The representation scheme of the BB task contains four entity types:
and two relation types:
In addition, Microorganisms are normalized to taxa from the NCBI taxonomy, and Habitat and Phenotype entities normalized to concepts from the OntoBiotope ontology.
Annotation example
The task corpus includes two types of documents:
The annotated corpus will be provided in the BioNLP-ST standoff annotation format.
Annotation guidelines can be found here.
The BB task is composed of three subtasks. Each subtask has two modalities: one where entities are given as input, and one where entities are not be provided. Teams are free to participate in the subtask(s) of their choice.
The evaluation will focus on the accuracy of the predicted categories compared to gold reference. A concept distance measure has been designed in order to sanction over-generalization or over-specialization with a fair penalty. Note that if an entity has several categories, then it is a conjunction: all categories must be predicted.
For norm+ner, boundary accuracy will be factored in the evaluation since the inclusion or exclusion of modifiers can change the meaning and the normalization of phrases.
The evaluation measures will be Recall and Precision of predicted events against gold events.
For rel+ner, boundary accuracy will be factored in the evaluation.
Participant systems are evaluated for their capacity to build a knowledge base from the corpus. The knowledge base is the set of Lives_in and Exhibits relations with the concepts of their Microorganism, Habitat and Phenotype arguments. The goal of the task is to measure how much of the information content of the corpus can be extracted automatically. It can be viewed as a combination of the first two subtasks, with results aggregated at the corpus level (i.e., not all occurrences need to be predicted).
The evaluation measures will be Recall and Precision of predicted events against gold events.
For kb+ner, boundary accuracy will be factored in the evaluation.