# "Is Whole Word Masking Always Better for Chinese BERT?": Probing on Chinese Grammatical Error Correction

arXiv: https://arxiv.org/abs/2203.00286 · HTML: https://mlsanity.com/paper/2203.00286/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Builds on 16 of its 26 references that are in the map.
- First posted on arXiv 1 Mar 2022.

## Builds on (most cited first, 16 of 16)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Adam: A Method for Stochastic Optimization](https://mlsanity.com/paper/1412.6980/) (2014)
- [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://mlsanity.com/paper/1907.11692/) (2019)
- [Neural Machine Translation of Rare Words with Subword Units](https://mlsanity.com/paper/1508.07909/) (2015)
- [Language Models as Knowledge Bases?](https://mlsanity.com/paper/1909.01066/) (2019)
- [Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation](https://mlsanity.com/paper/1609.08144/) (2016)
- [ERNIE: Enhanced Representation through Knowledge Integration](https://mlsanity.com/paper/1904.09223/) (2019)
- [CLUE: A Chinese Language Understanding Evaluation Benchmark](https://mlsanity.com/paper/2004.05986/) (2020)
- [Revisiting Pre-Trained Models for Chinese Natural Language Processing](https://mlsanity.com/paper/2004.13922/) (2020)
- [Commonsense Knowledge Mining from Pretrained Models](https://mlsanity.com/paper/1909.00505/) (2019)
- [oLMpics -- On what Language Model Pre-training Captures](https://mlsanity.com/paper/1912.13283/) (2019)
- [ChineseBERT: Chinese Pretraining Enhanced by Glyph and Pinyin Information](https://mlsanity.com/paper/2106.16038/) (2021)
- [AMBERT: A Pre-trained Language Model with Multi-Grained Tokenization](https://mlsanity.com/paper/2008.11869/) (2020)
- [Lattice-BERT: Leveraging Multi-Granularity Representations in Chinese Pre-trained Language Models](https://mlsanity.com/paper/2104.07204/) (2021)
- [TexSmart: A Text Understanding System for Fine-Grained NER and Enhanced Semantic Analysis](https://mlsanity.com/paper/2012.15639/) (2020)
- [Intrinsic Knowledge Evaluation on Chinese Language Models](https://mlsanity.com/paper/2011.14277/) (2020)

## Built on by (most cited first, 4 of 4)

- [uChecker: Masked Pretrained Language Models as Unsupervised Chinese Spelling Checkers](https://mlsanity.com/paper/2209.07068/) (2022)
- [NaSGEC: a Multi-Domain Chinese Grammatical Error Correction Dataset from Native Speaker Texts](https://mlsanity.com/paper/2305.16023/) (2023)
- [BIOptimus: Pre-training an Optimal Biomedical Language Model with Curriculum Learning for Named Entity Recognition](https://mlsanity.com/paper/2308.08625/) (2023)
- [ExLM: Rethinking the Impact of [MASK] Tokens in Masked Language Models](https://mlsanity.com/paper/2501.13397/) (2025)

## Abstract (arXiv)

Whole word masking (WWM), which masks all subwords corresponding to a word at once, makes a better English BERT model. For the Chinese language, however, there is no subword because each token is an atomic character. The meaning of a word in Chinese is different in that a word is a compositional unit consisting of multiple characters. Such difference motivates us to investigate whether WWM leads to better context understanding ability for Chinese BERT. To achieve this, we introduce two probing tasks related to grammatical error correction and ask pretrained models to revise or insert tokens in a masked language modeling manner. We construct a dataset including labels for 19,075 tokens in 10,448 sentences. We train three Chinese BERT models with standard character-level masking (CLM), WWM, and a combination of CLM and WWM, respectively. Our major findings are as follows: First, when one character needs to be inserted or replaced, the model trained with CLM performs the best. Second, when more than one character needs to be handled, WWM is the key to better performance. Finally, when being fine-tuned on sentence-level downstream tasks, models trained with different masking strategies perform comparably.