Iranzo-Sánchez, Jorge ; Mas-Mollà, Gerard ; Gimenez, Adrià ; Civera Saiz, Jorge ; Sanchis, Albert ; Juan, Alfons MLLP-VRAIN UPV System for the IWSLT 2026 Simultaneous Speech Translation Task Inproceedings Proc. 23rd Intl. Conf. on Spoken Language Translation (IWSLT 2026), pp. 212–226, San Diego (USA), 2026. Abstract | Links | BibTeX | Tags: Cascade System, Context-Augmented SimulST, Simultaneous Speech Translation @inproceedings{Iranzo-Sánchez2026,
title = {MLLP-VRAIN UPV System for the IWSLT 2026 Simultaneous Speech Translation Task},
author = {Iranzo-Sánchez, Jorge AND Mas-Mollà, Gerard AND Gimenez, Adrià AND Civera Saiz, Jorge AND Sanchis, Albert AND Juan, Alfons},
doi = {10.18653/v1/2026.iwslt-1.24},
year = {2026},
date = {2026-01-01},
booktitle = {Proc. 23rd Intl. Conf. on Spoken Language Translation (IWSLT 2026)},
pages = {212--226},
address = {San Diego (USA)},
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→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→De test set shows a substantial quality improvement of +5.82 XCOMET-XL. Our context track processing further improves performance by +1.03.},
keywords = {Cascade System, Context-Augmented SimulST, Simultaneous Speech Translation},
pubstate = {published},
tppubtype = {inproceedings}
}
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→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→De test set shows a substantial quality improvement of +5.82 XCOMET-XL. Our context track processing further improves performance by +1.03. |
