Focused Attention Improves Document-Grounded Generation

TL;DR

This paper introduces attention-based adaptations of large pre-trained models for document-grounded text generation, significantly improving performance on Wikipedia update and dialogue response tasks.

Contribution

It proposes novel attention mechanisms in encoder-decoder models for better document representation and relevance, along with a stronger BART baseline for these tasks.

Findings

At least 48% increase in BLEU-4 scores over previous methods

Improved human evaluation scores for relevance and closeness to references

Manual error analysis provides insights for future research

Abstract

Document grounded generation is the task of using the information provided in a document to improve text generation. This work focuses on two different document grounded generation tasks: Wikipedia Update Generation task and Dialogue response generation. Our work introduces two novel adaptations of large scale pre-trained encoder-decoder models focusing on building context driven representation of the document and enabling specific attention to the information in the document. Additionally, we provide a stronger BART baseline for these tasks. Our proposed techniques outperform existing methods on both automated (at least 48% increase in BLEU-4 points) and human evaluation for closeness to reference and relevance to the document. Furthermore, we perform comprehensive manual inspection of the generated output and categorize errors to provide insights into future directions in modeling these tasks.