CCNet: Extracting High Quality Monolingual Datasets from Web Crawl Data

Abstract

Pre-training text representations have led to significant improvements in many areas of natural language processing. The quality of these models benefits greatly from the size of the pretraining corpora as long as its quality is preserved. In this paper, we describe an automatic pipeline to extract massive high-quality monolingual datasets from Common Crawl for a variety of languages. Our pipeline follows the data processing introduced in fastText (Mikolov et al., 2017; Grave et al., 2018), that deduplicates documents and identifies their language. We augment this pipeline with a filtering step to select documents that are close to high quality corpora like Wikipedia.

Cited by in corpus (12)

• Multilingual Denoising Pre-training for Neural Machine Translation
• The Pile: An 800GB Dataset of Diverse Text for Language Modeling
• XGLUE: A New Benchmark Dataset for Cross-lingual Pre-training, Understanding and Generation
• Self-training Improves Pre-training for Natural Language Understanding
• Autoregressive Linguistic Steganography Based on BERT and Consistency Coding
• Unsupervised Neural Machine Translation with Generative Language Models Only
• A Survey on Low-Resource Neural Machine Translation
• Multi-task Learning for Multilingual Neural Machine Translation
• Is the Number of Trainable Parameters All That Actually Matters?
• Improving Pretrained Cross-Lingual Language Models via Self-Labeled Word Alignment
• Multilingual AMR-to-Text Generation
• RuSentEval: Linguistic Source, Encoder Force!