Using Second-order Vectors in a Knowledge-based Method for Acronym Disambiguation

Proceedings of the Fifteenth Conference on Computational Natural Language Learning, pages 145–153, Portland, Oregon, USA, 23–24 June 2011. c⃝2011 Association for Computational Linguistics Using Second-order Vectors in a Knowledge-based Method for Acronym Disambiguation Bridget T. McInnes∗ College of Pharmacy University of Minnesota Minneapolis, MN 55455 Ted Pedersen Department of Computer Science University of Minnesota Duluth, MN 55812 Ying Liu College of Pharmacy University of Minnesota Minneapolis, MN 55455 Serguei V. Pakhomov College of Pharmacy University of Minnesota Minneapolis, MN 55455 Genevieve B. Melton Institute for Health Informatics University of Minnesota Minneapolis, MN 55455 Abstract In this paper, we introduce a knowledge-based method to disambiguate biomedical acronyms using second-order co-occurrence vectors. We create these vectors using information about a long-form obtained from the Unified Medical Language System and Medline. We evaluate this method on a dataset of 18 acronyms found in biomedical text. Our method achieves an overall accuracy of 89%. The results show that using second-order features provide a dis- tinct representation of the long-form and po- tentially enhances automated disambiguation. 1 Introduction Word Sense Disambiguation (WSD) is the task of automatically identifying the appropriate sense of a word with multiple senses. For example, the word culture could refer to anthropological culture (e.g., the culture of the Mayan civilization), or a laboratory culture (e.g., cell culture). Acronym disambiguation is the task of automat- ically identifying the contextually appropriate long- form of an ambiguous acronym. For example, the acronym MS could refer to the disease Multiple Scle- rosis, the drug Morphine Sulfate, or the state Missis- sippi, among others. Acronym disambiguation can be viewed as a special case of WSD, although, un- like terms, acronyms tend to be complete phrases or expressions, therefore collocation features are not as easily identified. For example, the feature rate when disambiguating the term interest, as in ∗Contact author : bthomson@umn.edu. interest rate, may not be available. Acronyms also tend to be noun phrases, therefore syntactic features do not provide relevant information for the purposes of disambiguation. Identifying the correct long-form of an acronym is important not only for the retrieval of information but the understanding of the information by the re- cipient. In general English, Park and Byrd (2001) note that acronym disambiguation is not widely studied because acronyms are not as prevalent in lit- erature and newspaper articles as they are in specific domains such as government, law, and biomedicine. In the biomedical sublanguage domain, acronym disambiguation is an extensively studied problem. Pakhomov (2002) note acronyms in biomedical lit- erature tend to be used much more frequently than in news media or general English literature, and tend to be highly ambiguous. For example, the Uni- fied Medical Language System (UMLS), which in- cludes one of the largest terminology resources in the biomedical domain, contains 11 possible long- forms of the acronym MS in addition to the four examples used above. Liu et al. (2001) show that 33% of acronyms are ambiguous in the UMLS. In a subsequent study, Liu et al. (2002a) found that …

Content truncated to 500 words