Opus Medical En De: German-English Parallel Medical Text

A multi-domain German-English parallel corpus introduced by Aharoni and Goldberg in 2020. The data split was created to avoid duplicate examples and leakage from the train split to the dev/test splits. The original multi-domain data first appeared in Koehn and Knowles (2017) and consists of five datasets from the Opus website.

Use Cases

• Train machine translation models based on the described German-English parallel medical text.
• Benchmark translation model performance on a multi-domain medical corpus based on the described data split.
• Fine-tune large language models for cross-lingual medical NLP tasks based on the parallel text data.
• Study domain adaptation in translation models using the multi-domain structure mentioned in the description.

Strengths

• Data split designed to avoid duplicate examples and leakage between train and evaluation sets.
• Derived from five established datasets available on the Opus website.
• Last updated on the platform on 2023-07-16 07:37:53.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license information are unknown, which may limit suitability assessment.

Provenance

Source

Opus website, as referenced in Koehn and Knowles (2017).

Collection Method

New data split created from original multi-domain parallel data.

Freshness

Last updated 2023-07-16 07:37:53; freshness should be verified.