On the Influence of Masking Policies in Intermediate Pre-training

Qinyuan Ye, Belinda Z. Li, Sinong Wang, Benjamin Bolte, Hao Ma, Wen-tau Yih, Xiang Ren, Madian Khabsa


Abstract
Current NLP models are predominantly trained through a two-stage “pre-train then fine-tune” pipeline. Prior work has shown that inserting an intermediate pre-training stage, using heuristic masking policies for masked language modeling (MLM), can significantly improve final performance. However, it is still unclear (1) in what cases such intermediate pre-training is helpful, (2) whether hand-crafted heuristic objectives are optimal for a given task, and (3) whether a masking policy designed for one task is generalizable beyond that task. In this paper, we perform a large-scale empirical study to investigate the effect of various masking policies in intermediate pre-training with nine selected tasks across three categories. Crucially, we introduce methods to automate the discovery of optimal masking policies via direct supervision or meta-learning. We conclude that the success of intermediate pre-training is dependent on appropriate pre-train corpus, selection of output format (i.e., masked spans or full sentence), and clear understanding of the role that MLM plays for the downstream task. In addition, we find our learned masking policies outperform the heuristic of masking named entities on TriviaQA, and policies learned from one task can positively transfer to other tasks in certain cases, inviting future research in this direction.
Anthology ID:
2021.emnlp-main.573
Volume:
Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing
Month:
November
Year:
2021
Address:
Online and Punta Cana, Dominican Republic
Venue:
EMNLP
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
7190–7202
Language:
URL:
https://aclanthology.org/2021.emnlp-main.573
DOI:
10.18653/v1/2021.emnlp-main.573
Bibkey:
Cite (ACL):
Qinyuan Ye, Belinda Z. Li, Sinong Wang, Benjamin Bolte, Hao Ma, Wen-tau Yih, Xiang Ren, and Madian Khabsa. 2021. On the Influence of Masking Policies in Intermediate Pre-training. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 7190–7202, Online and Punta Cana, Dominican Republic. Association for Computational Linguistics.
Cite (Informal):
On the Influence of Masking Policies in Intermediate Pre-training (Ye et al., EMNLP 2021)
Copy Citation:
PDF:
https://preview.aclanthology.org/paclic-22-ingestion/2021.emnlp-main.573.pdf
Video:
 https://preview.aclanthology.org/paclic-22-ingestion/2021.emnlp-main.573.mp4
Data
KILTNatural QuestionsQuaRTzROPESTriviaQAWebQuestions