This directory contains question and answer pairs for our testing and training data sets. Each file follows the following standard:

• The name of the file is .txt where appears in name_indexer.json (e.g. jon_snow.txt)
• the first line is a question
• The second line is an answer
• The third line is the position of the answer in the context paragraph given by characters.json (e.g. characters[65][paragraphs][0] is the first context paragraph for Jon Snow)
• All nonempty lines up until the first empty line reference the first context paragraph. After a new line, all nonempty lines up until the second empty line reference the second context, and so on.
• Last line must be an empty line

The format_data.py script will take the data in qa_data and format the data so that the BERT script from huggingface will accept it. This format is how SQuAD formats its data.