Mitigating Political Bias in Language Models through Reinforced Calibration

Current large-scale language models can be politically biased as a result of the data they are trained on, potentially causing serious problems when deployed in real-world settings. In this paper, we describe metrics for measuring political bias GPT-2 generation and propose reinforcement learning (RL) framework mitigating biases generated text. By using rewards from word embeddings or classifier, our RL guides debiased without having access to training requiring model retrained. empirical experiments on three attributes sensitive (gender, location, topic), methods reduced according both human evaluation, while maintaining readability semantic coherence.