This article investigates the process of o-fronting in Ukrainian and claims that it should be analysed as an effect of PAL-e, a constraint mandating that the consonant and the vowel agree in backness. The operation of PAL-e is opaque, due to the interaction with a depalatalisation process applying before [ε]. The analysis establishes four points of theoretical interest. First, from a typological perspective, PAL-e parallels PAL-i, postulated by Rubach (2000a) for Russian, which strengthens the contention that PAL constraints should be stated as agreement constraints that bring under the same denominator the palatalisation of consonants and the fronting of vowels. Second, the opacity of o-fronting cannot be analysed by appealing to MAX[F] constraints. Third, sympathy theory cannot handle the data either. Fourth, Optimality Theory must be modified to admit derivational levels.

We argue that there is no need to split phonological representations into two worlds: one syllabic and another in which word stress is calculated. We show that both syllable- and stress-related phenomena can be accounted for with a single set of representations, if traditional syllabic analysis is modified in one central respect: what is traditionally taken to be a coda–onset cluster is interpreted as two independent onsets enclosing an empty nucleus. Accordingly, our proposal may be understood as a development of the idea that underlies classical metrical grids, i.e. that stress-relevant units project to higher levels and are therefore visible for stress. The units in the proposal made here, however, are uniformly nuclei. Contentful nuclei are always projected, while their empty counterparts (i.e. codas in traditional approaches) may or may not be. The weightlessness of onsets directly follows from this approach.

Function words in standard Serbian fall into two classes: prosodically free functional elements on the one hand, and prosodically bound functional elements, or clitics, on the other. The two classes have overlapping syntactic distributions, yet exhibit clear prosodic differences. Free functional elements may become prosodic words if they are minimally disyllabic, or if focused, and in this they crucially differ from clitics, whose only prosodic option is to be included into the host prosodic word. It is argued that the prosodic properties of the two classes of function words cannot be captured by a single set of ranked constraints, and that one of the classes, the class of clitics, needs to be encoded lexically, by virtue of prosodic prespecification.