Hierarchical Learning for Generation with Long Source Sequences
This addresses a key bottleneck in NLP for applications requiring long-sequence processing, such as summarization and translation, though it appears incremental as an extension of Transformer-based methods.
The paper tackles the challenge of processing long sequences in seq2seq models for tasks like summarization and document-level machine translation by introducing a Hierarchical Attention Transformer (HAT) architecture, which achieves state-of-the-art ROUGE scores on four summarization tasks and outperforms baselines on WMT20 English to German translation.
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.