Quality Control & Accuracy in Transliteration Flashcards

A transliteration QC lead is designing an acceptance sampling plan for a 5,000-name batch. Which factor most directly determines the required sample size?

Answer: The acceptable quality level (AQL) and confidence level specified by the client

Statistical sampling theory ties sample size to the AQL threshold and the desired confidence level, which are set by the client's quality requirements.

A US court interpreter's transliterated transcript is disputed because a witness's name was rendered differently from the police report. The QC specialist should first:

Answer: Determine which romanization standard each document used and whether both renderings are compliant

Different agencies may use different standards; establishing each document's governing standard is the first step before judging whether either rendering is erroneous.

A QC specialist is reviewing transliterated Hebrew text for a US genealogical database. The shin (ש) was romanized as 'sh' in most entries but as 's' in a few. The best QC action is:

Answer: Standardize all entries to 'sh' per the ALA-LC Hebrew romanization table

The ALA-LC romanization standard distinguishes shin (sh) from samekh (s), and a genealogical database requires consistent application to enable reliable name searches.

Error propagation in a transliteration workflow occurs when:

Answer: A translator makes a single error that is then multiplied by copy-paste into multiple records

Error propagation happens when a single initial mistake is duplicated into many records, dramatically amplifying the impact of one error in batch workflows.

A QC analyst must evaluate transliteration of names containing the Urdu retroflex 'ڈ'. Which approach best reflects best practice?

Answer: Use the prescribed mapping from the applicable standard (e.g., ALA-LC or BGN/PCGN) and flag deviations

Standards like ALA-LC and BGN/PCGN provide explicit mappings for retroflex consonants; QC must verify that the transliterator followed the correct mapping.

In a transliteration QC workflow, a 'post-edit' differs from a 'correction' in that:

Answer: Post-edits involve improving machine-generated output while corrections address human transliterator errors

Post-editing refers specifically to improving machine or automated transliteration output, while correction addresses errors in human-produced transliterations.

A QC coordinator notices that error rates spike on names from a specific source script every Monday morning. The most productive investigative step is:

Answer: Analyze whether a specific transliterator handles that script on Mondays or if a system configuration changes on Mondays

A pattern tied to a specific time or day often signals a human factor (such as a particular staff member's schedule) or a system factor (such as a batch setting), both of which require root cause analysis.