Incomplete neutralization and pragmatics in German

Abstract

Listen

Translate article

Earlier studies suggest that the neutralization of the voicing contrast in word pairs like Bund and Bunt is incomplete, but further research is required to evaluate the competing interpretations. Subjects here were given a range of different speaking tasks: reading continuous text, repeating spoken sentences (in circumstances where the concern of the experiment was carefully hidden), dictating the potential homophones to a German-speaking writer and, finally, reading from a word list. Discriminant analysis was used to combine the five spectro-temporal variables measured from sound spectrograms of these productions to categorize the tokens as voiced or voiceless in each condition. Correct categorization for discriminant analysis varied between 55% and 78% depending on the communicative task but was significant in all conditions. The data show that speakers can control the degree of neutralization depending on pragmatics and that information about the underlying contrast is distributed over much of the word. In Experiment 2, recorded productions from some conditions of Experiment 1 were played for a group of listeners. Through use of signal detection theory (and the statistic d ′), it is shown that listeners discriminated the intended word with accuracy very similar to that of discriminant analysis. They even tended to make errors on the same tokens. Apparently, the variables we measured capture information that is roughly equivalent to that employed by native listeners. The absence of complete neutralization implies that the German syllable-final devoicing rule cannot be stated in terms of the same [− voice] feature that is employed in the lexical specification of words. Instead, syllable-final devoicing is an effect that resembles implementation rules (since it is graded) and operates directly upon a syllable-like representation.