graciax452/lute-xhosa

# lute-xhosa An isiXhosa language parser plugin for Lute3, splitting agglutinative noun/verb morphology into separately trackable tokens. Fork of lute-shona, following that project's `DESIGN.md` (section 10) forking guidance. **Read lute-shona's `DESIGN.md` first** — it's the actual design methodology (lexicon-gated stripping to avoid overstemming, the token model, the closed-construction pattern); this project's own `DESIGN.md` only covers what's isiXhosa-specific. ## What this solves Same problem as Shona: isiXhosa is space-delimited (Lute's stock "Space Delimited" parser gets word boundaries right, including click-consonant letters — `c`, `q`, `x` — which are just ordinary letters within a word), but a single space-delimited word is often several grammatical morphemes glued together, leaving the whole inflected word as one unclickable blob. **Genuinely different from Shona**, not just a data swap: isiXhosa nouns have a two-layer prefix (an augment vowel plus a class prefix, e.g. `abantu` = `a-` + `ba-` + `-ntu`), and object concords use different surface forms from subject concords in several classes (unlike Shona, where they mostly overlap). See `DESIGN.md` for how the augment+prefix structure turned out to need only more data, not new engine code. ## Status Working noun-class splitting (all confirmed classes with singular/plural pairs, cross-class shared roots, derivational examples) and verb splitting (infinitives, present-tense `-ya-` infix, class-1 `a-past`). **Not yet implemented**: the `-ile` perfect suffix (structurally different from a prefix TAM marker — needs its own design pass), verbal extensions (causative/applicative/passive — not yet sourced), future tense (reportedly periphrastic, unconfirmed), and most persons' `a-past` forms beyond class 1. See `DESIGN.md` for the full list of what's confirmed vs. deferred, with page citations to the source grammar. Sources: J.C. Oosthuysen, *The Grammar of isiXhosa* (SUN PRESS, 2016), extracted directl …