abbreviation difference from scispacy
dakinggg opened this issue · comments
Hi! scispacy developer here. Could you share what changes you made to our abbreviation detector? I am curious what issues you encountered/fixed (obviously not bothered at all that you based yours off of ours).
Hi! Thanks for asking, I really love your work with scispacy, it's impressive.
Actually, I didn't fix any issue on your abbreviation detector, at least not on purpose.
Erre Quadro works on patents, which are very technical and heterogeneous documents, and I needed to have an our own abbreviation detector to be able to fix special cases as soon as encountered.
As main differences, I added:
- squared brackets in matching short/long forms, together with a no brackets pattern
- detection more strict in some cases, forcing to match the first letter of a word
- forced to skip one word max in checking short and long forms
There may be others, but would be minor.
On next updates, I would like to better handle:
- long/short forms punctuation separators (for example allowing dashes and not dots)
- long/short forms overlapping (one cancel other in global matching)
Of course I welcome any contribution or feedback, and thanks again for being inspiring.
Thanks for the info and glad you like scispacy! If you do make changes that you think would be general improvements, we'd love to have a look at a PR and incorporate them! (I'll close this since it isn't really an issue persay)