Articulatory dynamics of degemination in Dutch

In Dutch, derived clusters of identical consonants (“fake geminates”) are traditionally thought to undergo categorical degemination, though for fricative clusters it has been shown to be gradient durationally. Since Standard Dutch displays a pattern of strongly categorical /r/ allophony with coda-r a post-alveolar approximant and onset-r a uvular fricative/trill, /r#r/ clusters in a fake geminate context should show phonetically strongly distinct allophones (e.g. [ɻ#ʀ]). We investigate the spatial and temporal characteristics of coarticulation and possible degemination in these phonemically identical but phonetically disparate sequences.
Articulatory (high-speed ultrasound) and acoustic data from 8 speakers of Standard Dutch were analysed using dynamic (principal components analysis of pixel intensity data) and static (SS-ANOVA comparisons of tongue contours) measures, plus linear mixed-effects regression modelling of vowel+rhotic duration.
The results show that fake geminate-r entails an intermediate articulation between coda-r and onset-r, combining coda-like bunching with onset-like dorsal raising (gestural blending), with spatial reduction of both gestures vis-à-vis non-geminate onsets and codas. Durationally, however, degemination appears categorical, as the fake geminate context and singleton onset /r/ are non-distinct.
We discuss the problems these facts raise for both an Articulatory Phonology framework without a “phonemic” level, and more abstractionist models with discrete phonological and phonetic levels.