Graph-Based Bilingual Recurrent Autoencoder

Abstract

The invention discloses a bilingual recursion auto-encoder based on a graph, and relates to natural language processing based on deep learning. Bilingual phrases are extracted from parallel corpora as training data to calculate a translation probability between the bilingual phrases; on the basis of a method of pivot language, a retelling probability is calculated; a semantic relation graph for the bilingual phrases is constructed; and on the basis of the semantic relation graph for the bilingual phrases, a model target function is quantified, and model parameter training is carried out. A purpose of favorably learning the bilingual phrase embedding representation is taken as a target, and the bilingual recursion auto-encoder based on the graph is put forward by aiming at a situation that a traditional method lacks the consideration of a more sufficient semantic constraint relation in natural language. The bilingual recursion auto-encoder has the advantages of specific algorithm and clear thought, the learnt bilingual phrase embedding representation can be improved so as to more favorably act to a natural language processing task. Firstly, the semantic relation graph for the bilingual phrases is constructed, and two implicit semantic constraints are defined through a graph structure for learning more accurate bilingual phrase embedding representation so as to be better applied to the natural language processing tanks, such as machine translation.

Description

Technical field [0001] The present invention relates to natural language processing based on deep learning, in particular to a bilingual recursive autoencoder based on graphs. Background technique [0002] Natural language processing is an important research direction of artificial intelligence in computer science. It studies how to enable effective communication between humans and computers using natural language. It is a subject integrating linguistics, computer science, and mathematics. [0003] The present invention mainly involves the construction of a graph-based bilingual recursive autoencoder and its use in bilingual phrase embedding representation modeling. A neural network is a mathematical model that uses a synaptic connection structure similar to the brain's nerves to process information. In recent years, the research of natural language processing based on neural networks has become the main trend in the development of this discipline, and various neural networks ha...

AI Technical Summary

Problems solved by technology

However, the traditional method only considers the reconstruction error and the semantic correspondence of bilingual phrases in the modeling process, but lacks more adequate consideration of the semantic constraint relationship

Therefore, there are still deficiencies in the existing methods, and how to learn better bilingual phrase embedding representations is still a problem worth studying

Images