Continuous versus discrete information processing in pattern recognition.

A discrete feature model (DFM) and the fuzzy logical model (FLMP) were formulated to predict the distribution of rating judgments in a pattern recognition task. The distinction was between the spoken vowels /i/ and /I/, as in beet and bit. Subjects were instructed to rate the vowel on a nine-point scale from /i/ to /I/. Two features, the first formant frequency (F1) and the vowel duration, were orthogonally varied: The vowel /i/ has a lower (F1) and a longer duration compared to a somewhat higher (F1) and shorter duration for /I/. The DFM predicts that the separate features are recognized discretely, whereas the FLMP assumes that continuous information is available about each feature. Tests of these models on the observed data indicated that the continuous information assumption of the FLMP gave a significantly better description of the distribution of rating judgments.