Abstract

Structural Constraints and Real-World Plausibility in Analogical Inference Linsey A. Smith ([email protected]) Dedre Gentner ([email protected]) Department of Psychology, Northwestern University 2029 Sheridan Road, Evanston, IL 60208 USA Abstract deep matching systems are preferred over shallow matches (the systematicity principle), which reflects a preference for coherence and inductive power in analogical processing (Clement & Gentner, 1991; Falkenhainer, Forbus & Gentner, 1989). Candidate inferences are generated by completing the pattern in the (initially) less-structured member of the pair, based on the common structure. Models of analogy have largely converged on a set of assumptions like those outlined above (Falkenhainer, Forbus & Gentner, 1989; Gentner, Holyoak & Kokinov, 2001; Holyoak and Thagard, 1989; Hummel & Holyoak, 1997; Kokinov & French, 2003; Larkey & Love, 2003). Further, there is substantial empirical evidence in support of the idea that analogical reasoning obeys these constraints. A variety of studies have provided evidence that analogical matching is constrained by both structural consistency (including one-to-one mapping) (e.g., Krawczyk, Holyoak, & Hummel, 2005; Markman, 1997; Markman & Gentner, 1993; Spellman & Holyoak, 1992) and systematicity (e.g., Clement & Gentner, 1991). For example, Clement and Gentner (1991) showed participants analogous scenarios and asked them to judge which of two lower-order assertions shared by the base and target was most important to the match. Participants chose the assertion that was connected to matching causal antecedents – their choice was based not only on the goodness of the local match, but also on whether it was connected to the larger matching system. Thus, matching lower-order relations that are interconnected by higher-order relations were considered more important to the analogy. In sum, people demonstrate considerable structural sensitivity in analogical mapping. Theoretical accounts of analogy have largely agreed that structural constraints play a substantial role in the mapping process. Less is known, however, about the robustness of these constraints in the inference process and the way in which particular content influences the use of structural constraints in analogical inference. We conducted two studies testing whether the plausibility (or implausibility) of an inference influences adherence to general structural principles in analogical reasoning. We found substantial reliance on the predicted structural constraints, but also an influence of the plausibility of the inference. Introduction Our goal in this research is to explore the stability of analogical inference under different conditions: specifically, whether analogical inference is a domain-general reasoning process, governed by structural constraints inherent to the analogical process, or whether it is a loosely constrained process whose outcome is strongly influenced by the plausibility of the potential inferences in particular domains. This question is important not only for what it can tell us about basic analogy processes, but also because the use of analogy in scientific discovery (and even in science learning) sometimes requires making initially implausible inferences. We first review research on this issue in the arena of analogical mapping and alignment, which has been extensively studied, and then turn to analogical inference. Structural Constraints on Analogical Mapping Reasoning by analogy involves identifying a common system of relations between two domains and generating further inferences driven by these commonalities (Gentner, 1983; Holyoak & Thagard, 1989; Hummel & Holyoak, 1997; Kokinov & French, 2003). According to structure- mapping theory, the comparison process involves aligning a pair in such as way as to achieve a consistent structural alignment between two domains (Falkenhainer, Forbus & Gentner, 1989; Gentner, 1983; Gentner & Markman, 1997). The structural alignment process is guided by a set of tacit constraints that lead to structural consistency and inferential clarity: parallel connectivity, which requires that arguments of matching predicates must also be placed into correspondence; and one-to-one correspondence, which requires that each element of a representation match, at most, one element of the other representation. Importantly, Analogical Inference There is some research on the degree to which structural constraints hold in analogical inference. In the Clement and Gentner (1991) research just described, a second study found evidence for systematicity in inference projection. People generated inferences that were part of a shared system, rather than equally applicable inferences that were not. Markman (1997) also found evidence for systematicity in inference generation. In addition, he found that people based their inferences on one-to-one mappings. When given analogies with two possible sets of correspondences, people noticed both possibilities, but drew inferences from only one of them. These findings suggest a role for structural consistency in inference, as in alignment.