Difference between revisions of "User:Qfeng1/Final project"
From LING073
(→Corpus For Evaluation) |
|||
| (8 intermediate revisions by the same user not shown) | |||
| Line 1: | Line 1: | ||
| − | Our project is to expand what we have accomplished in class, the monolingual transducer for the Chechen language. | + | Our project is to expand what we have accomplished in class, the monolingual transducer for the Chechen language. The main goal is to increase the coverage rate of the transducer over a large corpus (we have a large corpus extracted from Wikipedia with all pages in Chechen and therefore there is a wide range of vocabularies included). |
==Major Steps== | ==Major Steps== | ||
*To expand the morphology (in lexc and towl files) | *To expand the morphology (in lexc and towl files) | ||
| − | *To generate the top list of unknown words in corpus and add more word stems to lexc file | + | *To generate the top list of unknown words in corpus and add more word stems to lexc file |
| − | * | + | |
| + | ==Code at GitHub== | ||
| + | https://github.com/sfeng233/Chechen_Transducer | ||
| + | |||
| + | ==Corpus For Evaluation == | ||
| + | https://github.swarthmore.edu/Ling073-sp19/ling073-che-corpus | ||
| + | *Wikipedia Corpus: It is the large corpus that we run coverage test on. | ||
| + | *Hand Annotated Corpus: This corpus is what we run precisionRecall test on. It consists example sentences that have been glossed in the grammar book we use for this project as main consulting resource. | ||
==Evaluation== | ==Evaluation== | ||
| Line 10: | Line 17: | ||
{| class="wikitable" | {| class="wikitable" | ||
|# of stems in transducer | |# of stems in transducer | ||
| − | |||
|# of words in Wiki corpus | |# of words in Wiki corpus | ||
|Coverage Rate | |Coverage Rate | ||
| Line 16: | Line 22: | ||
|Recall | |Recall | ||
|- | |- | ||
| − | | | + | |608 |
| − | + | |14,093,835 | |
| − | | | + | |~80.10% |
| − | | | + | |~75.34% |
| − | | | + | |~56.97% |
| − | | | ||
|} | |} | ||
| + | |||
==Further Improvement== | ==Further Improvement== | ||
| + | *some augment base for nouns and verbs are not clear due to the limitation of resources | ||
| + | *more verb morphology needs to be added: | ||
| + | :*proverb | ||
| + | :*deictic | ||
| + | :* agreement on noun class (when happen and when do not is still no clear) | ||
[[Category:sp19_FinalProjects]] | [[Category:sp19_FinalProjects]] | ||
Latest revision as of 21:04, 14 May 2019
Our project is to expand what we have accomplished in class, the monolingual transducer for the Chechen language. The main goal is to increase the coverage rate of the transducer over a large corpus (we have a large corpus extracted from Wikipedia with all pages in Chechen and therefore there is a wide range of vocabularies included).
Major Steps
- To expand the morphology (in lexc and towl files)
- To generate the top list of unknown words in corpus and add more word stems to lexc file
Code at GitHub
https://github.com/sfeng233/Chechen_Transducer
Corpus For Evaluation
https://github.swarthmore.edu/Ling073-sp19/ling073-che-corpus
- Wikipedia Corpus: It is the large corpus that we run coverage test on.
- Hand Annotated Corpus: This corpus is what we run precisionRecall test on. It consists example sentences that have been glossed in the grammar book we use for this project as main consulting resource.
Evaluation
| # of stems in transducer | # of words in Wiki corpus | Coverage Rate | Precision | Recall |
| 608 | 14,093,835 | ~80.10% | ~75.34% | ~56.97% |
Further Improvement
- some augment base for nouns and verbs are not clear due to the limitation of resources
- more verb morphology needs to be added:
- proverb
- deictic
- agreement on noun class (when happen and when do not is still no clear)
