The annotated terms are replaced through the anonymizer, which is built via core/src/lib/anonymizer/buildAnonymizer.ts.
A mapper is first generated: for every entityId in the annotations array, the specifier (%c or %d, which stands respectively for a character or a digit) will be replaced by the corresponding text, specified in the settings.json file.