Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

For the Misgendered Chinese in Gender Bias Research: Multi-Task Learning with Knowledge Distillation for Pinyin Name-Gender Prediction

Xiaocong Du, Haipeng Zhang

TL;DR

The paper tackles misgendering risks in gender-bias studies arising from Pinyin-only gender inference for Chinese names. It introduces a Pinyin name-gender prediction task and a knowledge-distillation aided multi-task learning framework in which a Pinyin-to-Character auxiliary task and a Hanzi-based teacher guide the Pinyin student via feature and response distillation. Empirical results on large-scale data show the method outperforms commercial tools by up to ~$20 ext{.}08 extrm{%}$ relative and achieves strong gains over baselines, with open-source release facilitating adoption. The work also discusses an estimated upper bound on accuracy and potential extensions to other languages with many-to-one romanizations, highlighting practical impact for more reliable gender analyses in cross-cultural datasets.

Abstract

Achieving gender equality is a pivotal factor in realizing the UN's Global Goals for Sustainable Development. Gender bias studies work towards this and rely on name-based gender inference tools to assign individual gender labels when gender information is unavailable. However, these tools often inaccurately predict gender for Chinese Pinyin names, leading to potential bias in such studies. With the growing participation of Chinese in international activities, this situation is becoming more severe. Specifically, current tools focus on pronunciation (Pinyin) information, neglecting the fact that the latent connections between Pinyin and Chinese characters (Hanzi) behind convey critical information. As a first effort, we formulate the Pinyin name-gender guessing problem and design a Multi-Task Learning Network assisted by Knowledge Distillation that enables the Pinyin embeddings in the model to possess semantic features of Chinese characters and to learn gender information from Chinese character names. Our open-sourced method surpasses commercial name-gender guessing tools by 9.70\% to 20.08\% relatively, and also outperforms the state-of-the-art algorithms.

For the Misgendered Chinese in Gender Bias Research: Multi-Task Learning with Knowledge Distillation for Pinyin Name-Gender Prediction

TL;DR

The paper tackles misgendering risks in gender-bias studies arising from Pinyin-only gender inference for Chinese names. It introduces a Pinyin name-gender prediction task and a knowledge-distillation aided multi-task learning framework in which a Pinyin-to-Character auxiliary task and a Hanzi-based teacher guide the Pinyin student via feature and response distillation. Empirical results on large-scale data show the method outperforms commercial tools by up to ~ relative and achieves strong gains over baselines, with open-source release facilitating adoption. The work also discusses an estimated upper bound on accuracy and potential extensions to other languages with many-to-one romanizations, highlighting practical impact for more reliable gender analyses in cross-cultural datasets.

Abstract

Achieving gender equality is a pivotal factor in realizing the UN's Global Goals for Sustainable Development. Gender bias studies work towards this and rely on name-based gender inference tools to assign individual gender labels when gender information is unavailable. However, these tools often inaccurately predict gender for Chinese Pinyin names, leading to potential bias in such studies. With the growing participation of Chinese in international activities, this situation is becoming more severe. Specifically, current tools focus on pronunciation (Pinyin) information, neglecting the fact that the latent connections between Pinyin and Chinese characters (Hanzi) behind convey critical information. As a first effort, we formulate the Pinyin name-gender guessing problem and design a Multi-Task Learning Network assisted by Knowledge Distillation that enables the Pinyin embeddings in the model to possess semantic features of Chinese characters and to learn gender information from Chinese character names. Our open-sourced method surpasses commercial name-gender guessing tools by 9.70\% to 20.08\% relatively, and also outperforms the state-of-the-art algorithms.
Paper Structure (22 sections, 10 equations, 4 figures, 6 tables)

This paper contains 22 sections, 10 equations, 4 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Name-based Gender Prediction
  4. Deep Multi-task Learning
  5. Knowledge Distillation
  6. Problem Statement
  7. Method
  8. Pinyin Preprocessing
  9. Pinyin-to-Character Prediction
  10. Response-Based and Feature-Based Knowledge Distillation
  11. Experiments
  12. Experimental Setup
  13. Datasets
  14. Train/Test Split
  15. Evaluation Metrics
  16. ...and 7 more sections

Figures (4)

  • Figure 1: The gender classification error rates of the commercial gender detection tools for names of different origins.
  • Figure 2: The possible correspondences between Pinyin and Chinese characters, as well as the gender information embedded in Chinese characters names, take 'yan' as an example. All the Chinese characters above have the same Pinyin 'yan'.
  • Figure 3: System structure. Orange represents the Pinyin name-gender prediction model (i.e., the student model) that incorporates the Multi-Task Learning module, while blue denotes the Chinese character name-gender prediction model (i.e., the teacher model).
  • Figure 4: Model performance under different training set sizes. (The x-axis is in log scale.)