Document Annotation Specialist (Japanese) for AI Training Data

About the Role

The role focuses on creating high-quality training data for AI models, specifically for languages like Japanese that are less frequently encountered by parsing and vision-language models. The project involves annotating real, publicly available PDF pages to produce a structural map and transcription of every text region.

The dataset emphasizes challenging document types such as handwritten content, dense multi-column layouts, vertical text, tables, diagrams, and mixed-script pages. Documents include newspapers, textbooks, examinations, flyers, forms, manuals, menus, brochures, notices, and worksheets to ensure a diverse and realistic representation of Japanese documents.

All annotation and transcription work is performed by humans, ensuring accuracy and consistency.

The Role

What You'll Do

• Open and verify tasks by confirming the page is in Japanese, legible, contains real content, and does not include personal details
• Annotate page structure by identifying and bounding regions such as document titles, section headings, paragraphs, lists, tables, figures, diagrams, captions, formulas, questions, and answer fields. Assign each region a component type and reading-order index
• Record relationships between regions, such as linking a region to its parent figure or table
• Transcribe text faithfully, including kanji, hiragana, katakana, furigana, and handwritten content. Flag any illegible regions
• Capture page metadata including language, document type, source, page dimensions, and flags for tables, formulas, and handwriting
• Review a colleague’s work to ensure quality and accuracy

Requirements

Who You Are

• A native Japanese speaker with full proficiency in kanji, hiragana, katakana, furigana, and variant character forms
• Professional experience in interpretation, journalism, transcription, translation, editorial work, or similar document-intensive fields
• Highly detail-oriented with a focus on character-level accuracy over speed
• Systematic and consistent in applying a taxonomy across numerous documents
• Comfortable with diverse and unfamiliar document layouts, including vertical text and multi-column formats

Nice-to-Have Specialties

• Experience with AI training data annotation, labeling, grading, or bilingual evaluation
• Transcription and localization skills, including MTPE (Machine Translation Post-Editing), subtitling, bilingual QA, OCR correction, or post-editing
• Document production expertise such as typesetting, copy-editing, proofreading, or digitization of Japanese-language material
• Familiarity with script and encoding, including Unicode normalization, Japanese input methods, full-width and half-width forms, and kanji variant handling

What Success Looks Like

• Every meaningful region on the page is accurately bounded and typed
• Reading order reflects the actual reading sequence, including vertical text and multi-column layouts
• Transcriptions match the source text character-for-character in Japanese script, not romaji
• Tasks pass second-expert review on the first attempt
• Unsuitable pages are identified and flagged immediately rather than after extensive work