nlprule

Crates.ionlprule
lib.rsnlprule
version0.6.4
sourcesrc
created_at2021-01-17 10:41:28.588653
updated_at2021-04-24 09:41:23.226048
descriptionA fast, low-resource Natural Language Processing and Error Correction library.
homepage
repositoryhttps://github.com/bminixhofer/nlprule
max_upload_size104857600
id343082
size349,311
Benjamin Minixhofer (bminixhofer)

documentation

README

nlprule

PyPI Crates.io Docs.rs PyPI Downloads License

A fast, low-resource Natural Language Processing and Error Correction library written in Rust. nlprule implements a rule- and lookup-based approach to NLP using resources from LanguageTool.

Python Usage

Install: pip install nlprule

Use:

from nlprule import Tokenizer, Rules

tokenizer = Tokenizer.load("en")
rules = Rules.load("en", tokenizer)
rules.correct("He wants that you send him an email.")
# returns: 'He wants you to send him an email.'

rules.correct("I can due his homework.")
# returns: 'I can do his homework.'

for s in rules.suggest("She was not been here since Monday."):
    print(s.start, s.end, s.replacements, s.source, s.message)
# prints:
# 4 16 ['was not', 'has not been'] WAS_BEEN.1 Did you mean was not or has not been?
for sentence in tokenizer.pipe("A brief example is shown."):
    for token in sentence:
        print(
            repr(token.text).ljust(10),
            repr(token.span).ljust(10),
            repr(token.tags).ljust(24),
            repr(token.lemmas).ljust(24),
            repr(token.chunks).ljust(24),
        )
# prints:
# 'A'        (0, 1)     ['DT']                   ['A', 'a']               ['B-NP-singular']       
# 'brief'    (2, 7)     ['JJ']                   ['brief']                ['I-NP-singular']       
# 'example'  (8, 15)    ['NN:UN']                ['example']              ['E-NP-singular']       
# 'is'       (16, 18)   ['VBZ']                  ['be', 'is']             ['B-VP']                
# 'shown'    (19, 24)   ['VBN']                  ['show', 'shown']        ['I-VP']                
# '.'        (24, 25)   ['.', 'PCT', 'SENT_END'] ['.']                    ['O']
Rust Usage

Recommended setup:

Cargo.toml

[dependencies]
nlprule = "<version>"

[build-dependencies]
nlprule-build = "<version>" # must be the same as the nlprule version!

build.rs

fn main() {
    println!("cargo:rerun-if-changed=build.rs");

    nlprule_build::BinaryBuilder::new(
        &["en"],
        std::env::var("OUT_DIR").expect("OUT_DIR is set when build.rs is running"),
    )
    .build()
    .validate();
}

src/main.rs

use nlprule::{Rules, Tokenizer, tokenizer_filename, rules_filename};

fn main() {
    let mut tokenizer_bytes: &'static [u8] = include_bytes!(concat!(
        env!("OUT_DIR"),
        "/",
        tokenizer_filename!("en")
    ));
    let mut rules_bytes: &'static [u8] = include_bytes!(concat!(
        env!("OUT_DIR"),
        "/",
        rules_filename!("en")
    ));

    let tokenizer = Tokenizer::from_reader(&mut tokenizer_bytes).expect("tokenizer binary is valid");
    let rules = Rules::from_reader(&mut rules_bytes).expect("rules binary is valid");

    assert_eq!(
        rules.correct("She was not been here since Monday.", &tokenizer),
        String::from("She was not here since Monday.")
    );
}

nlprule and nlprule-build versions are kept in sync.

Main features

  • Rule-based Grammatical Error Correction through multiple thousand rules.
  • A text processing pipeline doing sentence segmentation, part-of-speech tagging, lemmatization, chunking and disambiguation.
  • Support for English, German and Spanish.
  • Spellchecking. (in progress)

Goals

  • A single place to apply spellchecking and grammatical error correction for a downstream task.
  • Fast, low-resource NLP suited for running:
    1. as a pre- / postprocessing step for more sophisticated (i. e. ML) approaches.
    2. in the background of another application with low overhead.
    3. client-side in the browser via WebAssembly.
  • 100% Rust code and dependencies.

Comparison to LanguageTool

|Disambiguation rules| |Grammar rules| LT version nlprule time LanguageTool time
English 843 (100%) 3725 (~ 85%) 5.2 1 1.7 - 2.0
German 486 (100%) 2970 (~ 90%) 5.2 1 2.4 - 2.8
Spanish Experimental support. Not fully tested yet.

See the benchmark issue for details.

Projects using nlprule

  • prosemd: a proofreading and linting language server for markdown files with VSCode integration.
  • cargo-spellcheck: a tool to check all your Rust documentation for spelling and grammar mistakes.

Please submit a PR to add your project!

Acknowledgements

All credit for the resources used in nlprule goes to LanguageTool who have made a Herculean effort to create high-quality resources for Grammatical Error Correction and broader NLP.

License

nlprule is licensed under the MIT license or Apache-2.0 license, at your option.

The nlprule binaries (*.bin) are derived from LanguageTool v5.2 and licensed under the LGPLv2.1 license. nlprule statically and dynamically links to these binaries. Under LGPLv2.1 ยง6(a) this does not have any implications on the license of nlprule itself.

Commit count: 368

cargo fmt