Optimizing software engineering English translation using an enhanced Grey Wolf Optimization with self-attention and...

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Computational Linguistics

Background:

• Traditional neural machine translation models like Transformer and LSTM face challenges in dynamic hyperparameter optimization and domain adaptation.
• Suboptimal accuracy and efficiency hinder the application of existing models in specialized fields such as software engineering.
• Bridging language gaps effectively requires advanced machine translation solutions.

Purpose of the Study:

• To propose an enhanced machine translation model, Adaptive Grey Wolf Optimization with Self-Attention and LSTM (AGWO-SALSTM), for improved Chinese-to-English translation.
• To dynamically optimize hyperparameters and enhance contextual understanding for greater translation accuracy and efficiency.
• To address the limitations of traditional models in specialized domains.

Main Methods:

• Developed the AGWO-SALSTM model, integrating adaptive Grey Wolf Optimization (AGWO) for hyperparameter tuning (learning rates, attention weights, network configurations).
• Incorporated self-attention mechanisms and bidirectional Long Short-Term Memory (LSTM) networks to improve contextual understanding and sequential data processing.
• Validated the AGWO-SALSTM model against Transformer, LSTM-Seq2Seq, and MT5 baselines using PARACRAWL, WMT, UM-Corpus, and OPUS datasets.

Main Results:

• AGWO-SALSTM demonstrated superior performance over baseline models across all tested datasets.
• Achieved an average translation accuracy of 95.56%, significantly outperforming MT5 (90.94%).
• Required fewer iterations (16-20) for convergence compared to the Transformer model (up to 57), indicating enhanced efficiency.

Conclusions:

• The AGWO-SALSTM model offers a significant advancement in machine translation accuracy and efficiency, particularly for Chinese-to-English translation in specialized domains.
• Dynamic hyperparameter optimization via AGWO and enhanced contextual processing through self-attention and LSTM are key to the model's improved performance.
• The proposed model effectively overcomes the limitations of traditional neural machine translation approaches, paving the way for more robust and efficient language bridging solutions.