PDF Annotation & Transcription Experts – Japanese

The position of PDF Annotation & Transcription Expert – Japanese is offered by Mercor, a company that creates training data for document‑understanding models, especially for languages that current systems struggle with. The project concentrates on Japanese material, gathering real‑world PDFs such as newspapers, textbooks, exams, flyers and manuals, and turning each page into a fully annotated structural map together with an exact transcription of every text element. Each day you will locate a publicly available Japanese PDF that contains at least one multimodal feature—such as an image, table, diagram or handwritten segment—and note its source. You will then delineate every meaningful region on the page, assign it a component type, and give it a reading‑order index. Relationships between regions and their parent figures or tables must be recorded, and all visible text—including kanji, hiragana, katakana, furigana and any handwritten characters—must be reproduced verbatim, flagging any illegible parts. Page‑level metadata like language, document type and dimensions are also captured. After completing a task, a second Japanese specialist reviews the work end‑to‑end, and seasoned annotators may perform these reviews themselves. Candidates must be native Japanese speakers with complete mastery of kanji, hiragana, katakana and variant character forms. Relevant professional background includes interpretation, journalism, transcription, translation, editorial duties or comparable document‑heavy roles. The job demands character‑level precision, a systematic approach to applying a fixed taxonomy across many pages, and comfort with unconventional layouts such as vertical text, multi‑column newspapers or handwritten forms. Additional experience that is advantageous includes work on AI training data annotation, bilingual quality evaluation, subtitling, OCR correction, typesetting, copy‑editing, or any tasks involving Unicode normalization and Japanese input methods. Success is measured by capturing every region accurately, preserving the true reading order, delivering flawless transcriptions, and passing the peer review on the first attempt, while contributing diverse page layouts to the growing corpus.