Chinese names: hanzi improves gender prediction
Summary and headline written by AI from the source article. How we work
Researchers developed new models to better determine the gender of Chinese scientists from their names.
The team created both a regression model, producing a score from 0 to 1 indicating maleness to femaleness, and a classification model assigning male, neutral, or female. They trained these models using Chinese-language physics publications and ratings from 11 native Chinese speakers. The challenge lies in the dual representation of Chinese names: Hanzi characters and their Romanized form, Pinyin.
A single Pinyin name can be written with multiple Hanzi characters, each carrying different gender associations. The researchers addressed this by linking Pinyin names to possible Hanzi versions, then combining predictions based on frequency and a measure of ambiguity. When tested against existing tools like Namsor, GPT, and Claude, the new models outperformed them in accuracy for both gender scores and labels, working with both Hanzi and Pinyin.
Applying the models to authors in APS journals, the system classified every name, while other methods left 20% to 100% unassigned. The code and data are available to the public.
