A cross-temporal contrastive disentangled model for ancient Chinese understanding

Area of Science:

• Computational Linguistics
• Historical Linguistics
• Natural Language Processing

Background:

• Ancient Chinese is vital for understanding Chinese history and culture.
• Existing methods struggle with semantic and syntactic differences between ancient and modern Chinese.
• These differences lead to misunderstandings in ancient Chinese text analysis.

Purpose of the Study:

• To propose a novel language pre-training framework for ancient Chinese understanding.
• To bridge the semantic and syntactic gaps between ancient and modern Chinese.
• To improve the accuracy of ancient Chinese text analysis and interpretation.

Main Methods:

• Developed a Cross-temporal Contrastive Disentanglement Model (CCDM).
• Utilized a parallel ancient-modern corpus for cross-temporal data augmentation.
• Employed disentangling and reconstructing techniques for corpus enhancement.
• Applied cross-temporal contrastive learning for model training.

Main Results:

• The CCDM framework significantly outperforms state-of-the-art baselines.
• Achieved superior performance on six diverse ancient Chinese understanding tasks.
• Demonstrated the effectiveness of cross-temporal contrastive learning and data augmentation.

Conclusions:

• The proposed CCDM framework effectively addresses the challenges of ancient Chinese understanding.
• The model successfully bridges the linguistic gap between ancient and modern Chinese.
• The framework shows potential for application to other evolutionary languages.