Stabilizing determinants in the transmission of phonotactic systems: on the emergence

The diachronic development of Afrikaans phonotactics is characterized by a considerable number of processes that affect consonant clusters, i.e. sequences of two or more consonants, especially when occurring in the word-final coda position (cf. Watermeyer 1996; Roberge 2002). For example, clusters were reduced in direk /dɪrək/ ('direct', Dutch direct) or nag /nɑx/ ('night', Dutch nacht). Nevertheless, consonant clusters still surface word finally in numerous lexical items (e.g. /rk/ in kerk, 'church' or /ns/ in mens, 'man'). Several linguistic, cognitive and social factors have been considered to be involved in such phonotactic repair processes. In particular, it has been hypothesized that generally, clusters are more likely to get stably transmitted the greater the phonological difference between the respective consonants involved is. That is, clusters are expected to get reduced if their building blocks share the same place and manner of articulation or if they agree in voicing (Dziubalska-Kołaczyk 2014). Which of these factors has the largest impact on consonant clusters, is, however, not a priori evident, although research on sonority (Berent et al. 2007) and the neuro-cognitive organization of phonemes (Mesgarani et al. 2014) seems to imply that manner-of-articulation differences are particularly important for the stability of clusters. The present study seeks to determine the relative importance of (i) manner-of-articulation differences (ΔMOA), (ii) place-of-articulation differences (ΔPOA) and (iii) differences in consonant voicing (ΔVOICE) for the establishment of the Afrikaans inventory of word-final morpheme-internal biconsonantal coda clusters, by comparing Afrikaans to its sibling Dutch. This is done in two steps. First, a preliminary study will assess the impact of the three phonological factors on the learnability of consonant clusters. In order to do so, the respective age of acquisition (AOA) is estimated for each cluster type based on age-of-acquisition ratings of Dutch core vocabulary lexemes ending in a cluster (Brysbaert et al. 2014). The variable ΔMOA is operationalized by calculating pairwise differences between sonority scores (cf. Dziubalska-Kołaczyk 2014). In a similar way, ΔPOA values are calculated as the pairwise differences of ordinal scores reflecting the place of articulation, while ΔVOICE is operationalized as a binary categorical variable. The (Box-Cox) transformed estimated AOA is then implemented as an outcome variable into a generalized linear model (GLM) in which ΔMOA, ΔPOA, and ΔVOICE serve as predictors, additionally controlling for potentially intervening factors such as token frequency and word length. A top-down model-optimization procedure reveals that depending on the interacting control variable, ΔPOA and ΔVOICE …