Re: issue-60 (Re: Comment on ITS 2.0 specification WD)

For QT Launchpad this is rather distinct. We would need a way to say, in essence, “this segment is so screwed up that we can't tell what in the hell has gone wrong with it.” This is rather distinct from saying that we haven't categorized the issue according to our scheme (uncategorized) or that it is some sort of issue that isn't addressed. This is a statement that the error is fundamentally uncategorizable because something has broken down so badly that no diagnosis is possible.

A concrete example would be (from a real MT system) when I got a translation of "1862" in English (with 100% confidence indicated!) as "deutsches-ungarisches". Now in this case I happened to know that there was an underlying corpus problem responsible for the result, but a typical user would only be able to say that this was so messed up that they couldn't say what the actual nature of the error was. We don't have an error category for "number inappropriate replaced by totally unrelated text" (which could be "other" if a system had it), nor is this uncategorized (which would imply that there is a relevant error categorization but we jut haven't mapped it to the right thing yet). Maybe illegible is the wrong term, but here, without knowing what is wrong in the system, one cannot diagnose the problem other than to say "the system messed up too badly to categorize this."

In QT Launchpad we suspect that a lot of really bad MT results will be like this: to try to pin down what exactly is bad is more work than just tagging it and saying we can't figure it out.

-Arle

On 2012 Dec 11, at 17:04 , Pedro L. Díez Orzas <pedro.diez@linguaserve.com> wrote:

> Felix,
> 
> The question here is to agree whether values like "uncategorized"  or
> "other" cover the "unintelligible" one. 
> 
> Here, we can discuss the point of the note: "This value is to be used as a
> fall-back only in cases where a more specific issue type (including /other/)
> cannot be identified." 
> 
> According to me:
> 
> - "Uncategorized" is a concrete quality issue, but not listed and neither
> unintelligible.
> - "Other" means all the rest, including unintelligible, but we could make
> the distinction and adding this if it is important for QT Launchpad project.
> 
> Just my two cents.
> Pedro