By Gabriele Barbieri, François Pachet, Pierre Roy, and Mirko Degli Esposti.
In the Proceedings of the 20th European Conference on Artificial Intelligence (ECAI 2012), pp. 115–120, 2012.
About the paper
This paper applies constrained Markov generation to song lyrics. It uses a corpus to model style and imposes rhyme and meter as explicit conditions on the generated text.
Abstract summary
The authors generate lyrics that fit the structure of an existing song. They compare the combined approach with approaches based on Markov models or constraints alone.
Related work
Enforcing Meter in Finite-Length Markov Sequences · PDF (AAAI 2013) develops a constraint for enforcing metrical rules in generated sequences.
See Markov Constraints for the sequence generation methods and Flow Machines for related work on computer-assisted songwriting.