The interaction between the English regular plural affix (PL) and possessive clitic (POSS) presents a theoretical puzzle (Zwicky, 1975). Both have the form /z/, and so the OCP (Yip, 1998) predicts their combination (PL+POSS) should trigger epenthesis. Yet, in cases like my friends’ /fɹenz/ car, only PL is overtly realized. Why does the OCP fail to apply?
Two previous theories address this non-application of the OCP in PL+POSS constructions. The POSS-suppression theory (Stemberger, 1981; Zwicky, 1987) claims that POSS essentially inspects the morphological composition of its host and is actively suppressed by adjacent PL /z/, without exception. The alternative POSS-allomorphy theory (Bernstein & Tortora, 2005; Nevins, 2011) claims that POSS has a phonologically null allomorph, which is chosen when the possessor has the plural feature. Either POSS allomorph may be chosen for a singular possessor with embedded PL; thus, contra the suppression theory, epenthesis may be triggered in cases like the son of my friends’s /fɹenz ~ fɹenzəz/ car.
To address the contrasting claims of variability in PL+POSS constructions, we conducted a split-rating experiment (following Bresnan, 2007) in which participants compared the naturalness of written PL+POSS pronunciations with and without epenthesis. The results show that the inclination to realize POSS is variable across individuals and is gradient, depending on multiple factors. Crucially, participants rated the epenthesis strategy higher with all embedded possessors, regardless of their number feature: /bɔɪzəz/ was judged equally natural in both [one of [the boys]]’s dog and [two of [the boys]]’s dog, and more natural in both cases than in [the boys]’s dog.
The fact that epenthesis was not judged differently for singular and plural possessors is inconsistent with the POSS-allomorphy theory. By contrast, the POSS-suppression theory can successfully account for the results, if extended with a locality constraint. Under this extension, the ability of POSS to inspect the morphological composition of its host is hampered by the presence of an intervening constituent boundary. In instances like [one of [the boys]]’s dog, the structure of the internal constituent [the boys] may be invisible, meaning that the final /z/ is not registered as PL and does not enforce POSS-suppression. POSS can thus be realized as /z/ and trigger epenthesis, following the OCP. The observed structured variability implies that this is one among many interacting soft constraints.
This analysis has implications for the interleaving of phonology with morphology and syntax. To enact POSS-suppression, phonological processing at the attachment of POSS must not be blind to existing morphological structure, as is expected from Bracketing Erasure (Pesetsky, 1979). However, existing structure cannot remain equally available at all levels of derivation, as is assumed in standard Optimality Theory (Prince & Smolensky, 2004). The present results suggest that, at least at the phrasal level of phonology, an element attaching at the nth derivational cycle must have access to structure from the (n-1)th level, but may not have access to structures from previous levels.