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

By Emily M. Bender, Batya Friedman. 2018

1,089 citations recorded by the source corpus.

In this paper, we propose data statements as a design solution and professional practice for natural language processing technologists, in both research and development. Through the adoption and widespread use of data statements, the field can begin to address critical scientific and ethical issues 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 results, protect companies from public embarrassment, and ultimately lead to language technology that meets its users in their own preferred linguistic style and furthermore does not misrepresent them to others.

Subfields: Artificial Intelligence, Safety Research, Sociology and Political Science, Natural Language Processing, Evaluation & Benchmarks

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Safety Research

In Sociology and Political Science

In Natural Language Processing

In Evaluation & Benchmarks

Citations in the corpus

GLaM: Efficient Scaling of Language Models with Mixture-of-Experts (2021, background): “…et al., 2021; Bommasani et al., 2021), including bias in representations (Blodgett et al., 2020), proper handling (Rogers, 2021) and documentation (Bender & Friedman, 2018) of training data, privacy (Abadi et al., 2016b; Carlini et al., 2021), and environmental concerns (Strubell et al., 2019;…”
RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models (2020, background): “Therefore, echoing calls for transparency in NLP research (Bender and Friedman, 2018; Mitchell et al., 2019; Dodge et al., 2019), we recommend researchers publicly release all relevant information during data collection (e.”
Repairing the Cracked Foundation: A Survey of Obstacles in Evaluation Practices for Generated Text (2022, background): “…the choices of dataset creators, it is imperative to report the curation decisions alongside limitations of datasets in structured format to allow for a better interpretation and contextualization of performance results (Bender and Friedman, 2018; Gebru et al., 2021; McMillan-Major et al., 2021).”
AI and the Everything in the Whole Wide World Benchmark (2021, background): “…challenge of addressing this lack of context is proper documentation for these datasets, which is often underdeveloped [Gebru et al., 2020, Bender and Friedman, 2018] and the devaluation of the data work [Paullada et al., 2020, Sambasivan et al., 2021, Hutchinson et al., 2020, Jo and Gebru, 2020].”
Deduplicating Training Data Makes Language Models Better (2021, background): “While recent work focused on documenting the potential harms that could arise from problematic datasets (Bender and Friedman, 2018; Gebru et al., 2020), less work has been done”
Documenting Large Webtext Corpora: A Case Study on the Colossal Clean Crawled Corpus (2021, background): “EN dataset. structured, task-speciﬁc NLP datasets, best practices have emerged around documenting the collection process, composition, intended uses, and other characteristics (Bender and Friedman, 2018; Gebru et al., 2018; Hutchinson et al., 2021).”
The GEM Benchmark: Natural Language Generation, its Evaluation and Metrics (2021, methodology): “cards (Bender and Friedman, 2018; Gebru et al., 2018) we developed and for which we release a template and guide.”
Re-imagining Algorithmic Fairness in India and Beyond (2021, background): “, [34, 77, 136]), openly discussing limitations, failure cases, and lessons learnt can help move from the ‘magic pill’ role of fairness as a checklist for ethical issues in India—to a more pragmatic, flawed, and evolving scientific function.”
HateCheck: Functional Tests for Hate Speech Detection Models (2020, methodology): “Following Bender and Friedman (2018), we provide a data statement, which documents the generation and provenance of test cases in H ATE C HECK .”
WikiLingua: A New Benchmark Dataset for Multilingual Abstractive Summarization (2020, neutral): “The Data Statement (Bender and Friedman, 2018) for”
Datasheets for datasets (2018, neutral): “As another example, Bender and Friedman [2] outline a proposal similar to datasheets for datasets specifically intended for language-based datasets.”
CrowS-Pairs: A Challenge Dataset for Measuring Social Biases in Masked Language Models (2020, background): “2 The full data statement is in Appendix A (Bender and Friedman, 2018).”
Data Portraits: Recording Foundation Model Training Data (2023, methodology): “Figure 1: Output from ChatGPT [29] when asked to explain Bloom filters using text from Wikipedia.”
The ROOTS Search Tool: Data Transparency for LLMs (2023, background): “This highlights the general problem with the lack of documentation in NLP datasets of all sizes (Bender and Friedman, 2018; Gebru et al., 2020), and the fact that data work has generally not been a priority in NLP recently (Sambasivan et al., 2021).”