关键词: 关联相似性模型, 归纳强度, 关联强度, 关联相似性, 交互作用效应

Abstract: There are two kinds of conventional accounts for category-based feature inductions: similarity accounts and knowledge accounts. The former claim that feature inductions are based on overall similarities between premise and conclusion categories, and that inductive strength increases with overall similarity. The latter accounts emphasize the role of the knowledge of feature relevance in feature inductions, regardless of the role of similarities. The two accounts are separate from each other. Wang (2006) proposed a relevance similarity model of feature induction to describe an inductive reasoning based on similarity in terms of features relevant to induction features (namely relevance similarity). The model integrates similarity with the knowledge of feature relevance, and claims that size orders of inductive strength can be predicted by the product of relevance similarity and the connection strength between induction features and the relevance features of induction features.
According to the claim that size orders of inductive strength can be predicted by product of relevance similarity and connection strength between induction features and relevance features of induction features, there would be an interaction effect of connection strength and relevance similarity on inductive strength in feature induction. Three experiments with college students used two within-subject factors (connection strength and relevance similarity) designed to examine the interaction effect. In Experiment 1, each induction feature had one definite relevance feature. Experiment 1 used a 2 (connection strength: weak vs. strong connection) × 2 (relevance similarity: 0/1 vs. 1/1) within-subject design. In Experiment 2, each induction feature had four relevance features. Experiment 2 used a 2 (connection strength: weak vs. strong connection) × 3 (relevance similarity: 1/4, 2/4, and 3/4) within-subject design. In Experiment 3, each induction feature had ten relevance features. Experiment 3 used a 2 (connection strength: weak vs. strong connection) × 5 (relevance similarity: 8/10, 6/10, 5/10, 4/10, and 2/10) within-subject design.
The results of these three experiments showed the presence of the interaction effect of connection strength and relevance similarity on inductive strength in feature induction, based on relevance similarity. The influence of relevance similarity on inductive strength increased with connection strength. The influence of connection strength on inductive strength depended on relevance similarity: in high relevance similarity (above 60%), inductive strength increased with connection strength; in middle relevance similarity (40% to 60%), connection strength did not influence inductive strength; and in low relevance similarity (below 40%), inductive strength decreased as connection strength increased.
The above phenomena are new findings in the studies on feature induction

Key words: relevance similarity model, inductive strength, connection strength, relevance similarity, interaction effect