AlexandriaX Subtask 3: Dialectal Arabic Machine Translation Error Annotations

A dataset for AlexandriaX Subtask 3, a dialectal Arabic machine translation evaluation task. It contains train and development splits with machine-translated outputs, source text, and reference-side information. The dataset was created by UBC-NLP and its record was last updated on 2026-05-20.

Use Cases

• Train models for word-level error span detection in machine-translated text based on the described task objective.
• Classify types of translation errors using the LQM-inspired annotations mentioned in the description.
• Benchmark machine translation systems on dialectal Arabic based on the provided evaluation splits.
• Study error patterns in dialectal Arabic machine translation based on the annotated error spans.

Strengths

• Provides dedicated train and development splits for a structured evaluation task.
• Includes annotations for error spans and classification, following an LQM-inspired framework.
• Focuses on the specific and challenging domain of dialectal Arabic machine translation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

UBC-NLP

Collection Method

Likely created for a shared task on dialectal Arabic machine translation evaluation.

Freshness

Last updated 2026-05-20 22:00:46; freshness should be verified.