Hi,
On Sat, Feb 9, 2013, at 18:45, Jacob Nordfalk wrote: > 2013/2/5 Per Tunedal <[email protected]> > > > Hmm > > 1. Man blir glad när ens barn ger en blommor. > > > > en is analysed correctly in the first step: > > > > ^en/en<det><ind><ut><sg>/en<prn><pers><p3><ut><sg><nom>/man<prn><pers><p3><ut><sg><acc>$ > > > > but the analysis is abandoned in step 2: > > ^en<prn><pers><p3><ut><sg><nom>$ > > > > Why? > > > > This is a tagger error. > It should select man<prn><pers><p3><ut><sg><acc>, but this is *way* > outside > of what the bigram tagger is able to and will ever be able to, as it only > looks at 2 words when it decides what to select. What to words? Sounds like bad news to me. > > Try to open it in apertium-viewer. There you can manually remove the > incorrect ones and check if the rest of the stages works. A bit tricky to do that. ^Man<prn><pers><p3><ut><sg><nom>$ ^blir/*blir$ ^glad/glad<adj><pst><ut><sg><ind>$ ^när/*när$ ^ens/man<prn><pers><p3><ut><sg><gen>$ ^barn/barn<n><nt><sg><ind><nom>$ ^ger/*ger$ ^man<prn><pers><p3><ut><sg><acc>$ ^blommor$ gives me: Man *blir glad *när ens #barn *ger en \@blommor At least the forms of "man" are now produced correctly. > > > > > > > 2. Ankaret på 3 kg håller för minst 50 kp om man ankrat på en > > sandbotten. > > > > en gets two analysis in step 1: > > > > ^en/en<det><ind><ut><sg>/en<prn><pers><p3><ut><sg><nom>/man<prn><pers><p3><ut><sg><acc>$ > > > > but the wrong one is choosen in step 2: > > > > $ ^en<prn><pers><p3><ut><sg><nom>$ > > > > Why? > > > > Again, a tagger error. > For this it might be worthwhile to train the tagger to select > en<det><ind><ut><sg> when there is an <ut> noun just after. > > You could also do some pre-disambiguation (like CG) on the input before > it > is handlet over to the tagger. I would like to avoid the CG, because I would like to use the pair with OmegaT and Apertium-caffeine. > > Most surely the taggers could need a re-train, but it will probably > introduce other errors as the tragger is a really simple bigram tagger. > We have the same (big) problem with English->Esperanto. > Is it easier to translate between languages of Latin origin? Due to the original design (and goal) of Apertium? > > > > > the pronoun "man" should be easy but I get strange results. Is it the > > fault of the tagger or is something else wrong? > > Even if 'en' and 'man' are very similar pronouns, I think that probaly > swedish 'en' corresponds to danish 'en' and swedish 'man' corresponds > always to Danish 'man'. > If that is the case the best would be to remove entries that translates > 'en' to 'man' or inverse in the bidix. > No, not really! The pronoun "man" is exactly the same in Danish and Swedish, according to my dictionaries. The object form ("accusative") is "en" and the genitive is "ens". In other words: eg. the object form "en" of "man" should always translate to the object form of "man" i.e. "en"! Sounds easy, doesn't it? The dialectal and now popular variant "en" is similar: en - en - ens (instead of: man - en - ens). And the same in Danish, as far as I know. Maybe I would better treat the genitive as a possessive pronoun? Exactly as for "jag" - "min" etc? > > -- > Jacob Nordfalk <http://profiles.google.com/jacob.nordfalk> Yours, Per Tunedal ------------------------------------------------------------------------------ Free Next-Gen Firewall Hardware Offer Buy your Sophos next-gen firewall before the end March 2013 and get the hardware for free! Learn more. http://p.sf.net/sfu/sophos-d2d-feb _______________________________________________ Apertium-stuff mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/apertium-stuff
