An Empirical Evaluation of Resources for the Identif cation of Diseases and Adverse Effects in Biomedical Literature

Spelling correction in a large collection of clinical notes is a relatively lesser explored research area and serious problem as typically 30% of tokens are unknown. In this paper we explore techniques to devise a spelling suggestion generator for clinical data with an emphasis on the accuracy of the first suggestion. We use a combination of a rule-based suggestion generation system and a context-sensitive ranking algorithm to obtain the best results. The ranking is achieved using a language model created by bootstrapping from the corpus. The described method is embedded in a practical work-flow that we use for Knowledge Discovery and Reuse in clinical records.