Enhancement: Let text editable for word segmentation or fix typo about doccano HOT 9 CLOSED

+1

I'm doing sequence to sequence labeling right now, but sometimes there are errors in grammar or some kind of typos from the data. It would be really great if we are able to fix errors directly from doccano, but instead right now I have to export data from doccano into my computer, edit, then import it back into doccano.

from doccano.

Yes. We already have an API for editing document so we can implement this feature by implementing the frontend.

But one problem is that if we edit a document with annotation in a sequence labeling project, the annotated data will become useless. An example is as follows:

Original Text: Plsident Obama
Annotation: {'start_offset': 9, 'end_offset': 14}

Edited Text: Plesident Obama
Annotation: {'start_offset': 9, 'end_offset': 14}  # incorrect
True annotation: {'start_offset': 10, 'end_offset': 15} 

One solution is that when we edit the document, we delete all their annotation. This is my thought.

from doccano.

yes, it is A very clear and straight demands

from doccano.

Lack of information.

from doccano.

Is this feature on planing?

from doccano.

well, other possible options might be:

1. only delete labels that have start_offset after edited character position
2. compare what are actual values for old_doc[start_offset:end_offset] == new_doc[start_offset:end_offset] and if they are equal, keep them (covers 1) as well as cases when edits are swapping chars, without adding/removing them outside of labeled items

But this might still not work for everyone

from doccano.

I think it is not a big issue.
Alternative solution is, we save both original text and modified text

from doccano.

Implementing data editing feature on doccano is a little overwhelming. It would be good to identify the territory of the tool.

from doccano.

I support this suggestion, I am seeing incorrectly spelled text, wrong punctuations, etc, etc. Would love to have the capability to fix typos, grammatical errors, etc, as I see them.

from doccano.