| Popis: |
Ideally one would like to select symbols for use in visual displays which are both meaningful representations of the objects they stand for and maximally dissimilar to one another. In this paper a new, simple-to-compute metric for rating the similarity of symbols is proposed. The metric is based on the degree of overlap between “fuzzy pictures” of the symbols, and thus (unlike other current metrics) does not require a prior determination of relevant features. The similarity ratings produced by the metric are shown to be highly correlated with subjective ratings of similarity (rs = .82, p < 0.01). Thus, the similarity ratings derived from the metric can be used to select the best (most discriminable) subset of symbols from a set of meaningful symbols, bypassing the need for a time consuming and expensive empirical determination of similarities. |
