A semantic triple is the smallest unit of machine-processable knowledge: subject, relation and object, for example "Slovakia Programme – has allocation – amount". Both knowledge graphs and language models actually take away exactly such statements from text, not style. This gives a practical rule for writing: formulate key facts as short, complete sentences with a named subject, a clear relation and a specific value, ideally also with a time validity. Sentences like "there are more of them than last year" cannot be converted into a triple, because both the subject and the value are missing. Long compound sentences that mix three statements together are equally harmful. Where it makes sense, support the text with structured data and tables that record statements explicitly and don't leave them to be inferred from context.
See also: Entity Disambiguation, Content chunking, Author entity and sameAs.
