Iria de-Dios-Flores

Iria de-Dios-Flores, Marcos Garcia

This paper explores the ability of Transformer models to capture subject-verb and noun-adjective agreement dependencies in Galician. We conduct a series of word prediction experiments in which we manipulate dependency length together with the presence of an attractor noun that acts as a lure. First, we evaluate the overall performance of the existing monolingual and multilingual models for Galician. Secondly, to observe the effects of the training process, we compare the different degrees of achievement of two monolingual BERT models at different training points. We also release their checkpoints and propose an alternative evaluation metric. Our results confirm previous findings by similar works that use the agreement prediction task and provide interesting insights into the number of training steps required by a Transformer model to solve long-distance dependencies.

Pablo Gamallo, Pablo Rodríguez, Iria de-Dios-Flores et al.

Large language models (LLMs) have transformed natural language processing. Yet, their predominantly English-centric training has led to biases and performance disparities across languages. This imbalance marginalizes minoritized languages, making equitable access to NLP technologies more difficult for languages with lower resources, such as Galician. We present the first two generative LLMs focused on Galician to bridge this gap. These models, freely available as open-source resources, were trained using a GPT architecture with 1.3B parameters on a corpus of 2.1B words. Leveraging continual pretraining, we adapt to Galician two existing LLMs trained on larger corpora, thus mitigating the data constraints that would arise if the training were performed from scratch. The models were evaluated using human judgments and task-based datasets from standardized benchmarks. These evaluations reveal a promising performance, underscoring the importance of linguistic diversity in generative models.

Nora Graichen, Iria de-Dios-Flores, Gemma Boleda

We present a systematic review of 337 articles evaluating the syntactic abilities of Transformer-based language models, reporting on 1,015 model results from a range of syntactic phenomena and interpretability methods. Our analysis shows that the state of the art presents a healthy variety of methods and data, but an over-focus on a single language (English), a single model (BERT), and phenomena that are easy to get at (like part of speech and agreement). Results also suggest that TLMs capture these form-oriented phenomena well, but show more variable and weaker performance on phenomena at the syntax-semantics interface, like binding or filler-gap dependencies. We provide recommendations for future work, in particular reporting complete data, better aligning theoretical constructs and methods across studies, increasing the use of mechanistic methods, and broadening the empirical scope regarding languages and linguistic phenomena.

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.