RAMZ / LANGUAGES
Language-by-task evaluation.
A language name on a homepage does not tell you what a model can do. We will report progress separately for transcription, speech generation, translation and assistant tasks.
Our language work connects speech recognition, translation and language understanding with task-specific evaluation.
| Language | Transcription | Speech generation | Translation | Assistant tasks |
|---|---|---|---|---|
| HassaniyaInitial focus | ○ Research area | ○ Research area | ○ Research area | ○ Research area |
| Pulaar | ○ Research area | ○ Research area | ○ Research area | ○ Research area |
| Soninké | ○ Research area | ○ Research area | ○ Research area | ○ Research area |
| Wolof | ○ Research area | ○ Research area | ○ Research area | ○ Research area |
| Arabic | ○ Research area | ○ Research area | ○ Research area | ○ Research area |
| French | ○ Research area | ○ Research area | ○ Research area | ○ Research area |
Evaluation accounts for dialectal and regional variation in each task.