Sample Efficient Text Summarization Using a Single Pre-Trained Transformer

TL;DR

This paper introduces a novel approach for text summarization using a pre-trained decoder-only Transformer, achieving high-quality results with limited training data by leveraging the same model for encoding and generation.

Contribution

The paper proposes using a pre-trained decoder-only Transformer for summarization, ensuring all parameters are pre-trained, which improves sample efficiency over traditional encoder-decoder models.

Findings

Achieves 13.1 ROUGE-2 with only 1% of training data.

Outperforms pre-trained encoder-decoder models in limited-data scenarios.

Demonstrates the effectiveness of decoder-only models for sequence-to-sequence tasks.

Abstract

Language model (LM) pre-training has resulted in impressive performance and sample efficiency on a variety of language understanding tasks. However, it remains unclear how to best use pre-trained LMs for generation tasks such as abstractive summarization, particularly to enhance sample efficiency. In these sequence-to-sequence settings, prior work has experimented with loading pre-trained weights into the encoder and/or decoder networks, but used non-pre-trained encoder-decoder attention weights. We instead use a pre-trained decoder-only network, where the same Transformer LM both encodes the source and generates the summary. This ensures that all parameters in the network, including those governing attention over source states, have been pre-trained before the fine-tuning step. Experiments on the CNN/Daily Mail dataset show that our pre-trained Transformer LM substantially improves over pre-trained Transformer encoder-decoder networks in limited-data settings. For instance, it achieves 13.1 ROUGE-2 using only 1% of the training data (~3000 examples), while pre-trained encoder-decoder models score 2.3 ROUGE-2.