Hi!
I'm developing a pet project where I would need to fine-tune a DPR model. I was wondering if I could use prodigy as an annotation tool to label my data. I would have to build a custom prodigy recipe where it has a question and a span categorizer to categorize answers, positive or negative context. (I think this part is easy since you can create a new recipe). But my question is, it is possible to customize the output json format?
To better understand what I need, this is the output format that I'm looking forward to having:
[
{
"question": "....",
"answers": ["...", "...", "..."],
"positive_ctxs": ["...."],
"negative_ctxs": ["..."],
"hard_negative_ctxs": ["..."]
},
...
]
For the same question, each span would be appended to the corresponding list.
LG,
João