Annotation manuelle d'occurrences de candidats termes et écrit scientifique

This paper compares two successive annotation campaigns aimed at manually identifying the occurrences of candidate terms that actually fall within the scientific domain of the annotated document. The two campaigns are distinguished by their objectives. The first aimed the enrichment of existing terminological resources. The second had the dual objective of comparing several annotation tools (BRAT, GATE, GLOZZ) and measuring the difficulty of the annotation task in the human and social sciences compared to the so-called hard sciences. A direct comparison between both campaigns is not possible on the basis of the produced corpora. To do this, we use these corpora like learning corpus in the context of a test task. The role of this task is to automatise the manual annotation. The goal is to determine if the second corpus is of better quality than the first one with regards to the test task performances.