subsetpark/welsh-mutation

This project is a working, testable model of a grammar rule in Welsh called soft mutation, where the first letter of a word changes depending on what comes before it in a sentence. The repository contains two connected parts: a written explanation of the linguistic theory behind why these letter changes happen, and a text processing tool that actually applies the theory to real Welsh sentences, checking each word to see whether it was mutated correctly and flagging any place where the theory's prediction disagrees with the real text. The tool can take a Welsh sentence and mark which letters were changed and why, name the specific grammatical rule responsible, and even predict what the correctly mutated form of a sentence should look like if you feed it unmutated text. It can output its reasoning as a readable tree showing each word's evidence and verdict, or as structured data for other programs to use. A separate written report walks through worked examples, and the project is set up so that report is automatically checked against the program's actual output when it is built, so the documentation cannot describe behavior the code does not actually have. Setting it up involves installing dependencies with npm and downloading a Welsh language dataset called Universal Dependencies, which the tool uses to build its word list. An optional, larger word list can also be built from a separate open source Welsh dictionary project, though that dataset carries a different license and is deliberately kept out of the repository itself. The project's own code is released under the BSD-3-Clause license. The main Welsh word list is licensed CC BY-SA 4.0 because it is built from another project's data, and the optional larger word list uses GPL-3.0 and is never included directly, only fetched locally when needed.