El dj 27 de 02 de 2014 a les 22:11 +0800, en/na wei2912 Support va
escriure:
> I believe sushain and firespeaker had discussed about this previously,
> but I'll post it here to get more feedback.
>
> The idea is to chain language pairs together to provide support for a
> greater variety of languages. The shortest path is selected and out of
> the options, the highest quality path is selected. If the quality of
> that path does not meet a threshold, it is not included. Otherwise, the
> path is included as a language pair in apertium-apy.
>
> Was this idea already previously considered or has no one implemented
> it? If it is accepted, I will work on this during my 1 week holidays.
Hey there!
Something similar was implemented by Enrique Benimeli a while ago.
Although he did it in PHP on the main Apertium website. I think it would
be a nice thing to add to APY, and checking the quality is an excellent
idea.
How to check the quality:
* Find parallel corpora for the languages. Around 1,000 sentences.
* If we want to make e.g. fr->en vi fr->es and es->en, what we do is
calculate the WER/BLEU for the es->en pair, then calculate the WER
for the fr->es->en pair. If the automatic score for the indirect
pair is above 80% of the score for the direct pair then we include it.
* To start with, I would recommend performing tests on:
fr->es->en , pt->es->en , en->es->pt , it->es->en.
* For these, you can use texts from Europarl to do the evaluation.
Also, this should be an option to pass to APY in the listPairs, you
should set it to False by default.
See you on IRC :)
Fran
------------------------------------------------------------------------------
Flow-based real-time traffic analytics software. Cisco certified tool.
Monitor traffic, SLAs, QoS, Medianet, WAAS etc. with NetFlow Analyzer
Customize your own dashboards, set traffic alerts and generate reports.
Network behavioral analysis & security monitoring. All-in-one tool.
http://pubads.g.doubleclick.net/gampad/clk?id=126839071&iu=/4140/ostg.clktrk
_______________________________________________
Apertium-stuff mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/apertium-stuff
