Unreferenced English articles' translation quality-oriented automatic evaluation technology using sparse autoencoder...

Area of Science:

• Natural Language Processing
• Machine Learning
• Artificial Intelligence

Background:

• Current translation quality assessment (TQA) methods lack a strong theoretical-evaluative link.
• Existing automatic TQA lacks precision for unreferenced English articles.

Purpose of the Study:

• To propose a novel deep learning-based automatic TQA model for unreferenced English articles.
• To enhance the accuracy and reliability of TQA by addressing the limitations of current evaluation techniques.

Main Methods:

• Utilizing Sparse AutoEncoder (SAE) within a Deep Learning (DL) framework for TQA.
• Employing AutoEncoder (AE) for unsupervised learning of bilingual word representations and feature extraction.
• Integrating reconstructed translation language vector features into the DL-based TQA model.

Main Results:

• The proposed model demonstrates increasing evaluation scores with a larger number of sentences.
• The model achieves high-precision TQA for unreferenced English articles.
• The Bilingual Evaluation Understudy (BLEU) score shows consistent high performance, increasing from 96 to 98 with increased sentence count.

Conclusions:

• The developed SAE-based DL model offers a high-precision solution for automatic TQA of unreferenced English articles.
• The model effectively reconstructs and utilizes bilingual word vector features for improved translation evaluation.
• This approach addresses the deficiency in current TQA methods by establishing a stronger link between evaluation theory and practice.