Perceptual structure of 22 prevocalic English consonants.

The similarities of 22 English consonants were judged by three different methods: (1) equal‐appearing interval scale (SF), (2) magnitude estimation (ME), and (3) triadic judgments (ABX). These judgments were analyzed mainly by multidimensional (MD‐SCAL and IND‐SCAL), and multiple regression analysis techniques. In the city‐block MD‐SCAL space the SF and ME data were “best” analyzed in three dimensions and ABX data in four dimensions. In the Euclidean space the SF and ME data were analyzed in five dimensions and ABX in four. The best MD‐SCAL dimensional spaces were used to test how well the spaces were predicted by the four different feature systems, namely, (1) Miller‐Nicely (MN), (2) Singh‐Black (SB), (3) Wickelgren (W), and (4) Chomsky‐Halle (CH). The CH feature system in Euclidean space was the “best” on the criteria of consistency (Freedman's m‐ranking statistics) and prediction (tau correlation). The three data sets were also analyzed in a five‐dimensional IND‐SCAL space with the following feature interpretations: dimension 1—sibilant, dimension 2—place (front/back), dimension 3—voicing, dimension 4—plosive, and dimension 5—nasality. The feature system composed of the above features (IND‐SCAL system) was considered “best.” The features in CH system in order of perceptual importance were: (1) high; (2) voicing; (3) anterior; (4) vocalic; (5.5) nasality and strident; (7) continuant; (8) coronal; (9) low; (10) consonantal; and (11) back. The features in IND‐SCAL system in order of perceptual importance were: (1) place (front/back); (2) nasality; (3) sibilant; (4) voicing; (5) plosive. Implication of the findings of this study for phonetic theory is discussed.