CO-LOCATED WITH EMNLP IN BUDAPEST, HUNGARY OCTOBER 24th - 29th 2026

Miryam de Lhoneux
KU Leuven

Jaap Jumelet
University of Groningen

Ahmet Üstün
Cohere

Alham Fikri Aji
MBZUAI

The MRL Workshop will be held on October 28, 2026 (Room: TBA).

Multilingual NLP is English-centric: what can we do about it?
by Miryam de Lhoneux (KU Leuven)

NLP has a long history of focusing mainly on English. While increasing efforts are being made towards making language technology more multilingual, English remains the language on which NLP technology is developed first, and applied to other languages next, which inevitably leads to degraded performance compared to English. This talk is about reversing this trend and putting multilinguality at the center of NLP development, rather than being relegated to an after-thought. I describe how typology can inform NLP experimental paradigms, using our recently proposed typologically diverse language sampling framework. I then describe how typology can inform model development, by studying the interaction between architectural design choices and typological properties of languages. I finally discuss a bottleneck to these approaches: multilingual data availability, authenticity, and quality, together with some mitigating strategies.

Towards Multilingual Linguistic Evaluation at Scale
by Jaap Jumelet (University of Groningen)

Language model evaluation is becoming increasingly more multilingual and culturally aware, but resources measuring their linguistic abilities remain limited in size and scale. Developing methods that measure the quality of the language that LLMs generate is important, since this ensures that users can communicate with these systems fluently in their native language. We approach linguistic evaluation from a "minimal pair" perspective, comparing model probabilities on a grammatical and minimally different ungrammatical counterpart. Our methodology extends linguistic evaluation to over 100 languages, providing detailed insights in LLM abilities on a wide range of languages. I will present our findings on the first version of MultiBLiMP, as well as our approach in its follow-up work, in which we focus on word order. I end with a future outlook on the state of linguistic evaluation, and how we can extend this approach to evaluating open-ended generation.

TBA
by Ahmet Üstün (Cohere)

TBA

Multilingual Benchmarking in the Frontier LLM Era
by Alham Fikri Aji (MBZUAI)

Frontier models are increasingly used globally, but it is becoming harder to know which languages they were trained on, how well they work across them, and where they fail. As model releases become less transparent, benchmarks are among the few forms of public evidence we have. In this talk, we examine how multilingual benchmarks have evolved, which ones are used in frontier-model releases, and what their use reveals about current evaluation practices. We discuss what future benchmarks for frontier models should measure. We end with a call for our community to make multilinguality a core dimension of how AI progress is evaluated.

Multilingual representation learning methods have recently been found to be extremely efficient for transfer learning between languages, and they demonstrate potential for adapting natural language processing (NLP) models into languages or tasks with little to no training resources. On the other hand, there are many aspects of such models that have potential for further development and analysis in order to prove their applicability in various contexts. These contexts include different NLP tasks and also understudied language families, which face important obstacles in achieving practical advances that could improve the state-of-the-art in NLP for various low-resource or underrepresented languages.

This workshop aims to bring together the research community of scientists studying different aspects of multilingual representation learning, currently the most promising approach to improve NLP in low-resource or underrepresented languages, and provide the rapidly growing number of researchers working on the topic with a means of communication and an opportunity to present their work and exchange ideas. The main objectives of the workshop will be:

•    • To construct and present a wide array of multilingual representation learning methods, including their theoretical formulation and analysis, practical aspects such as the application of current state-of-the-art approaches in transfer learning to different tasks, or studies on adaptation into previously under-studied contexts;
•    • To provide a better understanding of how language typology may impact the applicability of these methods and motivate the development of novel methods that are more generic or competitive in different languages;
•    • To promote collaborations in developing novel software libraries or benchmarks for implementing or evaluating multilingual models, which would accelerate progress in the field.

By allowing a communication means for research groups working on machine learning, linguistic typology, or real-life applications of NLP tasks in various languages to share and discuss their recent findings, our ultimate goal is to support rapid development of NLP methods and tools that are applicable to a wider range of languages.

Topics of interest include, but are not limited to:

   • Understanding the learning dynamics of multi-lingual representation learning methods

   • Multilingual pretraining for discriminative and generative downstream tasks

   • Probing and analysis of multilingual representations

   • New methods for multi-lingual representation learning

   • New approaches to language adaptation of NLP systems

   • Zero-shot and few-shot learning for multilingual NLP

   • Investigating and understanding transfer learning methods for adaptation of NLP systems into previously under-studied languages, such as morphologically-rich languages

   • Data sets, benchmarks or libraries for implementing and evaluating multi-lingual models

Research papers: We invite all potential participants to submit their novel research contributions in the related fields as short or long papers following the EMNLP 2025 paper format (anonymized with 4 or 8 pages excluding the references, and an additional page for the camera-ready versions for the accepted papers). All accepted research papers will be published as part of our workshop proceedings and presented either as oral or poster presentations. Our research paper track will accept submissions through our own submission system available at MRL 2026 OpenReview Submission.

ARR Submissions: We accept papers previously submitted to ARR. Papers must have been submitted to the ARR May 2026 cycle or an earlier cycle and have received reviews and a meta review. We accept the submission via OpenReview, accompanied by a PDF copy of the reviews and meta-review (on the ARR Review page, press Print → Save as PDF).

Extended abstracts: Besides full paper submissions, we also invite previously published or ongoing and incomplete research contributions to our non-archival extended abstract track. All extended abstracts can use the same EMNLP template with a 2-page limit, excluding the bibliography.

As part of the MRL 2026 workshop, we are organizing another collaborative shared task. We are asking contributors to write culturally specific question-answer pairs in their language, leading to an ECLekTIc-style Multimodal and Multilingual Benchmark. All contributors to the dataset will be invited to be authors on the dataset paper, which we will submit for publication next year.

If you are interested in contributing or getting updates, please fill out the Interest Form. You can also see the Shared Task Website for up-to-date information and join our Discord to meet potential collaborators!

   • Aug 10, 2026: Regular paper submission deadline

   • August 1, 2026: Shared Task Phase 1 submission deadline

   • August 14, 2026: Shared Task Phase 2 submission deadline

   • Sept 4, 2026: ARR submission deadline

   • Sept 20, 2026: Notification of acceptance

   • Sept 25, 2026: Camera-ready papers due

   • Oct. 28, 2026: MRL Workshop (9:00-17:30)

   • Oct. 24-29, 2026: Main conference

(All deadlines are 11.59 pm UTC -12h (“anywhere on Earth”))

Interested in being a Sponsor? Contact us!