Karen Jesney (Carleton University) will give an in-person colloquium.
Important scheduling note: This will take place at 3:00, not at the usual time of 3:30.
Segmental markedness and the acquisition of consonant clusters: Learning predictions vs. corpus data
Consonant clusters introduce multiple sources of markedness, all of which must be mastered in the course of phonological acquisition. Weighted constraint-based models of phonological grammar (Legendre, Miyata & Smolensky 1990, Smolensky & Legendre 2006) predict that these sources of markedness will interact in a cumulative fashion, with two specific results. First, clusters that contain marked elements – such as the fricative [χ] or [θ] – will be acquired more slowly than those that lack these elements. Second, when clusters with marked elements are repaired, the preferred repair will be deletion of the marked element. This talk examines the basis of these predictions and tests them against corpus data. Two main case studies are presented. The first case study draws on data from the acquisition of Dutch in the CLPF database (Fikkert 1994, Levelt 1994), and demonstrates that clusters containing marked elements are, in fact, acquired more slowly than other clusters. At the same time, the precise repair patterns observed – specifically the choice of which segment is deleted – are, to the extent possible, governed by sonority rather than segmental markedness. The second case study looks at data from the acquisition of English clusters involving target [θ] vs. [f] in the Providence database (Demuth, Culbertson & Alter 2006). Here too, clusters containing more marked elements are, in fact, acquired more slowly than other clusters, but again the repair patterns diverge from those predicted by the models. The talk concludes with discussion of what these complex patterns mean for the interpretation of simulation results and for our models of grammar.
