Neural codec for natural language understanding

In recent years, distributed representation and codec technologies have been widely used in natural language processing. At the same time, they have also provided new approaches and paradigms for language cognitive computing research. So, how does the human brain represent language? How are the neural encoding and decoding performanced in different distributed representation methods? Can language information be reconstructed from neuroimaging data? This report will introduce part of the exploratory work that our team has carried out in recent years on those issues, and give preliminary prospects for the future development of this direction.

近年来，分布式表示（distributed representation）与编解码（codec）技术已在自然语言处理领域得到广泛应用。与此同时，二者亦为语言认知计算研究提供了全新的路径与范式。那么，人类大脑如何表征语言？针对不同的分布式表示方法，神经编解码机制的运作方式是怎样的？能否从神经影像数据中重构语言信息？本报告将介绍我们团队近年来针对上述问题开展的部分探索性工作，并对该研究方向的未来发展作出初步展望。