Iterative Multilingual Spectral Attribute Erasure
This addresses bias in multilingual AI models, enabling debiasing transfer between languages, which is incremental as it builds on existing spectral methods but extends them to a multilingual context.
The paper tackles the problem of debiasing multilingual representations by proposing IMSAE, which identifies and mitigates joint bias subspaces across multiple languages through iterative SVD-based truncation, showing effectiveness across eight languages and five demographic dimensions with improved performance over monolingual and cross-lingual approaches.
Multilingual representations embed words with similar meanings to share a common semantic space across languages, creating opportunities to transfer debiasing effects between languages. However, existing methods for debiasing are unable to exploit this opportunity because they operate on individual languages. We present Iterative Multilingual Spectral Attribute Erasure (IMSAE), which identifies and mitigates joint bias subspaces across multiple languages through iterative SVD-based truncation. Evaluating IMSAE across eight languages and five demographic dimensions, we demonstrate its effectiveness in both standard and zero-shot settings, where target language data is unavailable, but linguistically similar languages can be used for debiasing. Our comprehensive experiments across diverse language models (BERT, LLaMA, Mistral) show that IMSAE outperforms traditional monolingual and cross-lingual approaches while maintaining model utility.