Impact of Discourse Marker Accuracy on Translation Quality: Fluency, Coherence, and Patterns of Misuse in Machine Translation

This research explored the pivotal role of discourse marker (DM) accuracy in machine transla-tion (MT) vs. human translation (HT) quality prediction in terms of fluency, coherence, and misuse patterns. The research, based on a mixed-methods design, quantified DM accuracy as precision, recall, and F1 scores, and qualitatively assesses text quality through human judgments and BERT-based coherence models. Findings showed that HT is much more accurate in DM (85–88% correlation with fluency/coherence) than MT (62–65%), with MT systems tending to overuse additive markers (and, so) and underuse contrastive/causal markers (but, therefore), and misuse however. These tendencies compromise discourse coherence, contribute to post-editing effort, and demonstrate the limits of BLEU-based measures in detecting discourse-level errors. The research calls for discourse-sensitive MT models, more informed evaluation metrics (e.g., Coh-Metrix, RST parsing), and pedagogical innovation in translator education to detect DM subtleties. Findings also pointed to ethical practice in MT-mediated communication and ex-tend an invitation to cross-lingual research in low-resource language translation development. By combining theoretical linguistics and computational practice, the research takes steps for-ward in balancing DM-based errors and facilitating multilingual communication in a world that is progressively digitalized.