Deep Learning for Natural Language Processing
- Communication between Neural Network Systems and Human
The goal of deep learning for natural language processing (DL4NLP) at Noah’s Ark Lab is to build neural network systems that can communicate with humans and help them on real-world tasks. This kind of systems can understand human language, talk in human language that is both grammatically and semantically correct, and improve its ability of language through interactions with people. More specifically, we focus on neural machine translation and neural dialogue systems.
To this end, we need to make breakthrough on various aspects of natural language processing, including analysis, representation, and generation of natural language. In addition, to deal with the complex and diverse language phenomena, our systems should be able to plan, reason, and access knowledge-bases like a human being, in a fast and effective way. This calls for not only new data structures and new architectures of neural models, but also new learning mechanisms to accomplish the tasks with previously unseen complexity and scale.
Noah's Ark Lab has been extensively working on these topics, with recent highlights on the first purely neural dialogue system and a powerful neural reasoning system that greatly advance the state-of-the-arts. It is however just the first step, and we wish to jointly work with the community to move forward step by step toward the goal.
We have made several datasets used in our research work publicly available, which can be freely downloaded here.