A text-reuse test set built only from open sources
Researchers release a bilingual set for testing plagiarism detectors, with every source passage openly licensed or in the public domain.
Picture made for ABUREAU.
Researchers have published PlagiaCorpus, a dataset for testing tools that detect copied text, in Arabic and English. It includes reuse made by language models and reuse that crosses languages.
The set holds 198,643 records, of which 34,885 are source passages and 163,758 are queries. Every copied query links back to its source passage, and the training and test splits are grouped by source, so no passage appears in both.
The sources are English and Arabic Wikipedia, a collection of Arabic books, and a public-domain English collection. The full paper says the dataset is released under CC BY-SA 4.0, which means anyone who adapts it must keep the credit and share alike. It also lists its limits: the human check covered only 300 queries, and some model outputs are flagged as weak.
For people who own or use data, the point is the chain of rights. Because every input was open or public domain, the authors can state one clear licence for the whole set. Data built on sources with unclear terms cannot make that promise, and a buyer can tell the difference.
