Application of neural network word slicing method in Mongolian-Chinese ma-chine translation

Abstract

Mongolian word slicing is a major task in Mongolian information processing. The accuracy and reasonableness of the word slicing can alleviate the data sparsity problem and directly affect the subsequent work of Mongolian information processing. The paper firstly presents a pre-processing method based on neural network for Mongolian word slicing to address the stickiness of Mongolian and the rich variation of word morphology. The experimental results show that the accuracy of the neural network word slicing method reaches 99.47%. Secondly, the Transformer Mongolian-Chinese neural machine translation model based on Mongolian word segmentation is constructed. Finally, various Mongolian word -sorting methods on the Transformer translation model were compared. The experimental results show that the word slicing method of BiLSTM-CRF-CNN neural network with filtering of semantically missing concatenated letters and deactivated words achieves a BLEU value of 73.30% in machine translated translations, which can improve the quality of machine tr anslation to a certain extent.