“Gender Roles in the AI World” – Pascale Fung (Hong Kong University of Science and Technology)

Language representations from language models to word embeddings are known to reflect gender bias, where “doctor” is more closely associated with “men” and “nurse” with women. Intelligent system performances, ranging from that of voice recognition to recruitment tools, are known to favor male users. Virtual assistants such as Alexa, and humanoid robots such as Sophia, are often in female form whereas robots that perform physical tasks are often in male form. These gender roles of AI and robotics systems mirror the gender roles in the physical world, where deep seated bias, at times unconscious, has led to the acceptance of such roles. The cultural influence of science fiction and the gender roles of AI and robots in this genre are also non-negligible. Historical data also shows that the creators of both fictional and non-fictional AI and robot systems are predominantly men. Do NLP algorithms exacerbate gender bias or can they be used to mitigate such bias? 

In this talk, I will describe our work in detecting gender bias in both traditional media and social media. I will also describe my experience and insight with gender roles in AI in my research career, where I started  from working on the world’s first continuous speech recognizer, to creating virtual assistants and conversational systems. I will describe my experience as a researcher in countries with very different cultures.  I hope to share with the audience that, at the cusp of the 4th Industrial Revolution, the need for creating a more gender balanced world of AI has never been so urgent and our responsibilities as NLP researchers never so great. 


Pascale Fung is a Professor in the Department of Electronic & Computer Engineering and Department of Computer Science & Engineering at The Hong Kong University of Science & Technology (HKUST). She is an elected Fellow of the Institute of Electrical and Electronic Engineers (IEEE) and an elected Fellow of the International Speech Communication Association her contributions to “the interdisciplinary area of spoken language human-machine interactions”. She is a past president of SIGDAT, ACL and technical program chair of ACL, EMNLP and ICASSP. She is the Director of HKUST Center for AI Research (CAiRE), an interdisciplinary research center on top of all four schools at HKUST. She is the founding chair of the Women Faculty Association at HKUST. She is an expert on the Global Future Council, a think tank for the World Economic Forum. She represents HKUST on Partnership on AI to Benefit People and Society. Prof. Fung was born in Shanghai to professional artist parents but found her calling in artificial intelligence when she became interested in science fiction as a child. She has done research in the US, France, Japan and Hong Kong from the 1989 to present where she participated in the creation of the world’s first continuous speech recognition engine, the first English-Chinese statistical machine translation system, the first Chinese natural language search engine, and the first empathetic speaker. Today, her research interest lies in building intelligent systems that can understand and empathize with humans. She is also keenly interested in using AI  for public good and for the benefit of humanity. 

Website / @pascalefung

“Providing Gender-Specific Translations in Google Translate and beyond” – Melvin Johnson (Google AI)

While Google Translate has made significant improvements in translation quality over the past few years, we realized that translations from our models can reflect societal biases, such as gender bias.Recently, we announced that we’re taking the first step at reducing gender bias in our translations.  In this talk, I will highlight the research required to support gender-specific translations for sentences and the challenges we had to overcome to get to our initial launch. Finally, I will discuss our current work in two directions: scaling to more language-pairs quickly and handling gender beyond sentences (e.g. documents). Specifically, I will provide initial results from using a sentence-level gender rewriting system for handling gender in both sentence-level and document-level scenarios.


Melvin Johnson joined Google in 2015 where he works on Machine Translation and Natural Language Processing. Melvin’s research interests include providing gender-specific translations in Google Translate, Multilingual Neural Machine Translation [12], cross-lingual transfer learning, and end-to-end speech translation among others. Before Google, Melvin obtained a Masters degree in Computer Science from Stanford University where he worked with Prof. Chris Manning.