This paper introduces a new way to integrate gradient language redundancy effects and their acoustic correlates in LFG. Based on data from a production experiment that showed how semantic priming and lexical frequency affect target word duration, the approach models the inverse relationship between language redundancy and acoustic redundancy, with p-structure as a pivotal point between grammar and signal. To this end, all redundancy measures are re-scaled to a common, meaningful scale, while their gradient nature is retained as part of the system. The result not only allows for gradient data to be integrated into the architecture of LFG, but also for the prediction of concrete acoustic measures, thus taking a large step towards the modelling of the phonology-phonetic interface and the generation of spoken language.
Bögel et al. (Thu,) studied this question.