Cross-category harmony is one of the most well-known typological universals. It describes the tendency for syntactic heads to be ordered consistently relative to their dependents across different phrase types. Explanations for this universal vary as to whether cognitive factors play a role, or instead the tendency is due to mechanisms of language change alone. In this paper we report a series of artificial language learning experiments that aim to test a hypothesized link between cognition and cross-category harmony. As with the typological tendency itself, we find mixed evidence for harmony across different types of phrases. Specifically, learners are biased in favor of harmony between the order of verb and object and the order of adposition and noun. However, the bias for harmony between the order of verb and object and the order of adjective and noun depends on the semantic similarity between adjectives and verbs. When adjectives are active and therefore more verb-like (e.g., ‘broken'), we find harmony; when they are stative and therefore less verb-like (e.g., ‘blue'), we do not. These results suggest that the bias for cross-category harmony is not purely based on the syntactic notions of head and dependent, but reflects the interaction between a general cognitive bias for simplicity and cross-category similarity (Culbertson and Kirby, 2016; Culbertson and Kirby, 2022).