Data Statements for Natural Language Processing: Toward Mitigating System Bias and Enabling Better Science

Data Statements for Natural Language Processing: Toward Mitigating System Bias and Enabling Better Science Emily M. Bender Department of Linguistics University of Washington ebender@uw.edu Batya Friedman The Information School University of Washington batya@uw.edu Abstract In this paper, we propose data statements as a design solution and professional practice for natural language processing technolo- gists, in both research and development. Through the adoption and widespread use of data statements, the field can begin to address critical scientific and ethical is- sues that result from the use of data from certain populations in the development of technology for other populations. We present a form that data statements can take and explore the implications of adopting them as part of regular practice. We argue that data statements will help alleviate issues related to exclusion and bias in language technology, lead to better precision in claims about how natural language processing research can generalize and thus better engineering re- sults, protect companies from public embar- rassment, and ultimately lead to language technology that meets its users in their own preferred linguistic style and furthermore does not misrepresent them to others. 1 Introduction As technology enters widespread societal use it is important that we, as technologists, think critically about how the design decisions we make and sys- tems we build impact people—including not only users of the systems but also other people who will be affected by the systems without directly in- teracting with them. For this paper, we focus on natural language processing (NLP) technology. Potential adverse impacts include NLP systems that fail to work for specific subpopulations (e.g., children or speakers of language varieties that are not supported by training or test data) or systems that reify and reinforce biases present in train- ing data (e.g., a resume-review system that ranks female candidates as less qualified for computer programming jobs because of biases present in training text). There are both scientific and ethical reasons to be concerned. Scientifically, there is the issue of generalizability of results; ethically, the potential for significant real-world harms. Although there is increasing interest in ethics in NLP,1 there remains the open and urgent question of how we integrate ethical considerations into the everyday practice of our field. This question has no simple answer, but rather will require a constellation of multi-faceted solutions. Toward that end, and drawing on value sen- sitive design (Friedman et al., 2006), this paper contributes one new professional practice—called data statements—which we argue will bring about improvements in engineering and scientific outcomes while also enabling more ethically re- sponsive NLP technology. A data statement is a characterization of a dataset that provides con- text to allow developers and users to better under- stand how experimental results might generalize, how software might be appropriately deployed, and what biases might be reflected in systems built on the software. In developing this practice, we draw on analogous practices from the fields of psychology and medicine that require some stan- dardized information about the populations stud- ied (e.g., …

Content truncated to 500 words