Greek Root Allomorphy without Spans

In this paper, we focus on a set of data from Greek root-allomorphy that has been recently presented by Merchant (2015) as a counterexample to the strong hypothesis from Embick (2010) that linear adjacency between the trigger and the target is required for morphosyntactically-conditioned allomorphy to arise. We take issue with Merchant’s proposed alternative hypothesis which essentially eliminates adjacency as a locality condition for allomorphy, due to the fact that one such solution leads to the loss of a striking (and not language-specific) implicational generalization, namely, that root-allomorphy never occurs in the presence of overt verbalizers, a straightforward blocking effect under the linear adjacency hypothesis. We propose an alternative account of the problematic cases of Greek root-allomorphy, whereby ASP and VCE form a single node via post-syntactic re-bracketing, allowing us to save the linear adjacency hypothesis.