Pull request to integrate the friends qa data with the original json files. The format is like this:
At each scene level, a new entry called span_qa is created and it contains the information about the questions and answers for each scenes. Empty span_qa means the scene is not used for turking. Note that if the answer is the speaker, instead of the start and end range it will show up as the actual name.
span_qa: [ ... { "question": "What action is Joey doing ?" "answers" : [ [utterance_id, begin_token_index (inclusive), end_token_index (exclusive)], [ ... ], ] }, { "question": "Who is Joey dating?" "answers" : [ [utterance_id, Katey], // if the answer is the speaker's name, instead of answer range, it is the name [ ... ], ] }, ... ]
Pull request to integrate the friends qa data with the original json files. The format is like this: At each scene level, a new entry called span_qa is created and it contains the information about the questions and answers for each scenes. Empty span_qa means the scene is not used for turking. Note that if the answer is the speaker, instead of the start and end range it will show up as the actual name.
span_qa: [ ... { "question": "What action is Joey doing ?" "answers" : [ [utterance_id, begin_token_index (inclusive), end_token_index (exclusive)], [ ... ], ] }, { "question": "Who is Joey dating?" "answers" : [ [utterance_id, Katey], // if the answer is the speaker's name, instead of answer range, it is the name [ ... ], ] }, ... ]