Meaning of a poetic meter and historical randomness: how topic modeling could help us to generalize the relations between a verse form and semantics

Artjoms Šela

The effects of random processes in history of human culture are poorly understood and often underestimated. Humanities historically underlined foresight and omniscience of the individual human genius in works of art and tended to celebrate artists as brilliant inventors. We know, however, that this meritocratic picture of cultural history may be not correct: network effects place people in centers of cultural production by chance; cultural marketplaces dramatically enhance the slight formal differences between “the few best” and “many others”; “brilliant inventions” tend to be only a tip of an iceberg of a (forgotten) history of innovation.

This talk presents one more potential case for random effects in culture: the relations between poetic meter and meaning (also known as “the semantic halo of a meter”). Using word2vec semantic space to reduce lexical variance in corpus of Russian poetry (18-20th century) and LDA topic models to represent each meter as distribution of aggregated topic probabilities, we show that this semantic information is enough to organize metrical forms by “lineages”: iambs are more “similar” to other iambs, trochees – to other trochees, etc. This confirms assumptions of previous scholars, but also establish grounds to suggest that the emergence of “semantic halo” in meters could be understood in simple terms of “copy random things in small population” (so-called “founder effect”).