Paper CN Reader: Academic Paper Translation and Annotation

Problem

English paper PDFs often mix translation, figures, equations, and page context. Summary-style translation can omit constraints and experimental details. paper-cn-reader is aimed at close reading: it turns a paper into a self-contained Chinese HTML file and PDF while preserving original visual elements and adding concise AI annotations.

How It Works

• Parsing and extraction: detects sections, single/double-column reading order, images, tables, and equations, then exports figures to assets/; vector graphics can be captured by region.
• Faithful translation and annotation: translates sentence by sentence, uses Term (English Term) on first appearance, keeps model, dataset, and variable names in English, and reserves annotations for key contributions, structures, equations, tables, and ablations.
• Generation and QA: builds a MathJax-enabled page from template.html and styles.css, with navigation and print CSS, then exports an A4 PDF and checks completeness, translation, figures, equations, and annotations.

Boundaries

It does not replace the source or invent conclusions. Unclear text is marked, extraction failures are disclosed, and speculative AI analysis must stay in the annotation area rather than the translated body.