https://bugs.documentfoundation.org/show_bug.cgi?id=155087
--- Comment #34 from Tex2002ans <tex2002ans+libreoff...@gmail.com> --- @Gabriel I agree completely with your: - AutoCorrect analysis - + 3 principles for "DEFAULT replacements". :) @Ákos Thanks for the input again after all these years. :) @cipricus Definitely follow Mike's advice in Comment #31. > The autocorrection tool for any language must be > prepared to require the least possible effort from > user: the replacements that the tool makes must > be correct on 100% cases (well, 99.998% would probably be OK). No! Nothing below 99.999% should be allowed!!! Okay, okay, we can compromise—I'll take 99.998%. :P > Consider English replacement i->I. There *is* a non-zero > probability, that the author actually wanted to have the > "i" in their text. Yes, but then it will hopefully be caught at the other layers too! (Like grammarchecking!) - "i went to the park." vs. - "The variable i says..." Grammarcheck will see the word "variable" before 'i', and know that lowercase 'i' was probably intended! No green squiggly! You can't 100% rely on any 1 of the layers! You need to use all 3 together! - - - Side Note: For more info on that, see Daniel Naber's fantastic talk: FOSSDEM 2014: "How we found a million style and grammar errors in the English Wikipedia" - https://www.youtube.com/watch?v=2xmPwefktXI (He's the original creator of LanguageTool!) - - - Funny Side Note: With LanguageTool, I was pulling my hair out over: - AI Of course, everyone will be speaking about Artificial Intelligence... but there's actually a *very rare* English word: - ai which is a type of "three-toed sloth" in South America. While I was saying 99.99% of people want: - AI + AIs + AI's the LanguageTool developer wanted to also add: - ai + ais + ai's because "it's valid English"... and "people MIGHT be taking about the sloths"!!! - - - The 3 different layers can have different tolerances for what constitutes "an error". In my mind, it's like an inverse pyramid: - AutoCorrect should be very narrow/strict. - Spellchecking can be medium. - Grammarchecking could be wide/lax, allowing all sorts of valid words + parts of speech. If grammarchecking gives you a bad green squiggly or a not 100% correct suggestion, that's tolerable. But if AutoCorrect is constantly "correcting" you with wrong—and automatic—"fixes"... that gets frustrating as a user REAL fast. (So, like Gabriel/MikeKaganski said, AutoCorrect should lean heavily towards the 100% correct side by default.) - - - Anyway, I'll be watching this bug from the sidelines now. Looks like the Romanian AutoCorrect will be fixed after all! :) (Hopefully this can inspire others to look at updating AutoCorrect in other lesser-used languages too! Just like sophie said in Comment #28.) Thanks for the great comments, everyone. :) -- You are receiving this mail because: You are the assignee for the bug.
