Finding Difficult-to-Disambiguate Words: Towards an Efficient Workflow to Implement Word Sense Disambiguation

In the biomedical and clinical domain, valuable information is frequently represented in free-text documents. Natural language processing (NLP) is a powerful tool that can extract structured information from theses documents. Word sense disambiguation (WSD) is a critical component in an NLP pipeline that increases the accuracy of the extracted information. However, WSD is expensive to apply for all known ambiguous words. Given limited time and resources, one practical strategy is to prioritize easy-to-disambiguate words and efficiently maximize the coverage of disambiguation. To aid prioritization efforts, we studied two quantitative indicators that are associated with how easy/difficult it is to disambiguate any given word.