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 focused on 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. Datasets, models and code are publicly available under open licenses.

The current best practice to measure the performance of base Large Language Models is to establish a multi-task benchmark that covers a range of capabilities of interest. Currently, however, such benchmarks are only available in a few high-resource languages. To address this situation, we present IberoBench, a multilingual, multi-task benchmark for Iberian languages (i.e., Basque, Catalan, Galician, European Spanish and European Portuguese) built on the LM Evaluation Harness framework. The benchmark consists of 62 tasks divided into 179 subtasks. We evaluate 33 existing LLMs on IberoBench on 0- and 5-shot settings. We also explore the issues we encounter when working with the Harness and our approach to solving them to ensure high-quality evaluation.

Using psycholinguistic and computational experiments we compare the ability of humans and several pre-trained masked language models to correctly identify control dependencies in Spanish sentences such as ‘José le prometió/ordenó a María ser ordenado/a’ (‘Joseph promised/ordered Mary to be tidy’). These structures underlie complex anaphoric and agreement relations at the interface of syntax and semantics, allowing us to study lexically-guided antecedent retrieval processes. Our results show that while humans correctly identify the (un)acceptability of the strings, language models often fail to identify the correct antecedent in non-adjacent dependencies, showing their reliance on linearity. Additional experiments on Galician reinforce these conclusions. Our findings are equally valuable for the evaluation of language models’ ability to capture linguistic generalizations, as well as for psycholinguistic theories of anaphor resolution.

The development of language technologies (LTs) such as machine translation, text analytics, and dialogue systems is essential in the current digital society, culture and economy. These LTs, widely supported in languages in high demand worldwide, such as English, are also necessary for smaller and less economically powerful languages, as they are a driving force in the democratization of the communities that use them due to their great social and cultural impact. As an example, dialogue systems allow us to communicate with machines in our own language; machine translation increases access to contents in different languages, thus facilitating intercultural relations; and text-to-speech and speech-to-text systems broaden different categories of users’ access to technology. In the case of Galician (co-official language, together with Spanish, in the autonomous region of Galicia, located in northwestern Spain), incorporating the language into state-of-the-art AI applications can not only significantly favor its prestige (a decisive factor in language normalization), but also guarantee citizens’ language rights, reduce social inequality, and narrow the digital divide. This is the main motivation behind the Nós Project (Proxecto Nós), which aims to have a significant contribution to the development of LTs in Galician (currently considered a low-resource language) by providing openly licensed resources, tools, and demonstrators in the area of intelligent technologies.