A baseline feature set for learning rhetorical zones using full articles in the biomedical domain Tony Mullen, Yoko Mizuta, and Nigel Collier   View Paper (PDF)   Return to June 2005, Volume 7, Issue 1 Abstract At a time when experimental throughput in the field of molecular biology is increasing, it is necessary for biologists and people working in related fields to have access to sophisticated tools to enable them to eficiently process large amounts of information in order to stay abreast of current research. Rhetorical zone analysis is an application of natural language processing in which areas of text in scientific papers are classified in terms of argumentation and intellectual contribution in order to pinpoint and distinguish certain types of information. Such analysis can be employed to assist in information extraction, helping to assess and integrate data generated by experiments into the scientific community's store of knowledge. We present results for several experiments in automatic zone identification on the ZAISA-1 dataset, a new dataset composed of full biomedical research papers hand-annotated for rhetorical zones. We concentrate on general purpose and linguistically motivated features, and report results for a variety of sets of features. It is our intention to provide a baseline feature set for modeling, which can be extended in future work using combinations of heuristics and more sophisticated and task-specific modeling techniques. ©2006 Association for Computing Machinery