A clean retake on a procedural generation technique developed to generate vast amounts of interesting, unique names for a procedural universe. Designed to run very fast, very lean, ’themed’ on underlying language styles. Ideal for naming your next dropshipping company! Enatozaba! Zero Google search results, apart from probably this page soon!
An offline training process digests a corpus of name examples, such as an enormous list of Japanese place names. These are processed into trigram frequency tables. A code generation phase produces a set of C++ decision sequencers (lookup tables and vast switch statements) for building results a character at a time.
A full write-up of how it works is on the GitHub page, linked above.
There is also a basic Javascript code generator backend provided, the results of which are fairly trivial to embed into a webpage. The following demo uses two separate frequency tables, one trained from rōmaji and one from a mixture of English names and simple words. It has some basic ‘fix-up’ to improve the flow of the results but is missing some more advanced tuning and post-processing, so the output can be quite variable.