Unsupervised Acquisition of Machine Translation Corpus Combining LSTM-BiRNN Classifiers

The field of machine translation has advanced significantly due to artificial intelligence technology. To further enhance the quality of the parallel corpus, the study first achieved bilingual semantic acquisition, created parallel sentence pairs, and built a classifier for extracting parallel sentences using recurrent neural networks. The experimental results show that compared wit h the existing state-of-the-art models, the parallel corpus mining method designed by the study has the best performance in terms of accuracy and F1 value, achieving an accuracy of 0.981 and an F1 value of 0.928. Mining different bilingual corpora all obtained high Jaccard similarity, and mining 10,000 pairs of utterances only took 41.64 s. The machine translation system achieved the best translation quality on the parallel corpus mined by the research-designed method, with bilingual substitution measure of 47.46, character F1 score of 76.06, and translation editing rate of 65.14. The design of the study can further improve the quality of parallel corpus and promote the development of machine translation technology.