1{2 "document_id": "lb_qasper_bee3efb92996570d990ab04bfdf054835ef720b9dd1c6f5a",3 "family": "public_qasper",4 "split": "dev",5 "document_hash": "b41a0ed6ba75547c8fc1949f5cc9a23387b3c67230d0c0754a0b00315c0d9741",6 "questions": [7 {8 "question_id": "q00",9 "qtype": "qasper",10 "question": "How is the quality of the data empirically evaluated? ",11 "answer": "Validated transcripts were sent to professional translators., various sanity checks to the translations, sanity check the overlaps of train, development and test sets",12 "aliases": [13 "computed sentence-level BLEU, We manually inspected examples where the source transcript was identical to the translation, measured the perplexity of the translations, computed the ratio of English characters in the translations, calculate similarity scores between transcripts and translations"14 ],15 "answer_type": "qa_f1",16 "tolerance": 0.0,17 "supporting_fact_ids": [],18 "answer_position_frac": -1.019 }20 ],21 "facts": []22}