類似性構造に基づく図形分類反応の検討

吉川 左紀子

doi:10.4992/jjpsy.51.267

ON THE RELATIONSHIP BETWEEN THE SIMILARITY STRUCTURE AND THE CLASSIFICATION PERFORMANCE FOR SCHEMATIC FACES

SAKIKO YOSHIKAWA

Author information

JOURNAL FREE ACCESS

1980 Volume 51 Issue 5 Pages 267-274

DOI https://doi.org/10.4992/jjpsy.51.267

[Correction] The Japanese Journal of Psychology Vol.52 No.4 (1981) pp.258-258

Details

Published: January 20, 1981 Received: March 04, 1980 Available on J-STAGE: July 16, 2010 Accepted: - Advance online publication: - Revised: -
Correction information
Date of correction: July 16, 2010 Reason for correction: - Correction: ABSTRACT Details: Wrong : Using the schematic faces composed of three attributes (Eye separation, Eye height and Mouth height) as stimuli, the relationship between the classification RT and the similarity structure yielded by multidimensional scaling (MDS) was examined. In Exp. I, seven observers judged the pairwise dissimilarity of two sets of schematic faces with different degree of category discriminability (the Easy set and the Difficult set, 26 stimuli each), which were produced by manipulating the extent of overlap of attribute distributions between two classes of stimuli for each set. MDS procedure applied to the dissimilarity judgments yielded a three dimensional similarity space for each set. The configuration of stimuli in the similarity space revealed that (1) two classes of stimuli were distinctly separated from each other along dimension 1, regardless of their degree of category discriminability, and that (2) two classes of stimuli of the Easy set were more clearly separated than those of the Difficult set. The coordinates of dimension 1 were highly correlated with all three attribute values of the stimuli. For the Easy set, class A stimuli were more widely dispersed along dimension 2 than class B stimuli (see Fig. 3a). In Exp. II, following the classification learning trials for 18 stimuli, 16 subjects assigned to each set were asked to classify 26 stimuli (18 plus 8 new stimuli) into two classes as fast as possible. Mean classification RTs were analysed in relation to the spatial configuration for each set. The main results were as follows: (1) The classification RTs for the Easy set were significantly shorter than those for the Difficult set, as predicted by the configulation. (2) The high correlation between the classification RTs and the coordinates of dimension 1 regardless of the category discriminability suggests that subjects used this dimension as a judgment criterion to classify the stimuli. The correlation coefficient excecded those between RTs and any attribute value. (3) The fact that whether the stimuli were familiar to the subjects or not was not an important determinant of the RTs (see Fig. 4) suggests that the classification judgment criterion which subjects used can be generalized to the new stimuli as well. (4) For Easy set, the inequality of the variances between two classes in dimension 2 was reflected in the RTs; i.e., the RTs for class A stimuli were significantly longer than those for class B stimuli.
Based on the above findings, it was concluded that the overall similarity relationships among stimuli were the important factors in classifying stimuli such as schematic faces and that in this similarity structure the subjects set up the judgment criterion suited for a given task.
Right : The purpose of this study was to examine the relationship between the similarity structure of the schematic faces yielded by multidimensional scaling and the classification RTs. By manipulating the distribution of three attribute values, two sets of stimuli which differed in category discriminability were produced. Following the classification learning trials, subjects were asked to classify the stimuli into two classes as fast as possible. The classification RTs were highly correlated with the coordinates of dimension one of the similarity structure regardless of the category discriminability, and the correlation coefficient exceeded those between the classification RTs and any attribute value. The evidence suggests that the overall similarity relationships among stimuli are important factors in classifying stimuli, such as schematic faces.

Abstract

The purpose of this study was to examine the relationship between the similarity structure of the schematic faces yielded by multidimensional scaling and the classification RTs. By manipulating the distribution of three attribute values, two sets of stimuli which differed in category discriminability were produced. Following the classification learning trials, subjects were asked to classify the stimuli into two classes as fast as possible. The classification RTs were highly correlated with the coordinates of dimension one of the similarity structure regardless of the category discriminability, and the correlation coefficient exceeded those between the classification RTs and any attribute value. The evidence suggests that the overall similarity relationships among stimuli are important factors in classifying stimuli, such as schematic faces.

Corresponding author

Correction information

Register with J-STAGE for free!