Khresmoi manually annotated reference corpus

The Manually Annotated Reference Corpus is a collection of English web documents annotated with key entities (such as disease, drug), built in the framework of the Khresmoi project, funded by the European Commission. It has been constructed by first annotating these entities with an imperfect automatic process through GATE pipelines. Then, automatic annotations have been corrected by several human operators. The differences between these human operators were resolved to give the final reference corpus. <p><p>The corpus is divided into two parts:<p>1. The initial corpus: 625 documents from the Genetics Home Reference data set, automatically annotated with anatomical locations and diseases, and manually corrected by 3-4 annotators. Size of documents: between 26 and 8,306 tokens each. <p>2. The main corpus: 6,950 English documents from the Khresmoi crawl and 5,518 English Wikipedia pages, automatically annotated through the GATE Platform for Anatomy, Disease, Drug and Investigation. Size of documents: between 200 and 2,000 tokens each. Of these, 1,783 documents have been manually corrected by up to 7 people, to give 68,799 agreed consensus annotations. <p><p>The corpus is using the GATE XML format.