Using “A Survey on Gender Bias in Natural Language Processing” by Karolina Stańczak and Isabelle Augenstein as reference, the paper presents a survey of 304 papers on gender bias in natural language processing (NLP). The authors highlight several key findings. Firstly, they note that research on gender bias in NLP often treats gender as a binary variable, overlooking its fluidity and continuity. Additionally, the research primarily focuses on monolingual English or other high-resource languages. The authors also observe that many newly developed algorithms do not test for bias and disregard ethical considerations. Moreover, the methodologies used in this research are limited in scope and lack evaluation baselines and pipelines. The authors argue that addressing these limitations is crucial for the advancement of future research on gender bias in NLP.

Regarding the definition of gender bias, the paper does not provide an explicit definition. However, it discusses how gender is defined in the social sciences and how this relates to formal definitions of gender bias in NLP research. Generally, gender is considered a social construct that encompasses the attitudes, behaviors, and expectations associated with being male or female in a particular society. Gender bias, in the context of NLP, refers to the ways in which language can perpetuate and reinforce harmful stereotypes and biases based on gender. It is important to note that most research on gender bias in NLP treats gender as a binary variable, disregarding its fluid nature.

While the specific statement you mentioned is not directly quoted in the passage, it accurately summarizes the information provided in the paper about the definitions of gender and gender bias. The authors discuss how gender is defined within the social sciences, emphasizing its connection to attitudes, behaviors, and societal expectations associated with being male or female. They also highlight how language can be used to perpetuate harmful stereotypes and biases based on gender.