It is often assumed that prosody and segmental inventory of languages do not interact – while some authors claim that those kinds of interactions are impossible (De Lacy 2007, Rasin 2017), others argue that stress can choose the type of vowel it will be associated with (Becker and Jurgec 2017, Trommer 2011, van Oostendorp and d'Alessandro 2015). This talk provides evidence from vowel reduction patterns in three different dialects of Serbo-Croatian that prosody can interact with segments, creating cumulative effects. Patterns from other languages will also be presented as additional evidence in support of the claim that prosody can prefer features.  A sketch of the analysis will be given in Harmonic Grammar (Smolensky and Legendre 2006, McCarthy and Pater 2016).