Traditional theories of grammar posit that sentence formation begins with the underlying hierarchical structure generated by the syntactic component, which is then given meaning and sound interpretations by the semantic and phonological components (Chomsky, 1965). This dissertation provides evidence that word ordering is conditioned by the avoidance of phonologically-marked structures, challenging strictly feed-forward theories of grammar. Previous work in various languages shows phonological effects on word ordering, such as: binomials and sentence formation in English (Morgan, 2016; Breiss and Hayes, 2020); compounds in Navajo and English (Martin, 2011); topicalization in Serbo-Croatian (Inkelas and Zec, 1995); and noun-adjective ordering in Tagalog (Shih and Zuraw, 2017). This work examines the effects of six phonological markedness constraints on flexible noun-adjective ordering in five languages based on the analysis of large speech corpora. Constraints include stress clash and lapse, vowel hiatus, consonant clusters that disagree in voicing, consonant clusters that agree in place of articulation, and relative word length. The languages analyzed are French and Italian (Romance), Polish (Slavic), Hindi (Indo-Aryan), and Modern Standard Arabic (West Semitic). The first of two central hypotheses of this dissertation is that only those phonologically-marked phenomena that are avoided with phonological repairs will also be avoided via word ordering. This is tested by looking at the avoidance of two types of phonological constraints in each language, those that are active in the language (e.g., vowel hiatus in French), and those that are inactive in the language (e.g., consonant clusters at the same place of articulation in French). In general, results support this hypothesis. For example, vowel hiatus was found to be avoided via word ordering in French noun-adjective pairs, but not in Italian where it is not active; and, there is evidence that consonant clusters across the word boundary that share a place of articulation are not avoided in French, but are avoided in Polish. The second central hypothesis of this work is that phonological effects on ordering are stronger if semantic differences between orders are minimal. To test this hypothesis, semantic difference between the prenominal and postnominal form of an adjective is quantified by the similarity between positional embeddings; noun-adjective pairs with an adjective above a determined similarity threshold are analyzed separately from those below the threshold. Results did not support this hypothesis. There were no phonological effects that were greater or present only in the semantically-similar dataset. Overall, this dissertation argues that noun-adjective ordering is conditioned by phonological markedness effects at the prosodic, syllabic, and segmental levels. This finding is discussed with respect to existing theories of the interface between phonology and syntax, concluding that at a minimum, phonological factors compete with others to determine the output of the grammar.