Difference between revisions of "Shan and Thai"
(→Final Evaluation) |
(→Additions) |
||
| (One intermediate revision by the same user not shown) | |||
| Line 81: | Line 81: | ||
==Additions== | ==Additions== | ||
| − | *Add | + | *Add 50 words to the bilingual dictionary (100 stems) |
*Add 2 transfer rules | *Add 2 transfer rules | ||
*Add 2 lexical selection rules | *Add 2 lexical selection rules | ||
| Line 110: | Line 110: | ||
Percentage of unknown words: | Percentage of unknown words: | ||
*150.00 % | *150.00 % | ||
| + | |||
| + | Coverage of Large Corpus: | ||
| + | *18.8% | ||
| + | |||
| + | Tokens in Large Corpus: | ||
| + | *1827736 | ||
| + | |||
| + | Coverage of Longer Corpus: | ||
| + | *41.0% | ||
| + | |||
| + | Tokens in Longer Corpus: | ||
| + | *290 | ||
[[Category:Sp23_TranslationPairs]][[Category:Shan]][[Category:Thai]] | [[Category:Sp23_TranslationPairs]][[Category:Shan]][[Category:Thai]] | ||
Latest revision as of 20:23, 7 May 2023
Contents
Resources for machine translation between Shan and Thai
Resources for Shan: https://wikis.swarthmore.edu/ling073/Shan
Resources for Thai: https://wikis.swarthmore.edu/ling073/Thai
Lexical Selection: https://wikis.swarthmore.edu/ling073/Shan_and_Thai/Lexical_selection
Contrastive Grammar: https://wikis.swarthmore.edu/ling073/Shan_and_Thai/Contrastive_Grammar
Structural Transfer: https://wikis.swarthmore.edu/ling073/Shan_and_Thai/Structural_transfer
External Resources
Link to our translation system: https://github.swarthmore.edu/Ling073-sp23/ling073-shn-tha
Shan transducer: https://github.swarthmore.edu/Ling073-sp23/ling073-shn
Thai transducer: https://github.swarthmore.edu/Ling073-sp23/apertium-tha
shn -> tha evaluation
Coverage of Shan setences by the Shan monolingual transducer: 107/199 (~53%)
Coverage of Shan setences by the Shan Thai bilingual transducer: 47/226 (~20%)
10 setences from the setences we are testing:
Is the last line in each item the current output? What's up with the @ signs? - JNW
the last line is the output. This was from before we fixed the punctuation things. - Bua
- မႃတူဝ်ဢၼ်လႅၼ်ႈယူႇ
- หมาตัวที่วิ่งอยู่
- ^မႃ/မႃ<n>/@မႃ/မႃ<n>$ ^တူဝ်/တူဝ်<clf>/@တူဝ်/တူဝ်<clf>$ ^ဢၼ်/ဢၼ်<clf>/@ဢၼ်/ဢၼ်<clf>$ ^လႅၼ်ႈ/လႅၼ်ႈ<vblex>/@လႅၼ်ႈ/လႅၼ်ႈ<vblex>$ ^ယူႇ/*ယူႇ/@ယူႇ/*ယူႇ$ .
- @မႃ/မႃ @တူဝ်/တူဝ် @ဢၼ်/ဢၼ် @လႅၼ်ႈ/လႅၼ်ႈ @ယူႇ/*ယူႇ .
- ၵိၼ်မႃး
- กินมา
- ^ၵိၼ်/ၵိၼ်<vblex>/@ၵိၼ်/ၵိၼ်<vblex>$ ^မႃး/မႃး<vblex>/@မႃး/မႃး<vblex>$ .
- @ၵိၼ်/ၵိၼ် @မႃး/မႃး .
- ဝႆႉၵိၼ်
- ไว้กิน
- ^ဝႆႉ/ဝႆႉ<vblex>/ဝႆႉ<particle>/ဝႆႉ<adv>/@ဝႆႉ/ဝႆႉ<vblex>/ဝႆႉ<particle>/ဝႆႉ<adv>$ ^ၵိၼ်/ၵိၼ်<vblex>/@ၵိၼ်/ၵိၼ်<vblex>$ .
- @ဝႆႉ/ဝႆႉ @ၵိၼ်/ၵိၼ် .
- ၵိၼ်ဝႆႉ
- กินไว้
- ^ၵိၼ်/ၵိၼ်<vblex>/@ၵိၼ်/ၵိၼ်<vblex>$ ^ဝႆႉ/ဝႆႉ<vblex>/ဝႆႉ<particle>/ဝႆႉ<adv>/@ဝႆႉ/ဝႆႉ<vblex>/ဝႆႉ<particle>/ဝႆႉ<adv>$ .
- @ၵိၼ်/ၵိၼ် @ဝႆႉ/ဝႆႉ .
- ၶႂ်ႈၵိၼ်
- อยากกิน
- ^ၶႂ်ႈ/*ၶႂ်ႈ/@ၶႂ်ႈ/*ၶႂ်ႈ$ ^ၵိၼ်/ၵိၼ်<vblex>/@ၵိၼ်/ၵိၼ်<vblex>$ .
- @ၶႂ်ႈ/*ၶႂ်ႈ @ၵိၼ်/ၵိၼ် .
- ၵိၼ်မႅၼ်ႈ
- กินอย่างไม่ได้ตั้งใจ
- ^ၵိၼ်/ၵိၼ်<vblex>/@ၵိၼ်/ၵိၼ်<vblex>$ ^မႅၼ်ႈ/*မႅၼ်ႈ/@မႅၼ်ႈ/*မႅၼ်ႈ$ .
- @ၵိၼ်/ၵိၼ် @မႅၼ်ႈ/*မႅၼ်ႈ .
- ၵိၼ်လီလီ
- กินดีดี
- ^ၵိၼ်/ၵိၼ်<vblex>/@ၵိၼ်/ၵိၼ်<vblex>$ ^လီ/*လီ/@လီ/*လီ$ ^လီ/*လီ/@လီ/*လီ$ .
- @ၵိၼ်/ၵိၼ် @လီ/*လီ @လီ/*လီ .
- ၵိၼ်ၸွတ်ႇ
- กินหมด
- ^ၵိၼ်/ၵိၼ်<vblex>/@ၵိၼ်/ၵိၼ်<vblex>$ ^ၸ/*ၸ/@ၸ/*ၸ$ွ^တ်ႇ/*တ်ႇ/@တ်ႇ/*တ်ႇ$ .
- @ၵိၼ်/ၵိၼ် @ၸ/*ၸွ@တ်ႇ/*တ်ႇ .
- ၵိၼ်တူၺ်း
- กินดู
- ^ၵိၼ်/ၵိၼ်<vblex>/@ၵိၼ်/ၵိၼ်<vblex>$ ^တူၺ်း/တူၺ်း<vblex>/@တူၺ်း/တူၺ်း<vblex>$ .
- @ၵိၼ်/ၵိၼ် @တူၺ်း/တူၺ်း .
- ၵိၼ်လဵၼ်ႈ
- กินเล่น
- ^ၵိၼ်/ၵိၼ်<vblex>/@ၵိၼ်/ၵိၼ်<vblex>$ ^လဵၼ်ႈ/လဵၼ်ႈ<vblex>/@လဵၼ်ႈ/လဵၼ်ႈ<vblex>$ .
- @ၵိၼ်/ၵိၼ် @လဵၼ်ႈ/လဵၼ်ႈ .
Our setences are rather short but it was hard to find long sentences with pre-existing translation into any language.
Additions
- Add 50 words to the bilingual dictionary (100 stems)
- Add 2 transfer rules
- Add 2 lexical selection rules
Note: Lexical selection rules were just default rules. There was not not enough data to conclude on a definate pattern.
Final Evaluation
Monolingual Evaluation
- Precision: 68.87417%
- Recall: 61.90476%
Large Corpus Coverage:
- 66.7%
Number of Words in Large Corpus:
- 1943840
Stems in Transducer:
- 379
shn->tha Bilingual Evaluation
Word error rate (WER):
- 120.43 %
Position-independent word error rate (PER):
- 117.20 %
Percentage of unknown words:
- 150.00 %
Coverage of Large Corpus:
- 18.8%
Tokens in Large Corpus:
- 1827736
Coverage of Longer Corpus:
- 41.0%
Tokens in Longer Corpus:
- 290
