As with a surprising number of similar issues relating to human language, there's already a module on CPAN for this: Lingua::EN::Inflect -- behold:

#!/usr/bin/env perl

use strict;
use warnings;

use Lingua::EN::Inflect 'A';

while (<DATA>) {
    while ( /\b (an? \s (\w+))/gx ) {
        my ( $phrase, $word ) = ( $1, $2 );
        if ( $phrase eq A( $word )) {
            print "RIGHT: $phrase\n";
        }
        else {
            print "WRONG: $phrase\n";
        }
    }
}

__DATA__
I am a howling maniac in an hospital.
It takes a hour to win an honorable mention.
Calling him a umpire is a euphemism.
This is a united front against a evil empire.
[download]

That just shows how to use the module to check for errors. It should be easy to tweak the example a bit so as to fix them.

(updated to fix a typo in the cpan link.)

One more update: a bit more logic is needed to handle various kinds and amounts of punctuation that might occur between the article and the word that follows:

while (<DATA>) {
    while ( /\b ((an?) \s\W* (\w+))/gx ) {
        my ( $phrase, $article, $word ) = ( $1, $2, $3 );
        my $bigram = "$article $word";
        if ( $bigram eq A( $word )) {
            print "RIGHT: $phrase\n";
        }
        else {
            print "WRONG: $phrase\n";
        }
    }
}

__DATA__
Suitable for a (older) child or an 'early adult'.
[download]