In this work, we study how semantic narrative enrichment and syntactic encoding affect the performance of knowledge graph embedding models. Narratives are central to human understanding, yet their structured representation in knowledge graphs remains challenging due to their semantic and structural complexity. Although traditional knowledge graphs use binary relations, they struggle to capture richer narrative elements, such as roles and causality. More complex knowledge graph syntaxes such as rdf-star, reification, singleton or n-ary properties offer greater modeling flexibility, but their impact on downstream tasks such as link prediction remains unclear. We define a six-level categorization of narrative semantics and use it to construct a suite of structured knowledge graphs using four syntactic representations. Using different embedding models, we evaluate how semantic and syntactic factors influence the embedding quality. We find that semantic features such as properties and subevents generally enhance performance, while roles tend to have a detrimental effect. On the syntactic side, although differences were not statistically significant across all metrics, reification and rdf-star achieved the strongest results on average.