Whatlang

Natural language detection for Rust with focus on simplicity and performance.

Try online demo.

Content

• Features
• Get started
• Documentation
• Supported languages
• Feature toggles
• How does it work?
  • How language recognition works?
  • How is_reliable calculated?
• Running benchmark
• Comparison with alternatives
• Ports and clones
• Derivation
• License
• Contributors

Features

• Supports 78 languages
• 100% written in Rust
• Lightweight, fast and simple
• Recognizes not only a language, but also a script (Latin, Cyrillic, etc)
• Provides reliability information

Get started

Add to you Cargo.toml:

[dependencies]

whatlang = "0.12.0"

Example:

extern crate whatlang;

use whatlang::{detect, Lang, Script};

fn main() {
    let text = "Ĉu vi ne volas eklerni Esperanton? Bonvolu! Estas unu de la plej bonaj aferoj!";

    let info = detect(text).unwrap();
    assert_eq!(info.lang(), Lang::Epo);
    assert_eq!(info.script(), Script::Latin);
    assert_eq!(info.confidence(), 1.0);
    assert!(info.is_reliable());
}

For more details (e.g. how to blacklist some languages) please check the documentation.

Feature toggles

Feature	Description
enum-map	Lang and Script implement Enum trait from enum-map

How does it work?

How does the language recognition work?

The algorithm is based on the trigram language models, which is a particular case of n-grams. To understand the idea, please check the original whitepaper Cavnar and Trenkle '94: N-Gram-Based Text Categorization'.

How is_reliable calculated?

It is based on the following factors:

• How many unique trigrams are in the given text
• How big is the difference between the first and the second(not returned) detected languages? This metric is called rate in the code base.

Therefore, it can be presented as 2d space with threshold functions, that splits it into "Reliable" and "Not reliable" areas. This function is a hyperbola and it looks like the following one:

For more details, please check a blog article Introduction to Rust Whatlang Library and Natural Language Identification Algorithms.

Running benchmarks

This is mostly useful to test performance optimizations.

cargo bench

Comparison with alternatives

	Whatlang	CLD2	CLD3
Implementation language	Rust	C++	C++
Languages	87	83	107
Algorithm	trigrams	quadgrams	neural network
Supported Encoding	UTF-8	UTF-8	?
HTML support	no	yes	?

Ports and clones

• whatlang-ffi - C bindings
• whatlanggo - whatlang clone for Go language
• whatlang-py - bindings for Python
• whatlang-rb - bindings for Ruby

Derivation

Whatlang is a derivative work from Franc (JavaScript, MIT) by Titus Wormer.

License

MIT © Sergey Potapov

Contributors

• greyblake Potapov Sergey - creator, maintainer.
• Dr-Emann Zachary Dremann - optimization and improvements
• BaptisteGelez Baptiste Gelez - improvements
• Vishesh Chopra - designed the logo