FRACAS: FRench Annotated Corpus of Attribution relations in newS

A human-annotated corpus for French quotation extraction containing 1676 newswire texts with 10 965 annotated attribution relations (quotes attributed to its speaker). Data: 1676 newswire texts in French from Reuters annotated with 10 965 attribution relations Date: April 1995 to April 1996 Data structure: { "text": text of the newswire, "entities": a list of each entity in the following format ["id": unique_id, "text": text of entity, "label": entity label, "gender": gender (if labelled), "char_span": a list of character index span] "relations": a list of each relation in the following format [id of relation, label of relation, id of first entity, id of second entity } Labels: Entities: Quotation (Direct, Indirect or Mixed) Speaker (Agent, Organization, Group of People, Source Pronoun) Cue Attributes: Speaker Gender (Male, Female, Mixed, Unknown, Other) Relations: Speaker Quoted in Quotation Cue Indicates Quotation Source Pronoun Refers to Speaker