A computer identifies a sample document and annotates the sample document using a plurality of annotators to produce annotations associated with the sample document. The computer determines a plurality of patterns in the sample document based on the annotations. The computer populates a template using the patterns, and varies parts of the patterns in the template to generate test data. The computer identifies a knowledge domain of the sample document and identifies at least one of the annotators based on the knowledge domain of the sample document. The computer determines the observed occurrence count of at least one of the patterns in the sample document, and the likelihood of populating the template using the at least one of the patterns can be proportional to the observed occurrence count.