Core library for working with UniMorph morphological data.

This crate provides types, storage, and query capabilities for UniMorph datasets. It is designed for high-performance lookups (conjugation/declension queries) while also supporting bulk export for ML pipelines.

Quick Start

use unimorph_core::{Repository, Store};

#[tokio::main]
async fn main() -> unimorph_core::Result<()> {
    // Initialize repository (handles downloads and caching)
    let repo = Repository::new()?;

    // Ensure Italian dataset is available
    repo.ensure("ita").await?;

    // Open the store for queries
    let store = repo.store()?;

    // Look up all forms of "parlare"
    for entry in store.inflect("ita", "parlare")? {
        println!("{} -> {} ({})", entry.lemma, entry.form, entry.features);
    }

    // Reverse lookup: what lemmas produce "parlo"?
    for entry in store.analyze("ita", "parlo")? {
        println!("{} <- {} ({})", entry.form, entry.lemma, entry.features);
    }
    Ok(())
}

Architecture

• SQLite backend: All data stored in a single file at ~/.cache/unimorph/datasets.db
• Pre-computed stats: Aggregate statistics cached at import time
• Iterator-based queries: Results stream from SQLite, won't OOM on large datasets
• Parquet export: For users who need DataFrame integration