Multilingual instruction tuning boosts open-ended question answering accuracy
Task-Preserving Multilingual Instruction Tuning for Open-Ended QA in WMT 2026 MIST
Computation and Language
Summary
Open-ended question answering in many languages is challenging because models often struggle to perform well across different tasks and languages. This paper describes how the authors improved a language model by fine-tuning it with special multilingual instruction data that keeps the tasks consistent while adapting to different languages. Their approach showed better performance on questions and summarization for certain languages, but less so on others and on different types of question tasks. This shows that improving models across many languages and tasks requires careful data design.
What this means in practice
- •For multilingual ai developers: Develop improved multilingual open-ended QA systems by fine-tuning models on task-preserving localized instruction data.
- •For machine translation teams: Enhance cross-lingual QA capabilities by using cross-lingual augmentation that links instructions and responses in different languages.
Authors
Aleix Sant, Jordi Luque, Carlos Escolano
Abstract
This paper describes our submission to the WMT 2026 Multilingual Instruction Shared Task. We adapt Qwen3-8B with LoRA using organiser-provided data, EuroAlpaca, a multilingual instruction dataset created through task-preserving localisation, and CrossEuroAlpaca, a cross-lingual augmentation that assigns different source and target languages to instructions, contexts and responses. Our system targets open-ended question answering (QA-OEG) in ten shared-task languages represented in the auxiliary data. Under matched decoding, adaptation improves the mean Gemma-4 judge score on target-language QA-OEG by 7.8% relative to the base model. On an additional evaluation using the Aya Evaluation Suite across all shared-task languages, it yields relative gains of 41.5% in macro-averaged ROUGE-L and 2.9% in BERTScore-F1, with larger improvements in target languages. However, these gains are task- and language-specific. Gemma-4 scores improve for QA-OEG and summarisation in the target languages but decline for context-grounded QA there and across all three tasks in the non-target languages.