| Resumo: |
We are creating a gold-standard corpus of manually annotated full-text biomedical journal articles toward natural-language-processing applications. Central to this is our use of entire ontologies of the Open Biomedical Ontologies initiative as well as other terminologies as term sources, in contrast to most other such annotation projects, which have used small, ad hoc schemas. In addition to the standard difficulties in such annotation projects, each of the terminologies we have used has idiosyncrasies and ambiguities that present further challenges to consistent, high-quality annotation of these articles. In this paper we present and discuss the most salient of these with regard to the Gene Ontology that we have encountered and addressed in our annotation guidelines and training. The utility of these guidelines can be seen in the high and still-increasing interannotator-agreement statistics that we continually monitor.

|
