khmerthings 0.5.0

Deterministic Khmer language tools: word counter, segmentation primitives, and more.

khmerthings

Deterministic Khmer language tools for Python — built as community building blocks: small, correct, dependency-free primitives you can compose into bigger systems.

No machine-learning models, no third-party NLP dependencies, no network calls. Every result is reproducible and explainable. Khmer script writes no spaces between words, so even "simple" operations like counting or sorting need real language handling — khmerthings implements that from first principles.

Tools

Each tool is available both as a Python API and a CLI subcommand, and has its own detailed document:

Tool	CLI	Python	Docs
Word breaker — split Khmer text into words	khmerthings segment	break_words, mark_boundaries	docs/word-breaker.md
Word counter — count words in Khmer/mixed text	khmerthings count	count_words, analyze	docs/word-counter.md
Line sorter — Khmer dictionary-order sorting	khmerthings sort	sort_lines, khmer_sort_key	docs/line-sorter.md

Install

pip install khmerthings          # library
uv tool install khmerthings     # global CLI

A taste

$ echo "ខ្ញុំស្រឡាញ់ភាសាខ្មែរ" | khmerthings segment
ខ្ញុំ ស្រឡាញ់ ភាសា ខ្មែរ

from khmerthings import break_words, count_words, sort_lines

break_words("ខ្ញុំស្រឡាញ់ភាសាខ្មែរ")   # ['ខ្ញុំ', 'ស្រឡាញ់', 'ភាសា', 'ខ្មែរ']
count_words("ខ្ញុំមានឆ្កែ ២ ក្បាល and 3 cats")   # 8
sort_lines(["ក្រ", "កា", "កក"])                    # ['កក', 'កា', 'ក្រ']

Design principles

• Deterministic: same input, same output, always. Rule- and dictionary-based algorithms only; nothing probabilistic.
• Self-contained: zero runtime dependencies; all word data is our own hand-curated set of growable wordlists — words (core vocabulary), names (people's names & titles), and modern (slang, loanwords, trending terms) — 1,583 entries and growing, each verified entry by entry; no wordlist is imported wholesale.
• Lossless: no character is ever dropped — unknown Khmer spans are reported, not discarded.
• Tested first: every module ships with table-driven unit tests and invariant checks (258 tests as of v0.3.0).

Under the hood, the tools share deterministic primitives (character classification, character-cluster segmentation, a cluster-keyed lexicon trie, lossless tokenization) in src/khmerthings/ — see the module docstrings if you want to build on them directly.

Roadmap

• ✅ Word counter, line sorter, word breaker
• ⏳ Wordlist growth across all three sources (words, names, modern) — hand-curated batches each release; the accuracy lever for every dictionary-based tool
• 🔜 Spellchecker & spellfixer (engine is feasible today; waiting on lexicon coverage to make its verdicts trustworthy)
• Later: part-of-speech tagger, intent detection, paragraph categorization

Contributing

See DEVELOPMENT_GUIDE.md for setup, the architecture, the rules (determinism, self-owned data, tests first), and how to add words to the lexicon — the single most valuable contribution. Changes are tracked in CHANGELOG.md.

License

MIT