From Apertium
Jump to navigation Jump to search

My name is Ilnar Salimzyanov (Илнар Сәлимҗан) and I maintain the Kazakh-Tatar and Tatar-Russian language pairs.

My IRC nick is "selimcan".

Responsibilities as the maintainer of the Kazakh-Tatar pair[edit]

Making sure that the following things are constant:

  1. All (=100%) of the regression tests pass: ./ Regression kaz tat update
  2. Corpus test doesn't produce any debugging symbols: ./ kaz-tat-corp && grep -c '[*@#]' corpa/kaz-tat-nova.txt outputs 0.
  3. WER continuosly decreases or at least stays the same after every commit. So, the WER numbers output by the command
    kaz-tat$ perl ../apertium-eval-translator/ -test corpa/kaz-tat-nova.txt -ref corpa/corpus.tat.txt
    are less than or equal to numbers output by
    kaz-tat$ perl ../apertium-eval-translator/ -test corpa/kaz-tat-origina.txt -ref corpa/corpus.tat.txt

The same applies for the opposite direction:

  1. ./ Regression tat kaz update => 100%
  2. ./ tat-kaz-corp && grep -c '[*@#]' corpa/tat-kaz-nova.txt => 0
  3. The numbers output by
    kaz-tat$ perl ../apertium-eval-translator/ -test corpa/tat-kaz-nova.txt -ref corpa/corpus.kaz.txt
    are less than or equal to numbers output by
    kaz-tat$ perl ../apertium-eval-translator/ -test corpa/tat-kaz-origina.txt -ref corpa/corpus.kaz.txt

If and only if all test cases pass, and I don't actively work on another pair, I work towards achieving those goals and take care of documentation.

TODO's related to the above[edit]

  • ./qa in apertium-kaz-tat will compile the transcducers and the pair and run all the tests described above. Ideally it should output a single line indicating that all tests passed and do that as quickly as possible.
  • A good idea would be to set up some bot which compiles apertium-kaz, apertium-tat and apertium-kaz-tat modules if somebody commits to them, runs the tests above and emails me if they don't pass.