Difference between revisions of "Kurmanji and English/Final report"
| Line 27: | Line 27: | ||
| 9 |
| 9 |
||
| 83 |
| 83 |
||
| + | | 57% |
||
| − | | |
||
|- |
|- |
||
! After |
! After |
||
| Line 35: | Line 35: | ||
| 23 |
| 23 |
||
| 157 |
| 157 |
||
| − | | |
+ | | %85 |
|} |
|} |
||
Revision as of 10:16, 23 August 2016
This is the report for my 2016 Google Summer of Code project, Kurmanji-English Machine Translation.
What was done
My project was to improve significantly the preexisting pair, to around release quality. I have worked on adding vocabulary, disambiguation rules in CG, transfer rules and lexical selection.
The vocabulary was added from a number of sources, a few thousand were added from the work of Walther et al in their Kurmanji analyzer and POS tagger.
- Adherence to work plan
I have largely followed and met the goals of the work plan, however in adding vocabulary in some cases I face difficulty in meeting the goals, due to lack of digital resources I was required to add translations one by one, using Kurdish-Turkish dictionaries and translating the meanings into English.
Statistics
| Monodix | Bidix | CG Rules | Transfer | Paradigms | Bilingual Coverage | |
|---|---|---|---|---|---|---|
| Before | 1433 | 11421 | 3 | 9 | 83 | 57% |
| After | 17715 | 15597 | 97 | 23 | 157 | %85 |
Future work
The most immediate concern for future work would be adding more transfer rules, in order to improve the quality of translations, and improving the coverage of the bilingual dictionary a bit more, to around 90%.
List of commits
My commits are listed below, under the folders with the name kmr, which is the ISO code for Kurmanji Kurdish.
