We write a short and simple code for spell checking. For a given word, we simply check the distance between it and the words in NLTK corpus. Either Jaccard or Edit distance can be used to measure the distances.
import nltk
import pandas as pd
import numpy as np
from nltk.stem import WordNetLemmatizer
nltk.data.path.append("assets/")
nltk.download('words')
from nltk.corpus import words
correct_spellings = words.words()
from nltk.util import ngrams
from nltk.metrics import distance
[nltk_data] Downloading package words to
[nltk_data] /Users/mohsenkarkheiran/nltk_data...
[nltk_data] Package words is already up-to-date!
def spell_check(Word, dist_type, ng):
recoms = []
def sec(e):
return e[1]
ls = [ws for ws in correct_spellings if ws[0]==Word[0]]
label1 = set(ngrams(Word,ng))
dists = []
for ws in ls:
label2 = set(ngrams(ws,ng))
if dist_type == 'jaccard':
d = distance.jaccard_distance(label1, label2)
elif dist_type == 'edit':
d = distance.edit_distance(Word, ws)
dists = dists + [(ws, d)]
dists.sort(key=sec)
return dists[0][0], dists[1][0], dists[2][0]
spell_check('gooshh', dist_type = 'edit', ng = 3)
('goolah', 'goose', 'goosish')
spell_check('gooshh', dist_type = 'jaccard', ng = 3)
('goose', 'goosy', 'goosery')