Cognintive Natural Language Processing Link to heading

The cognitive procedure of human intelligence is essential for AGI. We aim to involve the cognitive procedure in the NLP tasks and build a cognitive NLP system. The procedure may unify several parts of language understanding, including linguistics, knowledge graphs, and language models, and result in a more robust, efficient, and explainable system.

Large Scale Reasoning System Link to heading

Reasoning is the core of human intelligence. However, reasoning algorithms play a minor role in current AI practices. This is due to two significant reasons: low efficiency for large data sets and the gap between pure symbolic and natural language. We are working on these two problems to build an efficient language-friendly reasoning system for large-scale knowledge bases.

World Models for Autonomous Driving Link to heading

Autonomous driving needs two types of world-understanding abilities: environment understanding and agent behavior understanding. We are building world models for both capabilities.