Missatge de Francis Tyers <fty...@prompsit.com> del dia ds., 23 de març
2019 a les 22:16:

> El 2019-03-23 12:28, Hèctor Alòs i Font escribió:
> > Missatge de Francis Tyers <fty...@prompsit.com> del dia ds., 23 de
> > març 2019 a les 9:58:
> >
> >> El 2019-03-23 06:40, Hèctor Alòs i Font escribió:
> >>> Hi all,
> >>>
> >>> I've put a preliminary proposal for GSoC 2019 at
> >>>
> >>
> >
> http://wiki.apertium.org/wiki/Hectoralos/GSOC_2019_proposal:_Catalan-Italian_and_Catalan-Portuguese
> >>> The goal is the improve the Catalan-Italian and Catalan-Portuguese
> >>> language pairs, including the development of a translator from
> >> Catalan
> >>> into Italian.
> >>>
> >>> I will be happy to receive your comments by email or on the IRC
> >>> channel.
> >>>
> >>
> >> Dear Hèctor!
> >>
> >> I think the proposal looks great, and you certainly have my support,
> >> but
> >> couldn't we persuade you to add in a prototype Corsican-->Catalan
> >> too?
> >> ))
> >>
> >> Fran
> >
> > Hi Fran,
> >
> > Thanks for you feedback!
> >
> > I don't like very much "prototypes". They tend to stay in a corner,
> > especially when we do not have dozens of developers. The two pairs
> > that I propose, for example, have been published for 10 years and are
> > clearly improvable, but nobody has done much for them in the last ten
> > years.
>
> This is true, but in addition they are already both fairly well served
> by existing translation systems.
>
> Oh wow, or maybe not, I just picked some random Wikipedia text for
> por->cat:
>
> Os machos medem geralmente entre 10 e 16,6 centímetros, enquanto as
> fêmeas medem entre 13,5 e 19 centímetros. É possível distinguir
> facilmente os machos das fêmeas, devido ao dimorfismo sexual, com os
> machos sendo verde-amarelados e as fêmeas sendo bicolores, apresentando
> o dorso bege com manchas escuras. Ambos possuem o ventre branco
> marmoreado com marrom, além de diversas saliências dorsais. Suas
> glândulas parotoides são bem desenvolvidas, podendo ser encontradas
> atrás do olho e possuindo formato elíptico.
>
> GOOGLE: Els mascles es troben generalment entre els 10 i els 16,6
> centímetres, en forma de fêmeas medem entre 13,5 i 19 cm. Possibilitat
> de distingir fàcilment els sexes sexuals sexuals sexuals sexuals sexuals
> sexuals sexuals sexuals sexuals sexuals sexuals sexuals sexuals sexuals
> sexuals sexuals sexuals com com com a com a com a com a com a com a com
> a f f f f enviar Ambos possibles i de branques de marbre amb marrom, de
> diverses maneres d’origen. Suas glândules parotoides são bem
> desenvolvidas, podent ser trobades darrere olho i possuindo formato
> elíptico.
>
> YANDEX: Els homes generalment a mesurar entre 10 i 16.6 centímetres,
> mentre que les femelles mesuren entre el 13,5 i 19 centímetres. Els
> mascles es pugui distingir fàcilment de les dones a causa sexual
> dimorphism, amb mascles de ser de color groguenc-verd i femelles, sent
> bicolor, amb beix dorsus amb punts foscos. Tots dos tenen color marró
> amb negre ventre, i diversos dorsal emergències. La seva parotoid
> glàndules estan ben desenvolupats, i es troba darrere de l'ull i tenen
> una forma el·líptica.
>
> Haha, classic NMT!!
>
> Perhaps a good thing to include in the proposal might be to give an idea
> of how good (WER) existing systems are for those pairs.
>

Yes, you are right, Fran. I have done it and I have put the results
(quantitative and qualitative) in my proposal:
http://wiki.apertium.org/wiki/Hectoralos/GSOC_2019_proposal:_Catalan-Italian_and_Catalan-Portuguese#Google_translation

In short, ita-cat has a WER of 14% and por-cat 22% (I don't know why there
is such a big difference). It is interesting to notice that I have found
the post-edition of the Google translation quite more time-expensive that
the Apertium one, although the quality is clearly better. I'm not sure that
this is because I'm much more used to post-edit Apertium translations.

Hèctor
_______________________________________________
Apertium-stuff mailing list
Apertium-stuff@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/apertium-stuff

Reply via email to