Knowledge extraction from text is a longstanding challenge and ambition in Natural Language Processing and AI in general. Success and failure in this area depends however on de nitions and use cases of 'knowledge', which I will address in this talk. Sa ron1 is a framework for knowledge extraction from text that has been developed over several years and was tested in a wide range of use cases. In the talk I will present the basic architecture and functionality of Sa ron and its use in several applications. In the second part of the talk I will address some of the shortcomings of Sa ron, which are the subject of our current research program.