Neural machine translation (NMT) has become an indispensable part of translation research and practice. Despite significant quality improvements, however, organizational and linguistic hurdles stand in the way of the successful integration of NMT into specialist translation processes: for example, insufficient consideration of pragmatic context factors as well as grammatical, lexical and stylistic consistency at document level. Therefore, negative effects on the cohesion and coherence of the target text cannot yet be ruled out. Rupcic is now examining whether and to what extent NMT systems are suitable for use in legal translation due to their handling of text cohesion and coherence. She outlines the specifics of the subject from a monolingual perspective and compares machine-generated, post-edited and human-produced contract translations. Their investigation into specific translation regularities reveals both the risks and the great potential of using NMT in this specialist area.