Textsetting as constraint conflict

Halle & Lerdahl (1993) lay out the problem of textsetting: when singers encounter a novel stanza for a song they know, they have consistent intuitions about where the syllables of each line should be aligned in time when the new stanza is sung. In other words, people have a productive ability for setting text to rhythm. Halle & Lerdahl offer the first explicit proposal for modeling this ability. The present paper likewise proposes a formal model of textsetting, but using a different theoretical approach. I argue that many well-formed textsettings represent the best possible resolution between conflicting metrical principles. These involve: (a) matching of stress to rhythmically strong position; (b) avoidance of long lapses (sequences where no syllable is initiated); (c) avoidance of extreme syllable compression; and (d) alignment of phonological phrase boundaries with line boundaries. A good textsetting often must sacrifice perfect realization of one of these goals in order to satisfy another goal that takes higher priority. For instance, many lines place stressed syllables in weaker rhythmic positions, and stressless syllables in stronger positions, in order to avoid a long syllable lapse, thus sacrificing (a) to satisfy (b). I formalize this approach under Optimality Theory (Prince & Smolensky 1993/2004). Using data from Hayes & Kaun (1996), in which native English speakers spontaneously set many lines of verse, I show that an approach based on constraint conflict offers considerable improvement in the accuracy with which the native speakers’ settings are predicted.