Text Generation by Learning from Demonstrations
This work addresses quality and bias issues in text generation for NLP applications, offering a novel method that improves performance across multiple tasks.
The paper tackles the problems of low-quality samples and exposure bias in text generation by framing it as an offline reinforcement learning problem with expert demonstrations, proposing the GOLD algorithm that upweights confident tokens and downweights unconfident ones during training, resulting in models that outperform MLE and policy gradient on tasks like summarization, question generation, and machine translation.
Current approaches to text generation largely rely on autoregressive models and maximum likelihood estimation. This paradigm leads to (i) diverse but low-quality samples due to mismatched learning objective and evaluation metric (likelihood vs. quality) and (ii) exposure bias due to mismatched history distributions (gold vs. model-generated). To alleviate these problems, we frame text generation as an offline reinforcement learning (RL) problem with expert demonstrations (i.e., the reference), where the goal is to maximize quality given model-generated histories. We propose GOLD (generation by off-policy learning from demonstrations): an easy-to-optimize algorithm that learns from the demonstrations by importance weighting. Intuitively, GOLD upweights confident tokens and downweights unconfident ones in the reference during training, avoiding optimization issues faced by prior RL approaches that rely on online data collection. According to both automatic and human evaluation, models trained by GOLD outperform those trained by MLE and policy gradient on summarization, question generation, and machine translation. Further, our models are less sensitive to decoding algorithms and alleviate exposure bias.