lemora

Local-first Latin study dictionary CLI for translating words and short phrases with dictionary-grounded output.

Features

• Offline-first CLI workflow
• Dictionary-first translation pipeline
• Whitaker and Lewis & Short adapters with local JSON override support
• Curated National Archives Stage 1/2 grammar reference for pronoun-form normalization
• Bundled National Archives stage Latin reference lexicon for common school phrases
• Optional CLTK sentence analysis and llama-cpp-python synthesis

Installation

Base install:

uv sync --group dev

With CLTK:

uv sync --group dev --extra nlp

For full CLTK parsing, download Latin Stanza assets once:

uv run python -c "import stanza; stanza.download('la')"

With local LLM support:

uv sync --group dev --extra llm

Usage

Translate a word or phrase:

uv run lemora translate "arma virumque"

Request JSON output:

uv run lemora translate "amo" --format json

Optional lexicon overrides:

export LEMORA_WHITAKER_PATH=/path/to/whitaker.json
export LEMORA_LEWIS_SHORT_PATH=/path/to/lewis_short.json
export LEMORA_MODEL_PATH=/path/to/model.gguf

• By default, lemora uses Lewis & Short. Whitaker is optional and enabled only when LEMORA_WHITAKER_PATH is set.
• Whitaker override expects JSON array entries with lemma, gloss, and optional forms, confidence, morphology.
• Lewis & Short override accepts the same JSON format or Perseus lat.ls.perseus-eng*.xml TEI files.
• Lewis & Short auto-detects ~/repos/lexica/CTS_XML_TEI/perseus/pdllex/lat/ls/lat.ls.perseus-eng2.xml when present.
• LEMORA_MODEL_PATH is only for optional synthesis models (for --synthesize workflows).

Development

uv run ruff format .
uv run ruff check .
uv run pytest

Current status

This repository is scaffolded for the first implementation phase:

• CLI entrypoint and command parsing
• Core data models
• Adapter interfaces for dictionary, NLP, and LLM integrations
• Service orchestration shell for dictionary-first translation