Hybrid Translation with Classification: Revisiting Rule-Based and Neural Machine Translation

• DOI: 10.3390/electronics9020201
• Publication Date: 2020-01-21T16:25:59Z
• Authors: Jin-Xia Huang; Kyung-Soon Lee; Young-Kil Kim
• Keywords: Training set

This paper proposes a hybrid machine-translation system that combines neural machine translation with well-developed rule-based to utilize the stability of latter compensate for inadequacy in rare-resource domains. A classifier is introduced predict which from two systems more reliable. We explore set features reflect reliability and its process, training data automatically expanded small, human-labeled dataset solve insufficient-data problem. series experiments shows system’s accuracy improved, especially out-of-domain translations, classification greatly improved when using proposed constructed set. comparison between feature- text-based also performed, results show feature-based model achieves better accuracy, even compared network text classifiers.