DocHero AI - Best paraphrasing and translation tool for academic and professional writing | DocHero AI - Best paraphrasing and translation tool for academic and professional writing

Comparing the semantic structures of lexicon of Mandarin and English

Yi Yang, R. Harald Baayen

2025

Language and Cognition Vol. 17

10.1017/langcog.2024.47

摘要

Abstract This paper presents a cross-language study of lexical semantics within the framework of distributional semantics. We used a wide range of predefined semantic categories in Mandarin and English and compared the clusterings of these categories using FastText word embeddings. Three techniques of dimensionality reduction were applied to mapping 300-dimensional FastText vectors into two-dimensional planes: multidimensional scaling, principal components analysis, and t-distributed stochastic ...

查看文献

问题

This paper explores the semantic structures of Mandarin and English lexicons using distributional semantics. It investigates similarities and differences in how semantic categories cluster in both languages, utilizing FastText word embeddings and dimensionality reduction techniques.

方法

The study used FastText word embeddings for Mandarin and English, applying multidimensional scaling, principal components analysis, and t-distributed stochastic neighbor embedding for dimensionality reduction. Procrustes analysis was used to align and compare the semantic spaces.

关键发现

T-SNE provided the clearest semantic category clustering. Similar differentiation between nouns, verbs, and adjectives was observed in both languages, as well as between concrete and abstract words. Procrustes analysis revealed subtle differences in semantic lexicon structure.

3个要点

T-SNE is effective for visualizing semantic clusters, outperforming PCA and MDS in differentiating semantic categories in both Mandarin and English.
Procrustes analysis is a useful tool for aligning and comparing semantic spaces across languages, highlighting both similarities and differences.
The study reveals both shared and distinct semantic organization patterns in Mandarin and English, reflecting universal cognitive structures and language-specific cultural influences.

学术详情点击展开

假设:The subtle differences in how words of different semantic categories are used in Mandarin and in English are likely to be reflected in the corpus-based semantic vectors of these words.

研究对象:Words from 21 partly culture-specific semantic categories for Mandarin and English.

研究设计:Cross-linguistic study of lexical semantics within the framework of distributional semantics.

结果指标:Clustering of semantic categories, distances between centroids, cosine similarities of centroids, and procrustes analysis residuals.

样本量:2173 Mandarin words and 2150 English words.

统计方法:Linear discriminant analysis, support vector machines, random forest, multidimensional scaling, principal component analysis, t-distributed stochastic neighbor embedding, Procrustes analysis, cosine similarity.

局限性:The number of words included is small, words assigned to only one category, subjective selection of words, FastText embeddings represent blends of context-specific senses, and Fasttext embeddings for Mandarin were trained on corpora written in both simplified and traditional Chinese.

未来研究方向:Comparing the semantic systems of other different languages and cultures, such as French, Estonian, Persian, Hindi, Japanese, Arabic, and Korean.

关键发现:The semantic spaces of Mandarin and English have many features in common, but also reveal some clear differences, which are highlighted most clearly by a graph-based analysis and procrustes analysis.

生成于 11/25/2025