Usage-based effects in Latin American Spanish syllable-final /s/ lenition.

摘要

Previous studies have identified factors that contribute to the weakening and deletion of syllable-final /s/ in Latin American Spanish, including the dialect and sex of the speaker, the phonetic environment and grammatical status of the /s/, and functional considerations. Proponents of a usage-based model of language claim that the structure of language is shaped by how it is used, and therefore factors such as word frequency, word predictability, and the context that words appear in have an effect on the form of language. This dissertation investigates the extent to which these usage-based factors contribute to syllable-final /s/ lenition in addition to the previously identified factors. Automated speech recognition methods were used to code three dependent variables for a corpus of over 50,000 tokens of syllable-final /s/: deletion or retention of /s/, duration of retained /s/, and the spectral center of gravity of retained /s/. Multiple regression was performed for each of the dependent variables, on all of the data combined and on several subsets of the data. For each multiple regression, usage-based factors were added to a base model to determine which of them improve the model. Word frequency and word predictability based on the following word both have an effect in the expected direction, with more frequent and more predictable words having higher levels of lenition. Word predictability based on the preceding word has the opposite effect, with more predictable words having lower levels of lenition. The phonetic context that words appear in most frequently also has an effect, with words that are more often followed by a consonant having more advanced lenition, even after taking into consideration the actual phonetic context. These usage-based factors contribute both to the categorical process of deletion and the gradient processes of shortening and weakening of articulation. This data supports the claim that these usage-based variables form part of the speaker's knowledge and that speakers have knowledge of low-level phonetic detail. This study suggests that the extent of lenition may be determined both within the lexical entry as in an exemplar model and by processing during production.