bezoku

An on-device, offline, trained-from-scratch dependency-parser grammar app — for learners and linguists. No pretrained embeddings, no external tokenizer, every prediction traceable to gold-annotated Universal Dependencies data.

Download

Direct download — no app store required. Pick your language, then the build that matches your device:

legacy = Android 5.0+ (minSdk 21), covers the widest range of devices. modern = Android 7.0+ (minSdk 24), 16 KB memory-page compliant for newer devices (Android 15+). Not sure which to pick? Use legacy.

Languagelegacymodern
Deutsch (German)DownloadDownload
Беларуская (Belarusian)DownloadDownload
Português (Brazilian Portuguese)DownloadDownload
Català (Catalan)DownloadDownload
Hrvatski (Croatian)DownloadDownload
Gaeilge (Irish Gaelic)DownloadDownload
NaijaDownloadDownload
Gàidhlig (Scottish Gaelic)DownloadDownload
سنڌي (Sindhi)DownloadDownload
Slovenščina (Slovenian)DownloadDownload
Español (Spanish)DownloadDownload
Cymraeg (Welsh)DownloadDownload

More languages (Turkish, Tagalog, Indonesian, Wolof, and others) are in progress — see the roadmap in our public repository.

What is Bezoku?

Bezoku parses your sentences into a full dependency tree — part of speech, grammatical relations, and morphological features — entirely on your device, with no internet connection required. It's built for two audiences: learners who want structured, gamified practice, and linguists who want an inspectable, gold-annotated parser they can trust.

Licensing & attribution

The Bezoku app and neural model are proprietary. Every language model is trained exclusively on its own Universal Dependencies CoNLL-U treebank, licensed CC BY-SA 4.0 (Irish Gaelic: CC BY-SA 3.0). Full per-language creator, copyright, and source-link attribution is shown in-app on the About screen for the language you're using, per the license deed at creativecommons.org/licenses/by-sa/4.0.

Legal