REDUNDANT-FEATURE CODING IN THE MENTAL LEXICON

It is commonplace in linguistics to make a distinction between redundant and nonredundant information. In current phonological theory, redundant features are taken to be the prime candidates for underspecification. One such feature is voice. It is of particular interest because it acts redundantly on sonorants but contrastively on obstruents (in English and German). This paper chooses to examine which role voice as a redundant and a distinctive feature takes during the process of language generation. To address this issue, the consonantal interaction patterns in two large collections of speech errors in English and German are analyzed. It can be established (a) that obstruents preferentially interact with obstruents of like voicing, (b) that sonorants interact more with voiced than with voiceless obstruents, and (c) that obstruents appear to be more sensitive to voicing than do sonorants. These results are interpreted as arguing against a categorical segregation of redundancy and contrastiveness, against the underspecification of voice on sonorants throughout their derivation, and against the monovalency of the voicing feature. Rather, the empirical data are compatible both with the full specification of voice on sonorants right from the beginning and with their underspecification at earlier points and full specification at later points in the derivation. Which of these alternatives is correct depends in part upon the question of whether more than one phonological-processing stage operating upon phoneme-sized units is actively employed during language production. © 1991 Walter de Gruyter