Introduction to Knowledge-augmented NLP

There has been tremendous progress on the research of Natural Language Processing since more than half a century ago. While the latest development of Large Language Models (LLMs) has brought unprecedented enthusiasm about achieving human-level language understanding and generation, there still exist considerable limitations of these language models, such as the lack of world knowledge, explainability and generalization. To solve these issues, it is important to augment NLP models with external knowledge sources, including both unstructured knowledge, e.g., free-form text, and structured knowledge, e.g., knowledge graphs. The integration of these knowledge sources consists of three steps: (1) Grounding language into related knowledge; (2) Representing knowledge; and (3) Fusing knowledge representation into language models.