Pre-training via Paraphrasing

Abstract

We introduce MARGE, a pre-trained sequence-to-sequence model learned with an unsupervised multi-lingual multi-document paraphrasing objective. MARGE provides an alternative to the dominant masked language modeling paradigm, where we self-supervise the reconstruction of target text by retrieving a set of related texts (in many languages) and conditioning on them to maximize the likelihood of generating the original. We show it is possible to jointly learn to do retrieval and reconstruction, given only a random initialization. The objective noisily captures aspects of paraphrase, translation, multi-document summarization, and information retrieval, allowing for strong zero-shot performance on several tasks. For example, with no additional task-specific training we achieve BLEU scores of up to 35.8 for document translation. We further show that fine-tuning gives strong performance on a range of discriminative and generative tasks in many languages, making MARGE the most generally applicable pre-training method to date.

References in corpus (7)

• Cross-lingual Language Model Pretraining
• Scaling Laws for Neural Language Models
• Multilingual Denoising Pre-training for Neural Machine Translation
• REALM: Retrieval-Augmented Language Model Pre-Training
• No Training Required: Exploring Random Encoders for Sentence Classification
• Generating Wikipedia by Summarizing Long Sequences
• Generalization through Memorization: Nearest Neighbor Language Models

Cited by in corpus (4)

• Neural Language Generation: Formulation, Methods, and Evaluation
• Better Fine-Tuning by Reducing Representational Collapse
• Effects of Word-frequency based Pre- and Post- Processings for Audio Captioning
• Current Limitations of Language Models: What You Need is Retrieval