Implicit Unlikelihood Training: Improving Neural Text Generation with Reinforcement Learning (2101.04229v1)

Abstract: Likelihood training and maximization-based decoding result in dull and repetitive generated texts even when using powerful LLMs (Holtzman et al., 2019). Adding a loss function for regularization was shown to improve text generation output by helping avoid unwanted properties, such as contradiction or repetition (Li at al., 2020). In this work, we propose fine-tuning a LLM by using policy gradient reinforcement learning, directly optimizing for better generation. We apply this approach to minimizing repetition in generated text, and show that, when combined with unlikelihood training (Welleck et al., 2020), our method further reduces repetition without impacting the LLM quality. We also evaluate other methods for improving generation at training and decoding time, and compare them using various metrics aimed at control for better text generation output.