Evan's 1968 ANALOGY system was the first computer model of analogy. This paper demonstrates that structure-mapping, when combined with high-level visual processing and qualitative representations, can do the same kinds of problems with hand-drawn sketched inputs.

Abstract. Anti-unification has often be used as a tool for analogy making. But while first-order anti-unification is too simple for many applications, general higher-order anti-unification is too complex and leads into theoretical difficulties. In this paper we present a restricted framework for higher-order substitutions and show that anti-unification is well-defined in this setting. A complexity measure for generalizations can be introduced in a quite natural way, which allows for selecting preferred generalizations. An algorithm for computing such generalizations is presented and the utility of complexity for anti-unifying sets of terms is discussed by an extended example. 1

We use SME, a domain-general model of analogy, to solve a set of problems from the Raven’s Progressive Matrices. SME is used in a two-stage mapping process which we have previously shown to be effective for solving geometric analogy problems. Each problem is drawn in PowerPoint and input into sKEA, our sketch understanding system. sKEA automatically computes qualitative representations of the drawings, using a spatial representation scheme motivated by research on human perception. We show that the representations generated by sKEA can be used with SME to solve the Raven’s Progressive Matrices problems, without using any processes specifically designed for the task.

We describe the use of analogies in 75 one-on-one tutoring sessions with first-year medical students carried on by two professors of physiology at Rush Medical College. Analogies were not very frequent, but were extremely effective when they were used. We have analyzed the goals, the topics, and the discourse strategies for proposing analogies. We have also studied the ways that the tutors follow up on these analogies and clarify them when necessary, with the goal of implementing analogies in our intelligent tutoring system, CIRCSIM-Tutor. Our knowledge representation scheme is based on Gentner’s theory and Forbus ’ MAC/FAC model, which allows for structural mappings between analogies.

We give a definition of reasoning by analogy, which is tailored to a setting of decision making under uncertainty. We present a model of decision making which is based on such a definition, and show that it is compatible with a large class of preferences.

The representation-building and the analogical mapping processes run in parallel in the AMBR model and thus they can influence each other. This paper describes how the AMBR model could explain the rerepresentation of the target when the analogical mapping forces it. The mechanisms are presented and a simulation is described which shows that depending on how strong the mapping is sometimes we arrive at one representation (the dominant one) of the target, sometimes another one (the unobvious one). This simulation replicates the psychological data that were obtained earlier. The simulation allows to track the dynamics of the process and to make another prediction that there will be inhibition of the alternative interpretation of the target. Thus we arrive at a very strange prediction that people may see consciously only the dominant interpretation, but unconsciously they may have built partially the alternative interpretation which will remain inhibited by the competing dominant one, i.e. the prediction is that there might be cases of inhibition of something the subjects would claim they have not seen. This prediction is supported in a pilot psychological experiment.

I show that virtually any model of decision making under uncertainty is associated to a certain structure. This contains three fundamental ingredients: (1) The domain of the acts; (2) Another set, which is called the set of models for the decision maker; and (3) The decision maker’s information about the set of models (an algebra of subsets of the set of models). A consequence of this finding is that that the decision maker’s choices can be viewed as the outcome of a two-stage process. First, the set of acts is mapped into a system of hypothetical bets on the set of models. Then, the latter are ranked by the decision maker. I show that this procedure can be thought of as describing a general form of analogical reasoning. I also observe that the appearance of two different sets implies that the decision maker is uncertain about two different objects and that he may receive information about any of them. In particular, information about the set of models affects the decision maker’s ranking of the

First, i would like to thank my supervisors Angela Schwering and Kai-Uwe Kühnberger for their valuable advice and support. Furthermore, i credit Helmar Gust and Ulf Krumnack for helpful discussions about the formal topics involved. Additionally, my gratitude goes to Egon Stemle and Jackie Griego for their help to typeset and proofread

Spatial problem-solving tasks are often used to evaluate people’s cognitive abilities. For example, Raven’s Progressive Matrices is a popular intelligence test. In it, an individual is shown an array of twodimensional images, with one image missing. The individual must compare the images and identify a pattern of differences between them, in order to solve for the missing image. Performance on tasks such as Raven’s and geometric analogy (“A is to B as C is to..?”) correlates strongly with performance on many other ability tasks, in the spatial, verbal, and mathematical domains. Thus, these tasks appear to depend on core, general-purpose representations and processes. However, it is as yet unclear what those representations and processes are. To better understand these tasks, we developed Spatial Routines for Sketches (SRS), a general framework for modeling spatial problem-solving. SRS is based on a set of psychological claims about how people perform spatial problem-solving: 1) When possible, people use qualitative representations describing features such as relative position or orientation, rather than exact numerical values. 2) Spatial representations are hierarchical. A given image might be represented as object groups, individual objects, or the parts within each object. 3) Qualitative spatial representations can be compared via structuremapping. Structure-mapping involves aligning the relational structure in two representations to find the