Rodrigo Agerri

Rodrigo Agerri, Eneko Agirre

Given the impact of language models on the field of Natural Language Processing, a number of Spanish encoder-only masked language models (aka BERTs) have been trained and released. These models were developed either within large projects using very large private corpora or by means of smaller scale academic efforts leveraging freely available data. In this paper we present a comprehensive head-to-head comparison of language models for Spanish with the following results: (i) Previously ignored multilingual models from large companies fare better than monolingual models, substantially changing the evaluation landscape of language models in Spanish; (ii) Results across the monolingual models are not conclusive, with supposedly smaller and inferior models performing competitively. Based on these empirical results, we argue for the need of more research to understand the factors underlying them. In this sense, the effect of corpus size, quality and pre-training techniques need to be further investigated to be able to obtain Spanish monolingual models significantly better than the multilingual ones released by large private companies, specially in the face of rapid ongoing progress in the field. The recent activity in the development of language technology for Spanish is to be welcomed, but our results show that building language models remains an open, resource-heavy problem which requires to marry resources (monetary and/or computational) with the best research expertise and practice.

Iker García-Ferrero, Rodrigo Agerri, German Rigau

Zero-resource cross-lingual transfer approaches aim to apply supervised models from a source language to unlabelled target languages. In this paper we perform an in-depth study of the two main techniques employed so far for cross-lingual zero-resource sequence labelling, based either on data or model transfer. Although previous research has proposed translation and annotation projection (data-based cross-lingual transfer) as an effective technique for cross-lingual sequence labelling, in this paper we experimentally demonstrate that high capacity multilingual language models applied in a zero-shot (model-based cross-lingual transfer) setting consistently outperform data-based cross-lingual transfer approaches. A detailed analysis of our results suggests that this might be due to important differences in language use. More specifically, machine translation often generates a textual signal which is different to what the models are exposed to when using gold standard data, which affects both the fine-tuning and evaluation processes. Our results also indicate that data-based cross-lingual transfer approaches remain a competitive option when high-capacity multilingual language models are not available.

Iker García-Ferrero, Rodrigo Agerri, German Rigau

In the absence of readily available labeled data for a given sequence labeling task and language, annotation projection has been proposed as one of the possible strategies to automatically generate annotated data. Annotation projection has often been formulated as the task of transporting, on parallel corpora, the labels pertaining to a given span in the source language into its corresponding span in the target language. In this paper we present T-Projection, a novel approach for annotation projection that leverages large pretrained text-to-text language models and state-of-the-art machine translation technology. T-Projection decomposes the label projection task into two subtasks: (i) A candidate generation step, in which a set of projection candidates using a multilingual T5 model is generated and, (ii) a candidate selection step, in which the generated candidates are ranked based on translation probabilities. We conducted experiments on intrinsic and extrinsic tasks in 5 Indo-European and 8 low-resource African languages. We demostrate that T-projection outperforms previous annotation projection methods by a wide margin. We believe that T-Projection can help to automatically alleviate the lack of high-quality training data for sequence labeling tasks. Code and data are publicly available.

Mikel Artetxe, Itziar Aldabe, Rodrigo Agerri et al.

The vast majority of non-English corpora are derived from automatically filtered versions of CommonCrawl. While prior work has identified major issues on the quality of these datasets (Kreutzer et al., 2021), it is not clear how this impacts downstream performance. Taking representation learning in Basque as a case study, we explore tailored crawling (manually identifying and scraping websites with high-quality content) as an alternative to filtering CommonCrawl. Our new corpus, called EusCrawl, is similar in size to the Basque portion of popular multilingual corpora like CC100 and mC4, yet it has a much higher quality according to native annotators. For instance, 66% of documents are rated as high-quality for EusCrawl, in contrast with <33% for both mC4 and CC100. Nevertheless, we obtain similar results on downstream NLU tasks regardless of the corpus used for pre-training. Our work suggests that NLU performance in low-resource languages is not primarily constrained by the quality of the data, and other factors like corpus size and domain coverage can play a more important role.

Rodrigo Agerri, Iñigo Alonso, Aitziber Atutxa et al.

Providing high quality explanations for AI predictions based on machine learning is a challenging and complex task. To work well it requires, among other factors: selecting a proper level of generality/specificity of the explanation; considering assumptions about the familiarity of the explanation beneficiary with the AI task under consideration; referring to specific elements that have contributed to the decision; making use of additional knowledge (e.g. expert evidence) which might not be part of the prediction process; and providing evidence supporting negative hypothesis. Finally, the system needs to formulate the explanation in a clearly interpretable, and possibly convincing, way. Given these considerations, ANTIDOTE fosters an integrated vision of explainable AI, where low-level characteristics of the deep learning process are combined with higher level schemes proper of the human argumentation capacity. ANTIDOTE will exploit cross-disciplinary competences in deep learning and argumentation to support a broader and innovative view of explainable AI, where the need for high-quality explanations for clinical cases deliberation is critical. As a first result of the project, we publish the Antidote CasiMedicos dataset to facilitate research on explainable AI in general, and argumentation in the medical domain in particular.

Nayla Escribano, German Rigau, Rodrigo Agerri

Detecting and normalizing temporal expressions is an essential step for many NLP tasks. While a variety of methods have been proposed for detection, best normalization approaches rely on hand-crafted rules. Furthermore, most of them have been designed only for English. In this paper we present a modular multilingual temporal processing system combining a fine-tuned Masked Language Model for detection, and a grammar-based normalizer. We experiment in Spanish and English and compare with HeidelTime, the state-of-the-art in multilingual temporal processing. We obtain best results in gold timex normalization, timex detection and type recognition, and competitive performance in the combined TempEval-3 relaxed value metric. A detailed error analysis shows that detecting only those timexes for which it is feasible to provide a normalization is highly beneficial in this last metric. This raises the question of which is the best strategy for timex processing, namely, leaving undetected those timexes for which is not easy to provide normalization rules or aiming for high coverage.

Irune Zubiaga, Aitor Soroa, Rodrigo Agerri

Large language models (LLMs) are increasingly used for the automatic evaluation of generated text, yet most prior work focuses on English. Despite the growing demand for multilingual evaluation, extending LLM-based evaluators to multilingual settings remains challenging, particularly for low-resource languages and scenarios where in-domain data is scarce. This work explores several strategies for developing multilingual LLMs-as-a-judge, considering whether in-domain data is available for fine-tuning or not. We systematically analyze English, Spanish, and Basque, representing high-, mid-, and low-resource languages, considering instruction translation, monolingual versus multilingual supervision, and model size. For evaluation, we extend two existing meta-evaluation datasets to Basque and Spanish. Our results reveal key trade-offs: When in-domain data is available, fine-tuned smaller models can achieve performance comparable to proprietary models, whereas zero-shot evaluation with larger models proves more effective in out-of-domain settings. We also observe that fine-tuning on out-of-domain data can adversely affect model performance. These findings provide practical guidance for building efficient, reliable multilingual evaluation pipelines. The data and code are publicly available at hitz-zentroa/mJudge.

Oscar Sainz, Iker García-Ferrero, Rodrigo Agerri et al.

Large Language Models (LLMs) combined with instruction tuning have made significant progress when generalizing to unseen tasks. However, they have been less successful in Information Extraction (IE), lagging behind task-specific models. Typically, IE tasks are characterized by complex annotation guidelines that describe the task and give examples to humans. Previous attempts to leverage such information have failed, even with the largest models, as they are not able to follow the guidelines out of the box. In this paper, we propose GoLLIE (Guideline-following Large Language Model for IE), a model able to improve zero-shot results on unseen IE tasks by virtue of being fine-tuned to comply with annotation guidelines. Comprehensive evaluation empirically demonstrates that GoLLIE is able to generalize to and follow unseen guidelines, outperforming previous attempts at zero-shot information extraction. The ablation study shows that detailed guidelines are key for good results.

Olia Toporkov, Rodrigo Agerri

Lemmatization is a natural language processing (NLP) task which consists of producing, from a given inflected word, its canonical form or lemma. Lemmatization is one of the basic tasks that facilitate downstream NLP applications, and is of particular importance for high-inflected languages. Given that the process to obtain a lemma from an inflected word can be explained by looking at its morphosyntactic category, including fine-grained morphosyntactic information to train contextual lemmatizers has become common practice, without considering whether that is the optimum in terms of downstream performance. In order to address this issue, in this paper we empirically investigate the role of morphological information to develop contextual lemmatizers in six languages within a varied spectrum of morphological complexity: Basque, Turkish, Russian, Czech, Spanish and English. Furthermore, and unlike the vast majority of previous work, we also evaluate lemmatizers in out-of-domain settings, which constitutes, after all, their most common application use. The results of our study are rather surprising. It turns out that providing lemmatizers with fine-grained morphological features during training is not that beneficial, not even for agglutinative languages. In fact, modern contextual word representations seem to implicitly encode enough morphological information to obtain competitive contextual lemmatizers without seeing any explicit morphological signal. Moreover, our experiments suggest that the best lemmatizers out-of-domain are those using simple UPOS tags or those trained without morphology and, finally, that current evaluation practices for lemmatization are not adequate to clearly discriminate between models.

Elisa Sanchez-Bayona, Rodrigo Agerri

The lack of wide coverage datasets annotated with everyday metaphorical expressions for languages other than English is striking. This means that most research on supervised metaphor detection has been published only for that language. In order to address this issue, this work presents the first corpus annotated with naturally occurring metaphors in Spanish large enough to develop systems to perform metaphor detection. The presented dataset, CoMeta, includes texts from various domains, namely, news, political discourse, Wikipedia and reviews. In order to label CoMeta, we apply the MIPVU method, the guidelines most commonly used to systematically annotate metaphor on real data. We use our newly created dataset to provide competitive baselines by fine-tuning several multilingual and monolingual state-of-the-art large language models. Furthermore, by leveraging the existing VUAM English data in addition to CoMeta, we present the, to the best of our knowledge, first cross-lingual experiments on supervised metaphor detection. Finally, we perform a detailed error analysis that explores the seemingly high transfer of everyday metaphor across these two languages and datasets.

Anar Yeginbergen, Rodrigo Agerri

Nowadays the medical domain is receiving more and more attention in applications involving Artificial Intelligence as clinicians decision-making is increasingly dependent on dealing with enormous amounts of unstructured textual data. In this context, Argument Mining (AM) helps to meaningfully structure textual data by identifying the argumentative components in the text and classifying the relations between them. However, as it is the case for man tasks in Natural Language Processing in general and in medical text processing in particular, the large majority of the work on computational argumentation has been focusing only on the English language. In this paper, we investigate several strategies to perform AM in medical texts for a language such as Spanish, for which no annotated data is available. Our work shows that automatically translating and projecting annotations (data-transfer) from English to a given target language is an effective way to generate annotated data without costly manual intervention. Furthermore, and contrary to conclusions from previous work for other sequence labelling tasks, our experiments demonstrate that data-transfer outperforms methods based on the crosslingual transfer capabilities of multilingual pre-trained language models (model-transfer). Finally, we show how the automatically generated data in Spanish can also be used to improve results in the original English monolingual setting, providing thus a fully automatic data augmentation strategy.

Nayla Escribano, Jon Ander González, Julen Orbegozo-Terradillos et al.

Parliamentary transcripts provide a valuable resource to understand the reality and know about the most important facts that occur over time in our societies. Furthermore, the political debates captured in these transcripts facilitate research on political discourse from a computational social science perspective. In this paper we release the first version of a newly compiled corpus from Basque parliamentary transcripts. The corpus is characterized by heavy Basque-Spanish code-switching, and represents an interesting resource to study political discourse in contrasting languages such as Basque and Spanish. We enrich the corpus with metadata related to relevant attributes of the speakers and speeches (language, gender, party...) and process the text to obtain named entities and lemmas. The obtained metadata is then used to perform a detailed corpus analysis which provides interesting insights about the language use of the Basque political representatives across time, parties and gender.

Iker García-Ferrero, Rodrigo Agerri, Aitziber Atutxa Salazar et al.

Research on language technology for the development of medical applications is currently a hot topic in Natural Language Understanding and Generation. Thus, a number of large language models (LLMs) have recently been adapted to the medical domain, so that they can be used as a tool for mediating in human-AI interaction. While these LLMs display competitive performance on automated medical texts benchmarks, they have been pre-trained and evaluated with a focus on a single language (English mostly). This is particularly true of text-to-text models, which typically require large amounts of domain-specific pre-training data, often not easily accessible for many languages. In this paper, we address these shortcomings by compiling, to the best of our knowledge, the largest multilingual corpus for the medical domain in four languages, namely English, French, Italian and Spanish. This new corpus has been used to train Medical mT5, the first open-source text-to-text multilingual model for the medical domain. Additionally, we present two new evaluation benchmarks for all four languages with the aim of facilitating multilingual research in this domain. A comprehensive evaluation shows that Medical mT5 outperforms both encoders and similarly sized text-to-text models for the Spanish, French, and Italian benchmarks, while being competitive with current state-of-the-art LLMs in English.

Maxime Masson, Rodrigo Agerri, Christian Sallaberry et al.

The rising influence of social media platforms in various domains, including tourism, has highlighted the growing need for efficient and automated Natural Language Processing (NLP) strategies to take advantage of this valuable resource. However, the transformation of multilingual, unstructured, and informal texts into structured knowledge still poses significant challenges, most notably the never-ending requirement for manually annotated data to train deep learning classifiers. In this work, we study different NLP techniques to establish the best ones to obtain competitive performances while keeping the need for training annotated data to a minimum. To do so, we built the first publicly available multilingual dataset (French, English, and Spanish) for the tourism domain, composed of tourism-related tweets. The dataset includes multilayered, manually revised annotations for Named Entity Recognition (NER) for Locations and Fine-grained Thematic Concepts Extraction mapped to the Thesaurus of Tourism and Leisure Activities of the World Tourism Organization, as well as for Sentiment Analysis at the tweet level. Extensive experimentation comparing various few-shot and fine-tuning techniques with modern language models demonstrate that modern few-shot techniques allow us to obtain competitive results for all three tasks with very little annotation data: 5 tweets per label (15 in total) for Sentiment Analysis, 30 tweets for Named Entity Recognition of Locations and 1K tweets annotated with fine-grained thematic concepts, a highly fine-grained sequence labeling task based on an inventory of 315 classes. We believe that our results, grounded in a novel dataset, pave the way for applying NLP to new domain-specific applications, reducing the need for manual annotations and circumventing the complexities of rule-based, ad-hoc solutions.

Jaione Bengoetxea, Itziar Gonzalez-Dios, Rodrigo Agerri

Recent research on dialectal NLP has identified data scarcity as a primary limitation. To address this limitation, this paper presents a catalog of contemporary Basque dialectal data and resources, offering a systematic and comprehensive compilation of the dialectal data currently available in Basque. Two types of data sources have been distinguished: online data originally written in some dialect, and standard-to-dialect adapted data. The former includes all dialectal data that can be found online, such as news and radio sites, informal tweets, as well as online resources such as dictionaries, atlases, grammar rules, or videos. The latter consists of data that has been adapted from the standard variety to dialectal varieties, either manually or automatically. Regarding the manual adaptation, the test split of the XNLI Natural Language Inference dataset was manually adapted into three Basque dialects: Western, Central, and Navarrese-Lapurdian, yielding a high-quality parallel gold standard evaluation dataset. With respect to the automatic dialectal adaptation, the automatically adapted physical commonsense dataset (BasPhyCowest) underwent additional manual evaluation by native speakers to assess its quality and determine whether it could serve as a viable substitute for full manual adaptation (i.e., silver data creation).

Jaione Bengoetxea, Itziar Gonzalez-Dios, Rodrigo Agerri

Physical commonsense reasoning represents a fundamental capability of human intelligence, enabling individuals to understand their environment, predict future events, and navigate physical spaces. Recent years have witnessed growing interest in reasoning tasks within Natural Language Processing (NLP). However, no prior research has examined the performance of Large Language Models (LLMs) on non-question-answering (non-QA) physical commonsense reasoning tasks in low-resource languages such as Basque. Taking the Italian GITA as a starting point, this paper addresses this gap by presenting BasPhyCo, the first non-QA physical commonsense reasoning dataset for Basque, available in both standard and dialectal variants. We evaluate model performance across three hierarchical levels of commonsense understanding: (1) distinguishing between plausible and implausible narratives (accuracy), (2) identifying the conflicting element that renders a narrative implausible (consistency), and (3) determining the specific physical state that creates the implausibility (verifiability). These tasks were assessed using multiple multilingual LLMs as well as models pretrained specifically for Italian and Basque. Results indicate that, in terms of verifiability, LLMs exhibit limited physical commonsense capabilities in low-resource languages such as Basque, especially when processing dialectal variants.

Anar Yeginbergen, Maite Oronoz, Rodrigo Agerri

Recent research on sequence labelling has been exploring different strategies to mitigate the lack of manually annotated data for the large majority of the world languages. Among others, the most successful approaches have been based on (i) the cross-lingual transfer capabilities of multilingual pre-trained language models (model-transfer), (ii) data translation and label projection (data-transfer) and (iii), prompt-based learning by reusing the mask objective to exploit the few-shot capabilities of pre-trained language models (few-shot). Previous work seems to conclude that model-transfer outperforms data-transfer methods and that few-shot techniques based on prompting are superior to updating the model's weights via fine-tuning. In this paper, we empirically demonstrate that, for Argument Mining, a sequence labelling task which requires the detection of long and complex discourse structures, previous insights on cross-lingual transfer or few-shot learning do not apply. Contrary to previous work, we show that for Argument Mining data transfer obtains better results than model-transfer and that fine-tuning outperforms few-shot methods. Regarding the former, the domain of the dataset used for data-transfer seems to be a deciding factor, while, for few-shot, the type of task (length and complexity of the sequence spans) and sampling method prove to be crucial.

Joseba Fernandez de Landa, Rodrigo Agerri

The large majority of the research performed on stance detection has been focused on developing more or less sophisticated text classification systems, even when many benchmarks are based on social network data such as Twitter. This paper aims to take on the stance detection task by placing the emphasis not so much on the text itself but on the interaction data available on social networks. More specifically, we propose a new method to leverage social information such as friends and retweets by generating Relational Embeddings, namely, dense vector representations of interaction pairs. Our experiments on seven publicly available datasets and four different languages (Basque, Catalan, Italian, and Spanish) show that combining our relational embeddings with discriminative textual methods helps to substantially improve performance, obtaining state-of-the-art results for six out of seven evaluation settings, outperforming strong baselines based on Large Language Models, or other popular interaction-based approaches such as DeepWalk or node2vec.

Olia Toporkov, Alan Akbik, Rodrigo Agerri

Lemmatization is the task of transforming all words in a given text to their dictionary forms. While large language models (LLMs) have demonstrated their ability to achieve competitive results across a wide range of NLP tasks, there is no prior evidence of how effective they are in the contextual lemmatization task. In this paper, we empirically investigate the capacity of the latest generation of LLMs to perform in-context lemmatization, comparing it to the traditional fully supervised approach. In particular, we consider the setting in which supervised training data is not available for a target domain or language, comparing (i) encoder-only supervised approaches, fine-tuned out-of-domain, and (ii) cross-lingual methods, against direct in-context lemma generation with LLMs. Our experimental investigation across 12 languages of different morphological complexity finds that, while encoders remain competitive in out-of-domain settings when fine-tuned on gold data, current LLMs reach state-of-the-art results for most languages by directly generating lemmas in-context without prior fine-tuning, provided just with a few examples. Data and code available upon publication: https://github.com/oltoporkov/lemma-dilemma

María Grandury, Javier Aula-Blasco, Júlia Falcão et al.

Leaderboards showcase the current capabilities and limitations of Large Language Models (LLMs). To motivate the development of LLMs that represent the linguistic and cultural diversity of the Spanish-speaking community, we present La Leaderboard, the first open-source leaderboard to evaluate generative LLMs in languages and language varieties of Spain and Latin America. La Leaderboard is a community-driven project that aims to establish an evaluation standard for everyone interested in developing LLMs for the Spanish-speaking community. This initial version combines 66 datasets in Basque, Catalan, Galician, and different Spanish varieties, showcasing the evaluation results of 50 models. To encourage community-driven development of leaderboards in other languages, we explain our methodology, including guidance on selecting the most suitable evaluation setup for each downstream task. In particular, we provide a rationale for using fewer few-shot examples than typically found in the literature, aiming to reduce environmental impact and facilitate access to reproducible results for a broader research community.

Iñigo Alonso, Maite Oronoz, Rodrigo Agerri

Large Language Models (LLMs) have the potential of facilitating the development of Artificial Intelligence technology to assist medical experts for interactive decision support, which has been demonstrated by their competitive performances in Medical QA. However, while impressive, the required quality bar for medical applications remains far from being achieved. Currently, LLMs remain challenged by outdated knowledge and by their tendency to generate hallucinated content. Furthermore, most benchmarks to assess medical knowledge lack reference gold explanations which means that it is not possible to evaluate the reasoning of LLMs predictions. Finally, the situation is particularly grim if we consider benchmarking LLMs for languages other than English which remains, as far as we know, a totally neglected topic. In order to address these shortcomings, in this paper we present MedExpQA, the first multilingual benchmark based on medical exams to evaluate LLMs in Medical Question Answering. To the best of our knowledge, MedExpQA includes for the first time reference gold explanations written by medical doctors which can be leveraged to establish various gold-based upper-bounds for comparison with LLMs performance. Comprehensive multilingual experimentation using both the gold reference explanations and Retrieval Augmented Generation (RAG) approaches show that performance of LLMs still has large room for improvement, especially for languages other than English. Furthermore, and despite using state-of-the-art RAG methods, our results also demonstrate the difficulty of obtaining and integrating readily available medical knowledge that may positively impact results on downstream evaluations for Medical Question Answering. So far the benchmark is available in four languages, but we hope that this work may encourage further development to other languages.

Nedjma Ousidhoum, Junho Myung, Carla Perez-Almendros et al.

We present our shared task on evaluating the adaptability of LLMs and NLP systems across multiple languages and cultures. The task data consist of an extended version of our manually constructed BLEnD benchmark (Myung et al. 2024), covering more than 30 language-culture pairs, predominantly representing low-resource languages spoken across multiple continents. As the task is designed strictly for evaluation, participants were not permitted to use the data for training, fine-tuning, few-shot learning, or any other form of model modification. Our task includes two tracks: (a) Short-Answer Questions (SAQ) and (b) Multiple-Choice Questions (MCQ). Participants were required to predict labels and were allowed to submit any NLP system and adopt diverse modelling strategies, provided that the benchmark was used solely for evaluation. The task attracted more than 140 registered participants, and we received final submissions from 62 teams, along with 19 system description papers. We report the results and present an analysis of the best-performing systems and the most commonly adopted approaches. Furthermore, we discuss shared insights into open questions and challenges related to evaluation, misalignment, and methodological perspectives on model behaviour in low-resource languages and for under-represented cultures.

Blanca Calvo Figueras, Rodrigo Agerri

The development of Large Language Models (LLMs) has brought impressive performances on mitigation strategies against misinformation, such as counterargument generation. However, LLMs are still seriously hindered by outdated knowledge and by their tendency to generate hallucinated content. In order to circumvent these issues, we propose a new task, namely, Critical Questions Generation, consisting of processing an argumentative text to generate the critical questions (CQs) raised by it. In argumentation theory CQs are tools designed to lay bare the blind spots of an argument by pointing at the information it could be missing. Thus, instead of trying to deploy LLMs to produce knowledgeable and relevant counterarguments, we use them to question arguments, without requiring any external knowledge. Research on CQs Generation using LLMs requires a reference dataset for large scale experimentation. Thus, in this work we investigate two complementary methods to create such a resource: (i) instantiating CQs templates as defined by Walton's argumentation theory and (ii), using LLMs as CQs generators. By doing so, we contribute with a procedure to establish what is a valid CQ and conclude that, while LLMs are reasonable CQ generators, they still have a wide margin for improvement in this task.

Anar Yeginbergen, Maite Oronoz, Rodrigo Agerri

This paper investigates Multilingual Medical Question Answering across high-resource (English, Spanish, French, Italian) and low-resource (Basque, Kazakh) languages. We evaluate three types of external evidence sources across models of varying size: curated repositories of specialized medical knowledge, web-retrieved content, and explanations from LLM's parametric knowledge. Moreover, we conduct experiments with multilingual, monolingual and cross-lingual retrieval. Our results demonstrate that larger models consistently achieve superior performance in English across baseline evaluations. When incorporating external knowledge, web-retrieved data in English proves most beneficial for high-resource languages. Conversely, for low-resource languages, the most effective strategy combines retrieval in both English and the target language, achieving comparable accuracy to high-resource language results. These findings challenge the assumption that external knowledge systematically improves performance and reveal that effective strategies depend on both the source of language resources and on model scale. Furthermore, specialized medical knowledge sources such as PubMed are limited: while they provide authoritative expert knowledge, they lack adequate multilingual coverage

Blanca Calvo Figueras, Eneko Sagarzazu, Julen Etxaniz et al.

We introduce a professionally translated extension of the TruthfulQA benchmark designed to evaluate truthfulness in Basque, Catalan, Galician, and Spanish. Truthfulness evaluations of large language models (LLMs) have primarily been conducted in English. However, the ability of LLMs to maintain truthfulness across languages remains under-explored. Our study evaluates 12 state-of-the-art open LLMs, comparing base and instruction-tuned models using human evaluation, multiple-choice metrics, and LLM-as-a-Judge scoring. Our findings reveal that, while LLMs perform best in English and worst in Basque (the lowest-resourced language), overall truthfulness discrepancies across languages are smaller than anticipated. Furthermore, we show that LLM-as-a-Judge correlates more closely with human judgments than multiple-choice metrics, and that informativeness plays a critical role in truthfulness assessment. Our results also indicate that machine translation provides a viable approach for extending truthfulness benchmarks to additional languages, offering a scalable alternative to professional translation. Finally, we observe that universal knowledge questions are better handled across languages than context- and time-dependent ones, highlighting the need for truthfulness evaluations that account for cultural and temporal variability. Dataset and code are publicly available under open licenses.

Olia Toporkov, Rodrigo Agerri

Modern contextual lemmatizers often rely on automatically induced Shortest Edit Scripts (SES), namely, the number of edit operations to transform a word form into its lemma. In fact, different methods of computing SES have been proposed as an integral component in the architecture of several state-of-the-art contextual lemmatizers currently available. However, previous work has not investigated the direct impact of SES in the final lemmatization performance. In this paper we address this issue by focusing on lemmatization as a token classification task where the only input that the model receives is the word-label pairs in context, where the labels correspond to previously induced SES. Thus, by modifying in our lemmatization system only the SES labels that the model needs to learn, we may then objectively conclude which SES representation produces the best lemmatization results. We experiment with seven languages of different morphological complexity, namely, English, Spanish, Basque, Russian, Czech, Turkish and Polish, using multilingual and language-specific pre-trained masked language encoder-only models as a backbone to build our lemmatizers. Comprehensive experimental results, both in- and out-of-domain, indicate that computing the casing and edit operations separately is beneficial overall, but much more clearly for languages with high-inflected morphology. Notably, multilingual pre-trained language models consistently outperform their language-specific counterparts in every evaluation setting.

Elisa Sanchez-Bayona, Rodrigo Agerri

Metaphors are a ubiquitous but often overlooked part of everyday language. As a complex cognitive-linguistic phenomenon, they provide a valuable means to evaluate whether language models can capture deeper aspects of meaning, including semantic, pragmatic, and cultural context. In this work, we present Meta4XNLI, the first parallel dataset for Natural Language Inference (NLI) newly annotated for metaphor detection and interpretation in both English and Spanish. Meta4XNLI facilitates the comparison of encoder- and decoder-based models in detecting and understanding metaphorical language in multilingual and cross-lingual settings. Our results show that fine-tuned encoders outperform decoders-only LLMs in metaphor detection. Metaphor interpretation is evaluated via the NLI framework with comparable performance of masked and autoregressive models, which notably decreases when the inference is affected by metaphorical language. Our study also finds that translation plays an important role in the preservation or loss of metaphors across languages, introducing shifts that might impact metaphor occurrence and model performance. These findings underscore the importance of resources like Meta4XNLI for advancing the analysis of the capabilities of language models and improving our understanding of metaphor processing across languages. Furthermore, the dataset offers previously unavailable opportunities to investigate metaphor interpretation, cross-lingual metaphor transferability, and the impact of translation on the development of multilingual annotated resources.

Anar Yeginbergen, Maite Oronoz, Rodrigo Agerri

This paper investigates the role of dynamic external knowledge integration in improving counter-argument generation using Large Language Models (LLMs). While LLMs have shown promise in argumentative tasks, their tendency to generate lengthy, potentially unfactual responses highlights the need for more controlled and evidence-based approaches. We introduce a new manually curated dataset of argument and counter-argument pairs specifically designed to balance argumentative complexity with evaluative feasibility. We also propose a new LLM-as-a-Judge evaluation methodology that shows a stronger correlation with human judgments compared to traditional reference-based metrics. Our experimental results demonstrate that integrating dynamic external knowledge from the web significantly improves the quality of generated counter-arguments, particularly in terms of relatedness, persuasiveness, and factuality. The findings suggest that combining LLMs with real-time external knowledge retrieval offers a promising direction for developing more effective and reliable counter-argumentation systems.

Elisa Sanchez-Bayona, Rodrigo Agerri

This paper presents a comprehensive evaluation of the capabilities of Large Language Models (LLMs) in metaphor interpretation across multiple datasets, tasks, and prompt configurations. Although metaphor processing has gained significant attention in Natural Language Processing (NLP), previous research has been limited to single-dataset evaluations and specific task settings, often using artificially constructed data through lexical replacement. We address these limitations by conducting extensive experiments using diverse publicly available datasets with inference and metaphor annotations, focusing on Natural Language Inference (NLI) and Question Answering (QA) tasks. The results indicate that LLMs' performance is more influenced by features like lexical overlap and sentence length than by metaphorical content, demonstrating that any alleged emergent abilities of LLMs to understand metaphorical language are the result of a combination of surface-level features, in-context learning, and linguistic knowledge. This work provides critical insights into the current capabilities and limitations of LLMs in processing figurative language, highlighting the need for more realistic evaluation frameworks in metaphor interpretation tasks. Data and code are publicly available.

Pietro Ferrazzi, Aitor Soroa, Rodrigo Agerri

Large Language Models (LLMs) with reasoning capabilities have recently demonstrated strong potential in medical Question Answering (QA). Existing approaches are largely English-focused and primarily rely on distillation from general-purpose LLMs, raising concerns about the reliability of their medical knowledge. In this work, we present a method to generate multilingual reasoning traces grounded in factual medical knowledge. We produce 500k traces in English, Italian, and Spanish, using a retrievalaugmented generation approach over medical information from Wikipedia. The traces are generated to solve medical questions drawn from MedQA and MedMCQA, which we extend to Italian and Spanish. We test our pipeline in both in-domain and outof-domain settings across Medical QA benchmarks, and demonstrate that our reasoning traces improve performance both when utilized via in-context learning (few-shot) and supervised fine-tuning, yielding state-of-the-art results among 8B-parameter LLMs. We believe that these resources can support the development of safer, more transparent clinical decision-support tools in multilingual settings. We release the full suite of resources: reasoning traces, translated QA datasets, Medical-Wikipedia, and fine-tuned models.

Ravi Kiran Chikkala, Tatiana Anikina, Natalia Skachkova et al.

False information poses a significant global challenge, and manually verifying claims is a time-consuming and resource-intensive process. In this research paper, we experiment with different approaches to investigate the effectiveness of large language models (LLMs) in classifying factual claims by their veracity and generating justifications in English and Telugu. The key contributions of this work include the creation of a bilingual English-Telugu dataset and the benchmarking of different veracity classification approaches based on LLMs.

Jaione Bengoetxea, Itziar Gonzalez-Dios, Rodrigo Agerri

In this paper, we evaluate the capacity of current language technologies to understand Basque and Spanish language varieties. We use Natural Language Inference (NLI) as a pivot task and introduce a novel, manually-curated parallel dataset in Basque and Spanish, along with their respective variants. Our empirical analysis of crosslingual and in-context learning experiments using encoder-only and decoder-based Large Language Models (LLMs) shows a performance drop when handling linguistic variation, especially in Basque. Error analysis suggests that this decline is not due to lexical overlap, but rather to the linguistic variation itself. Further ablation experiments indicate that encoder-only models particularly struggle with Western Basque, which aligns with linguistic theory that identifies peripheral dialects (e.g., Western) as more distant from the standard. All data and code are publicly available.

Irune Zubiaga, Aitor Soroa, Rodrigo Agerri

This paper proposes a novel approach to evaluate Counter Narrative (CN) generation using a Large Language Model (LLM) as an evaluator. We show that traditional automatic metrics correlate poorly with human judgements and fail to capture the nuanced relationship between generated CNs and human perception. To alleviate this, we introduce a model ranking pipeline based on pairwise comparisons of generated CNs from different models, organized in a tournament-style format. The proposed evaluation method achieves a high correlation with human preference, with a $ρ$ score of 0.88. As an additional contribution, we leverage LLMs as zero-shot CN generators and provide a comparative analysis of chat, instruct, and base models, exploring their respective strengths and limitations. Through meticulous evaluation, including fine-tuning experiments, we elucidate the differences in performance and responsiveness to domain-specific data. We conclude that chat-aligned models in zero-shot are the best option for carrying out the task, provided they do not refuse to generate an answer due to security concerns.

Joseba Fernandez de Landa, Rodrigo Agerri

Social media users express their political preferences via interaction with other users, by spontaneous declarations or by participation in communities within the network. This makes a social network such as Twitter a valuable data source to study computational science approaches to political learning inference. In this work we focus on three diverse regions in Spain (Basque Country, Catalonia and Galicia) to explore various methods for multi-party categorization, required to analyze evolving and complex political landscapes, and compare it with binary left-right approaches. We use a two-step method involving unsupervised user representations obtained from the retweets and their subsequent use for political leaning detection. Comprehensive experimentation on a newly collected and curated dataset comprising labeled users and their interactions demonstrate the effectiveness of using Relational Embeddings as representation method for political ideology detection in both binary and multi-party frameworks, even with limited training data. Finally, data visualization illustrates the ability of the Relational Embeddings to capture intricate intra-group and inter-group political affinities.

Jaione Bengoetxea, Yi-Ling Chung, Marco Guerini et al.

Counter Narratives (CNs) are non-negative textual responses to Hate Speech (HS) aiming at defusing online hatred and mitigating its spreading across media. Despite the recent increase in HS content posted online, research on automatic CN generation has been relatively scarce and predominantly focused on English. In this paper, we present CONAN-EUS, a new Basque and Spanish dataset for CN generation developed by means of Machine Translation (MT) and professional post-edition. Being a parallel corpus, also with respect to the original English CONAN, it allows to perform novel research on multilingual and crosslingual automatic generation of CNs. Our experiments on CN generation with mT5, a multilingual encoder-decoder model, show that generation greatly benefits from training on post-edited data, as opposed to relying on silver MT data only. These results are confirmed by their correlation with a qualitative manual evaluation, demonstrating that manually revised training data remains crucial for the quality of the generated CNs. Furthermore, multilingual data augmentation improves results over monolingual settings for structurally similar languages such as English and Spanish, while being detrimental for Basque, a language isolate. Similar findings occur in zero-shot crosslingual evaluations, where model transfer (fine-tuning in English and generating in a different target language) outperforms fine-tuning mT5 on machine translated data for Spanish but not for Basque. This provides an interesting insight into the asymmetry in the multilinguality of generative models, a challenging topic which is still open to research.

Yi-Ling Chung, Marco Guerini, Rodrigo Agerri

The growing interest in employing counter narratives for hatred intervention brings with it a focus on dataset creation and automation strategies. In this scenario, learning to recognize counter narrative types from natural text is expected to be useful for applications such as hate speech countering, where operators from non-governmental organizations are supposed to answer to hate with several and diverse arguments that can be mined from online sources. This paper presents the first multilingual work on counter narrative type classification, evaluating SoTA pre-trained language models in monolingual, multilingual and cross-lingual settings. When considering a fine-grained annotation of counter narrative classes, we report strong baseline classification results for the majority of the counter narrative types, especially if we translate every language to English before cross-lingual prediction. This suggests that knowledge about counter narratives can be successfully transferred across languages.

Elena Zotova, Rodrigo Agerri, German Rigau

Popular social media networks provide the perfect environment to study the opinions and attitudes expressed by users. While interactions in social media such as Twitter occur in many natural languages, research on stance detection (the position or attitude expressed with respect to a specific topic) within the Natural Language Processing field has largely been done for English. Although some efforts have recently been made to develop annotated data in other languages, there is a telling lack of resources to facilitate multilingual and crosslingual research on stance detection. This is partially due to the fact that manually annotating a corpus of social media texts is a difficult, slow and costly process. Furthermore, as stance is a highly domain- and topic-specific phenomenon, the need for annotated data is specially demanding. As a result, most of the manually labeled resources are hindered by their relatively small size and skewed class distribution. This paper presents a method to obtain multilingual datasets for stance detection in Twitter. Instead of manually annotating on a per tweet basis, we leverage user-based information to semi-automatically label large amounts of tweets. Empirical monolingual and cross-lingual experimentation and qualitative analysis show that our method helps to overcome the aforementioned difficulties to build large, balanced and multilingual labeled corpora. We believe that our method can be easily adapted to easily generate labeled social media data for other Natural Language Processing tasks and domains.

Elena Zotova, Rodrigo Agerri, Manuel Nuñez et al.

Stance detection aims to determine the attitude of a given text with respect to a specific topic or claim. While stance detection has been fairly well researched in the last years, most the work has been focused on English. This is mainly due to the relative lack of annotated data in other languages. The TW-10 Referendum Dataset released at IberEval 2018 is a previous effort to provide multilingual stance-annotated data in Catalan and Spanish. Unfortunately, the TW-10 Catalan subset is extremely imbalanced. This paper addresses these issues by presenting a new multilingual dataset for stance detection in Twitter for the Catalan and Spanish languages, with the aim of facilitating research on stance detection in multilingual and cross-lingual settings. The dataset is annotated with stance towards one topic, namely, the independence of Catalonia. We also provide a semi-automatic method to annotate the dataset based on a categorization of Twitter users. We experiment on the new corpus with a number of supervised approaches, including linear classifiers and deep learning methods. Comparison of our new corpus with the with the TW-1O dataset shows both the benefits and potential of a well balanced corpus for multilingual and cross-lingual research on stance detection. Finally, we establish new state-of-the-art results on the TW-10 dataset, both for Catalan and Spanish.

Rodrigo Agerri, Iñaki San Vicente, Jon Ander Campos et al.

Word embeddings and pre-trained language models allow to build rich representations of text and have enabled improvements across most NLP tasks. Unfortunately they are very expensive to train, and many small companies and research groups tend to use models that have been pre-trained and made available by third parties, rather than building their own. This is suboptimal as, for many languages, the models have been trained on smaller (or lower quality) corpora. In addition, monolingual pre-trained models for non-English languages are not always available. At best, models for those languages are included in multilingual versions, where each language shares the quota of substrings and parameters with the rest of the languages. This is particularly true for smaller languages such as Basque. In this paper we show that a number of monolingual models (FastText word embeddings, FLAIR and BERT language models) trained with larger Basque corpora produce much better results than publicly available versions in downstream NLP tasks, including topic classification, sentiment classification, PoS tagging and NER. This work sets a new state-of-the-art in those tasks for Basque. All benchmarks and models used in this work are publicly available.

Iker García-Ferrero, Rodrigo Agerri, German Rigau

This paper presents a new technique for creating monolingual and cross-lingual meta-embeddings. Our method integrates multiple word embeddings created from complementary techniques, textual sources, knowledge bases and languages. Existing word vectors are projected to a common semantic space using linear transformations and averaging. With our method the resulting meta-embeddings maintain the dimensionality of the original embeddings without losing information while dealing with the out-of-vocabulary problem. An extensive empirical evaluation demonstrates the effectiveness of our technique with respect to previous work on various intrinsic and extrinsic multilingual evaluations, obtaining competitive results for Semantic Textual Similarity and state-of-the-art performance for word similarity and POS tagging (English and Spanish). The resulting cross-lingual meta-embeddings also exhibit excellent cross-lingual transfer learning capabilities. In other words, we can leverage pre-trained source embeddings from a resource-rich language in order to improve the word representations for under-resourced languages.

Rodrigo Agerri, German Rigau

In this research note we present a language independent system to model Opinion Target Extraction (OTE) as a sequence labelling task. The system consists of a combination of clustering features implemented on top of a simple set of shallow local features. Experiments on the well known Aspect Based Sentiment Analysis (ABSA) benchmarks show that our approach is very competitive across languages, obtaining best results for six languages in seven different datasets. Furthermore, the results provide further insights into the behaviour of clustering features for sequence labelling tasks. The system and models generated in this work are available for public use and to facilitate reproducibility of results.

Iñaki San Vicente, Xabier Saralegi, Rodrigo Agerri

Talaia is a platform for monitoring social media and digital press. A configurable crawler gathers content with respect to user defined domains or topics. Crawled data is processed by means of the EliXa Sentiment Analysis system. A Django powered interface provides data visualization for a user-based analysis of the data. This paper presents the architecture of the system and describes in detail its different components. To prove the validity of the approach, two real use cases are accounted for: one in the cultural domain and one in the political domain. Evaluation for the sentiment analysis task in both scenarios is also provided, showing the capacity for domain adaptation.

Iñaki San Vicente, Xabier Saralegi, Rodrigo Agerri

This paper presents a supervised Aspect Based Sentiment Analysis (ABSA) system. Our aim is to develop a modular platform which allows to easily conduct experiments by replacing the modules or adding new features. We obtain the best result in the Opinion Target Extraction (OTE) task (slot 2) using an off-the-shelf sequence labeler. The target polarity classification (slot 3) is addressed by means of a multiclass SVM algorithm which includes lexical based features such as the polarity values obtained from domain and open polarity lexicons. The system obtains accuracies of 0.70 and 0.73 for the restaurant and laptop domain respectively, and performs second best in the out-of-domain hotel, achieving an accuracy of 0.80.

Iñaki San Vicente, Rodrigo Agerri, German Rigau

This paper presents a simple, robust and (almost) unsupervised dictionary-based method, qwn-ppv (Q-WordNet as Personalized PageRanking Vector) to automatically generate polarity lexicons. We show that qwn-ppv outperforms other automatically generated lexicons for the four extrinsic evaluations presented here. It also shows very competitive and robust results with respect to manually annotated ones. Results suggest that no single lexicon is best for every task and dataset and that the intrinsic evaluation of polarity lexicons is not a good performance indicator on a Sentiment Analysis task. The qwn-ppv method allows to easily create quality polarity lexicons whenever no domain-based annotated corpora are available for a given language.

Egoitz Laparra, Rodrigo Agerri, Itziar Aldabe et al.

In this paper we present an approach to extract ordered timelines of events, their participants, locations and times from a set of multilingual and cross-lingual data sources. Based on the assumption that event-related information can be recovered from different documents written in different languages, we extend the Cross-document Event Ordering task presented at SemEval 2015 by specifying two new tasks for, respectively, Multilingual and Cross-lingual Timeline Extraction. We then develop three deterministic algorithms for timeline extraction based on two main ideas. First, we address implicit temporal relations at document level since explicit time-anchors are too scarce to build a wide coverage timeline extraction system. Second, we leverage several multilingual resources to obtain a single, inter-operable, semantic representation of events across documents and across languages. The result is a highly competitive system that strongly outperforms the current state-of-the-art. Nonetheless, further analysis of the results reveals that linking the event mentions with their target entities and time-anchors remains a difficult challenge. The systems, resources and scorers are freely available to facilitate its use and guarantee the reproducibility of results.

Rodrigo Agerri, German Rigau

We present a multilingual Named Entity Recognition approach based on a robust and general set of features across languages and datasets. Our system combines shallow local information with clustering semi-supervised features induced on large amounts of unlabeled text. Understanding via empirical experimentation how to effectively combine various types of clustering features allows us to seamlessly export our system to other datasets and languages. The result is a simple but highly competitive system which obtains state of the art results across five languages and twelve datasets. The results are reported on standard shared task evaluation data such as CoNLL for English, Spanish and Dutch. Furthermore, and despite the lack of linguistically motivated features, we also report best results for languages such as Basque and German. In addition, we demonstrate that our method also obtains very competitive results even when the amount of supervised data is cut by half, alleviating the dependency on manually annotated data. Finally, the results show that our emphasis on clustering features is crucial to develop robust out-of-domain models. The system and models are freely available to facilitate its use and guarantee the reproducibility of results.