The released dataset comprises manually generated, localised and cross-lingually aligned TOD data in Arabic, Indonesian, Russian and Swahili, as well as the corresponding data in English from the SGD dataset, which served as the source of dialogue prompts. For details of our prompt-based language-specific dialogue generation method please see our paper.
Baseline code will be released shortly.
| ISO 639-2 | Name | Family | Area1 | Script |
|---|---|---|---|---|
| ar | Arabic | Afro-Asiatic | Northern Africa/Western Asia | Arabic |
| id | Indonesian | Austronesian | Southeastern Asia | Latin |
| ru | Russian | Indo-European | Eastern Europe | Cyrillic |
| sw | Swahili | Niger-Congo | Eastern Africa | Latin |
1 According to the United Nations geoscheme.