This is a synthetic e-commerce retrieval dataset containing 1,990 query-positive pairs and three supplied negatives per query. Queries cover English, Devanagari Nepali, Romanized Nepali, and mixed modes; the documents are almost entirely English. The most accurate description is therefore cross-lingual query-to-English retrieval, not a fully multilingual document corpus.
