Is there an API endpoint that we can hit that would essentially take verbs in the sentence and conjugate them correctly/fix articles? Like:
"You draws a estate"->"You draw an estate"
charles-toller
All 3 comments
hi!
yeah, this is something I've thought a lot about. Something that does this would be immediately useful - and also, a rule-based approach, like nlp_compromise does, to common english errors would be perfectly doable.
I had a beer a few weeks ago with an esl teacher that rhymed-off the most-common errors for each native-language speaker. It was pretty funny actually, (and a little cheeky) because every portuguese speaker makes the same 3 errors, and every japanese speaker makes the same 3 errors, etc. It seems like writing templates for these mistakes would be worthwhile to try. It's not my expertise.
nlp_compromise right now mostly assumes perfect grammar, and assumes any incongruity is the fault of its interpretation. Honestly, understanding vs fixing a sentence may be two incompatible projects. I'm not sure the best way to do them together. It's a very juicy idea though.
spencermountain
on 15 Aug 2016
Perhaps a series of correctional functions, then. Like, a conjugation function that fixes sentences (assuming nouns are correct), and its conjugate (pun intended), fixNouns that fixes the plurality of nouns and such assuming verbs are correct. An easy one to start with would be something that fixes articles, like a->an in certain cases.
By the way, I'm totally just bouncing around ideas to see what interest there would be in this.
charles-toller
on 15 Aug 2016
I feel that more attention should be brought to this idea and the comment by @spencermountain made earlier.
I also would appreciate having some sort of correction engine built into nlp_compromise as it is true that the module currently assumes perfect mechanics and grammar are being used in text processed by it. Of course, this isn't a simple feature to develop and may require many more "correction rules" and systems to be built.
A simple feature to build to start would be a typo correction engine using Levenshtein distance.
ethanent
on 6 Nov 2016
Related issues
Xetanai
路
4Comments
nelson-liu
路
3Comments
shamoons
路
4Comments
fhoffa
路
4Comments
struesdell
路
3Comments
Most helpful comment
I feel that more attention should be brought to this idea and the comment by @spencermountain made earlier.
I also would appreciate having some sort of correction engine built into nlp_compromise as it is true that the module currently assumes perfect mechanics and grammar are being used in text processed by it. Of course, this isn't a simple feature to develop and may require many more "correction rules" and systems to be built.
A simple feature to build to start would be a typo correction engine using Levenshtein distance.