A study evaluates whether leveraging linguistic relatedness enhances cross-lingual transfer from auxiliary languages to low-resource African languages in large automatic speech recognition (ASR) models. The research extends this framework through a systematic controlled experimental design spanning six factors, two Africa-centric corpora, and four large ASR models.
- Pre-adaptation on related auxiliary languages yields no practically meaningful transfer improvements given minimal target-language data.
- The findings suggest that linguistic relatedness alone may not reliably predict cross-lingual transfer gains in large multilingual ASR.
- Consequently, this approach does not constitute an effective strategy for extending such models to low-resource languages.
The authors conclude that while this strategy has shown improvements in small ASR models, it fails to provide reliable benefits in the context of large-scale multilingual systems.