A Machine Translation Framework Based on Neural Network Deep Learning: from Semantics to Feature Analysis

This paper uses an encoder-decoder framework based on semantic to feature analysis to construct a neural machine translation model, let the machine automatically perform feature learning, transform corpus data into word vectors in a distributed representation, and use neural networks to implement source language and Direct mapping between target languages. A word alignment method based on deep neural network is proposed, which effectively utilizes the similarity of vocabulary and context information to model word alignment more accurately. This method uses neural network dimensionality reduction method to learn from unlabeled data. The low-dimensional vector representation of the ordering feature is then used to combine the low-dimensional feature representation with other features using a multi-layer neural network and integrated into a linear ordering model.