Neurora/opus-tatoeba-eng-deu

Opus Tatoeba | English -> German

• dataset: opus
• model: transformer
• source language(s): eng
• target language(s): deu
• model: transformer
• pre-processing: normalization + SentencePiece (spm32k,spm32k)
• download: opus-2021-02-22.zip
• test set translations: opus-2021-02-22.test.txt
• test set scores: opus-2021-02-22.eval.txt

Benchmarks

testset	BLEU	chr-F	#sent	#words	BP
newssyscomb2009.eng-deu	23.3	0.539	502	11271	0.990
news-test2008.eng-deu	23.9	0.533	2051	47427	1.000
newstest2009.eng-deu	22.7	0.533	2525	62816	0.999
newstest2010.eng-deu	25.9	0.550	2489	61511	0.966
newstest2011.eng-deu	22.9	0.528	3003	72981	0.993
newstest2012.eng-deu	23.8	0.530	3003	72886	0.972
newstest2013.eng-deu	27.6	0.558	3000	63737	0.983
newstest2014-deen.eng-deu	29.6	0.595	3003	62964	1.000
newstest2015-ende.eng-deu	32.0	0.601	2169	44260	1.000
newstest2016-ende.eng-deu	37.9	0.644	2999	62670	0.992
newstest2017-ende.eng-deu	30.6	0.593	3004	61291	1.000
newstest2018-ende.eng-deu	46.4	0.697	2998	64276	0.999
newstest2019-ende.eng-deu	42.4	0.664	1997	48969	0.990
Tatoeba-test.eng-deu	45.8	0.655	10000	83347	0.995