English translation enhancement technology supported by deep learning and knowledge graphs

The globalization of information and the growing demand for multilingual communication have made accurate and context-aware English translation (ET) a critical research challenge. Existing ET models often suffer from limitations in context preservation, entity translation accuracy, and semantic consistency, particularly in domain-specific texts. To report these challenges, this research proposes an enhanced ET framework that integrates deep learning (DI) with Knowledge Graphs (KGs) to incorporate structured semantic and entity-level information into the translation process. The proposed approach performs linguistic preprocessing, including tokenization, lemmatization, stemming, and part-of-speech tagging, followed by feature extraction using Word2Vec and Bidirectional Encoder Representations from Transformers (BERT). KGs are employed to model entity relationships and domain-specific semantics, which are jointly learned using a Nomadic People Optimizer-driven Knowledge Graph Peephole Long Short-Term Memory (NPO-KG-Peephole LSTM) network. In addition, bilingual alignments between source and target KGs are utilized to enrich training data and improve cross-domain generalization. Experimental results obtained through Python-based implementation demonstrate that the proposed framework improves translation performance by enhancing entity translation accuracy and contextual understanding. The model achieves a precision of 0.9765, a recall of 0.9728, and an H-mean of 0.9746, indicating effective semantic integration. These results suggest that combining DL architectures with knowledge graph representations provides a scalable and semantically enriched solution for both general-purpose and domain-specific ET tasks.