Adapting knowledge graph embedding for neural machine translation
Nha Tran, Long Nguyen, Nam Nguyen, Tri Le
Vietnam Journal of Science and Technology/Science and Technology
Problems Identified (4)
NMT data scarcity: Neural machine translation models face challenges from needing large training corpora, especially for low-resource language pairs, causing corpus sparsity.
Rare and OOV word translation: Neural machine translation needs better handling of rare and out-of-vocabulary words.
NMT data scarcity: Neural machine translation models face challenges from needing large training corpora, especially for low-resource language pairs, causing corpus sparsity.
Rare and OOV word translation: Neural machine translation needs better handling of rare and out-of-vocabulary words.
Proposed Solutions (2)
KGE-NMT knowledge graph integration: The paper proposes KGE-NMT, which integrates knowledge graphs into NMT models and uses structured KG knowledge to improve entity semantic representations.
KGE-NMT knowledge graph integration: The paper proposes KGE-NMT, which integrates knowledge graphs into NMT models and uses structured KG knowledge to improve entity semantic representations.
Results (3)
Baseline outperformance on IWSLT:
Improved translation quality:
Baseline outperformance on IWSLT:
Research Domain
Neural machine translation / knowledge graph embeddings