asknowqa / vanilla Goto Github PK

Hi,

I went through your dataset, and there are several discrepancies between the question entity and the answer entity.
For example, the 4th and 24th example in training:

{"answer":"male","answer_sentence":"sterjo is a male.","question":"Which sex does Joseph Louis Watkins, Jr. belong to ?","question_entity_label":"Joseph Louis Watkins, Jr.","question_id":41188,"question_relation":"P21"}
{"answer":"male","answer_sentence":"peter muller is a male.","question":"What is the sex of William Bailey ?","question_entity_label":"William Bailey","question_id":12117,"question_relation":"P21"}

First statistics with 10 recurrent repeated names show that it stands for 5% of the training and testing data.
Can you correct those errors ? It directly impacts the BLEU and METEOR scores of the NLG model you trained since those errors also appear in the gold of the test file.

Hello,

Is it possible to provide answer URIs or SPARQL queries to retrieve them for the questions in the dataset?