Developing Specifications for Light Annotation Tasks in the Biomedical Domain

Biomedical texts pose an interesting challenge in natural language processing tasks. While the information contained in them is important to people of all backgrounds, often they are stylistically complex with specialized vocabularies, and require advanced degrees or other special training to interpret correctly. Because of this, researchers in Natural Language Processing are often at a disadvantage when it comes to extracting task-specific information from these texts: the experts who are best able to understand them may not have the time or interest in completing complicated and time-consuming annotations for use in corpus analysis and machine learning. This paper proposes a methodology for creating light annotation tasks for biomedical corpora that can be used to create useful annotations without requiring extensive training or exceptionally long annotation periods. The utility of the proposed methodology is examined in light of existing annotation projects, as well as through the lens of a case study using hospital discharge summaries for patient selection based on eligibility criteria