Perception of lowercase letters in peripheral vision: a discrimination matrix based on saccade latencies.

Abstract

The main method used to study the role of features for letter recognition is that of generating letter confusion matrices (Bouma, 1971; Geyer & DeWald, 1973; Gilmore, Hersh, Caramazza, & Griffin, 1979; Roethlein, 1912; Sanford, 1888; Townsend, 1971; van der Heijden, Malhas, & van den Roovart, 1984). The data of a confusion matrix are an indirect but useful measure of interletter similarity. Yet, the confusion matrix method has various problems. First, the identification task may evoke response bias phenomena due to different guessing strategies (Bouma, 1971; Sanford, 1888). Second, the data matrix is often incomplete: Some letters are not confounded at all, and the empty cells in the matrix weaken its utility. Third, it is also known that different methods used to degrade viewing conditions and to produce a sufficiently high number of confusion errors lead to different patterns of confusions (Bouma, 1971; Garner & Haun, 1978; Geyer, 1977). For researchers interested in the dynamics of letter perception, a final inconvenience is, of course, that the data of a confusion matrix tell us nothing about the timecourse of feature extraction during letter and word recognition.