CLIN 2005 Abstracts
  • Learning Medical Definitions (Ismail Fahmi and Gosse Bouma)
    Ismail Fahmi (Information Science, Groningen University)
    Gosse Bouma (Information Science, Groningen University)
    The ability to answer definition questions (What is the UNESCO, What is aphasia) is important for question answering systems. This is especially true for domain-specific QA systems, where users are typically confronted with highly domain-specific terminology.

    In this talk, we present an experiment in which we tried to recognize definitions of medical terms in free text. From the medical and health-care domain in Dutch Wikipedia, we extracted sentences which had the syntactic properties of definitions. Each sentence in the result was annotated as being a definition or not. Next, we trained classifiers which use a bag of words model, as well as various additional features, such as position of the sentence within the document, and presence of certain syntactic patterns. We report the accuracy of the resulting systems.