Data mining Mandarin tone contour shapes

Abstract

In spontaneous speech, Mandarin tones that belong to the same tone category may exhibit many different contour shapes. We explore the use of data mining and NLP techniques for understanding the variability of tones in a large corpus of Mandarin newscast speech. First, we adapt a graph-based approach to characterize the clusters (fuzzy types) of tone contour shapes observed in each tone n-gram category. Second, we show correlations between these realized contour shape types and a bag of automatically extracted linguistic features. We discuss the implications of the current study within the context of phonological and information theory.

Representation Learning for Discovering Phonemic Tone Contours

“…Levow (2006) performed unsupervised and semi-supervised tone clustering in Mandarin, using average pitch and slope as features, and k-means and asymmetric k-lines for clustering. Graph-based community detection techniques have been applied to group n-grams of contiguous contours into clusters in Mandarin (Zhang, 2019). In recent work concurrent to ours, Fry (2020) uses adversarial autoencoders and hierarchical clustering to identify tone inventories, and evaluate their method on Mandarin, Cantonese, Fungwa, and English data.…”

“…Levow [12] performed unsupervised and semisupervised tone clustering in Mandarin, using average pitch and slope as features, and k-means and asymmetric k-lines for clustering. Graph-based community detection techniques have been applied to group n-grams of contiguous contours into clusters in Mandarin [13]. Our work appears to be the first model to use unsupervised deep neural networks for phonemic tone clustering.…”