Examples
Worked examples
- Is an instance
A coreference dataset published with a data statement specifying US Mainstream English (white, college-educated, ~30-50yo annotators).
- Is an instance
A speech-recognition corpus's data statement specifying broadcast news from 1995-2005.
Counter-examples
Looks similar, but isn't
- Not an instance
A token count and vocabulary size summary.
- Not an instance
A model card (describes the model, not the data).
Editorial commentary
A data statement is a standardised description of a natural-language-processing dataset, proposed by Bender and Friedman in “Data Statements for Natural Language Processing: Toward Mitigating System Bias and Enabling Better Science” (Transactions of the ACL, 2018). It covers curation rationale, language variety, speaker and annotator demographics, the speech situation the text was produced in, text characteristics, and recording quality — fields chosen specifically because their absence had repeatedly let biases (most commonly, datasets skewed toward a narrow demographic or a single English variety) go unnoticed until a model trained on them failed for other populations.
How this differs from a datasheet
A datasheet for datasets (Gebru et al., 2018/2021) is the general-purpose ML-dataset documentation template, covering motivation, composition, collection process, and uses across any data modality. A data statement is the NLP-specific counterpart, developed independently and slightly earlier, with fields tuned specifically to the sociolinguistic properties — speaker demographics, language variety, speech situation — that matter for text data but have no equivalent in, say, an image dataset. In practice a project working with text data may complete a data statement instead of, or alongside, a general datasheet.
Why it matters
Reviewers assessing whether an NLP model or dataset is fit for a particular population should be able to check the data statement’s language-variety and demographic fields directly, rather than inferring coverage from the model’s stated aggregate performance alone.
References
- Bender, E.M. & Friedman, B. (2018). “Data Statements for Natural Language Processing.” Transactions of the ACL 6, 587-604.
- See also: Datasheet for datasets, Training data composition, Model card.
Also known as
NLP data statement · Bender-Friedman data statement
Machine-readable encodings
Use in your systems
<role vocab="credit"
vocab-identifier="https://casrai.org/dictionary/"
vocab-term="Data statement (NLP)"
vocab-term-identifier="https://casrai.org/dictionary/term/data-statement-nlp" />{
"@context": "https://schema.org",
"@type": "DefinedTerm",
"@id": "https://casrai.org/dictionary/term/data-statement-nlp",
"name": "Data statement (NLP)",
"identifier": "https://casrai.org/dictionary/term/data-statement-nlp",
"description": "A standardised description of an NLP dataset covering curation rationale, language variety, speaker and annotator demographics, speech situation, text characteristics, and recording quality.",
"inDefinedTermSet": "https://casrai.org/dictionary/domain/ai-ml-research-outputs#set",
"url": "https://casrai.org/dictionary/term/data-statement-nlp",
"alternateName": [
"NLP data statement",
"Bender-Friedman data statement"
],
"license": "https://creativecommons.org/licenses/by/4.0/",
"publisher": {
"@id": "https://casrai.org/#organization"
},
"author": {
"@id": "https://casrai.org/#editorial-team"
},
"datePublished": "2026-05-21T02:22:50",
"dateModified": "2026-09-05T17:31:40",
"inLanguage": "en-GB",
"isAccessibleForFree": true
}
