Gender bias in multilingual neural machine translation: The architecture matters

Multilingual Neural Machine Translation architectures mainly differ in the amount of sharing modules and parameters among languages. In this paper, and from an algorithmic perspective, we explore if the chosen architecture, when trained with the same data, influences the gender bias accuracy. Experi...

Descripción completa

Detalles Bibliográficos
Autores: Ruiz Costa-Jussà, Marta|||0000-0002-5703-520X, Escolano Peinado, Carlos|||0000-0001-6657-673X, Basta, Christine Raouf Saad, Ferrando Monsonís, Javier|||0000-0002-2637-0961, Batlle, Roser, Kharitonova, Ksenia
Tipo de recurso: informe técnico
Fecha de publicación:2020
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/348079
Acceso en línea:https://hdl.handle.net/2117/348079
Access Level:acceso abierto
Palabra clave:Sexism in language
Machine translating
Natural language processing (Computer science)
Sexisme en el llenguatge
Traducció automàtica
Tractament del llenguatge natural (Informàtica)
Àrees temàtiques de la UPC::Enginyeria de la telecomunicació::Processament del senyal::Processament de la parla i del senyal acústic
Descripción
Sumario:Multilingual Neural Machine Translation architectures mainly differ in the amount of sharing modules and parameters among languages. In this paper, and from an algorithmic perspective, we explore if the chosen architecture, when trained with the same data, influences the gender bias accuracy. Experiments in four language pairs show that Language-Specific encoders-decoders exhibit less bias than the Shared encoder-decoder architecture. Further interpretability analysis of source embeddings and the attention shows that, in the LanguageSpecific case, the embeddings encode more gender information, and its attention is more diverted. Both behaviors help in mitigating gender bias.