A corpus of tables in full-text biomedical research publications

The corpus was created to serve as a gold standard for two tasks of information extraction from biomedical tables: (1) mapping of table cells into fine-grained entity types, and (2) identification of relations between table cells. The corpus consists of a small set of full-text articles sourced from the Open Access subset of PubMed Central, for which four types of table annotations were created: (1) cell group, which splits each table into sets of homogeneous cell groups; (2) cell type, which represents the mapping of all cells in a homogeneous group into a single fine-grained named entity label; (3) concept, which represents the mapping of utterances inside table cells (i.e., the syntactic heads of the utterances expanded with their modifiers) into their semantic equivalents from a domain vocabulary; and (4) relation, which represents relations between cell groups.