Interpreting comparative constructions in biomedical text

We propose a methodology using underspecified semantic interpretation to process comparative constructions in MEDLINE citations, concentrating on two structures that are prevalent in the research literature reporting on clinical trials for drug therapies. The method exploits an existing semantic processor, SemRep, which constructs predications based on the Unified Medical Language System. Results of a preliminary evaluation were recall of 70%, precision of 96%, and F-score of 81%. We discuss the generalization of the methodology to other entities such as therapeutic and diagnostic procedures. The available structures in computable format are potentially useful for interpreting outcome statements in MEDLINE citations.