Figure 3:

A tree augmentation example of a constituency parse tree built for the sentence ”The patient had severe headache for three days”. (1) First, special entity nodes are generated for the patient and the symptom and replace their respective word token components. The original tokens and their pos tags are maintained in node attributes (2) Secondly, attachments that add additional information to the main sentence are detected and hidden in node attributes. Note that the POS tags and the full list of node attributes are omitted in the diagram for simplicity. The original parse tree structure is retained in node attributes.