Research on Korean Neural Machine Translation Model Based on Attention Mechanism
Xinfeng Wang, Qingsong Chang · 2025
The study examines a Korean Neural Machine Translation (NMT) model based on Attention Mechanism (AM). In the context of globalization, the demand for crosslingual communication is increasing, yet Korean machine translation faces challenges due to scarce data resources and complex grammatical structures. To improve translation quality, this paper proposes an innovative NMT model that integrates specially designed AM into the encoder-decoder framework. The model employs a bidirectional long short-term memory network (Bi-LSTM) as the encoder to capture bidirectional dependencies in sentences and utilizes multi-head AM in the decoder to dynamically focus on different parts of the source sentence, thereby providing more precise contextual information. Experimental results show that the proposed model outperforms baseline models based on RNN, CNN, and Transformer across multiple evaluation metrics such as BLEU, METEOR, ROUGE-N, and TER, demonstrating significant performance improvements. Particularly, when dealing with the complex grammar and morphological changes specific to the Korean language, the model exhibits stronger adaptability and accuracy, offering new insights for research in the field of Korean machine translation.