Autocorrect

Autocorrect is a lightweight, high-performance Rust library and CLI tool for on-device fuzzy spelling correction. It compiles a simple TSV word-frequency list into a memory-mapped finite-state transducer (FST) and streams Levenshtein-based suggestions in sub-millisecond time.

---

Features

• Offline Dictionary Build: Converts words.tsv into a minimal FST during build time using fst::MapBuilder.
• Memory-Mapped Lexicon: Loads the FST at runtime with memmap2, keeping RAM usage minimal.
• Real-Time Fuzzy Search: Provides top-K nearest-word suggestions on every keystroke with configurable edit distance.
• Interactive REPL Demo: A CLI demo (main.rs) that shows suggestions live as you type.
• Extensible: Easily bolt on touch-geometry weighting, quantized neural-LM reranking, or custom automata.
• Tiny Footprint: Binary size <2 MB, hot RAM <1 MB, latency <1 ms per lookup.

---

Getting Started

Prerequisites

• Rust (stable) installed via rustup
• A word-frequency TSV (data/words.tsv) with lines formatted as <word>\t<frequency>

Project Layout

autocorrect/
├─ Cargo.toml
├─ build.rs
├─ data/
│  └─ words.tsv        # word<TAB>frequency list
└─ src/
   ├─ lib.rs           # exposes dictionary & suggest modules
   ├─ dictionary.rs    # FST loader
   ├─ suggest.rs       # candidate generator
   └─ main.rs          # CLI REPL demo

Installation

Clone the repository and build in release mode:

git clone https://github.com/yourusername/autocorrect.git
cd autocorrect
cargo build --release

This will run build.rs to generate dict.fst from data/words.tsv and compile the binary to target/release/autocorrect.

Usage

CLI Demo

Launch the interactive REPL:

./target/release/autocorrect

Start typing—suggestions appear live:

Start typing… (Ctrl-D to quit)
t ▶ ["the", "to", "tea", "too", "ten"]
th ▶ ["the", "they", "then", "that", "thus"]

Library API

Include in your Cargo.toml:

[dependencies]
autocorrect = { path = "../autocorrect" }

Load the dictionary and query suggestions:

use autocorrect::{dictionary, suggest};
use std::path::Path;

fn main() -> anyhow::Result<()> {
    // Load the FST
    dictionary::load(Path::new("path/to/dict.fst"))?;

    // Get suggestions
    let s = suggest::candidates("exampl");
    println!("Suggestions: {:?}", s);
    Ok(())
}

Dictionary Format

The source TSV (data/words.tsv) must be in the format:

<word><TAB><frequency>

• <word>: UTF-8 string, usually lowercase.
• <frequency>: integer count or weight.

Lines should be sorted by descending frequency for optimal rank assignment.

Extending

• Touch-Geometry: Implement a custom fst::Automaton to weight edits by key proximity.
• Neural-LM Reranking: Post-filter the top-K candidates with a quantized Transformer or LSTM for context.

Contributing

1. Fork the repository.
2. Create a feature branch (git checkout -b feature/your-feature).
3. Commit your changes (git commit -am 'Add new feature').
4. Push to your branch (git push origin feature/your-feature).
5. Submit a pull request.

Please ensure all changes include tests and you’ve run cargo fmt and cargo clippy.

License

This project is licensed under the MIT License. See LICENSE for details.