Structual Alignment in Similarity and Differences of Simple Visual Stimuli - eScholarship
Zachary Estes, Uri Hasson · Proceedings of the Annual Meeting of the Cognitive Science Society · 2001
Structural A lignm ent in Sim ilarity and D ifference of Sim ple V isual Stim uli Zachary Estes (zcestes@ princeton.edu) U ri H asson (hasson@ princeton.edu) D epartm ent of Psychology, Princeton U niversity Princeton, N J 08544-1010, USA The present investigation tested the predictions of Structural A lignm ent theory (G entner & M arkm an, 1997) in sim ilarity and difference judgm ents of sim ple visual stim uli. A lignm ent theory explains com parison as a process of aligning the structure of one stim ulus w ith the structure of the other stim ulus. The theory m akes a critical distinction betw een alignable differences, w hich are related to com m onalities in the structures of tw o stim uli, and nonalignable differences, w hich have no structural correspondence in the tw o stim uli. Exam ples are show n in Figure 1. The difference betw een the standard stim ulus S and the N A D stim ulus is nonalignable because S has no elem ent that corresponds to the black triangle in N A D . The difference betw een S and the A D A stim ulus, on the other hand, is alignable because the w hite circle in S is aligned w ith, or corresponds to, the black circle in A D A . A lignable differences m ay also occur in the form of a different object (A D O ) or a different relation betw een objects (A D R). Figure 1: Stim uli used in Experim ents 1 and 2. Standard (S): N onalignable difference (NAD): A lignable difference-attribute (ADA): A lignable difference-relation (ADR): A lignable difference-object (ADO ): These distinctions are critical for predicting sim ilarity and difference judgm ents. A lignm ent theory predicts that “alignable differences count m ore against sim ilarity than nonalignable differences” (ibid, p. 50). That is, item s w ith an alignable difference (i.e., A D A , A D R, and A D O ) should be judged less sim ilar to (and m ore different from ) the standard than should item s w ith a nonalignable difference (i.e., N A D ). A second prediction is that the m ore different the alignable difference is from the standard, the m ore it w ill detract from sim ilarity (see M arkm an & G entner, 1996). Experim ents 1 and 2. Stim uli consisted of all possible pairs of item s show n in Figure 1 (excluding the standard stim ulus), thus creating 6 item pairs. For each item pair, participants judged w hich of the tw o stim uli w as m ore sim ilar to the standard stim ulus (Experim ent 1) or m ore different from the standard stim ulus (Experim ent 2). Table 1: Proportions of sim ilarity and difference choices. Item pair Sim ilarity D ifference (1) A D A & A D O A D A = .65 A D O = .70 (2) A D R & A D O ADR = .55 A D O = .58 (3) A D A & A D R A D A = .74 ADR = .57 (4) N A D & A D A A D A = .74 N A D = .72 (5) N A D & A D R ADR = .69 N A D = .68 (6) N A D & A D O A D O = .53 N A D = .50 D iscussion. Com parisons (1) and (2) in the Table above show that, of the item s w ith alignable differences, A D O w as m ost different from S. Com parison (3) show s that A D A w as the least different from S, w ith A D R falling in betw een. H aving established this hierarchy of alignable differences, w e next exam ined w hether the degree of difference of an alignable difference from S did affect the degree to w hich that alignable difference detracted from sim ilarity (w hen judged w ith a nonalignable difference). A s predicted, com parison (4) show s that the least different alignable difference detracted the least from sim ilarity judgm ents (i.e., A D A = .74), w hile (6) show s that the m ost different alignable difference detracted the m ost from sim ilarity judgm ents (i.e, A D O = .53). These findings extend and replicate those of M arkm an and G entner (1996). H ow ever, as apparent in the Table, in no case did an alignable difference (i.e., A D A , A D R, or A D O ) detract m ore from sim ilarity judgm ents than did a nonalignable difference (i.e., N A D ). O n the contrary, A D A and A D R actually counted less against sim ilarity (and conversely m ore against difference) than did N A D . This result does not support the prediction of alignm ent theory. Potential explanations of this failure to support alignm ent theory are that (i) alignm ent theory is not applicable to sim ple visual stim uli, (ii) the alignable differences used in these experim ents w ere not sufficiently different, or (iii) N A D w as not really a nonalignable difference, but rather w as an alignable difference in the num ber of elem ents in the item . W e w ould be delighted to discuss these and other possibilities w ith you at our poster. R eferences G entner, D . & M arkm an, A .B. (1997). Structure m apping in analogy and sim ilarity. A m erican Psychologist, 52, 45-56. M arkm an, A .B. & G entner, D . (1996). Com m onalities and differences in sim ilarity com parisons. M em ory & Cognition, 24, 235-249.