Okey, okey, i see. Indeed, the german pos-tagger works when using uppercase letters. Thanks for this hint Florian. So i have to ensure, that Uppercase/Lowercase is handled correctly.
Two final quesations: * is there stemming support in openNLP * is the spelling support in opneNLP Did not find any hints ... so guess that there is no support .. Dan -------- Original-Nachricht -------- > Datum: Tue, 26 Jun 2012 23:52:30 -0700 > Von: Lance Norskog <[email protected]> > An: [email protected] > Betreff: Re: Newby Question on German POS Tagging > I have no experience in this, I have just coded a little with the > libraries. > > On Tue, Jun 26, 2012 at 1:08 PM, Florian Kuhlmann <[email protected]> > wrote: > > Hello Daniel, > > > > I'm new to the list, but as i also work with german language and > POS-tags, i want to give my 2 cent: > > Please consider that capitalisation is probably one of the features used > to determine the correct tag. > > > > So if you have a noun like "Sonne" it should not begin with a small "s". > Most NLP tools are not very robust for this kind of errors (and so > sometimes out of the box not a good match for e.g. social media, because they > are > trained on news which mostly have correct spelling). > > > > Best, > > > > Florian > > > > ________________________________________ > > Von: daniel stieger [[email protected]] > > Gesendet: Dienstag, 26. Juni 2012 21:56 > > An: [email protected] > > Betreff: Re: Newby Question on German POS Tagging > > > > Hi Lance, > > > > thanks for your answer here. Is there a way to contact you (skype) for a > small talk. That would help me a lot. Or can you recommend someone who is > experienced? > > > > I m not into computer linguistic in detail. In past, i wrote my own > algorithms, compined them with stemming and levinstein but openNLP would offer > a totally different approach - i guess. > > > > Rright now, i just need someone who pushes me into the right direction. > > > > Any help appreciated .... :) > > Dan > > > > > > -------- Original-Nachricht -------- > >> Datum: Mon, 25 Jun 2012 01:27:07 -0700 > >> Von: Lance Norskog <[email protected]> > >> An: [email protected] > >> Betreff: Re: Newby Question on German POS Tagging > > > >> The Chunking tool might help here. Chunking means finding noun and > >> verb phrases. This can help you find recurring phrases. Because German > >> is agglutinative, this is probably a very different problem than in > >> English. Are there any de-agglutinizer algorithms? > >> > >> On Sun, Jun 24, 2012 at 11:43 PM, daniel stieger > <[email protected]> > >> wrote: > >> > Hi, > >> > > >> > thanks a lot for your answers. My goal is to identify adjectives and > >> nouns in association sentence. Eg. What do you associate with our > brand? > >> Answer: nice mountains, the mountains are very nice .. etc. > >> > > >> > > >> > If appropriate, i would use the openNLP posTagger (it seams to be the > >> most elaborated java postagger) in order to identify nouns and > adjectives. So > >> when i input the sentence "the", "mountains", "are", "nice" > >> > the output is correct - also when using single words: > >> > > >> >>> [DT, NNS, VBP, JJ] > >> >>> [DT] > >> >>> [NNS] > >> >>> [VBP] > >> >>> [JJ] > >> > > >> > > >> > Is the english model better than the german model? Do i have to build > my > >> own model - or is the de-maxent appropriate? > >> > > >> > Generally - is openNLP a good choice for my task? > >> > > >> > Thanks again, > >> > Dan > >> > > >> > > >> > -------- Original-Nachricht -------- > >> >> Datum: Sat, 23 Jun 2012 16:53:36 -0700 > >> >> Von: Lance Norskog <[email protected]> > >> >> An: [email protected] > >> >> Betreff: Re: Newby Question on German POS Tagging > >> > > >> >> What would you like to find out about your data? Until we know that > it > >> >> is difficult to recommend a technique. > >> >> > >> >> On Sat, Jun 23, 2012 at 4:15 AM, Thilo Goetz <[email protected]> wrote: > >> >> > On 22.06.2012 20:13, daniel stieger wrote: > >> >> >> > >> >> >> Hi List, > >> >> >> > >> >> >> i m looking for some suggestions and opinions for my task. The > >> >> situation > >> >> >> is this: > >> >> >> > >> >> >> In an online survey approx. 800 participants were asked a open > text > >> >> >> question like "What do you associate with our brand?". > Participants > >> can > >> >> then > >> >> >> enter 5 associations. Eg. > >> >> >> > >> >> >> - nature > >> >> >> - beautifull mountains > >> >> >> - relax > >> >> >> - family friendly > >> >> >> - very good service > >> >> >> > >> >> >> > >> >> >> Now i just want to run the openNLP Post tagger over all > >> associations. I > >> >> >> suppose that i can use one association just as one sentence. > Instead > >> of > >> >> the > >> >> >> english model, i used the de-maxent.bin model and some german > >> answers. > >> >> But > >> >> >> the tags are somehow wrong. Eg. > >> >> >> > >> >> >> sonne -> KON > >> >> >> familie -> ART (it is a noun, definitely not an aricle) > >> >> >> > >> >> >> Am I on a wrong path? Should i handle my data differently? Or > should > >> i > >> >> >> download an other model? Where can i get trainingdata ?? > >> >> >> > >> >> >> So many questions.. sorry.. but every hint appreciated, > >> >> >> > >> >> >> best, > >> >> >> Daniel > >> >> >> > >> >> >> > >> >> > > >> >> > I'm pretty sure the model was trained on complete sentences. The > >> >> > tagging takes context into account, and will not work properly > >> >> > without it. So just running it on a couple of words at a time > >> >> > will not work. > >> >> > > >> >> > If all your associations are NPs like your example, > >> >> > you can maybe fix things by always prefixing "I like the ". In > >> >> > German, maybe "Ich liebe ". > >> >> > > >> >> > HTH, > >> >> > Thilo > >> >> > > >> >> > > >> >> > >> >> > >> >> > >> >> -- > >> >> Lance Norskog > >> >> [email protected] > >> > > >> > -- > >> > NEU: FreePhone 3-fach-Flat mit kostenlosem Smartphone! > >> > Jetzt informieren: http://mobile.1und1.de/?ac=OM.PW.PW003K20328T7073a > >> > >> > >> > >> -- > >> Lance Norskog > >> [email protected] > > > > -- > > NEU: FreePhone 3-fach-Flat mit kostenlosem Smartphone! > > Jetzt informieren: http://mobile.1und1.de/?ac=OM.PW.PW003K20328T7073a > > > > -- > Lance Norskog > [email protected] -- NEU: FreePhone 3-fach-Flat mit kostenlosem Smartphone! Jetzt informieren: http://mobile.1und1.de/?ac=OM.PW.PW003K20328T7073a
