https://github.com/xwiz/spacy_symspell

Spacy symspell extension
https://github.com/xwiz/spacy_symspell

spacy spelling-correction spelling-suggestions symspell

Last synced: 24 days ago
JSON representation

Spacy symspell extension

• Host: GitHub
• URL: https://github.com/xwiz/spacy_symspell
• Owner: xwiz
• License: mit
• Created: 2019-04-18T09:16:32.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2019-08-26T16:12:59.000Z (over 5 years ago)
• Last Synced: 2024-11-16T17:10:46.282Z (about 1 month ago)
• Topics: spacy, spelling-correction, spelling-suggestions, symspell
• Language: Python
• Size: 3.41 MB
• Stars: 0
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# spaCy Symspell

## Spelling correction implementation in spaCy via Symspell

This package is a [spaCy 2.0 extension](https://spacy.io/usage/processing-pipelines#section-extensions) that adds sentnece/spelling corrections via Symspell to spaCy's text processing pipeline.

## Installation

`pip install spacy_symspell`

## Notes

This package is still in Alpha and there may be unforeseen errors. Dictionary loading time is also significant, can take up to 30 seconds on slow machines.

## Usage

Adding the component to the processing pipeline is relatively simple:

    import spacy

    from spacy_symspell import SpellingCorrector

    nlp = spacy.load('en_core_web_sm')

    corrector = SpellingCorrector()

    nlp.add_pipe(corrector)

    doc = nlp('What doyuoknowabout antyhing')

    for s in doc._.suggestions:#iterable

        print(s) #What doyon about anything

        

    doc._.segmentation  #::segmented_string - What doyouk now about antyhing ::corrected_string - that dook now about anything

spaCy_symspell operates on `Doc` and `Span` spaCy objects. When called on a `Doc` or `Span`, the object is given two attributes: `suggestions` (a list of all found spelling suggestions) and `segmentation` (a corrected sentence in the case of ommitted spaces).

## Todo

Symspell accuracy can be improved with the help of spaCy by extracting and analyzing resulting n-grams and cross-referencing with possible n-grams deductible from the character groups in the symspell result. For example the correction 'that dook now' leaves us with a verbless sentence, and on closer analysis will reveal that the character group 'now' is related with the verb 'know', and the verb know is associated with the n-gram 'you know'.

## Under the hood

[spacy_symspell](https://github.com/xwiz/spacy_symspell) is currently a wrapper of the [python port](https://github.com/mammothb/symspellpy) for [Symspell](https://github.com/wolfgarbe/SymSpell). For additional details, see the linked project pages.