A new kid on the block: Distributional semantics predicts the word-specific tone signatures of monosyllabic words in conversational Taiwan Mandarin

2025-11-21

AI Summary

This study investigates the systematic influence of lexical semantics on tone pitch contours in monosyllabic Mandarin words spoken in Taiwan, addressing a gap in tonal phonology research that has traditionally focused on phonetic and syntactic factors. Method: Using natural conversational speech data, we employ generalized additive models (GAMs) to model fine-grained pitch trajectories and integrate distributional semantic representations—including contextualized word embeddings (e.g., BERT)—to quantify lexical meaning. Crucially, phonological confounds (e.g., syllable structure, speaking rate, prosodic position) are rigorously controlled. Contribution/Results: We demonstrate that lexical semantics significantly predicts tone realization: homophonous words exhibit statistically distinguishable pitch contours, and contextualized semantic embeddings accurately predict speaker-specific tonal variation. This constitutes the first empirical evidence for a “semantics–tone interface,” revealing that lexical meaning exerts a robust, gradient influence on phonological implementation—thereby extending tonal theory beyond articulatory and syntactic constraints to incorporate semantic grounding.

Technology Category

• Natural Language Processing: Lexical Semantics and Morphology
• Knowledge Representation and Reasoning: Nonmonotonic Reasoning
• Machine Learning: Semi-Supervised Learning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Sentiment analysis and opinion mining

Abstract

We present a corpus-based investigation of how the pitch contours of monosyllabic words are realized in spontaneous conversational Mandarin, focusing on the effects of words' meanings. We used the generalized additive model to decompose a given observed pitch contour into a set of component pitch contours that are tied to different control variables and semantic predictors. Even when variables such as word duration, gender, speaker identity, tonal context, vowel height, and utterance position are controlled for, the effect of word remains a strong predictor of tonal realization. We present evidence that this effect of word is a semantic effect: word sense is shown to be a better predictor than word, and heterographic homophones are shown to have different pitch contours. The strongest evidence for the importance of semantics is that the pitch contours of individual word tokens can be predicted from their contextualized embeddings with an accuracy that substantially exceeds a permutation baseline. For phonetics, distributional semantics is a new kid on the block. Although our findings challenge standard theories of Mandarin tone, they fit well within the theoretical framework of the Discriminative Lexicon Model.

Problem

• Predicting tonal variations in monosyllabic words using semantic embeddings
• Investigating semantic influence on pitch contours beyond phonetic factors
• Demonstrating word meaning predicts tone realization better than lexical identity

Innovation

• Distributional semantics predicts word-specific tone signatures
• Generalized additive model decomposes pitch contours into components
• Contextualized embeddings accurately forecast individual word pitch contours