Evaluating WMT 2024 Metrics Shared Task Submissions on AfriMTE (the African Challenge Set)

Jiayi Wang, David Ifeoluwa Adelani, Pontus Stenetorp


Abstract
The AfriMTE challenge set from WMT 2024 Metrics Shared Task aims to evaluate the capabilities of evaluation metrics for machine translation on low-resource African languages, which primarily assesses cross-lingual transfer learning and generalization of machine translation metrics across a wide range of under-resourced languages. In this paper, we analyze the submissions to WMT 2024 Metrics Shared Task. Our findings indicate that language-specific adaptation, cross-lingual transfer learning, and larger language model sizes contribute significantly to improved metric performance. Moreover, supervised models with relatively moderate sizes demonstrate robust performance, when augmented with specific language adaptation for low-resource African languages. Finally, submissions show promising results for language pairs including Darija-French, English-Egyptian Arabic, and English-Swahili. However, significant challenges persist for extremely low-resource languages such as English-Luo and English-Twi, highlighting areas for future research and improvement in machine translation metrics for African languages.
Anthology ID:
2024.wmt-1.36
Volume:
Proceedings of the Ninth Conference on Machine Translation
Month:
November
Year:
2024
Address:
Miami, Florida, USA
Editors:
Barry Haddow, Tom Kocmi, Philipp Koehn, Christof Monz
Venue:
WMT
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
505–516
Language:
URL:
https://aclanthology.org/2024.wmt-1.36
DOI:
Bibkey:
Cite (ACL):
Jiayi Wang, David Ifeoluwa Adelani, and Pontus Stenetorp. 2024. Evaluating WMT 2024 Metrics Shared Task Submissions on AfriMTE (the African Challenge Set). In Proceedings of the Ninth Conference on Machine Translation, pages 505–516, Miami, Florida, USA. Association for Computational Linguistics.
Cite (Informal):
Evaluating WMT 2024 Metrics Shared Task Submissions on AfriMTE (the African Challenge Set) (Wang et al., WMT 2024)
Copy Citation:
PDF:
https://aclanthology.org/2024.wmt-1.36.pdf