This dissertation bridges the gap between scientific insights into how language models work and practical benefits for users of these systems, paving the way for better human-AI interaction practices for professional translators and everyday users worldwide.