Makes fake words that sound like real words!

Ola uses Markov chains, trained on a given word (train/2) or dictionary of words (train_dictionary/2), to determine statistically-likely combinations of letters.

One nice thing about the Markov chain approach is that it can be trained on a dictionary in any language (or combination of languages), and will spit out appropriate-sounding words.

A Note about N-Grams

"N-grams" are the number of combined letters that you look at when determining statistical letter combination likelihoods. These will be used as "windows" later when generating words. Training with a higher n_grams value creates more keys and generates a larger map of statistical probability.

For example:

Ola.train("balloon", n_grams: 3)

# The list of letter groups that are followed by other letters
Ola.Dictionary.keys()
%{
  1 => #MapSet<["a", "b", "l", "o"]>,
  2 => #MapSet<["al", "ba", "ll", "lo", "oo"]>,
  3 => #MapSet<["all", "bal", "llo", "loo"]>,
}

# The probability count for each letter that could follow
Ola.Dictionary.map()
%{
  "a" => %{"l" => 1},
  "al" => %{"l" => 1},
  "all" => %{"o" => 1},
  "b" => %{"a" => 1},
  "ba" => %{"l" => 1},
  "bal" => %{"l" => 1},
  "l" => %{"l" => 1, "o" => 1},
  "ll" => %{"o" => 1},
  "llo" => %{"o" => 1},
  "lo" => %{"o" => 1},
  "loo" => %{"n" => 1},
  "o" => %{"n" => 1, "o" => 1},
  "oo" => %{"n" => 1}
}

# Now generate a fake word from our training set:
Ola.word()
"loon"

Higher n-gram values allow for more accurate words, up to a certain point. The larger your values, the larger pieces-of-words that will be used for generating random values. However, if the n_grams value is higher than the average word length in your dictionary, obviously there will be no letter groups with n_gram sizes that large. In addition, if you only trained Ola using n-gram sizes up to 3, passing larger values such as 4, 5, 6, etc. won't make a difference.