Joint Generator-Ranker Learning for Natural Language Generation
This addresses a problem in natural language generation for researchers and practitioners by improving generation quality through joint optimization, though it is incremental as it builds on the generate-then-rank paradigm.
The paper tackles the limitation of separate training in generate-then-rank text generation by proposing JGR, a joint training algorithm that integrates generator and ranker, resulting in surpassing existing methods on four public datasets across three generation scenarios.
Generate-then-rank is a widely used mechanism for text generation, where a generator produces multiple text candidates and a ranker chooses the best one among the text candidates. However, existing methods usually train the generator and the ranker individually, neglecting the mutual feedback that could further enhance the generation quality. To tackle this limitation, we propose JGR, a novel joint training algorithm that integrates the generator and the ranker in a single framework. JGR optimizes the generator with a hybrid objective that combines data likelihood and ranker reward, and trains the ranker with a contrastive loss that compares the generator outputs. By iteratively updating the generator and the ranker, JGR can effectively harmonize their learning and enhance their quality jointly. We evaluate JGR on various text generation tasks and demonstrate that it surpasses existing methods on four public datasets across three common generation scenarios. Our code and models are publicly available at https://github.com/microsoft/ProphetNet/tree/master/JGR.