The Untranslatability Dataset is a multilingual resource designed to support research on untranslatability in machine translation. It contains source-language sentences that exhibit linguistic, figurative, or cultural phenomena whose meaning cannot be fully preserved through direct translation into English. Each source sentence is paired with multiple English translations generated using different compensation strategies… See the full description on the dataset page: https://huggingface.co/datasets/INK-USC/untranslatability-corpus.
