Perceptual validation of vowel normalization methods for variationist research

Abstract The evaluation of normalization methods sometimes focuses on the maximization vowel-space similarity. This focus can lead to adoption that erase legitimate phonetic variation from our data, is, overnormalization. First, a production corpus is presented highlights three types in formant patterns: uniform scaling, nonuniform and centralization. Then results two perceptual experiments are presented, both suggesting listeners tend ignore according while associating scaling centralization with differences. Overall, suggest remove not vowel data. As result, although these provide more similar spaces, they do so by erasing data may be socially linguistically meaningful, including potential male-female difference low vowels corpus.