Inspecting the concept knowledge graph encoded by modern language models

Simple item page
dc.catalogadorgrr
dc.contributor.authorAspillaga, Carlos
dc.contributor.authorSoto, Alvaro
dc.contributor.authorMendoza Rocha, Marcelo Gabriel
dc.date.accessioned2024-05-28T20:30:32Z
dc.date.available2024-05-28T20:30:32Z
dc.date.issued2021
dc.description.abstractThe field of natural language understanding has experienced exponential progress in the last few years, with impressive results in several tasks. This success has motivated researchers to study the underlying knowledge encoded by these models. Despite this, attempts to understand their semantic capabilities have not been successful, often leading to non-conclusive, or contradictory conclusions among different works. Via a probing classifier, we extract the underlying knowledge graph of nine of the most influential language models of the last years, including word embeddings, text generators, and context encoders. This probe is based on concept relatedness, grounded on WordNet. Our results reveal that all the models encode this knowledge, but suffer from several inaccuracies. Furthermore, we show that the different architectures and training strategies lead to different model biases. We conduct a systematic evaluation to discover specific factors that explain why some concepts are challenging. We hope our insights will motivate the development of models that capture concepts more precisely.
dc.fuente.origenScopus
dc.identifier.scopusidSCOPUS_ID:85115697837
dc.identifier.urihttps://arxiv.org/abs/2105.13471
dc.identifier.urihttps://repositorio.uc.cl/handle/11534/85934
dc.information.autorucEscuela de Ingeniería; Mendoza Rocha, Marcelo Gabriel; 0000-0002-7969-6041; 1237020
dc.language.isoen
dc.pagina.final3000
dc.pagina.inicio2984
dc.publisherAssociation for Computational Linguistics (ACL)
dc.relation.ispartofFindings of the Association for Computational Linguistics: ACL-IJCNLP 2021
dc.revistaarXive
dc.titleInspecting the concept knowledge graph encoded by modern language models
dc.typecomunicación de congreso
sipa.codpersvinculados1237020
sipa.trazabilidadSCOPUS;21-03-2022
Files
Collections
Artículos de conferencia