The Dangers of trusting Stochastic Parrots: Faithfulness and Trust in Open-domain Conversational Question Answering

Sabrina Chiesurin, Dimitris Dimakopoulos, Marco Antonio Sobrevilla Cabezudo, Arash Eshghi, Ioannis Papaioannou, Verena Rieser, Ioannis Konstas

Producción científica: Capítulo del libro/informe/acta de congresoContribución a la conferenciarevisión exhaustiva

7 Citas (Scopus)

Resumen

Large language models are known to produce output which sounds fluent and convincing, but is also often wrong, e.g. “unfaithful" with respect to a rationale as retrieved from a knowledge base. In this paper, we show that task-based systems which exhibit certain advanced linguistic dialog behaviors, such as lexical alignment (repeating what the user said), are in fact preferred and trusted more, whereas other phenomena, such as pronouns and ellipsis are dis-preferred. We use open-domain question answering systems as our test-bed for task based dialog generation and compare several open- and closed-book models. Our results highlight the danger of systems that appear to be trustworthy by parroting user input while providing an unfaithful response.

Idioma originalInglés
Título de la publicación alojadaFindings of the Association for Computational Linguistics, ACL 2023
EditorialAssociation for Computational Linguistics (ACL)
Páginas947-959
Número de páginas13
ISBN (versión digital)9781959429623
EstadoPublicada - 2023
Publicado de forma externa
Evento61st Annual Meeting of the Association for Computational Linguistics, ACL 2023 - Toronto, Canadá
Duración: 9 jul. 202314 jul. 2023

Serie de la publicación

NombreProceedings of the Annual Meeting of the Association for Computational Linguistics
ISSN (versión impresa)0736-587X

Conferencia

Conferencia61st Annual Meeting of the Association for Computational Linguistics, ACL 2023
País/TerritorioCanadá
CiudadToronto
Período9/07/2314/07/23

Huella

Profundice en los temas de investigación de 'The Dangers of trusting Stochastic Parrots: Faithfulness and Trust in Open-domain Conversational Question Answering'. En conjunto forman una huella única.

Citar esto