ReSeTOX: Re-learning attention weights for toxicity mitigation in machine translation
This addresses toxicity mitigation in machine translation for users without requiring re-training, but it is incremental as it builds on existing attention mechanisms.
The paper tackled the problem of Neural Machine Translation generating toxic words not in the input, and the result was a 57% reduction in added toxicity while maintaining 99.5% translation quality across 164 languages.
Our proposed method, ReSeTOX (REdo SEarch if TOXic), addresses the issue of Neural Machine Translation (NMT) generating translation outputs that contain toxic words not present in the input. The objective is to mitigate the introduction of toxic language without the need for re-training. In the case of identified added toxicity during the inference process, ReSeTOX dynamically adjusts the key-value self-attention weights and re-evaluates the beam search hypotheses. Experimental results demonstrate that ReSeTOX achieves a remarkable 57% reduction in added toxicity while maintaining an average translation quality of 99.5% across 164 languages.