Making Cognitive Summarization Agents Work In A Real-World Domain

The advantage of cognitively motivated automatic summarizing is that human users can better understand what happens. This improves acceptability. The basic empirical finding in human summarizers is that they combine a choice of intellectual strategies. We report here on SummIt-BMT (Summarize It in Bone Marrow Transplantation), a prototype system that applies a subset of human strategies to a real-world task: fast information supply for physicians in clinical bone marrow transplantation. The human strategies are converted to knowledge-based agents and integrated into a system environment inspired by user-centered information seeking research. A domain ontology provides knowledge shared by human users and system players. Users' query formulation is supported through empirically founded scenarios. Incoming retrieval results are first roughly checked by means of text passage retrieval before the agents apply strategies of competent human summarizers. The presumably relevant text clips are presented with links to their home positions in the source documents. SummIt-BMT has reached the state of a
prototype running on a Macintosh server (http://summit-bmt.fh-hannover.de/)