On 4/4/19 9:53 PM, Jonathan Washington wrote: > Сәлем Данияр, > > A couple important note, before I look at your proposal is the following: > > Evaluation for RBMT isn't to verify that every rule works, but to have a > passing "testvoc": http://wiki.apertium.org/wiki/Testvoc . We have a > particular strategy for performing testvoc for morphologically rich > languages like Kazakh, Turkish, and Tatar. Ilnar can tell you more > about the mechanics of it, but it would be good to interact with us > about it in real time—e.g. on IRC. @Daniyar, for details of this you should look at apertium-tat-rus [1].
What Jonathan refers to is in the `lite' folder. Both `lite testvoc' and `corpus testvoc' are important for us. `standard' testvoc is not feasible for Turkic pairs. Also, apertium-tat-rus's transfer rules might be helpful for learning about transfer module. They have INPUT-OUTPUT comments which are tested by the test-scripts/tNx.test scripts [2]. Writing such an INPUT-OUTPUT comment before writing a rule is a good idea because: 1) it makes it clear what you want to achieve 2) they serve as regression tests in the future Also, at [3] I started post-editing (well, as well as I could, with my limited knowledge of Kazakh...) apertium-kaz-tat's translation of a larger Tatar text. Any help with post-editing it is much appreciated. You can use that text as a test case, but for now, the official "Mary and James" story probably covers your needs for test sentences. Best, Ilnar [1] https://github.com/apertium/apertium-tat-rus/tree/master/testvoc [2] https://github.com/apertium/apertium-tat-rus/blob/master/test-scripts/t1x.test [3] https://github.com/taruen/apertiumpp/blob/master/data4apertium/corpora/lpp/tat-kaz.posted.txt > > Also, you should not be "implementing the transfer rules for tur-kaz > direction" and planning to "share it with [us] very soon". This is not > how open-source software development usually works. The mentors should > be able to virtually "look over your shoulder". That is, you should be > working in a public repository where we can see what you've been doing. > The main reason is for your benefit: we can step in to help you out if > we notice systematic problems or the like. > > And also, very importantly, while you're working, you should really be > in IRC. It's good practice for what we would expect over the summer, > and it also makes you more visible to the entire community, including a > wider range of potential mentors! > > -- > Jonathan > > пн, 1 апр. 2019 г. в 15:39, Daniyar Nariman via Apertium-stuff > <[email protected] > <mailto:[email protected]>>: > > Hello everyone, > > As we discussed earlier with some mentors, for now, I decided to > take kaz-tur and kaz-tat language pairs and make them work in the > opposite direction. So here is the draft of my proposal. > > > http://wiki.apertium.org/wiki/Narimann/GSOC_2019_proposal:_Kazakh-Turkish_and_Turkish-Kazakh > > > > In my workplan, I allocated a time(week 9,10,11,12) for testing > purposes, to check if everything is fine and bring some corrections > if needed. > > > > Is this kind of workplan fine or should I change it? > > > Regarding to the coding challenge, for now, I am implementing the > transfer rules for tur-kaz direction. I will share it with you very > soon. > > > Another point is that I can not guarantee that in my final > evaluation these two pairs will work stable, as I do not know how > much time it can take to cover all the rules. So for now, maybe I > should focus only on one pair, for this summer, to guarantee it > works stable? > > > If you have any suggestions or comments, let me know, I appreciate it. > > > Thanks in advance > > > Best Regards, > > Nariman > > > > _______________________________________________ > Apertium-stuff mailing list > [email protected] > <mailto:[email protected]> > https://lists.sourceforge.net/lists/listinfo/apertium-stuff > > > > _______________________________________________ > Apertium-stuff mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/apertium-stuff > -- GPG: 0xF3ED6A19 _______________________________________________ Apertium-stuff mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/apertium-stuff
