libkuraji

日本語

Japanese Braille translator originally developed for NVDAJP.

Translates mixed Kanji/Kana Japanese text into Japanese 6-dot braille (Unicode braille patterns), with morphological word segmentation (masuake / spacing). Returns a position map from each braille cell back to the source text.

Requirements

• Python 3.10 or later
• pip install libkuraji[integration] (fugashi)
• Environment variable LIBKURAJI_INTEGRATION=1 (enables the JTalk extended dictionary)

Installation

pip install 'libkuraji[integration]'
export LIBKURAJI_INTEGRATION=1   # Windows PowerShell: $env:LIBKURAJI_INTEGRATION=1

On macOS and Homebrew Python, system-wide installs may be restricted. Use a virtual environment in that case.

python3 -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install 'libkuraji[integration]'
export LIBKURAJI_INTEGRATION=1

For development (from the repository)

git clone https://github.com/nishimotz/libkuraji.git
cd libkuraji
python3 -m venv .venv
source .venv/bin/activate
pip install -e '.[dev,integration]'
export LIBKURAJI_INTEGRATION=1

In zsh, quote the extras: '.[dev,integration]'.

Quick start

On first run, dictionary binaries from libkuraji-jtalk-dic are downloaded automatically from GitHub Releases.

kuraji "私は点字を読みます。"
# => ⠄⠕⠳⠄ ⠟⠴⠐⠳⠔ ⠜⠷⠵⠹⠲

import libkuraji

cells, inpos, outpos, cursor = libkuraji.translate_kanji(
    "私は点字を読みます。",
    unicodeIO=True,
)
# => '⠄⠕⠳⠄ ⠟⠴⠐⠳⠔ ⠜⠷⠵⠹⠲'

Pass unicodeIO=True to get Unicode braille (the same format as the CLI).

Input is limited to 65,536 characters by default (InputTooLongError when exceeded). Override with LIBKURAJI_MAX_INPUT_CHARS; set 0 to disable the limit.

Usage

Python API

import libkuraji

cells, inpos, outpos, cursor = libkuraji.translate_kanji(
    "私は点字を読みます。",
    unicodeIO=True,
)
# inpos[i] is the input index for cells[i]
# outpos[j] is the braille index for input position j

To use a custom morphological analyzer, inject it with initialize. The analyzer must implement analyze(text, logwrite) and is_ready().

libkuraji.initialize(analyzer=my_analyzer)
cells, inpos, outpos, cursor = libkuraji.translate_kanji(
    "私は点字を読みます。",
    unicodeIO=True,
)

For NABCC (computer braille) mode: translate_kanji(text, nabcc=True, unicodeIO=True).

Foreign quotation marks and information processing braille

libkuraji automatically detects words containing Latin letters and symbols, wrapping them in one of two braille indicators.

Grade 2 English braille inside foreign quotation marks

Pass a louisTranslate function and louisTableList (e.g. ["en-ueb-g2.ctb"]) to translate_kanji to apply Grade 2 English braille translation to the inner text of foreign quotation marks ⠦...⠴. libkuraji does not provide a louis translate function — the caller must inject it.

import louis

def my_louis_translate(table_list, text, cursorPos=0, mode=0):
    return louis.translate(
        table_list, text,
        cursorPos=cursorPos, mode=mode,
    )

cells, inpos, outpos, cursor = libkuraji.translate_kanji(
    "これは ⠦English text⠴ です。",
    unicodeIO=True,
    louisTranslate=my_louis_translate,
    louisTableList=["en-ueb-g2.ctb"],
)

See docs/encoding.md for output encoding details (Unicode braille vs liblouis dotsIO, and what unicodeIO means).

MeCab setup

Override the dictionary release tag with LIBKURAJI_JTALK_DIC_TAG (default: DEFAULT_DIC_TAG in src/libkuraji/jtalk_dic.py). To skip download, point LIBKURAJI_JTALK_DIC_DIR at an already extracted dictionary directory.

CLI

kuraji "私は点字を読みます。"
⠄⠕⠳⠄ ⠟⠴⠐⠳⠔ ⠜⠷⠵⠹⠲

kuraji "私は点字を読みます。" --positions
{"text": "私は点字を読みます。", "braille": "⠄⠕⠳⠄ ⠟⠴⠐⠳⠔ ⠜⠷⠵⠹⠲", ...}

Main options:

Kana input only (supplement)

If the input is already in katakana, use translate without MeCab.

from libkuraji import translate

translate("ワタシワ テンジヲ ヨミマス。")

kuraji -k "ワタシワ テンジヲ ヨミマス。"

Testing

Default tests (same as CI)

Runs without MeCab. Replays recorded MeCab output from tests/mecabFixture.json to verify translator2.

pip install -e '.[dev]'
pytest

Test data in tests/harness.json and related files follow rules from the Japanese braille guide (Ten'yaku no Tebiki). To re-record tests/mecabFixture.json, use miscDepsJp/jptools/recordMecabFixture.py in the nvdajp repository.

Integration tests with the real dictionary (optional)

Opt-in tests that verify translation against the live JTalk extended dictionary and MeCab.

Prerequisites:

• pip install -e '.[dev,integration]'
• gh CLI (falls back to the GitHub REST API if unavailable)

Run:

export LIBKURAJI_INTEGRATION=1
pytest tests/test_integration.py -q

# Also run exact MeCab output parity checks
export LIBKURAJI_PARITY_CHECK=1
pytest tests/test_integration.py -q

On Windows PowerShell:

$env:LIBKURAJI_INTEGRATION=1
pytest tests/test_integration.py -q

Some integration tests also run in CI (GitHub Actions) on Windows and Linux. test_mecab_fixture_parity (internal MeCab output comparison across versions) is skipped by default and intended for dictionary update verification.

Integration tests also reproduce MeCab output correction via mecab_correct.py (ported from nvdajp's Mecab_correctFeatures under BSD relicensing). The libmecab.dll bundled with fugashi and the nvdajp MeCab used when recording fixtures may differ on a few symbol-only inputs (about 98% match overall). Final braille output consistency is preserved and CI passes.

Relationship to NVDA Japanese

• This library separates and standalone-izes the braille engine (translator1/translator2) from nvdajp. The translator1 equivalent (kana module) is a clean-room rewrite driven by tests; translator2 was ported under BSD relicensing by the copyright holder.
• Decoupling plan: projectDocs/jp/braille-engine-decoupling-plan.md in the nvdajp repository.

License

BSD 3-Clause License. See LICENSE.