I have trouble understanding how exactly the preprocessing script selects the answer spans from the answer text. From what I understand, the function fix_span loops over all matches of the answer text and tries to select the one that better "sticks" to individual tokens.
In the case where the answer perfectly matches some tokens, the first occurrence is returned.
Hello,
I have trouble understanding how exactly the preprocessing script selects the answer spans from the answer text. From what I understand, the function
fix_span
loops over all matches of the answer text and tries to select the one that better "sticks" to individual tokens. In the case where the answer perfectly matches some tokens, the first occurrence is returned.Is that right?
Thanks!