Multidimensional scaling representations and individual differences in concept learning of artistic style

Abstract

Several previous studies in which a concept learning task has been used to investigate the acquisition of the concept of artistic style, have noted a wide range of individual differences in performance. Tighe (1968) has proposed that these may reflect differential sensitivity to the relevant dimensions of stylistic variation. The INDSCAL technique of multidimensional scaling, which has previously been used to determine the predominant dimensions of variation in responses to visual art was used here to determine the salience of such dimensions for different individuals. These scores were then used to predict individual differences in performance on a concept formation task. The results offer support for a differential cue salience explanation for individual differences in learning about artistic style. Implications for aesthetic education are discussed.Recent research in experimental aesthetics can be characterized both by an increasing methodological sophistication, and by a growing concern with one or two topics of central importance rather than a scattered interest in a variety of disparate areas. One such topic is that of sensitivity to artistic style. The ability to recognize similarities and differences between the works of different painters has been investigated from a number of different viewpoints. Two of these, the developmental approach (e.g. Gardner, 1972) and the multidimensional scaling approach (e.g. O'Hare, 1976) have been outlined separately (O'Hare, 1978). The present article will be chiefly concerned with a third approach involving the concept learning of artistic style. the criteria that define the concept are hard to identify. Since two works of art may vary on an indefinite number of dimensions, an artist's style can only be defined with reference to positive instances. In Walk's experiment, subjects were presented with positive instances of an artistic style, and the ability of the subject to extend this to new instances was measured. Using this simple transfer technique, Walk was able to demonstrate concept learning of the styles of six Impressionist painters.Tighe (1968) modified the basic procedure to permit measures of the subject's concept formation after the presentation of each instance. Slides of samples of five Cubist painters' works were divided into sets so that each set contained one painting by each artist. Each set was then presented three times. On the first occasion the subject was required to guess the name of each artist. The second presentation, or training trial, consisted of the pairing of each painting with the appropriate artist's name. The final trial was the same as the first, except that after the subject had made a choice, the correct artist's name was provided.Once again, clear evidence of concept learning was obtained. The percentage of correct classifications improved from near-chance (17 per cent) on the first test trial to 60 per cent correct on the final test trial. These figures are all the more impressive in view of the fact that the slides of these closel...

Features of style classification: A multivariate experimental analysis of children's responses to drawings.

“…There is direct evidence in the present study that the dimensions underlying the children's judgments are also salient in organizing adults' discriminations of drawings representing the work of well-known artists from the renaissance to the present day. In addition, we found in a previous study of concept learning and artistic style that the MDS dimensions were predictive of individual differences in performance on the concept learning task (O'Hare, 1979). The few studies that have attempted to establish the generality of the MDS dimensions (e.g., Jones & Young, 1972) have supported the notion that the dimensions uncovered are of general significance in mediating other behaviors.…”

An Introduction to the Concepts and Methods of Multidimensional Scaling

“…As well as checking on the adequacy of the MDS solution, the clustering may directly aid in the interpretation of the configuration (see Shepard, 1972Shepard, , 1974. For a recent example of the combined use of MDS and hierarchial clustering techniques, see O'Hare (1979).…”

“…For example, the establishment of some kind of predictive validity for the model should substantially improve our belief in the appropriateness of the particular scaling model as a substantive model for the phenomena under study (see Shepard, 1972). An example of this approach can be found in O'Hare (1979).…”