Hi, In that light, Keld's hypothetical alternative lexical selection module looks feasible.
I've just read the GramTrans paper: see point 3.3 Contextual distinctors. According to that paper both the default statistical lexical selection module and Francis improved module are inferior (not to mention my simple domains and style!): They are considered " ... vulnerable to the "sparse data" problem " and are not "robust with respect to inflexion and lexical variation" etc. That was plain speaking! Anyhow, lexical selection interests me a lot, because that's an important task for a human translator. The first (and hardest) step is to understand the source text. The second, is to convey the same meaning without changing to much of the original style. In a naive way, I believe that the poor little computer has the same problem. We need to help it somehow. I consider e.g. statistical translation as an effective short cut, that avoids the main problem. Rule based translation can tackle it! Go ahead Keld! It might not be easy, but it's surely a very interesting path to follow. Yours, Per Tunedal On Wed, Dec 19, 2012, at 14:45, Kevin Brubeck Unhammer wrote: > Per Tunedal <[email protected]> > writes: > > > On Wed, Dec 19, 2012, at 13:10, Francis Tyers wrote: > >> El dc 19 de 12 de 2012 a les 13:02 +0100, en/na Per Tunedal va escriure: > >> > Hi, > >> > I'm a bit confused. See below. > >> > Yours, > >> > Per Tunedal > >> > > >> > On Wed, Dec 19, 2012, at 11:22, Francis Tyers wrote: > >> > > El dc 19 de 12 de 2012 a les 10:09 +0100, en/na Tino Didriksen va > >> > > escriure: > >> > > > On Wed, Dec 19, 2012 at 12:12 AM, > >> > > > <[email protected]> wrote: > >> > > > > GramTrans > >> > > > http://visl.sdu.dk/~eckhard/pdf/MTsummit07_final.pdf > >> > > > >> > --snip-- > >> > > > >> > > > Myself not being familiar with the code of Apertium at all, > >> > > > is > >> > > > this so? > >> > > > And could a module with use of concept reletions be easily > >> > > > included in the stack > >> > > > of translation modules? > >> > > > > >> > > > Someone more familiar with Apertium will have to answer that bit... > >> > > > >> > > I explained how you could do this in a previous email. You would need > >> > > to > >> > > write a module to go between the lexical-transfer output and the > >> > > structural transfer input. > >> > > >> > In that stage the lexical selection is already done, isn't it? > >> > >> In what stage ? In the lexical transfer stage ? no. In the structural > >> transfer stage yes. This is because, like I say above "write a module to > >> _go between_" this means "in the middle of". See my schematic below. > >> > >> > I supposed meaning (relations) was supposed to facilitate lexical > >> > selection. > >> > >> If you can make it facilitate lexical selection good luck!!! > >> > >> > What's the use of it later in the translations chain? > >> > >> ------------------- --------------------- ---------------------- > >> | lexical transf. | ==> | lexical selection | ==> | structural transf. | > >> |_________________| |___________________| |____________________| > >> > >> The lexical selection is done in the aptly-named "lexical selection" > >> stage, which sits between lexical transfer (which outputs all the > >> possible translations of each word) and structural transfer (which > >> performs structural changes, and only works on pairs of single source > >> language/target language words). > >> > >> Fran > > > > Hmm I don't get it. I'm comparing with the wiki page "Apertium for > > dummies" in an adjacent window: > > Tagger - Lexical Selection - Lexical transfer (look up disambiguated > > source-language baseword ...) - Structural transfer. Is transfer done > > before lexical selection in stead? > > > > If so, is the lexical selection done in the target language rather than > > in the source language? Thus, the "meaning" in the source language > > cannot be used at all? > > An example might help. > > Here's some possible output of lexical transfer, snatched from > http://wiki.apertium.org/wiki/Constraint-based_lexical_selection_module > : > > ^El<det><def><f><sg>/The<det><def><f><sg>$ > ^estació<n><f><sg>/season<n><sg>/station<n><sg>$ > ^més<preadv>/more<preadv>$ > ^plujós<adj><f><sg>/rainy<adj><sint><f><sg>$ > … > > This is the input to lexical selection. It contains both source and > target analyses. > > The first, third and fourth words have only one target analysis, the > second word has two possible target analyses (season vs station). The > goal of a lexical selection module is to pick the right target analysis > (or, remove the wrong ones target analyses) for each ambiguous unit. So > if 'season' was picked and 'station' removed, the output of lexical > selection would be: > > ^El<det><def><f><sg>/The<det><def><f><sg>$ > ^estació<n><f><sg>/season<n><sg>$ > ^més<preadv>/more<preadv>$ > ^plujós<adj><f><sg>/rainy<adj><sint><f><sg>$ > … > > This in turn goes into structural transfer. > > Keld's hypothetical alternative lexical selection module have to have > the same input and output as shown above. It should be perfectly > possible to use "meaning" (whatever that means) within that module; as a > part of a Unix pipeline its internals should be independent from the > rest of the pipeline. > > -- > Kevin Brubeck Unhammer > > Sent from my emacs > > > ------------------------------------------------------------------------------ > LogMeIn Rescue: Anywhere, Anytime Remote support for IT. Free Trial > Remotely access PCs and mobile devices and provide instant support > Improve your efficiency, and focus on delivering more value-add services > Discover what IT Professionals Know. Rescue delivers > http://p.sf.net/sfu/logmein_12329d2d > _______________________________________________ > Apertium-stuff mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/apertium-stuff ------------------------------------------------------------------------------ LogMeIn Rescue: Anywhere, Anytime Remote support for IT. Free Trial Remotely access PCs and mobile devices and provide instant support Improve your efficiency, and focus on delivering more value-add services Discover what IT Professionals Know. Rescue delivers http://p.sf.net/sfu/logmein_12329d2d _______________________________________________ Apertium-stuff mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/apertium-stuff
