MLLP-VRAIN UPV system for the IWSLT 2026 Simultaneous Speech Translation task

The MLLP-VRAIN research group announced their participation in the IWSLT 2026 Simultaneous Speech Translation task, utilizing the Parakeet and Qwen 3.5 models to develop a cascaded solution for long-form SimulST. They implemented adaptive "black-box" policies and explored relaxations to enhance quality-latency trade-offs, achieving a +5.82 improvement on the MCIF En→De test set and an additional +1.03 from context track processing. This work is significant for practitioners as it demonstrates advanced techniques for integrating ASR word-boosting and RAG mechanisms, potentially improving the efficiency and contextual relevance of simultaneous translation systems.