AfriSUD: A Dependency Treebank Collection for Evaluating Models on African Languages

In three linesAfriSUD is the first large-scale collection of syntactically annotated treebanks for nine African languages using the SUD framework. Evaluations on POS tagging and dependency parsing reveal a significant syntax gap: models (non-transformer baselines, multilingual encoders, LLMs) show clear limitations across the structural diversity of African-language syntax.

Summary generated by Claude — human-verified