- Free software: BSD license
- Documentation: https://vulyk.readthedocs.org.
We have results produced by different NLP software that should be checked and verified manually. So, we decided to generalize those tasks a bit and build a platform with basic support for crowdsourcing using knowledge and chunks of code from the unshred.it project. We don't use anything extraordinary: Flask, MongoDB, Bootstrap, jQuery etc (all trademarks are the property of their respective owners, don't forget that).
- improve the current state of Ukrainian NLP by creating a simple and robust solution for various NLP tasks that require human input.
- process and manually verify existing results from PullEnti (by Konstantin Kuznetsov) and morphological analyzer (by Andriy Rysin, Mariana Romanyshyn, et al.).
- build tagged corpora of different kinds using manually processed results.
- for the sake of The Great Justice of course!
We provide some kind of playground where any interested (or procrastination-addicted) person could spent some time checking correctness of NLP software processing results. Site will show you, mr. Solver, random text, which was processed by one of our NER solutions, with layered marks, that should highlight entities, extracted by our software, and their types. If you see that something is wrong with recognition you're able to:
- Delete false positive entities (i.e those where NER tagging software thinks it’s an entity but it’s not).
- Change boundary of the entity (i.e entity found only covers a part of a real one or entity found includes real one + neighbour words which aren’t a part of real one)
- Create new entities in a given text (i.e entity hasn’t been found by a NER tagging software).
- Provide free-form feedback to the author of the software.
In the best case we'll add some gamification (badges, achievement... you name it).
You just need to contact me mr_gambal@outlook.com or @dchaplinsky via chaplinsky.dmitry@gmail.com. One day we'll find one brave heart who'll create a list of issues so the process will besimplified. But not now...
To start developing you'll need a bootstrap data to work with. So... Here is it - 7zipped json files (circa 15 MB).
It must be loaded into your MongoDB using:
./cli.py db load <path_to_extracted_files>/*.json