Dr Luisa Bentivogli delivers lecture on gender bias and automatic translation
Title of the lecture: Gender bias and automatic translation
With language technologies entering widespread use and being deployed at a massive scale, their societal impact has raised concern both within and outside the scientific community. Indeed, while such technologies bring undeniable advantages in many contexts, it is also evident that they come with inherent risks, such as reproducing (or even amplifying) real-world asymmetries by codifying and entrenching various kinds of biases. One of them is gender bias, which affects automatic translation as well, in particular when systems are required to overtly express gender in the target languages while translating from languages that do not convey such information. In fact, previous studies showed that machine translation (MT) and speech translation (ST) models overproduce masculine references in their outputs (the so-called “masculine default”), and mirror traditional gender roles and stereotypes.
In this talk, I will present the research carried out so far in the Machine Translation group at FBK, aimed to understand, assess and mitigate gender bias in automatic translation. I will first introduce how the socio-cultural notions of gender interact with language(s) and translation, and frame which factors can contribute to the emergence of gender bias in automatic translation systems. Then, I will present the resources created to assess the 一 so far under-investigated 一 (biased) behaviour of ST systems, as well as the mitigation strategies developed to reduce feminine under-representation in their outputs. I will conclude by discussing potential research directions on the topic of gender bias in automatic translation.
Speaker's short bio: Dr Luisa Bentivogli is Senior Researcher in the Machine Translation Group at Fondazione Bruno Kessler (FBK). Her research interests include evaluation of human language technologies, translation technologies for translators, creation and annotation of multilingual corpora, crowdsourcing for natural language processing, computational lexicography in a multilingual environment, contrastive linguistics. Her latest research focuses on assessing and mitigating gender bias in MT and ST and on creating multilingual resources for ST. She contributed to several international and national projects, including MateCat, ModernMT and MateSub, which led to commercial products targeted to professional translators and translation companies. She is co-author of more than 70 scientific publications and she is regularly involved in the organization of international events related to machine translation, addressed to both the scientific community and the translators, such as the School of Advanced Technologies for Translators (SATT).
Event details: The event will take place online on Wednesday 13 October 2021, 15:30 BST. Registration is free and can be done on this Eventbrite webpage.
More information about the Convergence lecture series can be found in this news article.