README.md

metadata

language:
  - en
  - de
tags:
  - translation
  - opus-mt
license: cc-by-4.0
model-index:
  - name: opus-mt-tc-big-eng-deu
    results:
      - task:
          name: Translation eng-deu
          type: translation
          args: eng-deu
        dataset:
          name: Tatoeba-test.eng-deu
          type: tatoeba_mt
          args: eng-deu
        metrics:
          - name: BLEU
            type: bleu
            value: 45.7

Opus Tatoeba English-German

*This model was obtained by running the script convert_marian_to_pytorch.py - Instruction available here. The original models were trained by Jörg Tiedemann using the MarianNMT library. See all available MarianMTModel models on the profile of the Helsinki NLP group.

This is the conversion of checkpoint opus+bt-2021-04-13.zip *

---

eng-deu

• source language name: English

• target language name: German

• OPUS readme: README.md

• model: transformer-align

• source language code: en

• target language code: de

• dataset: opus+bt

• release date: 2021-02-22

• pre-processing: normalization + SentencePiece (spm32k,spm32k)

• download original weights: opus+bt-2021-04-13.zip

• Test set translations data: opus+bt-2021-04-13.test.txt

• test set scores file: opus+bt-2021-04-13.eval.txt

• Benchmarks

  Test set	BLEU	chr-F
  newssyscomb2009.eng-deu	22.8	0.538
  news-test2008.eng-deu	23.7	0.533
  newstest2009.eng-deu	22.6	0.532
  newstest2010.eng-deu	25.5	0.552
  newstest2011.eng-deu	22.6	0.527
  newstest2012.eng-deu	23.4	0.530
  newstest2013.eng-deu	27.1	0.556
  newstest2014-deen.eng-deu	29.6	0.599
  newstest2015-ende.eng-deu	31.6	0.600
  newstest2016-ende.eng-deu	37.2	0.644
  newstest2017-ende.eng-deu	30.6	0.595
  newstest2018-ende.eng-deu	45.6	0.696
  newstest2019-ende.eng-deu	41.3	0.659
  Tatoeba-test.eng-deu	45.7	0.654