Difference between revisions of "Apertium-kaz-kir/TODO"
Jump to navigation
Jump to search
Firespeaker (talk | contribs) |
Firespeaker (talk | contribs) |
||
(6 intermediate revisions by the same user not shown) | |||
Line 1: | Line 1: | ||
⚫ | |||
* total 6500 stems in dix |
|||
* azattyq_24455849 WER ≤7.5% |
|||
* trimmed coverage ≥72% |
|||
⚫ | |||
** {{tag|postadv}} {{tag|ij}} |
|||
** {{tag|num}} {{tag|post}} |
|||
** {{tag|cnjcoo}} {{tag|cnjadv}} {{tag|cnjsub}} |
|||
** {{tag|adv}} |
|||
* commit at least once every 24 hours! |
|||
== By 22 July == |
|||
⚫ | |||
⚫ | |||
* Work with JNW on testvoc for closed categories. |
|||
== By 14 July == |
== By 14 July == |
||
* '''add 800 stems''' |
* '''add 800 stems''' |
||
Line 26: | Line 42: | ||
** there's an issue with -ақ; I think we'll need to work on it together |
** there's an issue with -ақ; I think we'll need to work on it together |
||
[[Category:TODO lists]] |
|||
⚫ | |||
⚫ | |||
⚫ | |||
⚫ |
Latest revision as of 21:22, 19 August 2015
By midterm[edit]
- total 6500 stems in dix
- azattyq_24455849 WER ≤7.5%
- trimmed coverage ≥72%
- clean testvoc for the following categories:
<postadv>
<ij>
<num>
<post>
<cnjcoo>
<cnjadv>
<cnjsub>
<adv>
- commit at least once every 24 hours!
By 22 July[edit]
- Add another 1000 words
- Finish WER process for texts/azattyq_24455849.txt
- Work with JNW on testvoc for closed categories.
By 14 July[edit]
- add 800 stems
- mostly nouns, verbs, adjectives (i.e., simple categories)
- 100 top stems from wikipedia corpus
- 100 top stems from rferl/azattyq corpus
- 100 top stems from bible corpus
- 100 top stems from quran corpus
- any 400 words marked i="yes" in dix
- sort these into their appropriate sections
- fix the Kyrgyz translation when needed (many will need to be fixed)
- remove i="yes" part
- Start work on WER for texts/azattyq_24455849.txt
- Use kaz-kir and output to texts/azattyq_24455849.kaz-kir.txt
- Add words/etc. to transducer needed until there are no */#/@
- Copy to texts/azattyq_24455849.kaz-kir-postedited.txt
- Post-edit until the postedited Kyrgyz is clean
- Add lexical selection rules and transfer rules as needed
- Goal: get WER down to around 10%
- Fix the following minor problems:
- words should not be entered with different capitalisation:
- құран=куран / Құран=Куран (remove one of them)
- пайғамбар=пайгамбар / Пайғамбар=Пайгамбар (remove one of them)
- "шәксіз" is not a Kyrgyz word
- there's an issue with -ақ; I think we'll need to work on it together
- words should not be entered with different capitalisation: