Verify correct use of "a" and "an" in English texts - Python

Question

I want to create a program that reads text from a file and points out when "a" and "an" is used incorrect. The general rule as far as I know is that "an" is used when the next words starts with a vowel. But it should also take into consideration that there are exceptions which also should be read from a file.

Could someone give me some tips and tricks on how I should get started with this. Functions or so that could help.

I would be very glad :-)

I'm quite new to Python.

Answer 1

Here's a solution where correctness is defined as: an comes before a word that starts with a vowel sound, otherwise a may be used:

#!/usr/bin/env python
import itertools
import re
import sys

try:
    from future_builtins import map, zip
except ImportError: # Python 3 (or old Python versions)
    map, zip = map, zip
from operator import methodcaller

import nltk  # $ pip install nltk
from nltk.corpus import cmudict  # >>> nltk.download('cmudict')

def starts_with_vowel_sound(word, pronunciations=cmudict.dict()):
    for syllables in pronunciations.get(word, []):
        return syllables[0][-1].isdigit()  # use only the first one

def check_a_an_usage(words):
    # iterate over words pairwise (recipe from itertools)
    #note: ignore Unicode case-folding (`.casefold()`)
    a, b = itertools.tee(map(methodcaller('lower'), words)) 
    next(b, None)
    for a, w in zip(a, b):
        if (a == 'a' or a == 'an') and re.match('\w+$', w): 
            valid = (a == 'an') if starts_with_vowel_sound(w) else (a == 'a')
            yield valid, a, w

#note: you could use nltk to split text in paragraphs,sentences, words
pairs = ((a, w)
         for sentence in sys.stdin.readlines() if sentence.strip() 
         for valid, a, w in check_a_an_usage(nltk.wordpunct_tokenize(sentence))
         if not valid)

print("Invalid indefinite article usage:")
print('\n'.join(map(" ".join, pairs)))

Example input (one sentence per line)

Validity is defined as `an` comes before a word that starts with a
vowel sound, otherwise `a` may be used.
Like "a house", but "an hour" or "a European" (from @Hyperboreus's comment http://stackoverflow.com/questions/20336524/gramatically-correct-an-english-text-python#comment30353583_20336524 ).
A AcRe, an AcRe, a rhYthM, an rhYthM, a yEarlY, an yEarlY (words from @tchrist's comment http://stackoverflow.com/questions/9505714/python-how-to-prepend-the-string-ub-to-every-pronounced-vowel-in-a-string#comment12037821_9505868 )
We have found a (obviously not optimal) solution." vs. "We have found an obvious solution (from @Hyperboreus answer)
Wait, I will give you an... -- he shouted, but dropped dead before he could utter the last word. (ditto)

Output

Invalid indefinite article usage:
a acre
an rhythm
an yearly

It is not obvious why the last pair is invalid, see Why is it “an yearly”?

Answer 2

Maybe this can give you a rough guideline:

1. You need to parse the input text into prosodic units, as I doubt that the rules for "a/an" apply over prosodic boundaries (e.g. "We have found a (obviously not optimal) solution." vs. "We have found an obvious solution").

2. Next you need to parse each prosodic unit into phonological words.

3. Now you somehow need to identify those words, which represent the undefined article ("a house" vs "grade A product").

4. Once you have identified the articles, look at the next word in your prosodic unit and determine (here be dragons) the syllabic feature of the first phoneme of this word.

5. If it has [+syll] the article should be "an". If it has [-syll] the article should be "a". If the article is at the end of the prosodic unit, it should be maybe "a" (But what about ellipses: "Wait, I will give you an... -- he shouted, but dropped dead before he could utter the last word."). Except historical exceptions as mentioned by abanert, dialectal variance, etc, etc.

6. If the found article doesn't match the expected, mark this as "incorrect".

---

Here some pseudocode:

def parseProsodicUnits(text): #here be dragons
def parsePhonologicalWords(unit): #here be dragons
def isUndefinedArticle(word): #here be dragons
def parsePhonemes(word): #here be dragons
def getFeatures(phoneme): #here be dragons

for unit in parseProsodicUnits(text):
    for idx, word in enumerate (parsePhonologicalWords(unit)[:-1]):
        if not isUndefinedArticle(word): continue
        syllabic = '+syll' in getFeatures(parsePhonemes(unit[idx+1])[0])
        if (word == 'a' and syllabic) or (word == 'an' and not syllabic):
            print ('incorrect')

Answer 3

all_words = "this is an wonderful life".split()
for i in range(len(all_words)):
    if all_words[i].lower() in ["a","an"]:
       if all_words[i+1][0].lower() in "aeiou":
           all_words[i] = all_words[i][0]+"n"
       else:
           all_words[i] = all_words[i][0]
print " ".join(all_words)

that should get you started , however it is not a complete solution....

Answer 4

I'd probably start with an approach like:

exceptions = set(/*a whole bunch of exceptions*/)
article = None
for word in text.split():
    if article:
        vowel = word[0].lower() in "aeiou"
        if word.lower() in exceptions:
            vowel = not vowel
        if (article.lower() == "an" and not vowel) or (article.lower() == "a" and vowel):
            print "Misused article '%s %s'" % (article, word)
        article = None
    if word.lower() in ('a', 'an'):
       article = word