Danish Machine-Translated News Articles For Summarization

Encompassing 2,872 machine-translated news articles and corresponding summaries in Danish. It is derived from the CNN/DailyMail dataset and is intended for Danish text summarization tasks.

Use Cases

• Train sequence-to-sequence models on the 'article' and 'highlights' fields for Danish abstractive summarization.
• Benchmark the quality of machine-translated text for downstream NLP tasks using the translated 'article' content.
• Analyze the linguistic features and compression ratios between original English 'highlights' and their Danish translations.

Strengths

• Contains 2,872 translated article-summary pairs, providing a foundation for Danish NLP.
• Uses a specific translation model (Helsinki-NLP/opus-mt-en-da) for consistent machine translation.
• Derived from the established CNN/DailyMail dataset, which is a standard benchmark for summarization.

Limitations

• Small size of 2,872 rows limits training data for large language models.
• Potential for translation artifacts or errors introduced by the automated process.
• Lacks demographic or geographic metadata about the original news articles.

Provenance

Source

Subset of the CNN/DailyMail dataset, machine-translated to Danish.

Collection Method

Automated translation using the Helsinki-NLP/opus-mt-en-da model.

Freshness

Last updated in August 2022.