wiktionary-to-kindle

Turns Wiktionary into a MOBI dictionary your Kindle can use for tap-to-define — in any of a hundred-odd languages, for any Wiktionary edition kaikki.org publishes.

Available as a desktop app that bundles its own Java runtime, or as a command-line tool.

Install

The desktop app is the recommended way to run this — it bundles its own Java runtime, so there is nothing else to install.

macOS

brew install --cask nyg/tap/wiktionary-to-kindle

On macOS the app opens themed with AtlantaFX's Cupertino themes, following the system appearance and switching between light and dark as you do. Preferences has a Theme setting — JavaFX or AtlantaFX — Cupertino — that applies the moment you press OK, so either look is available on any platform; only the default differs.

Windows

scoop bucket add nyg https://github.com/nyg/scoop-bucket

scoop install wiktionary-to-kindle

Scoop installs per-user, needs no admin rights, and avoids the SmartScreen prompt.

Without Scoop, download wiktionary-to-kindle-<version>-windows-x64.exe from the latest release and double-click it. It also installs per-user and needs no admin rights, and it adds a Start Menu shortcut and a normal entry under Installed apps. Being unsigned, it does trip SmartScreen on first run — "Windows protected your PC", then More info and Run anyway.

Linux

sudo apt install ./wiktionary-to-kindle-<version>-linux-x64.deb

sudo dnf install ./wiktionary-to-kindle-<version>-linux-x64.rpm

Both install under /opt/wiktionary-to-kindle and register a desktop entry, so the app appears in your application menu. Remove it with sudo apt remove wiktionary-to-kindle or sudo dnf remove wiktionary-to-kindle.

On a distribution neither package suits, run the portable JAR instead, which needs Java 25 installed. See Command line.

Manual download

Grab the latest release:

File	For
wiktionary-to-kindle-<version>-macos-arm64.dmg	macOS (Apple Silicon)
wiktionary-to-kindle-<version>-windows-x64.exe	Windows — per-user installer, no admin rights
wiktionary-to-kindle-<version>-windows-x64-scoop.zip	Windows, no install — what Scoop consumes; extract it and run Wiktionary to Kindle.exe in place
wiktionary-to-kindle-<version>-linux-x64.deb	Debian, Ubuntu and derivatives
wiktionary-to-kindle-<version>-linux-x64.rpm	Fedora, RHEL, openSUSE and derivatives
wiktionary-to-kindle-<version>.jar	Any other distribution, or scripted use anywhere. Needs Java 25

The app is ad-hoc signed but not notarized, so macOS quarantines the DMG download and blocks the first launch — you may see "is damaged" or "Apple could not verify…", which both mean the same thing and do not mean the app is corrupted. Homebrew handles this for you. If you installed the DMG by hand, clear the flag once:

xattr -dr com.apple.quarantine "/Applications/Wiktionary to Kindle.app"

Using the app

1. Pick a Wiktionary edition — which Wiktionary the definitions are written in. The Greek Wiktionary defines words in Greek.
2. Pick a word language — which language's words to include. Any language may appear in any edition.
3. Press Build dictionary.

So edition: Greek, word language: English gives you an English–Greek dictionary: look up an English word, read the definition in Greek. The title updates as you choose, so you can see what you are about to build.

The first build for an edition downloads its dump, which is 100 MB to several GB. Progress and a live log are shown throughout, and Cancel stops the run cleanly at any stage. Subsequent builds reuse the downloaded dump.

When it finishes, Show in folder reveals the .mobi — w2k-dictionary-en-el.mobi for the example above. Copy it to your Kindle over USB, or send it to your device's Kindle email address.

The Dumps tab lists what has been downloaded, with the Wiktionary edition, generation date and size, and lets you delete dumps you no longer need — worth checking, since they are the largest thing this app puts on your disk.

On your Kindle

Once the .mobi is on the device, set it as the default dictionary for its language, and tap-to-define uses it everywhere you read. Titles are prefixed W2K, so the dictionaries this app builds group together in that list and stay apart from Amazon's own.

Where files go

What	Location
Dumps and dictionaries	~/Documents/wiktionary-to-kindle/ (change in Preferences; the CLI uses the same folders)
Build leftovers	dictionaries/intermediate/ — the HTML, OPF, NCX and cover files each build feeds to kindling-cli, one sub-folder per dictionary
Settings	~/.config/wiktionary-to-kindle/ — %LOCALAPPDATA%\wiktionary-to-kindle\Config on Windows
Log file	~/.local/state/wiktionary-to-kindle/logs/app.log, rolled daily to app-YYYY-MM-DD.log.gz and kept 14 days — %LOCALAPPDATA%\wiktionary-to-kindle\State on Windows
Cached kindling-cli	~/.cache/wiktionary-to-kindle/ — %LOCALAPPDATA%\wiktionary-to-kindle\Cache on Windows

macOS included, these follow the XDG Base Directory specification and honour XDG_CONFIG_HOME, XDG_CACHE_HOME and XDG_STATE_HOME. Dumps and dictionaries deliberately sit in your documents folder instead — they are multi-gigabyte and the .mobi has to be found and copied to a Kindle, so both belong somewhere a file manager shows by default. That folder is still resolved the XDG way, via XDG_DOCUMENTS_DIR.

Preferences also lets you delete the build leftovers as soon as a dictionary is built, point at a pre-installed kindling-cli, or pin a specific version. Maximum heap is shown there but cannot be changed: it is fixed when the app starts, and the bundle sizes it at 75% of your machine's RAM.

Command line

The same pipeline, for scripting and for Linux. Every release ships a portable JAR that runs anywhere with Java 25.

# Download a dump to the dumps folder (skipped if one for that edition already exists)
java -jar wiktionary-to-kindle-<version>.jar download el

# Generate into the dictionaries folder
# DUMP_LANG = which Wiktionary edition to read; WORD_LANG = ISO 639-1 filter
java -jar wiktionary-to-kindle-<version>.jar generate el en

# Read from and write to somewhere else, for one invocation
java -jar wiktionary-to-kindle-<version>.jar download el --dumps-dir ./dumps
java -jar wiktionary-to-kindle-<version>.jar generate el en --dumps-dir ./dumps --dictionaries-dir ./dictionaries

# Pin a kindling release, or use one already installed
java -jar wiktionary-to-kindle-<version>.jar generate el en --kindling-version vX.Y.Z
java -jar wiktionary-to-kindle-<version>.jar generate el en --kindling-cli /usr/local/bin/kindling-cli

java -jar wiktionary-to-kindle-<version>.jar --help
java -jar wiktionary-to-kindle-<version>.jar --version

dl and gen are short aliases. The CLI reads the same preferences file as the desktop app, so both front-ends share one dumps folder and one dictionaries folder: download in the app and generate from the command line, or the reverse, and the dump is found either way. --dumps-dir and --dictionaries-dir override that per invocation, for scripts that want their own working folders. download exits non-zero if the transfer fails.

Up to and including 2.0.3 the CLI resolved dumps/ and dictionaries/ relative to the working directory and ignored the app's preferences. To keep that behaviour, pass --dumps-dir ./dumps --dictionaries-dir ./dictionaries, or move the existing folders into ~/Documents/wiktionary-to-kindle/.

How it works

1. A kaikki.org pre-extracted Wiktionary JSONL dump is downloaded for the chosen edition. Dumps are produced weekly by wiktextract and include all languages with Lua templates fully expanded.
2. The compressed JSONL is streamed and filtered by language. Each entry's senses are rendered into an HTML definition, its inflected forms are collected as Kindle lookup targets, and the result is grouped in memory by normalised key.
3. Chunked MobiPocket HTML files, an NCX navigation map and an OPF manifest are written.
4. On first run, kindling-cli is downloaded and cached, verified against a pinned SHA-256.
5. kindling-cli build converts the OPF into a .mobi Kindle dictionary.

docs/program-flow.md has sequence diagrams for each stage.

Inflected forms

Wiktionary entries on kaikki include a forms array listing every inflected form of the lemma — plurals, declension cases, conjugations, gender-agreement forms. wiktionary-to-kindle exposes those forms in two ways:

• Lookup index — every form becomes a tap-to-lookup target on Kindle via <idx:iform> markup. Looking up συντρόφους (the accusative plural) resolves to the lemma σύντροφος. Emitted for every part of speech, including verbs.
• Visible paradigm table — for non-verb entries, a small table of the form {tag-abbrev}: {article} {form} is appended to the definition. Verb entries skip the table (Greek, Romance and Slavic verbs have 50–200+ forms, which would dwarf the definition). A forms.size() > 30 safety net also skips the table for pathological non-verb cases.

Gender-equivalent cross-references (e.g. συντρόφισσα listed as a feminine equivalent of σύντροφος, or ingénieure listed under ingénieur) are filtered out of the lookup index by a language-agnostic post-pass: any form whose normalised text already exists as a standalone headword in the same dump is dropped from the iform list, so long-press on συντρόφισσα resolves to its own entry instead of being shadowed by the lemma. The visible "Forms:" table in each entry's HTML body is untouched — readers still see the full paradigm (e.g. ingénieure, ingénieurs) under the lemma.

See docs/form-of-folding.md for the full design.

Examples of generated dictionaries

• English-English (96MB)
• French-English (27MB)
• Greek-English (6MB)

Building from source

Requires Java 25 and Apache Maven.

mvn package        # runs tests, produces target/wiktionary-to-kindle-<version>.jar
mvn javafx:run     # runs the desktop app

To build a native bundle for the host platform:

scripts/package.sh          # defaults per platform
scripts/package.sh deb rpm  # or name the types: dmg, app-image, exe, msi, deb, rpm

It jlinks a trimmed runtime, verifies it — TLS, locale data, the JavaFX modules and ImageIO all have to resolve inside the fresh image, and packaging fails if any of them do not — then runs jpackage once per type named. Output lands in target/dist. Packages can only be built for the platform you are on, and the Windows exe additionally needs the WiX Toolset on the PATH.

The release workflow ships the macOS dmg, the Windows exe and Scoop app-image, and the Linux deb and rpm. Per-PR CI only exercises the dmg and the app-image, so a regression in the other three surfaces when a release is cut rather than in review.

The shaded JAR deliberately excludes JavaFX so it stays cross-platform: JavaFX resolves to native, platform-specific artifacts, and the desktop app gets them from its bundled runtime instead.