Designing and Contextualising Probes for African Languages
It provides insights into model interpretability for African languages, which is incremental as it applies existing techniques to a new domain.
This paper systematically investigates how pretrained language models (PLMs) encode linguistic knowledge for African languages, finding that adapted PLMs capture more information than multilingual ones and that syntactic and semantic features are distributed across specific layers.
Pretrained language models (PLMs) for African languages are continually improving, but the reasons behind these advances remain unclear. This paper presents the first systematic investigation into probing PLMs for linguistic knowledge about African languages. We train layer-wise probes for six typologically diverse African languages to analyse how linguistic features are distributed. We also design control tasks, a way to interpret probe performance, for the MasakhaPOS dataset. We find PLMs adapted for African languages to encode more linguistic information about target languages than massively multilingual PLMs. Our results reaffirm previous findings that token-level syntactic information concentrates in middle-to-last layers, while sentence-level semantic information is distributed across all layers. Through control tasks and probing baselines, we confirm that performance reflects the internal knowledge of PLMs rather than probe memorisation. Our study applies established interpretability techniques to African-language PLMs. In doing so, we highlight the internal mechanisms underlying the success of strategies like active learning and multilingual adaptation.