Hi Anuradha, You need to update your proposal based on what Hèctor suggested, yeah it is better to work on both monodix and bidix simultaneously, but for a good lexicon, you need to take a small corpus and analysis the sentences and adding words.
Sevilay On Thu, Apr 8, 2021 at 9:24 AM Anuradha Pandey <anuradha200...@gmail.com> wrote: > Thank you for your response, Hèctor. I read the proposal for the > Hindi-Bengali translator. There aren't open-source dictionaries for the > Bhojpuri language (though there are resources for getting a Bhojpuri > corpus), so I was using a hardcopy of a BHO-HIN dictionary for manually > adding the pairs. I did some rough calculations, and I shall be able to add > at least 8,000 words to the monodix. And, based on my experience with > Apertium, I think simultaneously adding words in the bidix makes the work > easier, so I think roughly the same number of words in the bidix too. But, > I don't think I will be able to achieve a WER below 20% with 8000 words. > Should I aim for a WER of nearly 30% then? > > Since the time for GSoC has been reduced, I am planning to modify my > proposal and the inputs from mentors would be extremely helpful. > > On Wed, 7 Apr 2021 at 20:24, Hèctor Alòs i Font <hectora...@gmail.com> > wrote: > >> Hi, Anuradha. >> >> Thanks for your proposal draft. First, I would like to tell you that if >> Apertium is a rule-based translation system, it is because this paradigm >> still makes sense for many languages (indeed, for the vast majority of >> them). If Bhojpuri has extensive electronic language resources and, >> particularly, bilingual linguistic corpora, then Apertium is probably not >> the best approach. But this is probably not the case. If it was, it would >> probably already be on Google Translate. >> >> As for the project. I would advise you to look at Gourab Chakraborty's >> proposal for a Hindi-Bengali translator and the comments on it. Most of the >> comments apply to your proposal as well. The following message would be >> useful to you, for instance: >> https://sourceforge.net/p/apertium/mailman/message/37251899/ >> >> Your proposal seems to me unrealistic. 10,000 words in the monodix (and >> how many in the bidix?) are not enough for a WER below 20%, I think (maybe >> for two extremely close related languages). >> >> For better evaluation your proposal I'd like to find the answer for some >> basic questions: >> >> * Which is the current state of Bhojpuri language and, eventually, >> the Bhojpuri-Hindi language pair in Apertium? >> * Would you have to write a whole Bhojpuri morphological analyser from >> scratch and, afterwards, to add some 10,000 words manually assigning them >> to a given paradigm? How much time you'll need for this? >> * From where would you get the bilingual dictionary? Would you have to >> create it yourself? Are there freely available bilingual electronic >> dictionaries (like e.g. Wiktionary)? >> * Would you work on a Bhojpuri-to-Hindi translator or on a >> Hindi-to-Bhojpuri one? In any case there will be a quite a lot of work in >> the morphological disambiguation. But for one side you'll have it only >> once. If both Hindi-to-Bhojpuri and Hindi-to-Bengali are chosen (which is >> entirely possible), this work can be divided by the two projects. >> >> There is nothing wrong to this all this work by hand, if needed. It >> depends on the state of the language resources for the given language. But >> it is necessary to know to what extent you will have to do this >> time-consuming work. >> >> When we had twice the time in most of the cases the projects couldn't >> reach to create a working translator for a new language pair. In the >> current conditions, it is even more difficult. >> >> Hèctor >> >> >> >> >> Missatge de Anuradha Pandey <anuradha200...@gmail.com> del dia dc., 7 >> d’abr. 2021 a les 16:28: >> >>> Hello everyone, >>> I am Anuradha Pandey, a sophomore student at BITS Pilani. I am >>> interested I participating in GSoC 2021, on the project - "*Develop a >>> prototype MT system for a strategic language pair*". >>> >>> I have prepared a rough draft for the same and I am planning to build >>> Bhojpuri(BHO)-Hindi(HIN) MT pair. I am improving my translation system for >>> the coding challenge and I will update my work on the GitHub repository >>> mentioned in the draft. It would be really helpful if I could get some >>> feedback before I make the final submission. >>> >>> Link to the draft - >>> >>> https://docs.google.com/document/d/1U19gJ3TMKYkYsp-FRthrvXkCRJUnNYSYKi46XhvZGOE/edit?usp=sharing >>> >>> Thanks & Regards, >>> Anuradha Pandey >>> IRC: Anuradha_Pandey >>> _______________________________________________ >>> Apertium-stuff mailing list >>> Apertium-stuff@lists.sourceforge.net >>> https://lists.sourceforge.net/lists/listinfo/apertium-stuff >>> >> _______________________________________________ >> Apertium-stuff mailing list >> Apertium-stuff@lists.sourceforge.net >> https://lists.sourceforge.net/lists/listinfo/apertium-stuff >> > _______________________________________________ > Apertium-stuff mailing list > Apertium-stuff@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/apertium-stuff >
_______________________________________________ Apertium-stuff mailing list Apertium-stuff@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/apertium-stuff