Yamaguchi, A., Chrysostomou, G., Margatina, K. et al. (1 more author) (Submitted: 2021) Frustratingly simple pretraining alternatives to masked language modeling. arXiv. (Submitted)
Abstract
Masked language modeling (MLM), a self-supervised pretraining objective, is widely used in natural language processing for learning text representations. MLM trains a model to predict a random sample of input tokens that have been replaced by a [MASK] placeholder in a multi-class setting over the entire vocabulary. When pretraining, it is common to use alongside MLM other auxiliary objectives on the token or sequence level to improve downstream performance (e.g. next sentence prediction). However, no previous work so far has attempted in examining whether other simpler linguistically intuitive or not objectives can be used standalone as main pretraining objectives. In this paper, we explore five simple pretraining objectives based on token-level classification tasks as replacements of MLM. Empirical results on GLUE and SQuAD show that our proposed methods achieve comparable or better performance to MLM using a BERT-BASE architecture. We further validate our methods using smaller models, showing that pretraining a model with 41% of the BERT-BASE's parameters, BERT-MEDIUM results in only a 1% drop in GLUE scores with our best objective.
Metadata
Item Type: | Article |
---|---|
Authors/Creators: |
|
Copyright, Publisher and Additional Information: | © 2021 The Author(s). For reuse permissions, please contact the Author(s). |
Keywords: | cs.CL; cs.CL; cs.AI; cs.LG |
Dates: |
|
Institution: | The University of Sheffield |
Academic Units: | The University of Sheffield > Faculty of Engineering (Sheffield) > Department of Computer Science (Sheffield) |
Funding Information: | Funder Grant number ENGINEERING AND PHYSICAL SCIENCE RESEARCH COUNCIL EP/V055712/1 Engineering and Physical Sciences Research Council EP/V055712/1 |
Depositing User: | Symplectic Sheffield |
Date Deposited: | 15 Sep 2021 08:52 |
Last Modified: | 15 Sep 2021 08:52 |
Published Version: | https://arxiv.org/abs/2109.01819 |
Status: | Submitted |
Related URLs: | |
Open Archives Initiative ID (OAI ID): | oai:eprints.whiterose.ac.uk:178206 |