Analysis and computational modelling of Emirati Arabic intonation – A preliminary study

Abstract

This study is a preliminary investigation of intonation in Emirati Arabic (EA) (an under-researched Arabic dialect), using systematic acoustic analysis and computational modelling. First, we investigated the prosodic realisation of information focus and contrastive focus at sentence-initial, -penultimate and -final positions. The analysis of 1980 EA utterances produced by eleven EA native speakers revealed that (1) in focused words, only contrastive focus is realised with expanded excursion size, longer duration, and stronger intensity relative to their neutral focus counterparts, (2) post-focus words have a lower f0 and weaker intensity in both contrastive focus and information focus, and (3) pre-focus words have compressed excursion size and relatively short duration. We then used computational modelling to test how much of the EA intonation could be captured by the PENTA model, with focus-defined functional categories and a number of other, putative categories. PENTAtrainer was trained on syllable-sized multi-functional targets from a subset of the production data. The model then generated f0 contours with the learned targets and imposed them on resynthesised speech for perceptual evaluation. A comparison of the model-generated f0 contours with the natural f0 contours showed that not only focus but also weight, stress, position of word-level stressed syllable and prosodic word are important factors determining the fine details of EA intonation. A perceptual test with native EA listeners showed that the synthetic EA f0 contours sounded nearly as natural as the original intonation, and could convey focus nearly as accurately as natural intonation.