Poems, as aesthetic objects, generate a subjective experience, which can be different for different readers. In this paper, we propose a method to quantify these subjective experiences. We gave participants three parallel excerpts and asked them to describe, in free text, the perceived emotive qualities of these excerpts. The descriptions were analysed quantitatively according to the dimensions of the Valence-Arousal-Dominance model of emotion. With the help of additional rating tasks and a structural theory of phonetic symbolism, we attribute the perceived emotive qualities to an interaction between the meaning of words, patterns of alliteration, and metric deviation.

There is a growing research literature on phonetic symbolism in poetry, sometimes with incongruent results. Through a theoretical structural analysis we show that, (a) individual speech sounds have (sometimes conflicting) potentials to suggest elementary percepts, such as abruptness, hardness, smoothness; and (b) from these elementary percepts some general psychological atmosphere may be abstracted that may be individuated in specific emotions as ‘love,’ ‘joy,’ or ‘anger,’ by semantic feature-addition. This proposal can reconcile incongruous research results. Sound-symbolic lexical entries are governed by similar principles, but fossilized. Large-scale statistical investigation may reveal significant sound-symbolic effects only when the same phonemes repeat throughout the poem. They may, however, miss conspicuous local sound effects, revealed only by local analysis. Some sceptical conclusions in the research literature may be due to this phenomenon. The proposed method may account not only for statistical correlations, but also for the perception of a pervasive emotional atmosphere in a poem.