Hierarchical Learning for Generation with Long Source Sequences

TL;DR

This paper introduces a Hierarchical Attention Transformer (HAT) that effectively handles long sequences in tasks like summarization and translation, achieving state-of-the-art results and providing insights into hierarchical learning.

Contribution

The paper proposes a novel hierarchical architecture for sequence modeling that improves performance on long-sequence tasks and offers interpretability through attention visualization.

Findings

Outperforms standard Transformers on multiple seq2seq tasks

Achieves state-of-the-art ROUGE scores on four summarization datasets

Improves document-level machine translation performance

Abstract

One of the challenges for current sequence to sequence (seq2seq) models is processing long sequences, such as those in summarization and document level machine translation tasks. These tasks require the model to reason at the token level as well as the sentence and paragraph level. We design and study a new Hierarchical Attention Transformer-based architecture (HAT) that outperforms standard Transformers on several sequence to sequence tasks. Furthermore, our model achieves state-of-the-art ROUGE scores on four summarization tasks, including PubMed, arXiv, CNN/DM, SAMSum, and AMI. Our model outperforms document-level machine translation baseline on the WMT20 English to German translation task. We investigate what the hierarchical layers learn by visualizing the hierarchical encoder-decoder attention. Finally, we study hierarchical learning on encoder-only pre-training and analyze its performance on classification tasks.