Semantic Preprocessing and Selective Translation of PowerPoint Slides using Embeddings and GPT-4o

This article presents an artificial intelligence approach for automatic translation of presentation slides using semantic embedding’s and a lightweight neural network for intelligent preprocessing. The system aims to selectively translate only relevant slide content – excluding references, metadata, and technical terms – by semantically classifying text chunks before passing them to the GPT-4o model. The approach enhances translation quality, reduces token consumption, and maintains slide formatting integrity. The proposed method demonstrates significant gains in translation efficiency and contextual accuracy, especially for academic educational presentations. The obtained results confirm the potential of integrating semantic models and generative language systems in the field of automated translation of educational materials. Future research plans include extending the system for multilingual support and adaptation to various presentation formats. Figs.: 5. Ref.: 12 items.