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

2026-06-15

AI Summary

This work addresses the challenge of balancing translation quality and latency in long-form simultaneous speech translation by proposing a cascaded system that integrates the Parakeet speech model with the Qwen 3.5 large language model. The system employs an adaptive “black-box” strategy for efficient simultaneous translation and further refines the quality–latency trade-off through policy relaxation. It is the first to compete across all language directions and introduces a context-aware track for English-to-German, Italian, and Chinese, incorporating ASR word-level augmentation and a retrieval-augmented generation (RAG) mechanism grounded in pre-translated examples to inject domain-specific context. On the MCIF English-to-German test set, the system achieves a 5.82-point improvement in XCOMET-XL over last year’s submission, with an additional 1.03-point gain in the context track, substantially outperforming the baseline.

Abstract

This work describes the participation of the MLLP-VRAIN research group in the shared task of the IWSLT 2026 Simultaneous Speech Translation track. Our submission utilizes the recently released Parakeet and Qwen 3.5 models to create a robust, cascaded solution for long-form SimulST through the use of adaptive "black-box" policies. We explore relaxations of these policies to achieve better quality-latency trade-offs. Compared to last year, we participate on all language directions. In addition to this, for the En$\rightarrow${De, It, Zh} directions we also participate in this year's new context track employing a combination of ASR word-boosting and a RAG mechanism of offline pre-translated exemplars to guide generation and enrich our system with domain-specific context. Finally, we provide a detailed latency analysis of our system. Compared to last year, results on the MCIF En$\rightarrow$De test set shows a substantial quality improvement of +5.82 XCOMET-XL. Our context track processing further improves performance by +1.03.

Problem

• Simultaneous Speech Translation
• quality-latency trade-off
• context-aware translation
• long-form speech

Innovation

• adaptive black-box policies
• cascaded SimulST
• ASR word-boosting
• RAG
• quality-latency trade-off