<p>This study investigates whether cohesive and coherent patterns differ across human-translated, machine-translated and non-translated English texts, and whether these patterns remain consistent across four distinct registers. Drawing on five categories of metrics from Coh-Metrix 3.0, namely referential cohesion, personal pronouns, connectives, latent semantic analysis and situation model, the analysis employs principal component analysis, flexible discriminant analysis and Permutational Multivariate Analysis of Variance to triangulate results. The findings reveal that: (i) academic texts exhibit significantly higher levels of cohesion and coherence than other registers, particularly in coreference, semantic similarity, logical connectivity and intentionality, whereas fictional texts, shaped by story-telling conventions, tend to create cohesive chains through anaphoric reference to maintain narrative fluidity and character interaction; (ii) both human and machine translations show a general tendency toward explicitation in comparison to non-translated texts, although this trend is not consistent across all cohesive and coherent dimensions or registers; and (iii) register variation exerts a stronger influence on cohesive and coherent patterns than translation variety. These results underscore the importance of adopting nuanced, context-sensitive approaches to studying translated language and of situating such inquiries within both technological and functional frameworks. Moreover, the observed patterns provide evidence for the hypothesis of risk aversion, suggesting that human translators often adopt risk-averse strategies to reduce potential misunderstandings, while the explicitation in machine translations may reflect underlying algorithmic biases. Taken together, these findings contribute theoretical, methodological and practical insights to the ongoing investigation of translation universals and the evolving role of machine translation in translation practice and pedagogy.</p>