Abstract
The chapter highlights the role of AI and modern language models in the transformation of spoken translation, combining a historical overview with practical architectures. Central to the chapter is the model of a multi-agent translator assistant that performs speaker identification, segmentation, translation, and quality control in real time. The use of Retrieval Augmented Generation is considered for loading glossaries and reference materials from local storage. The application of AR glasses for subtitles and prompts without diverting attention is demonstrated. The concept of personal conferences with agent-based avatars delivering presentations from preloaded materials is proposed. This supports event preparation and skill development. Modern methods of translation quality evaluation are discussed, including the GEMBA metric, multi-agent evaluation, and new indicators integrated into INFERScore. The development of these approaches will enable cognitive alignment with human understanding.

This publication has 24 references indexed in Scilit: