Optimization of English Machine Translation by Deep Neural Network under Artificial Intelligence

Area of Science:

• Computational Linguistics
• Artificial Intelligence
• Natural Language Processing

Background:

• Current machine translation systems face challenges in global language adaptation and semantic understanding.
• Deep neural networks (DNNs) provide a foundational approach for advanced translation models.
• Transfer learning offers a method to leverage existing models for new tasks, improving efficiency.

Purpose of the Study:

• To enhance machine translation (MT) performance through deep neural network (DNN) modeling and transfer learning.
• To optimize the word alignment function within neural machine translation (NMT) systems.
• To improve semantic understanding and reduce errors in cross-lingual translation tasks.

Main Methods:

• Implemented a deep learning translation network for English.
• Designed a neural machine translation model incorporating transfer learning with a random shielding method for language training.
• Optimized word alignment in a transformer system using a word corpus.

Main Results:

• Achieved significant reductions in average word alignment error rates: 8.1% (EnRo), 24.4% (EnGe), and 22.1% (EnFr) compared to previous algorithms.
• Demonstrated lower word alignment error rates than traditional methods.
• Validated the feasibility of the modeling and optimization approach.

Conclusions:

• The proposed method effectively addresses issues of insufficient information utilization, large parameter scale, and storage difficulties in machine translation.
• The approach provides a viable direction for optimizing and improving neural machine translation (NMT) systems.
• Enhanced semantic understanding and translation accuracy were achieved.