People conceptualize auditory pitch as vertical space: low and high pitch correspond to low and high space respectively. The strength of this cross-modal correspondence, however, seems to vary across different cultural contexts and a debate on the different factors underlying this variation is currently taking place. According to one hypothesis, pitch mappings are semantically mediated. For instance, the use of conventional metaphors such as ‘falling’ or ‘rising’ melodies strengthens a pitch-height mapping to the detriment of other possible mappings (e.g. pitch as bright/dark color or small/big size). Hence, entrenched pitch terms shape specific conceptualizations. The deterministic role of language is called into question by the hypothesis that different pitch mappings share a less constraining conceptual basis. As such, conceptual primitives may be concretized ad hoc into specific domains so that more local variation is possible. This claim is supported, for instance, by the finding that musicians use language-congruent (conventional) and language-incongruent (ad hoc) mappings interchangeably. The present paper substantiates this observation by investigating the head movements of musically trained and untrained speakers of Dutch in a melody reproduction task, as embodied instantiations of a vertical conceptualization of pitch. The OpenPose algorithm was used to track the movement trajectories in detail. The results show that untrained participants systematically made language-congruent movements, while trained participants showed more diverse behaviors, including language-incongruent movements. The difference between the two groups could not be attributed to the level of accuracy in the singing performances. In sum, this study argues for a joint consideration of more entrenched (e.g. linguistic metaphors) and more context-dependent (e.g. musical training and task) factors in accounting for variability in pitch representations.