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

arXiv cs.CL Papers
simultaneous-speech-translation iwslt cascaded-system quality-latency-tradeoff rag

Summary

This paper describes the MLLP-VRAIN UPV system for the IWSLT 2026 Simultaneous Speech Translation task, using Parakeet and Qwen 3.5 models with adaptive 'black-box' policies and a RAG mechanism for context, achieving significant quality improvements.

arXiv:2606.17255v1 Announce Type: new 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.
Original Article
View Cached Full Text

Cached at: 06/17/26, 05:39 AM 

# MLLP-VRAIN UPV system for the IWSLT 2026 Simultaneous Speech Translation task
Source: [https://arxiv.org/abs/2606.17255](https://arxiv.org/abs/2606.17255)
[View PDF](https://arxiv.org/pdf/2606.17255)

> 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\.

## Submission history

From: Jorge Iranzo\-Sánchez \[[view email](https://arxiv.org/show-email/a30262f8/2606.17255)\] **\[v1\]**Mon, 15 Jun 2026 19:57:05 UTC \(464 KB\)

Similar Articles

NAVER LABS Europe Submission to the Instruction-following 2026 Short Track

arXiv cs.CL

This paper describes NAVER LABS Europe's submission to the IWSLT 2026 instruction-following short track, improving upon their previous winning system by using a new speech projector (SpeechMapper) trained solely on ASR data and augmenting training with a synthetic SQA dataset (fakACL). The resulting system ties for first place in the constrained track while using a weaker LLM backbone.

Streaming Speech-to-Text Translation with a SpeechLLM

arXiv cs.CL

Presents a SpeechLLM architecture for streaming speech-to-text translation that adaptively decides when to output tokens based on audio, achieving 1-2 second latency with quality close to non-streaming baselines.