PDF Annotation & Transcription – Korean at Mercor

Korean PDF Annotation & Transcription Expert | $33.58/hr | Remote (South Korea, Japan, India, US, Canada, or Western Europe)

Mercor is building high-quality document AI training data for Korean and other underrepresented scripts. This role involves sourcing real Korean PDF documents, annotating their structural components, and producing faithful character-level transcriptions in Hangul — all human-authored, no model-generated content.

What You'll Do

• Source publicly accessible Korean PDFs across assigned document types (newspapers, textbooks, forms, menus, manuals, etc.) that include at least one multimodal element such as tables, diagrams, or handwriting
• Annotate every meaningful page region — titles, headings, paragraphs, lists, tables, figures, captions, formulas, and answer fields — assigning component types and reading-order indices
• Record parent-child relationships between regions and their associated figures or tables
• Transcribe all text exactly as it appears in Hangul, including hanja and handwritten content, flagging illegible regions
• Capture page-level metadata: language, document type, source URL, dimensions, and flags for tables, formulas, and handwriting
• Perform second-expert review of a colleague's completed tasks (experienced annotators)

Who You Are

• Native Korean speaker with full command of Hangul, including hanja in formal or older documents
• Background in bilingual transcription, translation, editorial work, or AI training data — reviewer experience is a plus
• Precise and detail-oriented: character-level accuracy is the primary standard; a single incorrect jamo is considered a defect
• Systematic and consistent: able to apply a structured taxonomy across hundreds of pages
• Comfortable with complex layouts: multi-column newspapers, exam papers, handwritten forms

Nice-to-Have Experience

• AI training data: annotation, labeling, grading, or bilingual evaluation
• Transcription and localization: MTPE, subtitling, bilingual QA, or OCR post-editing
• Document production: typesetting, copy-editing, proofreading, or digitization of Korean materials
• Script and encoding: Unicode normalization, Hangul jamo composition, Korean input methods, hanja handling

What Success Looks Like

• All meaningful page regions are captured, correctly bounded, and correctly typed
• Reading order accurately reflects how the page is read, including across columns
• Transcriptions match the source character for character in Hangul (not romanization)
• Tasks pass second-expert review on the first submission
• Sourced documents contribute genuine layout diversity to the corpus

This is a remote contractor role open to contributors based in South Korea, Japan, India, the United States, Canada, or Western Europe only.