biu-nlp/qamr

Notes

* The `answer_ranges` feature here has a different meaning from that of the `qanom` and `qa_srl` datasets, although both are structured the same way; while in qasrl/qanom, each "answer range" (i.e. each span, represented as [begin-idx, end-idx]) stands for an independant answer which is read separately (e.g., "John Vincen", "head of marketing"), in this `qamr` dataset each question has a single answer who might be conposed of non-consecutive spans; that is, all given spans should be read successively.

• Another difference is that the meaning of `predicate` in QAMR is different and softer than in QASRL/QANom - here, the predicate is not necessarily within the question, it can also be in the answer; it is generally what the annotator marked as the focus of the QA.