In "cat" tasks participants are given the entity boundaries as input, the prediction output is normalization (categories) only. In "cat+ner" tasks, participants are given the text only, the prediction output is entity boundaries and their normalization (categories).