Korean PDF Annotation Expert for AI Training Data

Mercor; Wien; Vor Ort; EUR 42.000 - 64.000; Vollzeit

Zusammenfassung

Mercor is seeking a meticulous Korean annotator to build training data for Korean documents. You will identify and bound every meaningful region on a Korean page, ensure correct reading order across columns, and transcribe text in Hangul with high accuracy.

The role involves handling diverse layouts, including handwriting and tables, with cross-checks by a second expert. Ideal candidates are native Korean speakers with experience in bilingual transcription or editorial work, who are precise and

Qualifikationen

• Must be a native Korean speaker with full command of Hangul and hanja where it appears in documents.
• Character-level accuracy is required; one wrong jamo is a defect.
• Experience in bilingual transcription, editorial work, or AI training data is preferred.

Aufgaben

• Source publicly accessible Korean PDFs with multimodal elements.
• Annotate page structure: title, headings, paragraphs, lists, tables, figures, captions, formulas, questions, answers with types and reading-order indices.
• Link regions to related figures or tables via parent component identifiers.
• Transcribe faithfully in Hangul, marking unreadable regions.
• Capture page metadata: language, document type, source, page dimensions, and flags for tables, formulas and handwriting.
• Undergo peer review by a second Korean expert.

Kenntnisse

Native Korean

Hangul proficiency

Attention to detail

Tools

OCR tools

Jobbeschreibung

Mercor is seeking a meticulous Korean annotator to build training data for Korean documents. You will identify and bound every meaningful region on a Korean page, ensure correct reading order across columns, and transcribe text in Hangul with high accuracy.

The role involves handling diverse layouts, including handwriting and tables, with cross-checks by a second expert. Ideal candidates are native Korean speakers with experience in bilingual transcription or editorial work, who are precise and