Re: [Apertium-stuff] Lexical selection also in source language

2016-02-21 Thread Mikel L. Forcada 
Joonas,
well, in my message I was advocating more of a division of work (CG for
morphosyntactical disambiguation with lexical selection dealing with
problems related to lemmas). This does not mean that you cannot do
everything with CG — in fact, I see no principled reason why you can't.

A couple of things should however be taken into account:

(1) our CG processor is currently a bit slow, while Fran's lexical
selection processor is lots faster as rules are compiled as finite-state
processors

(2) when writing a morphosyntactical disambiguation rule in CG gets tough,
one can let a statistical part-of-speech tagger deal with the remaining
ambiguity (I recently added a Google Summer of Code idea to do this).

Hope this helps

Mikel


2016-02-21 17:42 GMT+01:00 Joonas Kylmälä :

> Thanks for the infromation, Tino. So as we can achieve 100%
> disambiguation with CG then there's no need for adding extra lexical
> selection module after the CG! :)
>
> On Sun, Feb 21, 2016 at 6:18 PM, Tino Didriksen 
> wrote:
> > On 21 February 2016 at 17:07, Joonas Kylmälä 
> wrote:
> >>
> >> I read from  that
> >> CG can leave 3-7% of all words ambiguous (not sure how reliable that
> >> information is..) and at the moment the language pairs that use
> >> vislcg3 don't have anything after vislcg3 in the pipeline that would
> >> resolve those ambiguities, and so the first analysis is selected
> >> whether or not it is the right one.
> >
> >
> > It is not a limitation of CG. You can achieve 100% disambiguation if you
> add
> > or improve the CG rules.
> >
> > -- Tino Didriksen
> >
> >
> --
> > Site24x7 APM Insight: Get Deep Visibility into Application Performance
> > APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
> > Monitor end-to-end web transactions and take corrective actions now
> > Troubleshoot faster and improve end-user experience. Signup Now!
> > http://pubads.g.doubleclick.net/gampad/clk?id=272487151=/4140
> > ___
> > Apertium-stuff mailing list
> > Apertium-stuff@lists.sourceforge.net
> > https://lists.sourceforge.net/lists/listinfo/apertium-stuff
> >
>
>
> --
> Site24x7 APM Insight: Get Deep Visibility into Application Performance
> APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
> Monitor end-to-end web transactions and take corrective actions now
> Troubleshoot faster and improve end-user experience. Signup Now!
> http://pubads.g.doubleclick.net/gampad/clk?id=272487151=/4140
> ___
> Apertium-stuff mailing list
> Apertium-stuff@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/apertium-stuff
>
>


-- 
Mikel L. ForcadaE-mail: m...@dlsi.ua.es
Departament de Llenguatges  Phone: +34-96-590-9776
i Sistemes Informàticsalso +34-96-590-3772.
UNIVERSITAT D'ALACANT   Fax:   +34-96-590-9326, -3464
E-03071 ALACANT, Spain.

URL: http://www.dlsi.ua.es/~mlf
--
Site24x7 APM Insight: Get Deep Visibility into Application Performance
APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
Monitor end-to-end web transactions and take corrective actions now
Troubleshoot faster and improve end-user experience. Signup Now!
http://pubads.g.doubleclick.net/gampad/clk?id=272487151=/4140___
Apertium-stuff mailing list
Apertium-stuff@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/apertium-stuff

Re: [Apertium-stuff] Lexical selection also in source language

2016-02-21 Thread Joonas Kylmälä 
Thanks for the infromation, Tino. So as we can achieve 100%
disambiguation with CG then there's no need for adding extra lexical
selection module after the CG! :)

On Sun, Feb 21, 2016 at 6:18 PM, Tino Didriksen  wrote:
> On 21 February 2016 at 17:07, Joonas Kylmälä  wrote:
>>
>> I read from  that
>> CG can leave 3-7% of all words ambiguous (not sure how reliable that
>> information is..) and at the moment the language pairs that use
>> vislcg3 don't have anything after vislcg3 in the pipeline that would
>> resolve those ambiguities, and so the first analysis is selected
>> whether or not it is the right one.
>
>
> It is not a limitation of CG. You can achieve 100% disambiguation if you add
> or improve the CG rules.
>
> -- Tino Didriksen
>
> --
> Site24x7 APM Insight: Get Deep Visibility into Application Performance
> APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
> Monitor end-to-end web transactions and take corrective actions now
> Troubleshoot faster and improve end-user experience. Signup Now!
> http://pubads.g.doubleclick.net/gampad/clk?id=272487151=/4140
> ___
> Apertium-stuff mailing list
> Apertium-stuff@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/apertium-stuff
>

--
Site24x7 APM Insight: Get Deep Visibility into Application Performance
APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
Monitor end-to-end web transactions and take corrective actions now
Troubleshoot faster and improve end-user experience. Signup Now!
http://pubads.g.doubleclick.net/gampad/clk?id=272487151=/4140
___
Apertium-stuff mailing list
Apertium-stuff@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/apertium-stuff

Re: [Apertium-stuff] Lexical selection also in source language

2016-02-21 Thread Mikel L. Forcada 
Hi Joonas:
I always thought that the role of Constraint Grammar (CG) in Apertium was
more of a morphosyntactic desambiguation. Therefore, even if the CG
processor completely solved the morphosyntactic ambiguity of each and every
source-language surface form, there could still well be the chance for a
given *lemma* to have more than one target-language equivalent. For
instance, the Spanish surface form "registro" can have two lexical forms:
it may be a verb (registrar.vblex.pri.1.sg) or a noun (registro.n.m.sg). A
CG rule could use the context to discard the noun lexical form, but when
translating into English, any verb lexical form with "registrar" as lemma
could mean "search" as when the police has a "search warrant" and enters a
house to look for some evidence, or "register", as in annotating something
in a register.  This second problem is clearly a candidate for a lexical
selection module.

Cheers
Mikel

2016-02-21 17:07 GMT+01:00 Joonas Kylmälä :

> Hey everyone,
>
> I read from  that
> CG can leave 3-7% of all words ambiguous (not sure how reliable that
> information is..) and at the moment the language pairs that use
> vislcg3 don't have anything after vislcg3 in the pipeline that would
> resolve those ambiguities, and so the first analysis is selected
> whether or not it is the right one. Could we use the
> apertium-lex-tools (lrx-proc) also after cg-proc to get better
> translations? I think we might need to do some changes to
> apertium-lex-tools (or possibly not) in order to get it working.
>
> I also understand that it slows down the translation process a bit but
> it would benefit those people who want more accurate translations, and
> it would be easy to make it optional for those that don't want it
> because of the pipeline architecture we use.
>
> What do you think about this?
>
> -Joonas
>
>
> --
> Site24x7 APM Insight: Get Deep Visibility into Application Performance
> APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
> Monitor end-to-end web transactions and take corrective actions now
> Troubleshoot faster and improve end-user experience. Signup Now!
> http://pubads.g.doubleclick.net/gampad/clk?id=272487151=/4140
> ___
> Apertium-stuff mailing list
> Apertium-stuff@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/apertium-stuff
>
>


-- 
Mikel L. ForcadaE-mail: m...@dlsi.ua.es
Departament de Llenguatges  Phone: +34-96-590-9776
i Sistemes Informàticsalso +34-96-590-3772.
UNIVERSITAT D'ALACANT   Fax:   +34-96-590-9326, -3464
E-03071 ALACANT, Spain.

URL: http://www.dlsi.ua.es/~mlf
--
Site24x7 APM Insight: Get Deep Visibility into Application Performance
APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
Monitor end-to-end web transactions and take corrective actions now
Troubleshoot faster and improve end-user experience. Signup Now!
http://pubads.g.doubleclick.net/gampad/clk?id=272487151=/4140___
Apertium-stuff mailing list
Apertium-stuff@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/apertium-stuff