SymSpell

A fast spelling correction and fuzzy search library for Rust. Given a misspelled word like "roket", it finds the correct spelling "rocket". It can also fix entire sentences, split concatenated words, and rank suggestions by how common each word is.

Rust implementation of the SymSpell algorithm originally written in C# by @wolfgarbe. It works by pre-computing all possible deletions of dictionary words and storing them in a hash map, which allows O(1) average lookup time at query time — orders of magnitude faster than traditional approaches like BK-trees or Norvig's algorithm.

Features

• Single-word correction (lookup) — find the closest matching dictionary word within a given edit distance
• Multi-word correction (lookup_compound) — correct entire sentences, handling word splits, joins, and replacements. Supports bigram frequency dictionaries for context-aware ranking
• Word segmentation (word_segmentation) — split concatenated words into their components (e.g. "thequickbrownfox" → "the quick brown fox")
• WebAssembly support — compile to WASM for use in JavaScript/browser environments
• Configurable string handling — pluggable string strategies for ASCII transliteration or full Unicode support

Usage

use symspell::{AsciiStringStrategy, SymSpell, Verbosity};

fn main() {
    let mut symspell: SymSpell<AsciiStringStrategy> = SymSpell::default();

    symspell.load_dictionary("data/frequency_dictionary_en_82_765.txt", 0, 1, " ");
    symspell.load_bigram_dictionary(
      "./data/frequency_bigramdictionary_en_243_342.txt",
      0,
      2,
      " "
    );

    let suggestions = symspell.lookup("roket", Verbosity::Top, 2);
    println!("{:?}", suggestions);

    let sentence = "whereis th elove hehad dated forImuch of thepast who couqdn'tread in sixtgrade and ins pired him";
    let compound_suggestions = symspell.lookup_compound(sentence, 2);
    println!("{:?}", compound_suggestions);

    let sentence = "whereisthelove";
    let segmented = symspell.word_segmentation(sentence, 2);
    println!("{:?}", segmented);
}

N.B. the dictionary entries have to be lowercase

Advanced Usage

Using Custom Settings

let mut symspell: SymSpell<AsciiStringStrategy> = SymSpellBuilder::default()
    .max_dictionary_edit_distance(2)
    .prefix_length(7)
    .count_threshold(1)
    .build()
    .unwrap()

String Strategy

String strategy is abstraction for string manipulation, for example preprocessing.

There are two strategies included:

• UnicodeStringStrategy
  • Doesn't do any preprocessing and handles strings as they are.
• AsciiStringStrategy
  • Transliterates strings into ASCII only characters.
  • Useful when you are working with accented languages and you don't want to care about accents, etc

To configure string strategy just pass it as a type parameter:

let mut ascii_symspell: SymSpell<AsciiStringStrategy> = SymSpell::default();
let mut unicode_symspell: SymSpell<UnicodeStringStrategy> = SymSpell::default();

Javascript Bindings

This crate can be compiled against wasm32 target and exposes a SymSpell Class that can be used from Javascript as follow. Only UnicodeStringStrategy is exported, meaning that if someone wants to manipulate ASCII only strings the dictionary and the sentences must be prepared in advance from JS.

const fs = require('fs');
const rust = require('./pkg');

let dictionary = fs.readFileSync('data/frequency_dictionary_en_82_765.txt');
let sentence = "whereis th elove hehad dated forImuch of thepast who couqdn'tread in sixtgrade and ins pired him";

let symspell = new rust.SymSpell({ max_edit_distance: 2,  prefix_length: 7,  count_threshold: 1});
symspell.load_dictionary(dictionary.buffer, { term_index: 0,  count_index: 1, separator: " "});
symspell.load_bigram_dictionary(bigram_dict.buffer, { term_index: 0,  count_index: 2, separator: " "});
symspell.lookup_compound(sentence, 1);

It can be compiled using wasm-pack (eg. wasm-pack build --release --target nodejs)