EvoPoem: Context Free Grammars for Automated Poetry Generation

Abstract

In this thesis, we discuss various poetry generators. We present our own text synthesis and natural language generation algorithm that can be used with various machine learning technologies, as well as a setup for the final machine learning enabled poetry generator, which we call EvoPoem. The algorithm is able to produce short, grammatically correct sentences, and create a visual spacing that suggests rhythm. This algorithm uses a grammar, a lexicon and a feature and unification algorithm enriched with constraint satisfaction, to parse a string of bits deterministically into a potential poem.