The state of the art in text summarization currently is sentence extraction. This method of producing summaries, in spite of its robustness, is rather limited since sentence extracts have well known problems of cohesion and coherence. The research proposed in SKATeR will overcome the limitation of current summarization systems by investigating, proposing, and implementing corpus-based empirical approaches to model the content of documents and abstractive summaries. The success of the technology developed in STAKeR will be demonstrated in the context of knowledge-based summarization.