Hierarchical Learning for Generation with Long Source Sequences

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.

References in corpus (12)

• Sequence to Sequence Learning with Neural Networks
• Teaching Machines to Read and Comprehend
• Longformer: The Long-Document Transformer
• A Deep Reinforced Model for Abstractive Summarization
• Multilingual Denoising Pre-training for Neural Machine Translation
• SAMSum Corpus: A Human-annotated Dialogue Dataset for Abstractive Summarization
• Big Bird: Transformers for Longer Sequences
• Hierarchical Recurrent Attention Network for Response Generation
• BP-Transformer: Modelling Long-Range Context via Binary Partitioning
• Mind The Facts: Knowledge-Boosted Coherent Abstractive Text Summarization
• Deep Reinforcement Learning with Stacked Hierarchical Attention for Text-based Games
• Efficient Attentions for Long Document Summarization

Cited by in corpus (5)

• An Empirical Survey on Long Document Summarization: Datasets, Models and Metrics
• Summ^N: A Multi-Stage Summarization Framework for Long Input Dialogues and Documents
• Abstractive Text Summarization Using the BRIO Training Paradigm
• Ascle: A Python Natural Language Processing Toolkit for Medical Text Generation
• An Exploratory Study on Long Dialogue Summarization: What Works and What's Next