Integrating Memory and Reasoning in Analogy-Making: The AMBR Model

Boicho N. Kokinov *# Alexander A. Petrov *
* Department of Cognitive Science
New Bulgarian University
# Institute of Mathematics and Informatics
Bulgarian Academy of Science

Reference to this publication:
Kokinov, B. & Petrov, A. (2001). Integrating memory and reasoning in analogy-making: The AMBR model. In D.Gentner, K.Holyoak, & B.Kokinov (Eds.), The analogical mind: Perspectives from cognitive science (pp. 59-124). Cambridge, MA: MIT Press.
A camera-ready version of this document is available in pdf format (436K). Download Free Acrobat Reader


The authors take an integrative approach that tries to bring analogy and memory together. Their chapter addresses phenomena emphasized by constructivist approaches to memory, such as memory distortions and memory illusions, and show how these phenomena interact with analogy-making. They provide evidence for omissions, blending of episodes, intrusions from generic knowledge, and effects of context, priming, and order in analogical reminding. The authors explain these phenomena in terms of interactions among memory, mapping and perception. The chapter presents the latest development of their AMBR model, which simulates these phenomena by the parallel work and interplay of many subprocesses. This model uses dynamic emergent representations and computations performed by a society of hybrid micro-agents. AMBR is built on a general cognitive architecture, which makes it possible to integrate analogy with other cognitive processes and to provide a basis for unified explanations of phenomena such as context-sensitivity that cut across virtually all cognitive processes.

  1. Reuniting Memory and Reasoning Research: An Appeal for a Second Marriage after Their Divorce
  2. Reconstructing the Dinosaur: Dynamic and Constructive Nature of Human Memory
  3. Analogy-Making in a DUAListic Society: The AMBR View of Analogy
  4. Integration of Memory and Reasoning in AMBR2
  5. Conclusions
  6. Acknowledgements and References

1. Reuniting Memory and Reasoning Research: An Appeal for a Second Marriage after Their Divorce

Three blind men were exploring an elephant. The first of them, who happened to reach the leg, described the elephant as something like a tree trunk--high and of cylindrical shape. The second one grasped the ear and described the elephant as something like a blanket--flexible, thin, and covering a large surface. The third grasped the trunk and therefore formed an image of a long and flexible pipe-shaped object like a hose. For a long time they were arguing about the right conception of the elephant.

blind men and elephant
Figure 1. Cognitive scientists study human cognition in small fractions and often do not recognize its underlying unity.

We cognitive scientists are often in the role of those blind researchers trying to understand human cognition. Since it is a huge and complex object of study each of us approaches it from a certain perspective and studies only a tiny bit of it. Although we do not misrepresent the whole of cognition with the particular object of study, say memory or analogy, we tend to think of mechanisms that could explain the tiny fraction we have focused on. To continue the elephant story, when "trunk specialists" observe the fluid that comes out when the trunk is cut, they tend to hypothesize it is an olfactory secretion. "Leg specialists" also observe a fluid coming out when the leg is cut but have a very different hypothesis about it--it must be some filling of the leg. The fact that this fluid is one and the same in all cases (blood) and has the same function can be discovered only when these scientists come together and consider the elephant as a whole. They need to explore the interactions between various parts (e.g. that an infection in the leg might cause complications in the trunk) and to postulate general principles and systems (like the cardio-vascular system).

There is nothing wrong with separating cognition into pieces and studying them. The practice of "carving nature at its joints" dates at least as far back as the dialogues of Plato. "Scientists try to study systems that are sufficiently closed to be predictable and sufficiently small to be understandable" (Hunt, 1999, p. 8). Big and complex systems are hardly manageable. Studies of isolated parts have lead to very important achievements in understanding the mechanisms of human cognition and analogy-making in particular.

However, studies of components should be done with awareness of the fact that the separation of human cognition into various processes is just a convenient tool and not a reality. They should be complemented with explorations of the interactions among various cognitive processes, that is, instead of being carved, the "joints of nature" have to be studied.

Early philosophers like Aristotle considered thinking and memory in an integrated way. The doctrine of associationism explained human thinking by means of the content and organization of human memory. Later on as science developed and psychology became an experimental science, researchers tended to analyze simple and separate faculties of the human mind in order to be able to study them experimentally. Nowadays we have a huge pile of facts about both memory and reasoning (and analogical reasoning in particular). The problem is that these two research communities do not speak to each other often. As a result, facts established in one of the fields are often neglected and ignored in the other.

We feel the time has come to try to put the pieces back together. This chapter makes an attempt to re-integrate research on analogy-making with research on memory. Holyoak and Hummel (this volume) present another attempt in a similar direction--they integrate analogy with memory and learning. Forbus (this volume) also appeals for integrating analogy models with models of large-scale cognitive processes. He presents an integrated model of commonsense thinking based on analogical reasoning and reasoning from first principles. Hofstadter (this volume) argues that analogy-making might be the core of many cognitive processes from perception to categorization to translation of poetry. Fauconnier (this volume) integrates analogy with conceptual blending. Thagard (this volume) integrates analogy with emotions. Markman (this volume) integrates analogy-making with decision-making. These are all small but important steps in the direction of re-integrating our knowledge about human cognition. It seems that cognitive science has matured enough to pursue these steps.

Modeling has too many degrees of freedom. A phenomenon can often be modeled in several different ways and it is difficult to evaluate the model based on this single phenomenon alone. That is why it is important to restrict the space of possible models by bringing to bear as many constraints as possible. Several types of constraints can be exploited:

In addition, we can differentiate between specific and general constraints. Typically when modeling a specific phenomenon we tend to concentrate on the constraints known to apply to that specific phenomenon. Thus when studying analogy we tend to collect data with respect to analogy. The utility of these data is clear and we try to draw from as many sources as we can: psychological, neurological, evolutionary, and developmental. Very often, however, we ignore data that are not directly related to analogy but are nevertheless very useful because of their relation to other cognitive processes that in turn relate to analogy. If we consider analogy as an integrated phenomenon in the complex structure of human mind, we need to pay attention to these general constraints as well.

This is, of course, an over-ambitious task that is clearly beyond the scope of this chapter. However, it is an important motivation of the current work. This chapter describes only a few steps on the way toward integrating analogy back again into human cognition. Special emphasis is put on some general behavioral and architectural constraints and particularly on the integration of analogy-making and memory.

Section 2 presents a highly selective and biased review of the literature on memory. It concludes with a summary of the behavioral and architectural constraints on analogy models as seen by the authors. Section 3 reviews the AMBR research program. Finally, Section 4 describes AMBR2--the current version of the AMBR model --which tries to bring memory and analogy back together.

Next page > > > Reconstructing the Dinosaur

[ Kokinov's Home Page ][ Kokinov's Publications ]
[ Petrov's Home Page    ][ Petrov's Publications    ]
[ DUAL Main Page ] [ Get pdf file ]
[ AMBR Main Page ] [ Back to top ] Check the validity of this page's XHTML Page maintained by Alex Petrov
Created 1999-11-05, last updated 2004-06-27.