Resolving Abbreviations in Clinical Texts Without Pre-existing Structured Resources

One of the most important topics in clinical text processing is the identification of relevant concepts. This includes the detection and resolution of abbreviations, acronyms, or other shortened forms in the documents. Even though the task of resolving abbreviations can be treated as a word sense disambiguation problem, such methods require structured lexical knowledge bases. However, for less-resourced languages such resources are not available. In this paper, a method is proposed for the disambiguation and resolution of abbreviations found in Hungarian clinical records. In order to achieve reasonable performance, a lexicon must be created for each domain. It is shown how the set of entries to be included in such a lexicon can be induced from the corpus, thus the manual effort of creating a lexicon can be reduced significantly. The results for resolving abbreviations in Hungarian clinical documents are also shown, which are achieved by using the corpus instead of non-existing structured resources.