92 exams from the 2006 ATA Spanish-to-English certification exam contain error data. Each error includes anonymized exam and grader IDs, MQM and ATA error labels, severity values, and source and target text segments. The dataset was authored by Will Carr and is hosted by Harvard Dataverse.
Use Cases
- Analyze the distribution of MQM and ATA error labels across 92 exams to identify common translation pitfalls.
- Correlate ATA severity values with specific error labels to understand grader assessment patterns.
- Compare source passage segments with target text segments to study error contexts and translation choices.
- Examine grader ID patterns in relation to severity levels to assess scoring consistency.
Strengths
- Data covers 92 distinct exams from a professional certification process.
- Includes multiple annotation schemes (MQM and ATA labels) for error classification.
- Contains both source and target text segments, enabling direct comparison of errors.
Limitations
- The dataset is limited to a single year (2006) and language pair (Spanish-to-English).
- Sample size is modest with only 92 exams, which may limit statistical power for some analyses.
- The anonymization of exam and grader IDs prevents linking to external demographic or performance metadata.
Provenance
- Source
- Harvard Dataverse
- Collection Method
- Collected from the 2006 ATA Spanish-to-English translation certification exam grading process.
- Time Range
- 2006
- Freshness
- null
- Geography
- null