An incorrect change introduced during the manual or automated conversion of text or metadata from a source into another representation (for example manual typing, OCR, automated parsing, or format conversion) that alters content, structure, semantics, encoding, or identifiers in a way that degrades data quality, discoverability, or interpretability.

Principle

Transcription errors propagate downstream: an uncorrected error in a primary record will be copied into derivative records, indexes, and analyses, producing systematic quality degradation unless detected and remediated.

Demonstration

Illustrative scenario — Situation: OCR processing of a late-19th-century directory misreads '1888' as '1988'. Recognition: a cataloger notices temporal inconsistency when comparing entries. Action: the cataloger verifies the original scan and corrects the date and records an edit with provenance. Consequence: search accuracy and chronological analyses are restored for that record.

Misapplication

Treating machine-generated transcriptions as authoritative without sampling or validation. The error is assuming automated output equals correct encoding rather than a probabilistic result that requires quality control.

Consequence

Transcription errors cause misindexing, failed or imprecise search results, incorrect metadata-driven decisions (such as mis-sorting or mis-filtering), compromised research findings, and increased curation costs to detect and correct propagated errors.

Reversal

When transcription is accompanied by confidence scores, provenance of the transcription process, human review, or retention of original source images, the negative impact of errors can be detected and mitigated; conversely, deliberate normalization or editorial correction documented as such is not a transcription error.

Boundary

Clearly within: a misread year or name produced by OCR that changes meaning (e.g., 'Smith' → 'Srnith' producing failed lookups). Boundary case: minor typographical variants that do not affect discoverability. Clearly outside: intentional editorial modernization or enrichment that is documented and reversible.

Semantic Tension

Trade-off between scale/efficiency (favoring automated bulk transcription) and accuracy/curation (favoring human review and verification).

Synthesis

A transcription error is an operational data-quality failure whose management requires both provenance of the transcription process and processes (sampling, confidence metrics, human review) that enable detection and efficient correction before errors propagate.