Grant: Canada Research Chairs

The objective of this Canada Research Chair in Natural Language Information Processing and Applications is to develop methods to understand and to represent the contents of texts so that they can be communicated and accessed effectively in various applications. Textual information in natural language takes a large part of our collective human knowledge. Effective access to and use of relevant information is essential in our daily lives, which can only be achieved when the information is well understood. Natural language is also a natural means for users to interact with systems. Effective communication with systems also requires deep understanding of the language. The research program of this chair focuses on language understanding and representation, as well as its applications in different domains. A key characteristic of this program is the exploitation of knowledge in text understanding. Instead of relying only on large language models built on large volumes of texts, we will incorporate knowledge to guide the understanding process. Our assumption is that text understanding should be guided by knowledge already acquired, as in human understanding. Knowledge will help us to recognize important concepts from texts, to relate them to other concepts, and to make reasoning based on it for decision making. In this research program, we intend to incorporate both general and domain knowledge in all steps of natural language information processing: information extraction from texts, text representation and text matching. Natural language information processing is required in many applications: for search of relevant information, for interacting with users in dialogues, or for answering user’s questions. The general natural language understanding methods we develop will be applied to these scenarios. In addition, specific domain knowledge, in particular, medical knowledge, will be incorporated to build domain-specific search engines, dialogue and question answering systems. With the strong role of knowledge in language understanding, this program differs from most current endeavours in natural language processing. It will provide new insights on the fundamental problems of language understanding. We also expect that knowledge will enable us to be less dependent on a huge amount of training data than the current approaches, thus the resulting methods can be more easily applied in real situations.