Modelling German Word Stress

Abstract

Standard linguistic and psycholinguistic approaches to stress assignment argue that the position of word stress is determined on the basis of abstract information such as syllable weight and number of syllables in the word. In the present study, we contrasted this approach with a perspective based on learning analogies according to which speakers learn to associate basic word form cues to stress position. To do so, we use a simple two-layer neural network trained with an error-driven learning mechanism to predict stress position in German morphologically simple and complex words. We find that networks trained on word forms outperformed networks trained on cues that represent abstract information. Moreover, most standard approaches assign stress from right to left. We tested this proposal and found that in morphologically simple words, assignment from right yielded better results than assignment from left, supporting the standard approach. By contrast, in morphologically complex words assignment from left outperformed assignment from right. We discuss the implications of our results for psycholinguistic theories of stress assignment by taking into account word form cues, abstract cues, assigning direction, and the representation of stress in the mental lexicon.