A Chinese/English mixed text proofreading algorithm for Central Asian universities based on transfer learning
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
The mixed Chinese and English texts in universities in Central Asia have unique language phenomena such as grammar confusion and semantic ambiguity, resulting in a high false positive and missed negative rate of traditional proofreading methods. To address this issue, this paper proposes a Chinese-English mixed text proofreading algorithm for Central Asian universities based on transfer learning. The core innovation lies in the design of a hierarchical transfer learning framework: Firstly, on the source domain task, a deep semantic representation model based on Transformer is pre-trained using large-scale general Chinese-English parallel corpora; Then, for the target domain task, a dynamic gated domain adapter module is introduced. This module can adaptively adjust the parameters of the source domain model and focus on learning the unique language blending patterns in the context of colleges and universities. Finally, by fusing the bidirectional gated recurrent unit with the self-attention mechanism, the long-distance context dependencies of the mixed text are captured to accurately identify and correct problems such as semantic incoherence and collocation errors caused by language switching. In the experimental section, the algorithm proposed in this paper is compared with the baseline method. The results show that in terms of the three key indicators of precision, recall and F1 value, the algorithm proposed in this paper achieves 92.1%, 89.7% and 90.8% respectively, significantly outperforming the baseline model. Especially in correcting specific error types caused by language habit interference, the algorithm proposed in this paper, with its powerful domain transfer and context awareness capabilities, has improved the proofreading accuracy by approximately 15.7% compared to the optimal baseline model, effectively demonstrating its superiority in practical applications.
Fields of Study & Subjects
Natural Language Processing Techniques
Topic Modeling
Text Readability and Simplification
Publication details
- Journal / source
- IEEE International Conference on Power, Electronics and Green Energy (ICPEGE)
- Pages
- 1326–1332
- Publication date
- 28 January 2026
- Language
- EN