Low-resource neural machine translation method based on multi-strategy prototype generation

• Main Authors: ZHU ENCHANG, YU ZHENGTAO, YU ZHIQIANG
• Format: Patent
• Language: Chinese; English
• Published: 28.06.2022
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention relates to a low-resource neural machine translation method based on multi-strategy prototype generation, and belongs to the technical field of natural language processing. The method comprises the steps that firstly, a prototype sequence is retrieved by combining keyword matching and distributed representation matching, and if matching is not obtained, an available pseudo prototype sequence is generated through a pseudo prototype generation method; secondly, in order to effectively utilize the prototype sequence, a traditional encoder-decoder framework is improved. The encoding end receives prototype sequence input by using an additional encoder; a decoding end uses an improved loss function to reduce the influence of a low-quality prototype sequence on a model while controlling information flow by using a gating mechanism. According to the method provided by the invention, the number and quality of retrieval results can be effectively improved based on a small amount of parallel corpora, and t