New language models developed for low-resource Angolan languages

Researchers have developed ANGOFA, a new approach to creating language models for Angolan languages, which are typically very low-resource. The method utilizes Multilingual Adaptive Fine-tuning (MAFT) combined with informed embedding initialization and synthetic data. This technique significantly improved upon existing models, outperforming the SOTA AfroXLMR-base by 12.3 points and OFA by 3.8 points on downstream tasks. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Addresses the gap in AI development for under-resourced languages, potentially enabling broader linguistic inclusion.

RANK_REASON Academic paper introducing a new method for low-resource language modeling. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• ANGOFA
• OFA
• AfroXLMR-base
• David Adelani

• paper
• other