Into the Deep End: An Approach to Generation of Formal Poetry

Depth-first searches (DFS) have enabled computers to beat human opponents at chess for decades. By constructing a search tree of future board states and evaluating them against a given rubric, chess-playing machines guide themselves toward a desirable outcome. In my paper, I examine how a DFS-based algorithm can be used to generate formal poetry (meaning verse that exists under metrical and phonological constraints) by advancing one word at a time through a library of potential choices and "scoring" the results based on formal characteristics, backtracking as necessary.

There are two primary challenges involved with such an approach: the design of a search tree and the construction of a suitable rubric against which to judge a poem's progress. To address the first of these, my paper presents a search tree/Markov chain hybrid as a viable means to store found language. Markov chains are already used in many contemporary electronic poetry installations, and should be familiar to most artists in the field. As such, I devote more attention to the problems encountered in creating a desirable rubric. First of these is developing a machine-based scansion algorithm to determine if a given line conforms to the correct meter. I present a recursive algorithm that uses a set of phonological and positional rules to locate stresses in a line based on phonetic data from the CMU pronouncing dictionary, then compares the resulting pattern to a given meter. In addition, I also discuss methods for detection of other aural characteristics common to formal verse (alliteration, rhyme, assonance, etc.).

As an initial case study, my paper presents the interactive sonnet sequence TheInternetIsAnEnglishSonnet.com, an online poetry installation I designed and released in September of 2009. The site harnesses a DFS-based algorithm to build Elizabethan sonnets from Google results using input from the reader as search terms.

My paper concludes by addressing the consequences of applying this type of model to poetry in regard to authorship. For example, given that the best computer algorithms are now superior to even elite chess players, will there come a point when electronic poetry becomes formally indistinguishable from that written by flesh-and-blood poets? Will there come a point when readers begin to prefer algorithmic poetry to traditional verse? (Source: Author).