Exploiting Biased Models to De-bias Text: A Gender-Fair Rewriting Model
This work addresses gender bias in AI for languages with complex morphology, offering a practical solution without handcrafted rules, though it is incremental as it builds on existing rewriting approaches.
The paper tackled the problem of gender bias in natural language generation for morphologically complex languages like German by proposing a rewriting model that uses machine translation to create training data from gender-fair text, achieving performance comparable to state-of-the-art methods for English and increasing gender-fairness in human evaluations.
Natural language generation models reproduce and often amplify the biases present in their training data. Previous research explored using sequence-to-sequence rewriting models to transform biased model outputs (or original texts) into more gender-fair language by creating pseudo training data through linguistic rules. However, this approach is not practical for languages with more complex morphology than English. We hypothesise that creating training data in the reverse direction, i.e. starting from gender-fair text, is easier for morphologically complex languages and show that it matches the performance of state-of-the-art rewriting models for English. To eliminate the rule-based nature of data creation, we instead propose using machine translation models to create gender-biased text from real gender-fair text via round-trip translation. Our approach allows us to train a rewriting model for German without the need for elaborate handcrafted rules. The outputs of this model increased gender-fairness as shown in a human evaluation study.