Researchers have created a collection of open-source machine translation resources for 19 Sub-Saharan African languages. The TranslatePsy-AfriSLM dataset includes curated parallel data and synthetic data tailored to the needs of these languages. Studies show that filtering out low-quality training tokens improves performance, and fine-tuning on this filtered data results in better translation quality compared to larger systems with more parameters. --- Why it matters: This matters because it addresses a significant gap in machine translation capabilities for African languages, which have been largely overlooked by AI research so far. By providing high-quality resources and techniques, researchers can develop more effective small language models that support language preservation and digital inclusion on the continent. Source: https://arxiv.org/abs/2608.18655