metadata
license: other
task_categories:
- sentence-similarity
language:
- nl
pretty_name: OpenSubtitles STS Dataset for Dutch
size_categories:
- 1M<n<10M
OpenSubtitles STS Dataset for Dutch
OS-STS.nl is an extensive Dutch STS dataset containing over two million sentence pairs and similarity scores. The dataset is automatically extracted from movie and documentary subtitles sourced from OpenSubtitles2018, a vast parallel corpus of aligned video subtitles.
Recognizing the high prevalence (>10%) of paraphrased statements and question-and-answer pairs in subtitled spoken language, we systematically extract the consecutive parallel sentence pairs from the subtitles that exhibit significant semantic overlap.
Content of the dataset
The dataset contains Dutch sentence pairs, as well as semtatic similarity scores derived from their English translation derived from sentence-transformers/all-mpnet-base-v2.
