gmw-rus

opus1m-2021-05-16.zip

• dataset: opus1m
• model: transformer-align
• source language(s): afr ang deu eng enm fry gos ltz nds nld
• target language(s): rus
• model: transformer-align
• pre-processing: normalization + SentencePiece (spm32k,spm32k)
• download: opus1m-2021-05-16.zip
• test set translations: opus1m-2021-05-16.test.txt
• test set scores: opus1m-2021-05-16.eval.txt

Benchmarks

testset	BLEU	chr-F	#sent	#words	BP
newstest2012.deu-rus	15.4	0.427	3003	64830	0.989
newstest2012.eng-rus	25.6	0.533	3003	64830	1.000
newstest2013.deu-rus	18.4	0.454	3000	58560	0.965
newstest2013.eng-rus	20.4	0.481	3000	58560	0.989
newstest2015-enru.eng-rus	22.8	0.517	2818	55915	1.000
newstest2016-enru.eng-rus	22.1	0.505	2998	62018	1.000
newstest2017-enru.eng-rus	23.8	0.524	3001	60255	1.000
newstest2018-enru.eng-rus	20.6	0.510	3000	61920	1.000
newstest2019-enru.eng-rus	22.3	0.486	1997	48153	0.939
Tatoeba-test.afr-rus	43.7	0.613	228	1211	0.990
Tatoeba-test.ang-rus	1.6	0.122	10	32	1.000
Tatoeba-test.deu-rus	40.0	0.614	10000	67173	0.969
Tatoeba-test.eng-rus	40.8	0.606	10000	66695	0.987
Tatoeba-test.enm-rus	29.6	0.313	15	52	1.000
Tatoeba-test.fry-rus	23.5	0.444	28	168	1.000
Tatoeba-test.gos-rus	2.8	0.278	6	32	1.000
Tatoeba-test.ltz-rus	30.2	0.461	1	5	1.000
Tatoeba-test.multi-rus	41.1	0.610	10000	66046	0.982
Tatoeba-test.nds-rus	21.2	0.409	894	5263	1.000
Tatoeba-test.nld-rus	44.5	0.645	2500	15641	0.992
Tatoeba-test.yid-rus	0.0	0.009	150	795	1.000
tico19-test.eng-rus	20.6	0.485	2100	55837	0.931