Re: [gutvol-d] Language free version of guiguts?

On Tue, 31 Jan 2006 19:30:55 +0000, Cori <hiddengreen@gmail.com> wrote: |On 1/31/06, Dave Fawthrop <hyphen@hyphenologist.co.uk> wrote: |> |Indeed :) Catching a dozen real errors is a definite win! |> |> 12 in 500plus is a resounding failure. | |I think there might be a misunderstanding about the purpose of the |Gutcheck ... if a text checking tool was provided that never gave me |any false errors, I'd be convinced that it wasn't catching all it |should be. Spellcheckers, or the barrage of regex checks that DP has |developed, all flag up false positives on my books, but they couldn't |be made much more effective without personalising them to each and |every text -- which would take more time than just clicking through |the false alarms..? The point of all these checks is to (hopefully) |be over-sensitive to problems, rather than under-sensitive (thus |leaving errors.) | |Or have I missed something in turn..? Do you have text checking tools |that only ever signal real errors..? Can they be shared..? With my other hat on I write "intelligent" language software, Low 90% correct is very bad, above 99% correct is acceptable. For a voluntary organisation I would be accept 50% correct. -- Dave Fawthrop <dave hyphenologist co uk> "Intelligent Design?" my knees say *not*. "Intelligent Design?" my back says *not*. More like "Incompetent design". Sig (C) Copyright Public Domain

On 1/31/06, Dave Fawthrop <hyphen@hyphenologist.co.uk> wrote:
12 in 500plus is a resounding failure.
Then don't use it. For many users, 12 out of 500 is good enough to make the program useful. No program is ever going to handle dialect well, because dialect doesn't follow the normal rules.
With my other hat on I write "intelligent" language software, Low 90% correct is very bad, above 99% correct is acceptable. For a voluntary organisation I would be accept 50% correct.
Intelligent language software is too broad; you write code to automatically hyphenate words. A concrete problem like that is significantly easier than a problem like "find errors in this text document". 50% of errata sent by humans to errata@pglaf.org is wrong; how do you expect a computer to do better?
participants (2)
-
Dave Fawthrop
-
David Starner