António Branco’s research while affiliated with University of Lisbon and other places

Instruction-guided image editing consists in taking an image and an instruction and deliverring that image altered according to that instruction. State-of-the-art approaches to this task suffer from the typical scaling up and domain adaptation hindrances related to supervision as they eventually resort to some kind of task-specific labelling, masking or training. We propose a novel approach that does without any such task-specific supervision and offers thus a better potential for improvement. Its assessment demonstrates that it is highly effective, achieving very competitive performance.

While language processing services are key assets for the science and technology of language, the possible ways under which they may be made available to the widest range of their end users are critical to support an Open Science policy for this scientific domain. Although providing such processing services under some web-based interface, at large, offers itself as an immediate and cogent response to that challenge, turning this view into an effective access to language processing services is an undertaking deserving a clear conceptual direction and a corresponding robust empirical validation. Based on an extensive overview of major undertakings towards making language processing tools available and on the design principles worked out and implemented in the PORTULAN CLARIN infrastructure, in this paper we advocate for a Research-Infrastructure-as-a-Service (RIaaS) model. This model unleashes accessibility to language processing services in as many web-based interface modalities as the current stage of technological development permits to support, in order to serve as many types of end users as possible, from IT developers to Digital Humanities researchers, and including citizen scientists, teachers, students and digital artists among many others.

Sentence encoder encode the semantics of their input, enabling key downstream applications such as classification, clustering, or retrieval. In this paper, we present Serafim PT*, a family of open-source sentence encoders for Portuguese with various sizes, suited to different hardware/compute budgets. Each model exhibits state-of-the-art performance and is made openly available under a permissive license, allowing its use for both commercial and research purposes. Besides the sentence encoders, this paper contributes a systematic study and lessons learned concerning the selection criteria of learning objectives and parameters that support top-performing encoders.

Retrieval-augmented generation resorts to content retrieved from external sources in order to leverage the performance of large language models in downstream tasks. The excessive volume of retrieved content, the possible dispersion of its parts, or their out of focus range may happen nevertheless to eventually have a detrimental rather than an incremental effect. To mitigate this issue and improve retrieval-augmented generation, we propose a method to refine the retrieved content before it is included in the prompt by resorting to meta-prompting optimization. Put to empirical test with the demanding multi-hop question answering task from the StrategyQA dataset, the evaluation results indicate that this method outperforms a similar retrieval-augmented system but without this method by over 30%.

To advance the neural encoding of Portuguese (PT), and a fortiori the technological preparation of this language for the digital age, we developed a Transformer-based foundation model that sets a new state of the art in this respect for two of its variants, namely European Portuguese from Portugal (PT-PT) and American Portuguese from Brazil (PT-BR). To develop this encoder, which we named Albertina PT-*, a strong model was used as a starting point, DeBERTa, and its pre-training was done over data sets of Portuguese, namely over a data set we gathered for PT-PT and over the brWaC corpus for PT-BR. The performance of Albertina and competing models was assessed by evaluating them on prominent downstream language processing tasks adapted for Portuguese. Both Albertina versions are distributed free of charge and under a most permissive license possible and can be run on consumer-grade hardware, thus seeking to contribute to the advancement of research and innovation in language technology for Portuguese.

CLARIN is a European Research Infrastructure Consortium developing and providing a federated and interoperable platform to support scientists in the field of the Social Sciences and Humanities in carrying-out language-related research. This contribution provides an overview of the entire infrastructure with a particular focus on tool interoperability, ease of access to research data, tools and services, the importance of sharing knowledge within and across (national) communities, and community building. By taking into account FAIR principles from the very beginning, CLARIN succeeded in becoming a successful example of a research infrastructure that is actively used by its members. The benefits CLARIN members reap from their infrastructure secure a future for their common good that is both sustainable and attractive to partners beyond the original target groups.

This chapter provides an analysis of the level of technological preparation of the Portuguese language for the digital age, as well as the actions necessary for the consolidation of Portuguese as a language of international communication with global projection.