Amazigh (Berber) languages present a wealth of intricate phonological alternations involving glides and high vowels, some of which still resist standard phonological analyses. Glides typically appear in the immediate vicinity of a vowel, in complementary distribution with the corresponding high vowels: e.g. Tashlhiyt gru ‘pick up’ vs agraw ‘assembly’, bri ‘crush’ vs abraj ‘crushed seeds’. Based on this kind of observations, standard theories analyse glides and high vowels as phonetic reflexes of the same underlying segments. The problem arises with the corresponding dative forms grujas ‘pick to him’ and brijas ‘crush seeds to him’. Followed by a vowel-initial morpheme –as, U and I should normally surface as glides, leading to *grwas and *brjas. The key to understanding this paradox lies, we argue, in the morphosyntactic structure of dative formations. In line with recent works at the syntax – phonology interface (Marvin 2002, Marantz 2007, Pigott & Newell 2014, Samuels 2010), we show that vP corresponds to a phase where U and I are spelled-out as high vowels before the enclitic –as is added. The resulting hiatus is then resolved by means of j epenthesis, leading to grujas and brijas. In agraw and abraj, U and I surface as glides since they belong to the same phase (nP) as the neighboring vowel. External evidence for the hypothesis that non-alternating high vowels and the dative enclitic reside in distinct syntactic domains is drawn from emphasis spread. We show that the projections of category-forming heads, including vP and nP, are the maximal domain of emphasis spread in Tashlhiyt.