2014-02-04 Francis Tyers <[email protected]>:
> Hello all!
>
> Out of the 39 or so language pairs that we have in trunk/, only two or
> three could be considered to offer "state of the art" performance with
> respect to vocabulary coverage and translation quality -- where there is
> a competitor.
>
> Spanish-Catalan
> Norwegian Bokmål-Norwegian Nynorsk
> (Spanish-Portuguese)
>
> The objective of this task would be to take another language pair, one
> that is already quite developed, and make it "Google-beating", and if
> not Google-beating then at least improve coverage by at least 10-15% on
> a range of corpora, and a WER reduction of at least 10-15%.
>
> This would be a lot of work, but I don't think that given a "well
> resourced" language (e.g. English, Spanish, French, Catalan, Italian,
> Portuguese) that it would be much more than the language pair students
> do each year.
>
> In general we try to have no more than 50% language pairs as GSOC
> projects, which I think is a pretty good idea. If we think that this
> task is a good idea then we could decide to make at least one language
> pair a "state of the art" one.
>
In my opinion, improving qualitity should be one of our main focus.
Creating resources for new language pairs is awesome, but the result (as of
today) is, as you said, that only 2-3 out of 40 pairs in trunk work really
well. For pairs where there's no competitor, this is better than nothing,
but for other pairs (I'm thinking of en-es, en-ca, fr-es, fr-ca, pt-ca,
es-it,...) other translators are better options (basically because of the
highest coverage). And as a result of that, people don't use Apertium that
much.
I really believe in Apertium as a "product" for the end-user, because I
really believe it can perform at least as good as proprietary translators
when the coverage is high enough.
So, as Mikel said... +1.
--
< Xavi Ivars >
< http://xavi.ivars.me >
------------------------------------------------------------------------------
Managing the Performance of Cloud-Based Applications
Take advantage of what the Cloud has to offer - Avoid Common Pitfalls.
Read the Whitepaper.
http://pubads.g.doubleclick.net/gampad/clk?id=121051231&iu=/4140/ostg.clktrk
_______________________________________________
Apertium-stuff mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/apertium-stuff
