Should You Mask 15% in Masked Language Modeling?

Alexander Wettig, Tianyu Gao, Zexuan Zhong, Danqi Chen

Research output: Chapter in Book/Report/Conference proceedingConference contribution

10 Scopus citations

Abstract

Masked language models (MLMs) conventionally mask 15% of tokens due to the belief that more masking would leave insufficient context to learn good representations; this masking rate has been widely used, regardless of model sizes or masking strategies. In this work, we revisit this important choice of MLM pre-training. We first establish that 15% is not universally optimal, and larger models should adopt a higher masking rate. Specifically, we find that masking 40% outperforms 15% for BERT-large size models on GLUE and SQuAD. Interestingly, an extremely high masking rate of 80% can still preserve 95% fine-tuning performance and most of the accuracy in linguistic probing, challenging the conventional wisdom about the role of the masking rate. We then examine the interplay between masking rates and masking strategies and find that uniform masking requires a higher masking rate compared to sophisticated masking strategies such as span or PMI masking. Finally, we argue that increasing the masking rate has two distinct effects: it leads to more corruption, which makes the prediction task harder; it also enables more predictions, which benefits optimization. Using this framework, we revisit BERT's 80-10-10 corruption strategy. Together, our results contribute to a better understanding of MLM pre-training.

Original languageEnglish (US)
Title of host publicationEACL 2023 - 17th Conference of the European Chapter of the Association for Computational Linguistics, Proceedings of the Conference
PublisherAssociation for Computational Linguistics (ACL)
Pages2977-2992
Number of pages16
ISBN (Electronic)9781959429449
StatePublished - 2023
Event17th Conference of the European Chapter of the Association for Computational Linguistics, EACL 2023 - Dubrovnik, Croatia
Duration: May 2 2023May 6 2023

Publication series

NameEACL 2023 - 17th Conference of the European Chapter of the Association for Computational Linguistics, Proceedings of the Conference

Conference

Conference17th Conference of the European Chapter of the Association for Computational Linguistics, EACL 2023
Country/TerritoryCroatia
CityDubrovnik
Period5/2/235/6/23

All Science Journal Classification (ASJC) codes

  • Computational Theory and Mathematics
  • Software
  • Linguistics and Language

Fingerprint

Dive into the research topics of 'Should You Mask 15% in Masked Language Modeling?'. Together they form a unique fingerprint.

Cite this