Natural language processing covers everything from extracting a diagnosis from a discharge summary to generating a note. The clinical value is that most of the useful information in a medical record is prose, not structured fields.

Modern clinical NLP is largely language model based, replacing earlier rule and dictionary driven pipelines that were brittle in the face of real clinical writing.

Where This Gets Misread

Clinical text is uniquely hostile to language models: dense abbreviation, negation ("no evidence of"), hedging ("cannot exclude"), and temporal ambiguity about whether a condition is current or historical. Misreading a negation inverts the clinical meaning entirely, and it is a documented failure mode rather than a theoretical one.