Skip to content

Acknowledgements

Built with
a little help.

Hamburger Keyboard is closed-source, but it is not built from nothing. This page credits the public data its correction runs on and the one piece of open-source software it ships.

Language data.

Correction is built from word frequency and n-gram tables, and lists of proper names: statistics about which words and names are common, and how words follow each other. Those tables ship as derived numbers and word lists, not as a copy of the original text.

How each source is used.

No source above does all the work by itself, and no source text ships with the app.

Word frequency comes from FrequencyWords. Sentence rhythm and which words tend to follow which come from Wikipedia, Simple English Wikipedia, and Tatoeba. Recognizing proper names, and telling a name from an ordinary word, draws on OpenSubtitles, FineWeb, Webis-TLDR-17, and the Enron email corpus. A small amount of Common Crawl text helps the app tell contractions like it's apart from possessives like its. A short ranking model is trained on Tatoeba sentences, and the built-in practice passages are either written by hand or drawn from public-domain text credited where they appear.

FrequencyWords and Wikipedia are share-alike licensed (CC BY-SA), which means anything adapted from them has to be offered back under the same terms. The word-frequency and n-gram data files adapted from those sources ship as separate files alongside the app, available under CC BY-SA 4.0. The app's own code and the rest of what you buy are unaffected and stay under Hamburger Keyboard's own license.

Software.

Update checks run on Sparkle, MIT-licensed. Sparkle also bundles bsdiff/bspatch, sais-lite, a portable Ed25519 implementation, and SUSignatureVerifier.m, each under its own permissive license; the full text is on GitHub.

Aside from Sparkle, the app links no other third-party code.

Something missing or wrong here?

Email [email protected]