Sound symbolism in Japanese names: Machine learning approaches to gender classification.

Area of Science:

• Computational Linguistics
• Psycholinguistics
• Sound Symbolism Research

Background:

• Investigates the linguistic phenomenon of sound symbolism, where phonemes carry inherent meaning.
• Examines whether sound-meaning mappings, prevalent in Indo-European languages, apply to Japanese names.
• Addresses the need to understand cultural and linguistic specificity in sound symbolism.

Purpose of the Study:

• To explore how gender is phonemically expressed in Japanese names using machine learning.
• To compare the efficacy of Random Forest and XGBoost algorithms in gender classification of names.
• To identify specific phonemes associated with masculine and feminine gender in Japanese names.

Main Methods:

• Trained Random Forest and XGBoost machine learning models on Japanese name phonemes and associated gender.
• Employed k-fold cross-validation (28 folds) for robust model evaluation.
• Utilized feature importance analysis to determine phonemic contributions to gender classification.

Main Results:

• Both algorithms achieved reasonable accuracy in classifying Japanese names by gender.
• XGBoost significantly outperformed the Random Forest algorithm in classification accuracy.
• Identified specific phonemes: /m/ and /k/ associated with femininity, and /i/ with masculinity, contrary to other languages.

Conclusions:

• Japanese exhibits unique sound symbolic expressions of gender, differing from typical cross-linguistic patterns.
• XGBoost demonstrates superior capability in capturing complex phoneme-gender relationships.
• Findings emphasize the importance of cultural context in sound symbolism and provide insights into linguistic gender markers.