Back to LanguageTool Homepage - Privacy - Imprint

Confidence for suggested replacements?


(Guest) #1

When I call getSuggestedReplacements, is there any way to see quantitatively how much better of a replacement the first one is compared to the second, or how good any of them are? I know that this is not a probabilistic system, but is there any way to find the rough "probability" of a match?

Thanks!


(Guest) #2

Bumping this since it's fairly important for what I'm trying to do. I've looked through the API but nothing stood out for me immediately; is there a way to intelligently pick which replacement to choose for a word (if only one can be picked), as well as automatically decide if any replacement should be used?

I apologize for the somewhat general question; any advice is appreciated.


(Daniel Naber) #3

If LanguageTool knew which suggestion was the correct one, it would only
suggest that one... depending on your text quality and the language,
LanguageTool's errors can contain quite some false alarms, so I don't
think it makes sense to apply suggestions automatically.


(Guest) #4

Well, a lot of spell checkers are probabilistic (see http://norvig.com/spell-correct.html), and in those cases there are often some suggestions which are much higher probability than others, and if there is a large difference it's probably reasonable to automatically choose the one with highest probability. In the cases where the probabilities are similar (or very low, or both), it might make sense to ignore the suggestions.

Since your software is rule-based, I realized this isn't very relevant, but perhaps you should consider some kind of "probability" calculation in future versions (even just based on word counts in a corpus).