Joint Generator-Ranker Learning for Natural Language Generation

TL;DR

The paper introduces JGR, a joint training framework for generator and ranker in text generation, improving quality by mutual feedback and outperforming existing methods across multiple datasets.

Contribution

It presents a novel joint training algorithm that integrates generator and ranker with hybrid and contrastive objectives, enhancing text generation performance.

Findings

JGR outperforms existing methods on four datasets.

Joint training improves generation quality.

Effective mutual feedback between generator and ranker.

Abstract

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.