Topic Title: Exploration of Neural Machine Translation Network Structures

 

Technical Area: Natural Language Processing

 

Background

Globalization is one of the three long-term strategies of Alibaba in the coming five to ten years. In order to achieve this goal, it is necessary for Alibaba to build a platform that can process, analyze, and translate local languages wherever it does business. Therefore, multilingual Natural Language Processing (NLP) and Machine Translation (MT) technologies play a key role to achieve this goal. Alibaba has extensive and in-depth application scenarios and has massive multilingual data and corpora for multilingual NLP and MT technologies.

 

Deep Learning has made remarkable progress and has had a profound influence on NLP and MT technologies in recent years. Since Neural Machine Translation (NMT) was proposed in 2014, it has become the mainstream in Machine Translation technology over the past three years; the translation quality of NMT is significantly better than that of traditional Statistical Machine Translation (SMT) technology in most languages and scenarios. Although NMT technology has made great progress, there are still some problems that need to be improved or remain unsolved, which will have a big impact on the practical application of NLP and MT technologies.

 

The mainstream NMT systems have three different network structures: Recurrent Neural Network (RNN), Convolutional Neural Network (CNN), and the Transformer model with Self-Attention Network (SAN). There are advantages and disadvantages of the three network structures in terms of translation performance and efficiency. Therefore, they are all used in academia and industry. It is possible to improve translation performance and efficiency at the same time by exploring new network structures or improving the existing network structures.

 

Target

We are looking for collaboration on this topic to improve the translation performance and decoding efficiency of NMT technology. It can achieve state-of-the-art performance in academia, increase the conversion rate in practical applications, accelerate iterative system updating, and increase the translation speed. We expect the outcomes of the collaboration would be NMT prototype systems with new algorithms, patents, and academic publications.

 

Related Research Topics

NMT is one of the sequence learning applications. Therefore, any new network structures which have proved success in other sequence-to-sequence applications, such as question answering, parsing, etc. could be adapted to machine translation.