Glossary
Attribution
Attribution identifies the source or actor responsible for a statement, dataset or derived work.
Provenance and citation
Attribution answers who made, published or contributed to a statement or information object. It may identify an organization, document author, dataset publisher or responsible process, depending on the claim. The named actor and its role must be distinguished: a publisher may reproduce another author's report, and a repository may host material it did not create. Attribution alone says nothing about correctness.
In practice, an evidence record separates immediate host, original creator and any intermediary when those are known. It also records publication date and version so the attributed statement can be found again. Where authorship is uncertain, that uncertainty should be expressed rather than assigning a convenient institution. Repeated claims can then be assessed for source dependence rather than counted as independent testimony.
For organizations defending values, attribution lets reviewers assess authority and responsibility in the relevant context. Provenance connects the attributed material to transformations and later uses; source reliability concerns observed accuracy for a defined task; a sentence-level citation connects a particular assertion to its supporting material. These are complementary questions, not synonyms.
Example
A Halden Therapeutics AG (fictional) article reposts a regulator notice; the article is the host, while the notice remains the attributed origin of the claim.
What attribution records
Attribution names the party responsible for a statement, a dataset or a derived work. In the PROV data model it is a relation between an entity, such as a document or a value, and an agent, such as a person, an organisation or a piece of software, and it is kept distinct from the activity that generated the entity and from the derivation of one entity from another. The distinction matters because the author of a document, the organisation that published it and the process that copied it into a database are three different parties, and each answers a different question. The Dublin Core terms give widely used names for these roles, among them creator, publisher, contributor and source.
For machine-generated text, attribution has a second meaning: whether a generated statement can be traced to an identified source that supports it. Rashkin and colleagues define attributable to identified sources as a property that human raters can judge, and propose a measure of it. Both senses share the requirement that the credit or responsibility attaches to a specific party and a specific statement, not to a general body of material.
Attribution in practice: whose claim is it
A statement in a document is often not the author's own. A report may quote a company, a regulator, a survey or an earlier report, and a reader who attributes the statement to the report's author may misjudge its status. Correct attribution distinguishes who said it, who reported it, and who is vouching for it now. A quoted forecast attributed to a company is a statement about what the company said, and it does not become a fact merely by appearing in a reputable publication. When such a statement is extracted into a record, the quoting layers should be preserved.
Attribution also supports fairness and legal compliance: licences and terms of use often require credit, and content-provenance initiatives such as the C2PA specification define signed manifests that record who created or edited a media asset and with what tools. For fact-checks, the schema.org ClaimReview vocabulary attributes a verdict to a named reviewer about a named claim. Each is a way of making responsibility explicit and machine-readable.
A worked example
A trade article reports that Halden Therapeutics AG (fictional) expects a regulatory decision in the autumn, citing an unnamed company spokesperson. An extractor stores the claim that a decision is expected in the autumn. Correct attribution records three parties: the spokesperson, as the original source of the expectation, unnamed; the company, on whose behalf the spokesperson spoke; and the trade publication, which reported it on a stated date. The weight a reviewer gives the claim depends on all three. When the company later publishes its own statement with a different timetable, the new statement is attributed to the company directly and is treated as superseding the earlier expectation, which remains on record with its attribution.
Checks a reviewer can run
For any stored statement, ask who said it, who reported it, and when. Ask whether the record distinguishes a quotation from the reporter's own assertion. Ask whether attribution is stored with the statement or only with the document, since a document can contain statements by several parties. Ask whether the attribution of a generated sentence points to a passage that actually supports it, and test a sample by reading the passages.
Related terms and boundaries
Attribution is one part of provenance, alongside derivation and activity. It affects source dependence, since knowing who originated a statement shows which repetitions are copies, and it bears on source reliability, which is judged for a named party. It is supported by sentence-level citations in generated text.
Common misreadings
- Confusing a host with the original author.
- Treating a byline as proof of accuracy.
- Omitting the role of an intermediary source.
Appears in
- Citing at the sentence
Bind each assertion to the exact source passage, its locator, reading date and retained copy.
- ORCID iD
A persistent identifier that distinguishes contributors to research.
References
- W3C (2013), PROV-O: The PROV Ontology.
- W3C (2013), PROV-DM: The PROV Data Model.
- Dublin Core Metadata Initiative, DCMI Metadata Terms.
- Rashkin et al. (2021), Measuring Attribution in Natural Language Generation Models, Computational Linguistics.
- Coalition for Content Provenance and Authenticity (C2PA), Content Credentials technical specification.
- Schema.org, ClaimReview.
Found an error on this page? See corrections for what counts as one, how to report it and the dated log.
Enquiry
Where does "Attribution" bear on a decision of yours?
This entry sits under provenance and citation. If a decision in your organization turns on it, send the entity concerned, the sources you fuse to describe it and who may read which of them. We reply with what a first record would cover.
Or write to [email protected] in your own words.