Word meaning co-determines vowel-inherent spectral change. A corpus-based investigation of conversational Mandarin.

This study investigates whether Vowel Inherent Spectral Change (VISC)—the dynamic movement of formant frequencies over the course of a vowel—is influenced by the meaning of the word in which the vowel appears. While traditional phonetics often treats vowels as having stable targets, this research explores whether the fine-grained articulatory details of vowels in spontaneous Mandarin are co-determined by the semantic context of the word.

The researchers analyzed a dataset of 5,828 tokens representing 87 word types from the Taiwan Mandarin Spontaneous Corpus. They employed Generalized Additive Mixed Models (GAMMs) to decompose formant trajectories (F1 and F2) into additive components. By controlling for a wide array of factors—including vowel duration, speaker identity, gender, utterance position, co-articulation with neighboring vowels, and fundamental frequency (F0)—the authors isolated the word-specific contributions to spectral change. They then tested whether these word-specific components could be predicted using contextualized word embeddings derived from distributional semantics.

The study confirms that VISC is a robust phenomenon in spontaneous conversational Mandarin, extending findings previously limited to laboratory-controlled speech. Crucially, the researchers found that formant trajectories contain word-specific components that cannot be explained by prosodic or articulatory variables alone. These components are significantly correlated with the word's meaning in context, as evidenced by the ability of contextualized embeddings to predict these trajectories with accuracy exceeding a permutation baseline. This suggests that speech production is not a modular process where semantics and articulation are strictly separated; rather, the semantic content of a word influences the physical realization of its vowels.