The choice between large-scale, multilingual, foundation models and specialized monolingual models for languages like Brazilian Portuguese (PT-BR) presents a complex trade-off between generalization and specialization. This paper investigates this trade-off through an empirical study across a diverse suite of tasks. We evaluate multiple families of language models under both linear probing and fine-tuning regimes. We find that monolingual encoders exhibit greater "adaptation plasticity" during fine-tuning, improving on both classification and semantic similarity, where global (multilingual) models degrade. However, this plasticity comes at a cost: our tokenization analysis suggests that monolingual models struggle with foreign terms, whereas modern multilingual tokenizers show surprising morphological competence, challenging a long-standing assumption in the field. We conclude that the optimal model choice is a task-dependent trade-off between vocabulary coverage and adaptation flexibility.

Quantization is key for efficient LLM inference, but its language-specific effects are understudied. We compare INT8 and FP8 (E4M3) quantization for Meta-Llama-3-8B on English and Brazilian Portuguese (PT-BR). INT8 with outlier handling preserves perplexity in both languages, while naive FP8 casting degrades English far more than PT-BR (+18% vs. +3.9%). Activation analysis shows rarer, larger English spikes (>35) that are more prone to saturation under unscaled E4M3, whereas PT-BR activations are more concentrated. Our FP8 results reflect a naive casting stress test (no calibration/scaling), not an optimized FP8 recipe.

Hate speech detection is often treated as a binary task, ignoring the hierarchical nature of toxicity, such as severity levels and specific target groups. This work presents a Multitask Learning (MTL) approach for the HateBR dataset, utilizing a shared BERTimbau encoder to simultaneously predict binary offensiveness, ordinal severity, and hate speech targets. Our experiments demonstrate that the MTL architecture outperforms Single-Task baselines on the primary offensive detection task, increasing the Matthews Correlation Coefficient from 0.80 to 0.82. Beyond predictive performance, we show that joint training implicitly enforces hierarchical sanity: the unified model yields a 0% target-inconsistency rate (i.e., no cases where a comment is predicted Non-offensive while still assigned a hate target). However, we observe negative transfer in the fine-grained multilabel target task (Micro-F1 drops from 0.59 to 0.42), highlighting a trade-off between logical consistency and target attribution under extreme imbalance.

Language resources are valuable assets, both for institutions and researchers. To safeguard these resources requirements for repository systems and data management have been specified by various branch organizations, e.g., CLARIN and the Data Seal of Approval. This paper describes these and some additional ones posed by the authors’ home institutions. And it shows how they are met by FLAT, to provide a new home for language resources. The basis of FLAT is formed by the Fedora Commons repository system. This repository system can meet many of the requirements out-of-the box, but still additional configuration and some development work is needed to meet the remaining ones, e.g., to add support for Handles and Component Metadata. This paper describes design decisions taken in the construction of FLAT’s system architecture via a mix-and-match strategy, with a preference for the reuse of existing solutions. FLAT is developed and used by the Meertens Institute and The Language Archive, but is also freely available for anyone in need of a CLARIN-compliant repository for their language resources.