3 ms·

How is this different from building a huge database of n-grams and doing probabilistic pattern matching on them?