Blockchain technology is nowadays a prominent solution to ensure trustworthiness in the enactment of multi-party business processes, typically described in terms of BPMN choreography models. These blockchain-based enactment approaches are based on the translation of choreographies into smart contracts. However, existing proposals aim at optimizing the execution costs, resulting in smart contract code that does not intuitively reflect the behavior of the corresponding choreography, thus hindering its understandability. To overcome this issue, we propose a translation approach that is modular and semantics-driven. The aim is to generate smart contracts that have a one-to-one correspondence with the elements of the originating BPMN choreographies, thus increasing the trust in the overall solution. Our approach is agnostic about the language used to write the smart contracts, which makes the approach general and applicable to different blockchain platforms. To assess the feasibility of our proposal, and the execution costs that it induces, we experimented with Solidity and PyTeal smart contracts resulting from the translation of a choreography from the healthcare application domain.