spaCy Symspell

Spelling correction implementation in spaCy via Symspell

This package is a spaCy 2.0 extension 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 is currently a wrapper of the python port for Symspell. For additional details, see the linked project pages.