Improve classification in case of spelling mistakes

[Aran30]

When using following training examples:

confirmation_intent = bot.new_intent('confirmation_intent', [
    'yes',
    'yeah',
    'of course',
    'i am happy to say yes'
])

rejection_intent = bot.new_intent('rejection_intent', [
    'no',
    'nah',
    'never',
    'no i am not happy'
])

created bots would fail to classify "simple" spelling mistakes of abbreviations such as ["yea","noo","n"]. Improving the classification for such examples would definitely improve the interaction overall.

[jcabot]

Let's keep in mind that the solution needs also to be language-dependent (and must not throw an exception when using the framework with a language for which no spell-checker solution is available)