TTLlab fine tunes arabic system to spot translation errors effectively
TTLab at AlexandriaX-2026: A Fine-Tuned Surface Tagger for Arabic Machine-Translation Error-Span Detection and Classification
Computation and LanguageArtificial IntelligenceMachine Learning
Summary
Finding mistakes in Arabic machine-translated text is important but tricky, especially because some errors happen rarely. The authors created a system that marks each word to identify error spots precisely, using a special way to deal with rare error types and dialect differences. They tested six Arabic language models and found one that worked best, ranking third among many teams. While their system locates errors well, it struggles to classify rare mistakes, suggesting more data could help.
What this means in practice
- •For machine translation developers: Identify exact error spans in Arabic translated texts to improve translation quality by focusing corrections on precise words and phrases.
- •For language quality assurance teams: Use token-level error detection in Arabic to prioritize reviews and corrections in dialect-specific translated content effectively.
Authors
Ali Abusaleh, Bhuvanesh Verma, Alexander Mehler
Abstract
We present TTLab's submission to the AlexandriaX-2026 Subtask~3 on Arabic MT error span detection and classification. Our system frames the task as token-level classification over surface forms, preserving character offsets to ensure exact alignment with the evaluation metric. To handle severe label imbalance, we employ a focal loss with class weighting and dialect-specific decoding thresholds. Among six Arabic pre-trained encoders, MARBERTv2 achieves the best overall performance of 40.8 and 40.91 on the development and test set, respectively, ranking $\nth{3}$ out of all participating teams. While our system localizes error spans effectively, classification of rare error types remains challenging, highlighting the need for data augmentation for tail categories. The code is available at ${\href{https://github.com/ENTAILab/arabic-dialectal-mt-error-span-detection}{\faGithub~ TTLab at AlexandriaX-2026}$