Linguistically Informed Evaluation of Multilingual ASR for African Languages
Fei-Yueh Chen, Lateef Adeleke, C.M. Downey · 2026
Read Linguistically Informed Evaluation of Multilingual ASR for African Languages by Fei-Yueh Chen, Lateef Adeleke, C.M. Downey (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-04716.txt.
Description
Fei-Yueh Chen, Lateef Adeleke, and C. M. Downey evaluate speech recognition for African languages with richer error measures in a 2026 paper. The standard word error rate lumps phonological, tonal, and other mistakes into one blunt score, hiding what models actually get wrong.
The authors supplement it with character and feature error rates, plus a new tone-aware extension. Feature-level scoring counts errors on individual sound properties. They tested three speech encoders on Yoruba and on Uneme, an endangered language absent from pretraining data. Models handled segmental sound features better than tones, which remained hardest. Even when word accuracy was near zero, feature scores revealed partial learning. The metrics expose linguistically meaningful patterns for improving African language technology.
Catalog record
| CREATOR | Fei-Yueh Chen, Lateef Adeleke, C.M. Downey |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 42 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-04716