Analysis of Chinese Machine Translation Training Based on Deep Learning Technology

Area of Science:

• Marine Science and Technology
• Computational Linguistics
• Artificial Intelligence

Background:

• Effective communication across languages is crucial in the globalized information age.
• Existing machine translation struggles with specialized domains like marine science.
• Neural network technology offers potential for improved cross-lingual communication.

Purpose of the Study:

• To develop a bidirectional English-Chinese machine translation model for marine science and technology.
• To enhance translation accuracy and fluency using deep learning techniques.
• To address communication barriers in specialized scientific literature.

Main Methods:

• Construction of a specialized Chinese-English corpus for marine science and technology.
• Implementation of local weight sharing in encoders within a bidirectional translation model.
• Fusion of Chinese and English encoder sublayer outputs.
• Evaluation using BLEU (Bilingual Evaluation Understudy) and PPL (Perplexity) metrics.

Main Results:

• The proposed model, with local weight sharing and encoder sublayer fusion, showed improved BLEU scores by 1.6 (Chinese-English) and 3.8 (English-Chinese) compared to the standard Transformer model.
• Perplexity (PPL) values decreased by 18.72% (Chinese-English) and 14.62% (English-Chinese), indicating enhanced translation quality.
• Demonstrated the effectiveness of deep learning for adaptive machine language translation in a specialized field.

Conclusions:

• The developed deep neural network-based translation model significantly improves bidirectional translation for marine science and technology literature.
• The integration of local weight sharing and encoder sublayer fusion enhances model performance.
• This approach offers a more effective alternative to traditional machine translation for specialized scientific domains.