Hello to all;
I’m testing out different ways to output randomly generated names. I’ve
toyed with some basic algorithms but recently I’ve become hooked on the
idea of using a Markov chain in the process to generate names based on a
text file. Below is a link to my first effort - I won’t even begin to
start on what a catastrophe it is and I don’t expect you to understand
it:
http://pastie.textmate.org/178820
It’s hideous. I’m never going near it again. It basically looks at each
line of a text file and throws the adjacent letter pairings into a big
hash with probability weightings, moves through the probability hash
coughing out likely letter results based on the previous entry. I think
the term for it is “first-order Markov chain”.
What I’d like to see if anyone has any suggestions, or more importantly
examples of how this should be done? I’m fairly certain there are more
elegant ways about doing this. Thanks.
I would suggest using combinations of phonemes rather than letters. It
would simplify the problem and generate more natural sounding
combinations.
On Thu, 10 Apr 2008 16:40:17 -0500, Dss Jss wrote:
It’s hideous. I’m never going near it again. It basically looks at each
line of a text file and throws the adjacent letter pairings into a big
hash with probability weightings, moves through the probability hash
coughing out likely letter results based on the previous entry. I think
the term for it is “first-order Markov chain”.
What I’d like to see if anyone has any suggestions, or more importantly
examples of how this should be done? I’m fairly certain there are more
elegant ways about doing this. Thanks.
http://rubyquiz.com/quiz74.html
The solutions are for word markov chains, but you could easily adapt any
of them for letter markov chains.
–Ken
On Fri, 11 Apr 2008 04:47:20 -0500, Sean Mehonoshen wrote:
I would suggest using combinations of phonemes rather than letters. It
would simplify the problem and generate more natural sounding
combinations.
Too hard, because then you have to specify what they are. Use a higher-
order markov model (like 4 or 5), and you'll get pronouncable results.
One of my colleagues wrote Linguistic Cognition Laboratory: About
gibberish.sh, which is a 4th-order markov model of English letters. It
shoudl be relatively pronouncable, and might suit your needs.