Training state-of-the-art large language models requires vast amounts of clean and diverse textual data. However, building suitable multilingual datasets remains a challenge. In this work, we present HPLT v2, a collection of high-quality multilingual monolingual and parallel corpora, extending prior work of the HPLT project. The monolingual portion of the data contains 8T tokens covering 193 languages, while the parallel data contains 380M sentence pairs covering 51 languages. We document the entire data pipeline and release the code to reproduce it. We provide extensive analysis of the quality and characteristics of our data. Finally, we evaluate the performance of language models and machine translation systems trained on HPLT v2, demonstrating its value.

Modular deep learning is the state-of-the-art solution for lifting the curse of multilinguality, preventing the impact of negative interference and enabling cross-lingual performance in Multilingual Pre-trained Language Models. However, a trade-off of this approach is the reduction in positive transfer learning from closely related languages. In response, we introduce a novel method called language arithmetic, which enables training-free post-processing to address this limitation. Extending the task arithmetic framework, we apply learning via addition to the language adapters, transitioning the framework from a multi-task to a multilingual setup. The effectiveness of the proposed solution is demonstrated on three downstream tasks in a MAD-X-based set of cross-lingual schemes, acting as a post-processing procedure. Language arithmetic consistently improves the baselines with significant gains, especially in the most challenging case of zero-shot application. Our code and models are available at https://github.com/mklimasz/language-arithmetic.

Large language models increasingly support multiple languages, yet most benchmarks for gender bias remain English-centric. We introduce EuroGEST, a dataset designed to measure gender-stereotypical reasoning in LLMs across English and 29 European languages. EuroGEST builds on an existing expert-informed benchmark covering 16 gender stereotypes, expanded in this work using translation tools, quality estimation metrics, and morphological heuristics. Human evaluations confirm that our data generation method results in high accuracy of both translations and gender labels across languages. We use EuroGEST to evaluate 24 multilingual language models from six model families, demonstrating that the strongest stereotypes in all models across all languages are that women are beautiful, empathetic and neat and men are leaders, strong, tough and professional. We also show that larger models encode gendered stereotypes more strongly and that instruction finetuned models continue to exhibit gendered stereotypes. Our work highlights the need for more multilingual studies of fairness in LLMs and offers scalable methods and resources to audit gender bias across languages.

Filtering data, particularly data scraped from the internet, has long been recognised as a means to improve model performance. Recent studies have shown that effective filters can be created by utilising Large Language Models (LLMs) to synthetically label data, which is then used to train smaller neural models for filtering purposes. However, this approach has been tested mainly in English. Our paper extends this approach to languages beyond English, including languages not officially supported by the LLM. We validate our results on the downstream task of NMT and demonstrate that our approach is effective at both filtering parallel text for translation quality and filtering for domain specificity. For training the filtering model, we experiment with two different objectives for finetuning pre-trained transformers, as well as an efficient approach based on *n*-gram language models.

Over the last few years, there has been great interest in applying large language models (LLMs) to problems in the finance industry, and the field needs a robust LLM benchmark to support this work. Current financial LLM benchmarks contain simple tasks which are not representative of real use cases and have test sets with licences that do not allow commercial use. In response, we release AveniBench, a permissively licensed benchmark that tests a group of six key finance-related skills: tabular reasoning, numerical reasoning, question answering, long context modelling, summarisation and dialogue. We refactor the test sets to ensure that metrics are comparable, providing a unified framework. Furthermore, AveniBench introduces two task difficulty modes, easy and hard, enabling scalable evaluation based on real-world deployment needs. We use our benchmark to evaluate a diverse set of 20 widely used LLMs, from small open-weight models to proprietary systems like GPT-4. This evaluation initiates our public leaderboard, providing valuable insights for future academic research and commercial development.

In this work, we present our submissions to the unconstrained track of the System subtask of the WMT 2025 Creole Language Translation Shared Task. Of the 52 Creole languages included in the task, we focus on translation between English and seven Lusophone Creoles. Our approach leverages known strategies for low-resource machine translation, including back-translation and distillation of data, fine-tuning pre-trained multilingual models, and post-editing with large language models and lexicons. We also demonstrate that adding high-quality parallel Portuguese data in training, initialising Creole embeddings with Portuguese embedding weights, and strategically merging best checkpoints of different fine-tuned models all produce considerable gains in performance in certain translation directions. Our best models outperform the baselines on the Task test set for eight out of fourteen translation directions. When evaluated on a more diverse test set, they surpass the baselines in all but one direction.

The rise of Modular Deep Learning showcases its potential in various Natural Language Processing applications. Parameter-efficient fine-tuning (PEFT) modularity has been shown to work for various use cases, from domain adaptation to multilingual setups. However, all this work covers the case where the modular components are trained and deployed within one single Pre-trained Language Model (PLM). This model-specific setup is a substantial limitation on the very modularity that modular architectures are trying to achieve. We ask whether current modular approaches are transferable between models and whether we can transfer the modules from more robust and larger PLMs to smaller ones. In this work, we aim to fill this gap via a lens of Knowledge Distillation, commonly used for model compression, and present an extremely straightforward approach to transferring pre-trained, task-specific PEFT modules between same-family PLMs. Moreover, we propose a method that allows the transfer of modules between incompatible PLMs without any change in the inference complexity. The experiments on Named Entity Recognition, Natural Language Inference, and Paraphrase Identification tasks over multiple languages and PEFT methods showcase the initial potential of transferable modularity.

This paper presents the system description of Samsung R&D Institute Poland participation in WMT 2022 for General MT solution for medium and low resource languages: Russian and Croatian. Our approach combines iterative noised/tagged back-translation and iterative distillation. We investigated different monolingual resources and compared their influence on final translations. We used available BERT-likemodels for text classification and for extracting domains of texts. Then we prepared an ensemble of NMT models adapted to multiple domains. Finally we attempted to predict ensemble weight vectors from the BERT-based domain classifications for individual sentences. Our final trained models reached quality comparable to best online translators using only limited constrained resources during training.

We introduce COMBO – a fully neural NLP system for accurate part-of-speech tagging, morphological analysis, lemmatisation, and (enhanced) dependency parsing. It predicts categorical morphosyntactic features whilst also exposes their vector representations, extracted from hidden layers. COMBO is an easy to install Python package with automatically downloadable pre-trained models for over 40 languages. It maintains a balance between efficiency and quality. As it is an end-to-end system and its modules are jointly trained, its training is competitively fast. As its models are optimised for accuracy, they achieve often better prediction quality than SOTA. The COMBO library is available at: https://gitlab.clarin-pl.eu/syntactic-tools/combo.

We introduce the COMBO-based approach for EUD parsing and its implementation, which took part in the IWPT 2021 EUD shared task. The goal of this task is to parse raw texts in 17 languages into Enhanced Universal Dependencies (EUD). The proposed approach uses COMBO to predict UD trees and EUD graphs. These structures are then merged into the final EUD graphs. Some EUD edge labels are extended with case information using a single language-independent expansion rule. In the official evaluation, the solution ranked fourth, achieving an average ELAS of 83.79%. The source code is available at https://gitlab.clarin-pl.eu/syntactic-tools/combo.

We present a system for cross-domain suggestion mining, prepared for the SemEval-2019 Task 9: Suggestion Mining from Online Reviews and Forums (Subtask B). Our submitted solution for this text classification problem explores the idea of treating different suggestions’ sources as one of the settings of Transfer Learning - Domain Adaptation. Our experiments show that without any labeled target domain examples during training time, we are capable of proposing a system, reaching up to 0.778 in terms of F1 score on test dataset, based on Target Preserving Domain Adversarial Neural Networks.