-
source language name: English
-
target language name: French
-
-
model: transformer-align
-
source language code: en
-
target language code: fr
-
dataset: opus
-
release date: 2021-02-22
-
pre-processing: normalization + SentencePiece (spm32k,spm32k)
-
download original weights:
opus-2021-02-22.zip
-
Training data:
- fra-eng: Tatoeba-train (180923857)
-
Validation data:
- eng-fra: Tatoeba-dev, 250098
- total-size-shuffled: 249757
- devset-selected: top 5000 lines of Tatoeba-dev.src.shuffled!
-
Test data:
- newsdiscussdev2015-enfr.eng-fra: 1500/27986
- newsdiscusstest2015-enfr.eng-fra: 1500/28027
- newssyscomb2009.eng-fra: 502/12334
- news-test2008.eng-fra: 2051/52685
- newstest2009.eng-fra: 2525/69278
- newstest2010.eng-fra: 2489/66043
- newstest2011.eng-fra: 3003/80626
- newstest2012.eng-fra: 3003/78011
- newstest2013.eng-fra: 3000/70037
- Tatoeba-test.eng-fra: 10000/80769
- tico19-test.eng-fra: 2100/64655
-
test set translations file:
test.txt
-
test set scores file:
eval.txt
-
BLEU-scores
|Test set|score|
|---|---|
|Tatoeba-test.eng-fra|50.8|
|tico19-test.eng-fra|41.8|
|newsdiscusstest2015-enfr.eng-fra|40.8|
|newstest2011.eng-fra|34.6|
|newsdiscussdev2015-enfr.eng-fra|33.9|
|newstest2013.eng-fra|33.5|
|newstest2010.eng-fra|33.0|
|newstest2012.eng-fra|32.0|
|newssyscomb2009.eng-fra|30.0|
|newstest2009.eng-fra|29.9|
|news-test2008.eng-fra|27.5|
-
chr-F-scores
|Test set|score|
|---|---|
|Tatoeba-test.eng-fra|0.671|
|newsdiscusstest2015-enfr.eng-fra|0.649|
|tico19-test.eng-fra|0.638|
|newstest2011.eng-fra|0.614|
|newsdiscussdev2015-enfr.eng-fra|0.606|
|newstest2010.eng-fra|0.599|
|newstest2012.eng-fra|0.593|
|newstest2013.eng-fra|0.591|
|newssyscomb2009.eng-fra|0.587|
|newstest2009.eng-fra|0.58|
|news-test2008.eng-fra|0.556|