Generating automated meeting summaries

The thesis at hand introduces a novel approach for the generation of abstractive summaries of meetings. While the automatic generation of document summaries has been studied for some decades now, the novelty of this thesis is mainly the application to the meeting domain (instead of text documents) as well as the use of a lexicalized representation formalism on the basis of Frame Semantics. This allows us to generate summaries abstractively (instead of extractively). The thesis begins with an overall motivation of the research domain, and a description of the central research questions. After that, the notion of a “summary” is discussed in general, and different dimensions of summaries are compared, before we give a broad overview over related work in the field of automatic summarization. Then, we introduce the necessary theories for this approach and the data sets used. Following that, we discuss the architecture of the MEESU system which has been developed in the course of this work, as well as the theory and implementation of the contained components. The system has been evaluated using a novel extrinsic evaluation approach which is detailed next. The thesis concludes with a summary and a discussion of possible points for future work.