Papers by Victor Fresno
Is anisotropy really the cause of BERT embeddings not being semantic? (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to train contextual language models for NLP use a lightweight approach called bi-encoder, which takes two sentences as input, but does not perform well with vanilla pre-trained Transformers. |
| Approach: | They conduct a set of experiments to improve our understanding of the lack of semantic isometry in contextualized word representations in BERT. |
| Outcome: | The proposed approach does not perform well with vanilla pre-trained Transformers. |
Bilingual Evaluation of Language Models on General Knowledge in University Entrance Exams with Minimal Contamination (2025.coling-main)
Copied to clipboard
Eva Sánchez Salido, Roser Morante, Julio Gonzalo, Guillermo Marco, Jorge Carrillo-de-Albornoz, Laura Plaza, Enrique Amigo, Andrés Fernandez García, Alejandro Benito-Santos, Adrián Ghajari Espinosa, Victor Fresno
| Challenge: | Existing benchmarks for Large Language Models have been proposed as single-task evaluations, but they are not fully comprehensive. |
| Approach: | They present a bilingual dataset that contains 1003 multiple-choice questions in Spanish and English. |
| Outcome: | The proposed model ranking is almost identical to the one obtained with MMLU . |