EXPLAINABLE AI FOR TRANSLATOR TRUST

This paper explores how explainable-AI (XAI) techniques – visual attention maps, feature-attribution graphs and confidence heat-lines – can be operationalised in real-time dashboards that let professional translators audit the reasoning of neural-machine-translation (NMT) engines. Building on recent work in MT interpretability and user-centred design, we synthesise design principles (granular salience, cognitive frugality, cross-modal alignment, and low-resource adaptability) and analyse three case studies – Translation Canvas, NMT Visualising Tools and an open-source XAI toolkit – demonstrating their impact on trust and revision effort. We also consider the Uzbek MT ecosystem, showing how locally authored corpora and Turkic-language initiatives expand XAI research. Recommendations are offered for future hybrid human-AI workflows.