Knowledge-based NLP is gaining momentum again after a period of relative calmness in the 1990s and early 2000s. This development is made possible by knowledge extraction and acquisition endeavours of various kinds: manual/ semi-automatic /automatic from structured, semi-structured or unstructured texts. In this talk I will present an overview of the work of my NLP group at EML Research, which focuses on Wikipedia as a source of knowledge. We focus mostly on the categories and category structure underlying Wikipedia. I will present the type of knowledge we extract, and how we use this for several NLP applications.