Can Embedding Similarity Predict Cross-Lingual Transfer? A Systematic Study on African Languages
Domain:
natural language processingRecord type:
paper
Creator:
IdrMit
Host:
Cross-lingual transfer is essential for building NLP systems for low-resource African languages, but practitioners lack reliable methods for selecting source languages. We systematically evaluate five embedding similarity metrics across 816 transfer experiments spanning three NLP tasks, three African-centric multilingual models, and 12 languages from four language families. We find that cosine gap and retrieval-based metrics (P@1, CSLS) reliably predict transfer success ($ρ= 0.4-0.6$), while CKA shows negligible predictive power ($ρ\approx 0.1$). Critically, correlation signs reverse when pooling across models (Simpson's Paradox), so practitioners must validate per-model. Embedding metrics achieve comparable predictive power to URIEL linguistic typology. Our results provide concrete guidance for source language selection and highlight the importance of model-specific analysis. 13 pages, 1 figure, 19 tables
Similar
Cross-lingual Transfer Accuracy and Task Similarity in Low-Resource LanguagesCross-lingual Transfer Accuracy in Low-Resource Languages via Task SimilarityDomain Similarity Impact on Zero-Shot Cross-Lingual Transfer Performance in Low-Resource LanguagesCross-lingual transfer of multilingual models on low resource African Languageskaranwxliaa/Cross-lingual-transfer-of-multilingual-models-on-low-resource-African-LanguagesDomain Similarity Impact on Multilingual Hate Speech Detection Generalization in Zero-Shot Cross-Lingual Transfer