Hora: 17:00
Ponente: Dra. Iria de Dios Flores, Proxecto Nós, CiTIUS-Centro Singular de Investigación
en Tecnoloxías Intelixentes, Univ. Santiago de Compostela.
Resumen: As large language models based on artificial neural networks have become increasingly effective, they have become more complex and difficult to analyze, since they encode words and sentences in numerical vectors that bear no transparent relationship to the symbolic representations typically enunciated in the field of linguistics. In recent years, a fruitful field of research has emerged inspired by the idea that the study of the linguistic abilities of language models can benefit extensively from the analytical foundations employed in the study of the linguistic abilities of the human mind, the "black box" par excellence as far as language is concerned.
Through this talk, I intend to illustrate the potential of using the analytical methods of language psychology to delve into the linguistic generalizations that emerge in language models, and thus contribute to a better understanding of these systems. To this end, I will present some examples and reflections, as well as the results of several experiments conducted in Spanish and Galician on the processing of different linguistic dependencies.