Learning ontological knowledge from the Web

Summary form only given. Two ongoing developments make the (semi-)automated construction of shallow semantic resources such as shallow ontologies increasingly feasible: advanced machine learning techniques and the growth of the Web. By decomposing the problem of knowledge acquisition for a shallow ontology into a series of steps, we apply the most appropriate resources and techniques to each step. Integrating the results, we begin to realize the decades-old dream of creating a single knowledge ´pool´ that not only captures all (relevant/important) knowledge known, but that is also easy (and perhaps one day even automatic) to update on a continuous basis. Naturally, we have to adopt a methodology that supports testing; for this, we use NLP applications such as text summarization and question answering. It is both very interesting and a lot of fun to see how much we learn, and what kinds of problems arise, when we try to do this. In this paper the author explores some of the experiences his colleagues, students, and he had over the past 5 years.