{"id":5222,"date":"2026-10-28T10:50:01","date_gmt":"2026-10-28T09:50:01","guid":{"rendered":"https:\/\/iriadacunha.com\/FundacionBBVA2015\/?page_id=5222"},"modified":"2026-10-28T10:50:01","modified_gmt":"2026-10-28T09:50:01","slug":"ministry-of-science-innovation-and-universities-2025-2028","status":"publish","type":"page","link":"https:\/\/iriadacunha.com\/FundacionBBVA2015\/en\/ministry-of-science-innovation-and-universities-2025-2028\/","title":{"rendered":"Ministry of Science, Innovation and Universities 2025-2028"},"content":{"rendered":"

Towards Robust and Linguistically Informed Evaluation of LLMs: from benchmarks to Real-world applications (TRAIL-LLMs) (AIA2025-163322-C65)<\/strong><\/p>\n

Research project developed at Universidad Nacional de Educaci\u00f3n a Distancia (UNED)<\/a> by the Departamento de Lenguajes y Sistemas Inform\u00e1ticos<\/a> of the Escuela T\u00e9cnica Superior de Ingenier\u00eda Inform\u00e1tica<\/a>, in the framework of the nlp.uned.es<\/a> group and the LENAR <\/a>research center, and by the Departamento de Filolog\u00edas Extranjeras y sus Ling\u00fc\u00edsticas<\/a> of the\u00a0 Facultad de Filolog\u00eda<\/a> in the framework of the ACTUALing <\/a>group.<\/p>\n

\"\"<\/a> \"\"<\/a> \"\"<\/a><\/p>\n

\u00a0<\/a>\"logo_UNED_Facultad_Filolog\u00eda\"<\/a>\"actualing\"<\/a><\/p>\n

Funding\u00a0<\/span><\/h6>\n

\"\"<\/p>\n

This project has been funded by the 2025 call \u00abProyectos de Investigaci\u00f3n en el \u00c1mbito de la Inteligencia Artificial\u00bb in the frame of the Programa Estatal para la Investigaci\u00f3n y el Desarrollo Experimental, del Plan Estatal de Investigaci\u00f3n Cient\u00edfica y T\u00e9cnica y de Innovaci\u00f3n 2024-2027.<\/p>\n

Budget: 297.500 euros.<\/p>\n

AIM AND METHODOLOGY\u00a0<\/span><\/h6>\n
\n

This project aims to evaluate large language models (LLMs)<\/strong> from two perspectives: assessing their core capabilities and evaluating their performance across three real-world application scenarios (use cases):<\/p>\n

USE CASE 1. Plain language in the Administration-legal domain.<\/p>\n

USE CASE 2. Data anonymisation for privacy protection.<\/p>\n

USE CASE 3. The biomedical domain.<\/p>\n

Specifically, the arText team is responsible for the USE CASE 1<\/strong>, for which the following methodology has been defined:<\/p>\n

 <\/p>\n<\/div>\n\n\n\n\n\n
METHODOLOGY<\/strong><\/td>\n<\/tr>\n
STAGE 1<\/td>\nSTAGE 2<\/td>\nSTAGE 3<\/td>\nSTAGE 4<\/td>\nSTAGE 5<\/td>\nSTAGE 6<\/td>\n<\/tr>\n
Compilation of a corpus of texts from the administrative domain.<\/td>\nAdapting the corpus to plain language using LLMs.<\/td>\nEvaluation of LLMs performance to clarify the corpus.<\/td>\nHuman post-editing of plain language adaptations.<\/td>\nDesign and implementation of a hybrid system that combines generative and symbolic AI to generate plain texts.<\/td>\nEvaluation of the hybrid system\u2019s performance.<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
TEAM<\/strong><\/span><\/h6>\n

From the nlp.uned.es group:<\/p>\n