Unified Detoxifying and Debiasing in Language Generation via Inference-time Adaptive Optimization
This addresses ethical concerns in natural language generation for users and developers by providing a practical solution to reduce harmful content, though it is incremental as it builds on prior separate detoxifying and debiasing methods.
The paper tackles the problem of language models generating toxic and biased text by proposing UDDIA, a unified framework that simultaneously detoxifies and debiases outputs during inference, achieving improved performance with minimal quality loss and acceptable computational cost.
Warning: this paper contains model outputs exhibiting offensiveness and biases. Recently pre-trained language models (PLMs) have prospered in various natural language generation (NLG) tasks due to their ability to generate fairly fluent text. Nevertheless, these models are observed to capture and reproduce harmful contents in training corpora, typically toxic language and social biases, raising severe moral issues. Prior works on ethical NLG tackle detoxifying and debiasing separately, which is problematic since we find debiased models still exhibit toxicity while detoxified ones even exacerbate social biases. To address such a challenge, we propose the first unified framework of detoxifying and debiasing called UDDIA, which jointly formalizes these two problems as rectifying the output space. We theoretically interpret our framework as learning a text distribution mixing weighted attributes. Besides, UDDIA conducts adaptive optimization of only a few parameters during decoding based on a parameter-efficient tuning schema without any training data. This leads to minimal generation quality loss and improved rectification performance with acceptable computational cost. Experimental results demonstrate that compared to several strong baselines, UDDIA achieves debiasing and detoxifying simultaneously and better balances efficiency and effectiveness, taking a further step towards practical ethical NLG.