Markov Word Generator
It's trivial for a computer program to generate random words. The tricky part is creating words that humans perceive as legible and pronounceable instead of mangled and cryptic. This web app solves the problem by applying a Markov chain.
View the live site here.
Why?
Because Markov chains are cool!
I originally wanted a program to help me generate science fiction names. There were plenty of websites out there that used Markov chains to generate random paragraphs, but I couldn't find any good ones to generate random words.
I probably got a little carried away with CoffeeScript and Bootstrap.
Contributions
Please contribute. I'm so lonely.
Report issues here and I'll gladly look into them.
Markov.coffee
The back-end logic for managing the Markov chain is available as a standalone CoffeeScript module. It's flexible enough for you to use in your own project, if you want to. Like the rest of this project, it's distributed under the permissive MIT license.
Below is a light introduction to the module. For more detailed documentation, see the documentation comments in the source.
Markov.coffee-specific terminology
- element: In Markov.coffee, an "element" is a basic, indivisible building block of the material that we are working with. For example, if the end goal is to generate random sentences, then the individual words are the elements.
- sequence: A "sequence" is simply a list of elements. Markov.coffee can accept sequences both as arrays and as strings. (If you pass a string as a sequence, then the individual characters are the elements.)
Although not strictly necessary, some familiarity with Markov chains and n-grams is also helpful here.
Setup
After including the script, Markov objects can be constructed like this:
markov = new Markov ["sassafras", "mississippi"], 1
# Or, on CommonJS:
# Markov = require "./markov"
# markov = new Markov ["sassafras", "mississippi"], 1
The first parameter to the constructor is an array of sequences. The sequences are combined together to form the corpus. The generator takes care not to link elements across sequence boundaries. In the example above, the last S in sassafras is not associated with the M in mississippi. If you really do want those letters to be associated, here's how to do it:
markov = new Markov ["sassafrasmississippi"], 1
The second parameter to the constructor is n, the Markov order - basically, how many previous elements the next element depends on. Low values make the Markov chain more random, while high values make it stick closer to the corpus.
If left unspecified, the array of sequences defaults to [] and the Markov order defaults to 2.
You can directly modify these properties later in your code, if you need to. They're not private variables.
markov.sequences.push "foo"
markov.n = 3
Generation
Make the Markov chain do something useful with .generate(). Note that it returns
an array, so if you want a string you'll have to use .join("").
markov = new Markov ["sassafras", "mississippi"]
alert markov.generate().join "" # Alerted "rassippi".
alert markov.generate().join "" # Alerted "frassissafrassippi".
.generate() takes an optional maximum length parameter, e.g. markov.generate(10) to
limit generated words to 10 characters long. If unspecified, it defaults to 20 elements.
There always needs to be a maximum length, because otherwise, things like this
could result in infinite loops:
markov = new Markov ["abba"], 1
alert markov.generate().join "" # "bbababbabbbbababababbbabababababbabbbbabbbabababab..."
Other Stuff
The Markov class supports other methods, too. I won't describe them all in
detail here, but you can read about how they work in the source
if you're interested.
.ngrams()gives you the raw list of n-grams used to build the chain..tree()gives you a probability tree representing the n-grams..continue(sequence)gives you a single next element to continuesequence.