Mishearing and the voice in poetry

Abstract

Speech recognition software can be used as a compositional tool in poetry and as such constitutes a form of homophonic translation, in which phonemes are substituted for each other based on near similarity. The end result bears a degree of similarity to the original based on phonemic patterning. One of the conclusions drawn in this article is that the phoneme is not an invariant fundamental particle of language, but can be thought of as bearing information in the sense that a voice can be defined as a distinctive patterning of phonemes, and thus the phoneme can be understood to bear information, in a probabilistic sense, about its neighbours. Drawing on Albert Bregman’s work on auditory scene analysis, as well as recent research into non-arbitrary aspects of the relation of signifier to signified, this article proposes stylometric analysis of misheard or mistranslated poetry to see if distinctive phonemic patterning can be measured.