bert-ancient-chinese
| Property | Value |
|---|---|
| Author | Jihuai |
| Paper | The Uncertainty-based Retrieval Framework for Ancient Chinese CWS and POS (2022) |
| Vocabulary Size | 38,208 tokens |
| Base Model | BERT-base-chinese |
What is bert-ancient-chinese?
bert-ancient-chinese is a specialized language model designed for processing ancient Chinese texts. Built upon BERT-base-chinese, it implements domain-adaptive pretraining to better handle classical Chinese literature. The model features an expanded vocabulary that includes rare and traditional Chinese characters, making it particularly effective for historical text analysis.
Implementation Details
The model was developed through further pretraining on a massive corpus approximately six times larger than the Siku Quanshu, encompassing various classical Chinese domains including Confucianism, Buddhism, Taoism, poetry, history, medicine, and arts. It employs a BERT+CRF architecture for downstream tasks and has demonstrated superior performance in Chinese Word Segmentation (CWS) and Part-of-Speech (POS) tagging.
- Expanded vocabulary of 38,208 tokens (compared to BERT-base-chinese's 21,128)
- Comprehensive training on diverse ancient Chinese texts
- Domain-adaptive pretraining approach
- State-of-the-art performance on classical Chinese NLP tasks
Core Capabilities
- Chinese Word Segmentation with up to 96.33% F1 score on Zuozhuan dataset
- POS Tagging with up to 92.50% F1 score on Zuozhuan dataset
- Processing of traditional and rare Chinese characters
- Handling of various classical Chinese text genres
Frequently Asked Questions
Q: What makes this model unique?
The model's distinctiveness lies in its significantly expanded vocabulary that includes rare and traditional Chinese characters, coupled with its specialized training on a diverse corpus of ancient Chinese texts. This makes it particularly effective for processing classical Chinese literature.
Q: What are the recommended use cases?
The model is ideal for digital humanities research, particularly in Sinology, historical text analysis, philology, and Chinese history education. It excels in tasks such as word segmentation and part-of-speech tagging for ancient Chinese texts.
