The addition bias in Dutch and Spanish phonological speech errors: The role of structural context

Abstract

In speaking, both content (words, segments) and structure (syntactic trees, metrical structures) need to be produced. There is controversy about which aspects of metrical structure are generated during phonological encoding. In particular, do speakers retrieve a unit that represents CV structure? The present study shows that the incorporation of units for syllable CV structures in a connectionist model of phonological encoding enables us to explain empirical patterns of speech errors. The model accounts for the finding of a bias towards additions of segments. First, corpus analyses in Dutch and Spanish showed an addition bias in both languages. A second corpus analysis showed that in apparently noncontextual deletion errors the resulting CV structure is often identical to that of syllables in the immediate context. Furthermore, simulations are reported with two connectionist models, one that incorporates representations for CV structures and one that does not. Only the first model correctly simulated the empirically obtained addition bias, as well as the general pattern of substitution errors. Further simulations showed that the model predicts effects of the structural context. A final simulation tested a second explanatory mechanism for the addition bias: feedback from segments to syllables.