Data that identifies a person, which is a wider category than most people assume. Not just names and addresses, but also combinations that become identifying together: a postcode, a date of birth and a job title will usually pin down one individual. Free text fields are where it hides.
Technical view
Direct identifiers name a person; quasi-identifiers identify in combination, which is why k-anonymity and related measures reason about combinations rather than fields. Detection in unstructured text relies on named entity recognition and pattern matching, both imperfect, so recall is never complete. Special categories under data protection law, including health, biometrics and beliefs, carry additional restrictions and can be inferred from apparently innocuous data.
Related
Part of AI Terminology Index, section: Data.