Knowledge fusion method applied to machine translation

Abstract

The invention discloses a knowledge fusion method applied to machine translation, which comprises the following steps: using a bilingual corpus to train one or more neural machine translation models as knowledge providers, namely teacher models; translating the source language S in the corpus and extracting teacher model knowledge; carrying out knowledge fusion; the source language S is copied and spliced, and finally a corpus after knowledge fusion is obtained; training a finally used neural machine translation model; source statements in the bilingual corpus are decoded, firstly, the corresponding source statements are separated by using the same sentence splicing method and using separation symbols, then the source statements are sent into a finally trained neural machine translation model to be decoded, then decoded results are separated, separation marks are separators used during sentence splicing, and finally the source statements are decoded. And knowledge fusion is realized. According to the method, the problem that the knowledge often has more errors is solved, and the knowledge of the teacher model and the knowledge in the real corpus can be mutually fused to form the corpus with rich knowledge.

Description

technical field [0001] The invention relates to a neural machine translation technology, specifically a knowledge fusion method, which can improve the performance of a trained model. Background technique [0002] Neural network machine translation usually adopts an encoder-decoder structure to model variable-length input sentences. The encoder realizes the "understanding" of the source language sentence, forming a floating-point vector of a specific dimension, and then the decoder generates the translation result of the target language word by word according to this vector. In the early stage of the development of neural network machine translation, recurrent neural network (RNN, Recurrent Neural Network) was widely used as the network structure of the encoder and decoder. The network is good at modeling natural language. The RNN network represented by the long-term short-term memory network LSTM (Long Short-Term Memory networks) and the Gated Recurrent Unit network GRU (Ga...

AI Technical Summary

Problems solved by technology

[0011] Aiming at the problem of limited knowledge during training in the existing machine translation, which may lead to insufficient learning of the model, the present invention provides a knowledge fusion method applied to machine translation, so that the model can learn from the knowledge learned by other models and then help itself To learn to obtain better performance, the knowledge learned in the training process of the model is more sufficient

Images