Tatar and Bashkir/GSOC 2018

From Apertium
Jump to navigation Jump to search

This is the report for Google Summer of Code 2018 project — Tatar-Bashkir machine translation.

List of commits[edit]

The list of all the commits can be found here: https://apertium.projectjj.com/gsoc2018/zu-ann/zu-ann.html.

tar.gz with commits can be downloaded here: https://apertium.projectjj.com/gsoc2018/zu-ann.tar.gz.

zip with commits can be downloaded here: https://apertium.projectjj.com/gsoc2018/zu-ann.zip.

What was done[edit]

Lexicons in bak.lexc were changed to correspond to the ones in tat.lexc, missing lexicons and tags were added to bak.lexc and new rules were added to bak.twol.

The stems from tat.lexc were translated into Bashkir and added to bak.lexc and bidix.

Words from the Bashkir frequency list http://lcph.bashedu.ru/index.php?go=wikilist_lemmas were translated into Tatar and added to tat.lexc and bidix.

Using Russian-Tatar and Russian-Bashkir dictionaries new stems were added to tat.lexc, bak.lexc and bidix.

Using Wikidata new toponyms were added to tat.lexc, bak.lexc and bidix.

Statistics[edit]

tat.lexc bak.lexc bidix Bilingual Coverage
Before ~ 26 400 lemmas ~ 2 800 lemmas ~ 2 600 lemmas tat-bak: 72.00%, bak-tat: 68.63%
After ~ 56 900 lemmas ~ 56 000 lemmas ~ 50 000 lemmas tat-bak: 87.56%, bak-tat: 88.98%

Bilingual coverage was calculated using aq-covtest (http://wiki.apertium.org/wiki/Apertium-quality/Application_Documentation) and Wikipedia dumps (https://dumps.wikimedia.org/ttwiki/20180801/ and https://dumps.wikimedia.org/bawiki/20180801/).

Future work[edit]

  • Continue improving Bashkir monolingual transducer to make more word forms analyzed.
  • Continue improving the coverage.
  • Check (and fix if necessary) the words, mostly proper nouns, which were added using auto translation.
  • Revise the dictionaries.