, 2006).

A good theory must also be extended to explain a wide range of phenomena within its domain. Laudan (1977) suggested that one way of evaluating a scientific theory is to examine its explanatory scope, that is, the number of problems for which it provides an empirically based explanation. The second aim of The Mental Models Theory of Reasoning: Refinements and Extensions is to extend MMT theory into new areas of cognitive science. The book describes three extensions of MMT.

1. The extension of models to include information beyond semantics. Early cognitive psychologists viewed representations as being purely abstract and amodal (e.g., Chomsky, 1965). However, there has been a move away from this position toward a view of representations that has broadened to include grounded information from multiple systems (Barsalou, 2005). Although MMT has never excluded such information, the inclusion of modal information into mental models (e.g., Handley & Feeney, Chapter 2) has the potential to increase the explanatory power of MMT. Evans, Over, and Handley (Chapter 4) suggest that models include information about the probabilities associated with various words (e.g., if) and reasoning outcomes. Such information may help reasoners develop expectations for conditions of truth or falsity and may help to explain common errors in reasoning.

2. New areas of cognition such as causality and probability. Girotto and Gonzalez (Chapter 8) extend MMT to explain naive probabilistic reasoning. The authors suggest that although people are relatively poor at computing formal probabilities (e.g., requiring a formula), they are surprisingly adept at simple probability judgments (e.g., considering simple combinations). In an experiment with adults and children, the authors demonstrate that reasoners set up simple and relatively accurate models of quantities in probability judgments. This finding fits well with recent evidence in both the efficacy of simple heuristic judgments (Todd & Gigerenzer, 2000) and intuitive quantitative judgments (e.g., Dehaene, 1997). Johnson-Laird and Goldvarg-Steingold (Chapter 9) propose that semantic models of cause and effect can be achieved by ordering events in time (i.e., Cause A is sufficient for Effect B, Effect B does not occur before Cause A). The authors suggest that such models of cause would distinguish between relations and evidence and thus allow an implicit encoding of possibility and necessity.

3. Training and informal argumentation. Does the process of model creation and representation show the same patterns in informal argumentation and after training? Green (Chapter 10) examines the utility of model-based representation applied to informal argumentation and suggests a dual representation of argumentation. One component is the model for each individual argument, an iterative process that is updated throughout the course of the argument itself. The second component is a model of the overall thesis, updated with counterarguments and new information. The process of argumentation is the change in both representations in real time across the course of the argument. The results again suggest that minimal model creation may be common outside of formal reasoning situations.

Can formal reasoning be improved through training? Klauer and Meiser (Chapter 11) suggest that training can improve reasoning performance; however, not all types of training are equally beneficial. One important component of successful training is to help reasoners better comprehend the language of logical reasoning (a point echoed in Chapter 5 by Roberts). Although training on constructing models of the premises also appeared to be helpful to participants in a study by Klauer and Meiser, training on formal syntax did not show similar improvement and learning was relatively task specific. These chapters suggest that strategies underlying model creation and search implement the most efficient operations given processing constraints (e.g., working memory capacity).

Laudan (1977) stated that "the adequacy or effectiveness of individual theories is a function of how many significant empirical problems they solve, and how many important anomalies and conceptual problems they generate" (p. 119). The high-quality work presented in The Mental Models Theory of Reasoning suggests that MMT is an effective theory. The authors and editors should be praised for including a large number of new experiments in the volume in addition to high-quality reviews of the literature. The book is essential for those researching higher-order cognition, particularly those working in deduction, causality, or probabilistic reasoning.

That said, the volume may have been more effective had it addressed three issues. One critical question is how MMT should be adapted to novel findings. For example, are the extensions and refinements presented in this volume compatible with the basic concept of the theory? If not, do the extensions increase explanatory power at the expense of parsimony? Finally, do the refinements and extensions add to the theory or do they require prior premises to be revised or deleted? One indication of the latter point is the chapter by Evans, Over, and Handley (Chapter 4), in which they challenge current MMT notions of how reasoners establish truth-falsity. To this end, the volume might have benefited from a more structured introduction and conclusion in which these issues could have been addressed. Similarly, although I understand the editorial decision to distribute introductory information across chapters, the book would have been more accessible to nonexperts if the first chapter had presented an overview of MMT. As written, the book is an excellent addition for those already well-versed in MMT but less accessible to audiences such as new graduate students.

A second issue is that the volume addresses only a small part of cognition. Although much research in the MMT tradition has focused on formal reasoning, particularly deductive reasoning, model-based representations are being used in many other areas of research on reasoning. For example, quantity representations may be model based. Number (and time) may be represented in models that encode "more or less" rather than as propositional representations of exact amount (see Mix, Levine, & Huttenlocher, 2002, for a discussion). Although early writings on MMT closely tied reasoning and language processing (Johnson-Laird, 1983), recent work focuses more on reasoning than language. Yet MMT holds great explanatory potential for language processing unrelated to formal reasoning. For example, MacWhinney (2008) suggested that model-based representations of language might better explain embodied representations used in language.

Finally, the book provides little information about how models may be instantiated neurologically. Although there has been some research examining brain function and model-based reasoning (see Goel, 2005, for a review), there is little discussion on this topic in this volume. Examining behavior at multiple levels of analysis may provide new evidence related to recurring questions (e.g., rules or models in deduction), would undoubtedly sharpen existing questions, and would likely provoke new issues. For example, recent research has demonstrated that brain regions active during spatial perception are active when comprehending sentences involving spatial imagery (Just, Newman, Keller, McEleney, & Carpenter, 2004). Communication with researchers investigating model-based representations in other cognitive domains would strengthen research in deduction and MMT itself.

These limitations, however, do not diminish the important contributions of this volume. The chapters are well written, filled with trenchant reviews of extant literature and well-designed experiments. Most important, the chapters are filled with interesting ideas regarding the nature of human cognition that will stimulate debate and generate ideas for future work. Therefore, I highly recommend this book.


BARSALOU, L.w. (2005). Abstraction as dynamic interpretation in perceptual symbol systems. In L. Gershkoff-Stowe & D. Rakison (Eds.), Building object categories (pp. 389-431). Carnegie Symposium Series. Mahwah, NJ: Lawrence Erlbaum.

BRAINE, M. D. S., & O'BRIEN, D. P. (1998). Mental logic. Mahwah, NJ: Lawrence Erlbaum.

CHOMSKY, N. (1965). Aspects of the theory of syntax. Cambridge, MA: MIT Press.

DEHAENE, S. (1997). The number sense. New York: Oxford University Press.

GENTNER, D. (2002). Mental models, psychology of. In N. J. Smelser & P. B. Bates (Eds.), International encyclopedia of the social and behavioral sciences (pp. 9683-9687). Amsterdam: Elsevier Science.

GIGERENZER, G. (2000). Adaptive thinking: Rationality in the real world. New York: Oxford University Press.

GOEL, V. (2005). Cognitive neuroscience of deductive reasoning. In K. Holyoak & R. Morrison (Eds.), The Cambridge handbook of thinking & reasoning (pp. 475-492). Cambridge, MA: Cambridge University Press.

JOHNSON-LAIRD, P. N. (1983). Mental models. Cambridge, MA: Harvard University Press.

JOHNSON-LAIRD, P. N. (2005). Mental models in thought. In K. Holyoak and R. J. Sternberg (Eds.). The Cambridge handbook of thinking and reasoning (pp. 179-212). Cambridge, MA: Cambridge University Press.

JOHNSON-LAIRD, P. N., BYRNE, R. M. J., & SCHAEKEN, W. S. (1992). Propositional reasoning by model. Psychological Review, 99, 418-439.

JUST, M. A., NEWMAN, S. D., KELLER, T. A., MCELENEY, A., & CARPENTER, P. A.(2004). Imagery in sentence comprehension: An fMRI study. NeuroImage, 21, 112-124.

KUHN, T. s. (1969). The structure of scientific revolutions. Chicago: University of Chicago Press.

LAUDAN, L. (1977). Progress and its problems: Towards a theory of scientific growth. Berkeley, CA: University of California Press.

MACWHINNEY, B. (2008). How mental models encode embodied linguistic perspectives. In R. Klatzky, B. MacWhinney, & M. Behrmann (Eds.), Embodied cognition (pp. 365-405). Mahwah, NJ: Lawrence Erlbaum.

MCCLELLAND, J. L. (2009). The place of modeling in cognitive science. Topics in Cognitive Science, 1, 11-38.

MIX, K., LEVINE, s. C, & HUTTENLOCHER, J. (2002). Quantitative development in infancy and early childhood. Oxford: Oxford University Press.

MORRIS, B. J., & SLOUTSKY, V. (2002). Children's solutions of logical vs. empirical problems: What's missing and what develops? Cognitive Development, 16(4), 907-928.

SLOUTSKY, V. M., & GOLDVARG, Y. (2004). Mental representation of logical connectives. Quarterly Journal of Experimental Psychology: Human Experimental Psychology, 574(4), 636-665.

STENNING, K., & OBERLANDER, J. (1995). A cognitive theory of graphical and linguistic reasoning: Logic and implementation. Cognitive Science, 19(1), 97-140.

TENENBAUM, J. B., GRIFFITHS, T. L., & KEMP, C. (2006). Theory-based Bayesian models of inductive learning and reasoning. Trends in Cognitive Sciences, 10(7), 309-318.

TODD, P. M., & GIGERENZER, G. (2000). Precis of simple heuristics that make us smart. Behavioral and Brain Sciences, 23, 727-741.

VAN DER HENST, J.-B., YANG, Y., & JOHNSON-LAIRD, P. N. (2002). Strategies in sentential reasoning. Cognitive Science, 26, 425-468.

Bradley J. Morris, Grand Valley State University
Gale Copyright: Copyright 2010 Gale, Cengage Learning. All rights reserved.