Simon Gonzalez, Gerard Docherty
Thursday, December 15th, 2022, 2.30pm – 3pm
The aim of this paper is to describe the initial development of a computational framework designed to automatically recognize and classify vowel-/l/ rhyme realisations produced by Australian English speakers as either consonantal or vocalised. We implemented a Random Forest model as the main classificatory technique. This allowed us to explore in a hierarchical way the contribution to the classification of a wide a range of potential predictors. The test classification accuracy of the Random Forest model was 82.1% overall, with its sensitivity estimated to be 73.7% (consonantal realisations) and the specificity to be 89.1% (vocalised realisations).