I built a free context-aware furigana tool for Japanese text, PDFs, and ebooks
Hi everyone,
I wanted to share a free Japanese reading tool I’ve been building called EZFurigana:
It adds furigana to Japanese text, but the main thing I’ve been focusing on is making the readings more context-aware rather than just applying a dictionary reading everywhere.
For example, simple furigana tools can struggle with words where the reading changes depending on context, such as:
- 市場: いちば / しじょう
- 大分: おおいた / だいぶ
- 人気: にんき / ひとけ
- 最中: さいちゅう / さなか / もなか
- 方: かた / ほう
EZFurigana combines Sudachi, expanded dictionary coverage, custom rules, and a lightweight ModernBERT ambiguity classifier to choose readings from sentence context, not dictionary lookup alone. It is not perfect, especially with names, unusual compounds, and ambiguous phrases, but I’m trying to make it more useful for real reading rather than just short example sentences.
I also ran an internal benchmark to check whether the context-aware approach was actually helping. Each engine was tested on the same evaluation set. The table below reports 1,000 manually checked target readings drawn from a larger 7,500-line Japanese text set.
| Tool | Wrong readings | Accuracy |
|---|---|---|
| EZFurigana | 12 / 1,000 | 98.8% |
| Sudachi | 54 / 1,000 | 94.6% |
| MeCab | 61 / 1,000 | 93.9% |
| Yahoo API | 98 / 1,000 | 90.2% |
This is still just one benchmark, so I don’t want to overclaim from it, but it was encouraging and helped me find cases where dictionary-based readings break down.
I’ve also been working on file-based reading tools:
- PDF furigana converter for Japanese PDFs and documents
- EPUB ebook reader for reading Japanese ebooks with furigana
- Browser extension for adding furigana while reading Japanese web pages
- Dictionary popups for checking words while reading
- JLPT-level filtering and reading settings
The tool is free and does not require signup. My main goal is to help learners move from textbook/kanji study into native Japanese material more smoothly.
I’d especially appreciate feedback on:
- whether the furigana readings feel trustworthy while reading
- examples where the reading is wrong
- whether the PDF or ebook workflows are useful
- what would make it better for reading native Japanese material
Thanks, and criticism is welcome.