Modeling Iu Mien tone with eigenpitch representations

To achieve adequate description of an understudied tone language, we argue for unsupervised computational modeling of lexical tone, even at the earliest stages of documentation. We apply a technique using eigenvalues and discriminant analysis to differentiate lexical tones on monosyllabic items in Iu Mien. The resulting ‘eigenpitch’ representations can be used to evaluate the differences and similarities between tones, to sharpen the impressions of field linguists, and to contribute to a richer understanding of phonemic tonal contrasts.