Save is a Shavian transliteration tool. Why does the world need yet another Shavian transliterator, I hear you ask? Well, this one isn't like the other transliterators. Inspired by Dave Coffin's transliterator, it uses rules to figure out the words it doesn't know yet. Also like his tool, it uses a ML engine to disambiguate words like 'read' and 'lead' by using part-of-speach (POS) tagging.

Unlike Dave's tool, shave has an 'after-burner' in the form of a word-sense-disambiguator. This is another ML that I trained specifically for this transliterator, on ambiguous cases that are hard in Shavian, like 'bow' and 'tear'. It also, quite uniquely, is quite honest about what it does and doesn't know. Each decision it makes is assigned a 'certainty', and if it isn't sure, it can surface multiple options to you so _you_ can choose.

Oh, and it also has a reverse transliteration mode. Try it in the 'text' mode on this page. It too uses some ML tricks to narrow down the odds and decide whether "I" is more likely than "Eye". It's fresh from the press, and still under active development, but you will find that it's already pretty usable.

The underlying engine is a work in progress, still evolving. The rules are being tuned and expanded, the vocabulary (Readlex + some supplements of my own) is ever growing.

My primary driver in building this transliterator has always been to read more in Shavian. I have been using this tool to transliterate e-books for a long time now. I hope you enjoy this newly developed front end around my toolchain as much as I have been doing!

Engine version: run-20260504215001