Notícies Notícies

Torna a la pàgina principal
Vés enrere

18/07/2022 Seminari de COLT, a càrrec de Núria Bel (també online)

18/07/2022 Seminari de COLT, a càrrec de Núria Bel (també online)

Comparing RoBERTa's masked item predictions as evidence of language understanding, a càrrec de Núria Bel (UPF)




Dia: dilluns 18 de juliol del 2022

Hora: de 14.30 a 15.30

Lloc: sala 52.737 (amb la ponent) i també online a:

I will present the LUTEST, Language Understanding Test Sets, project, whose objective is the creation of annotated test sets and an associated metric to provide evidence about the linguistic generalization capabilities of the deep learning based methods that claim to be natural language understanding systems. LUTEST is seeking a method to assess the capabilities of current and future language models with respect to the traditionally linguistically motivated questions about the expected capabilities of a language processor that ‘understands a language’. In particular, I will present the research for a method to compare the representations of sentences with different surface but the same meaning, that are expected to be similar or equivalent.  LUTEST project PID2019-104512GB-I00 is funded by Ministerio de Ciencia e Innovación (Spain).

Professor of Language Technologies (Catedràtica Contractada) at the Departament de Traducció i Ciències del Llenguatge of the Universitat Pompeu Fabra (UPF) and senior researcher at the Institut de Lingüística Aplicada (IULA) of the UPF.
My area of research is Natural Language Processing (NLP). My current interests are mainly related to the automatic acquisition and exploitation of Language Resources (LR). Currently, I am leading a research group focused on projects related to Technologies of Language Resources





ODS - Objectius de desenvolupament sostenible:

ODS - Objetivos de desarrollo sostenible:

SDG - Sustainable Development Goals:

Els ODS a la UPF Los ODS en la UPF The ODS at UPF