Parapaper

Open source · In development開源 · 開發中

Parapaper

Bilingual translation for academic papers that keeps layout and math intact. 學術論文雙語對照翻譯，保留排版與公式。

Drop in a PDF and get a side-by-side reader, plus a bilingual and a translation-only PDF with the original layout, formulas and figures in place. It runs on your own computer, with your own API key. 拖進一份 PDF，就能得到左右對照的閱讀器，以及保留原始排版、公式與圖表的雙語 PDF 和純譯文 PDF。它在你自己的電腦上執行，使用你自己的 API key。

View on GitHub在 GitHub 上查看 Run it yourself自行架設

The app as it is today目前的實際畫面

Parapaper's reader. Left: the English paper with its abstract highlighted and a tooltip reading biomarkers → 生物標記. Right: the Traditional Chinese translation with the same paragraph highlighted.
Interactive reader.互動閱讀器。 Clicking a word pairs its paragraph with the translation and glosses the word in context — here, biomarkers → 生物標記. 點一下單字，就會把該段落與譯文配對，並依上下文顯示釋義——這裡是 biomarkers → 生物標記。 Source:來源： Qi et al., Scientific Data 11:210 (2024) · CC BY 4.0 · machine-translated into Traditional Chinese已機器翻譯為繁體中文
Parapaper's start screen: original language English, translation 繁體中文, engine DeepL or Claude, and a drop zone for a PDF.
Upload and translate.上傳翻譯。 Pick the languages and the engine, then drop a PDF. 選好語言和翻譯引擎，拖進 PDF 即可。
A page of the bilingual PDF: the original English page with a multi-panel figure on the left, the same page translated into Traditional Chinese on the right, with the figure and layout unchanged.
Bilingual PDF.雙語 PDF。 An original page and its translated page. Figures, columns and page furniture stay where they were. 原文頁與它的翻譯頁。圖表、分欄和頁首頁尾都留在原位。 Source:來源： Qi et al., Scientific Data 11:210 (2024) · CC BY 4.0 · machine-translated into Traditional Chinese已機器翻譯為繁體中文

Screenshots show Qi, Y. et al. “A comprehensive database of exosome molecular biomarkers and disease-gene associations.” Scientific Data 11, 210 (2024), doi:10.1038/s41597-024-03015-7, licensed under CC BY 4.0. Changes: translated into Traditional Chinese by Parapaper using the DeepL engine. 截圖中的論文為 Qi, Y. 等，〈A comprehensive database of exosome molecular biomarkers and disease-gene associations〉，Scientific Data 11, 210 (2024)，doi:10.1038/s41597-024-03015-7，採 CC BY 4.0 授權。修改：由 Parapaper 以 DeepL 引擎翻譯為繁體中文。

What it does功能

  • Layout and formulas kept保留排版與公式

    Translation runs through pdf2zh (PDFMathTranslate), which puts the translated text back onto each page, so formulas, figures and columns stay where they were. 翻譯透過 pdf2zh（PDFMathTranslate）進行，譯文會排回每一頁的原位，公式、圖表和分欄都維持原樣。

  • Claude, with a glossaryClaude 後端與術語表

    Before translating, Claude reads the opening of the paper (up to 12,000 characters) and builds a terminology table and a short summary. Both go into the prompt for every paragraph, to keep key terms consistent from start to finish. 翻譯前，Claude 會先讀論文開頭（最多 12,000 字元），建立術語表和簡短摘要，並放進每個段落的翻譯提示中，讓關鍵術語從頭到尾保持一致。

  • Or DeepL也可選 DeepL

    Choose DeepL or Claude for each upload. A free DeepL API key is enough to get started. 每次上傳都能選擇 DeepL 或 Claude。用 DeepL 免費版 API key 就能開始。

  • Interactive reading and lookup互動閱讀與查詞

    Click a paragraph to pair it with its translation. Hover or click a word for a meaning picked from its paragraph’s context (word lookup uses your DeepL key). Mark up the original with pen and highlighter. 點一下段落就和譯文配對；滑過或點一下單字，會依段落上下文顯示釋義（查詞使用你的 DeepL key）。也能用畫筆和螢光筆在原文上標註。

  • Offline reading離線閱讀

    Download a paper to the browser and it still opens with the server stopped. Words you looked up before still gloss, and new marks wait on the device until the server is back. 把論文下載到瀏覽器後，即使伺服器關閉也能開啟。查過的單字仍會顯示釋義，新畫的標註會先存在裝置上，等伺服器恢復後再同步。

  • 16 interface languages16 種介面語言

    English, 繁體中文, 简体中文, 日本語, 한국어, Deutsch, Français, Español, Italiano, Português, Nederlands, Polski, Русский, Türkçe, Tiếng Việt and Bahasa Indonesia. 英文、繁體中文、簡體中文、日文、韓文、德文、法文、西班牙文、義大利文、葡萄牙文、荷蘭文、波蘭文、俄文、土耳其文、越南文與印尼文。

How it works運作方式

  1. Start it on your computer在自己的電腦上啟動

    Run start.bat on Windows or ./start.sh on macOS and Linux. The first run sets everything up; then the app opens at http://localhost:8000. Windows 執行 start.bat，macOS 與 Linux 執行 ./start.sh。第一次會自動完成安裝，之後在 http://localhost:8000 開啟。

  2. Drop a PDF拖進 PDF

    Choose the languages and DeepL or Claude, then drop the paper in. Layout analysis runs on your machine; only the text is sent to the engine you picked. 選好語言和 DeepL 或 Claude，把論文拖進來。版面分析在你的電腦上執行，只有文字會送到你選的翻譯引擎。

  3. Read side by side左右對照閱讀

    Read in the interactive reader, or download the bilingual and translation-only PDFs. Opening a paper again costs nothing: finished translations are reused. 在互動閱讀器中閱讀，或下載雙語與純譯文 PDF。重新開啟已翻譯的論文不會再花費額度，完成的翻譯會直接沿用。

Status狀態

  • Open source · AGPL-3.0開源 · AGPL-3.0
  • In development開發中
  • Claude vs DeepL evaluation in progressClaude 與 DeepL 評測進行中

Parapaper is open source under the AGPL-3.0 and still in development. A comparison of Claude and DeepL translations is under way. Parapaper 以 AGPL-3.0 開源，仍在開發中。Claude 與 DeepL 翻譯品質的比較評測正在進行。

Run it yourself自行架設

Parapaper runs on your own computer. From a clone you need Node.js 20 or newer to build the interface once; Python and the rest are set up for you. The guide on GitHub covers Windows, macOS and Linux. Parapaper 在你自己的電腦上執行。從 GitHub 下載原始碼時，需要 Node.js 20 以上來建置一次介面；Python 和其他套件會自動安裝。GitHub 上的說明涵蓋 Windows、macOS 和 Linux。

git clone https://github.com/LittleBlack0001/parapaper.git
cd parapaper
./start.sh          # Windows: start.bat
Installation guide on GitHubGitHub 上的安裝說明