Figure 1:

We begin with the text from primary care physician consultation notes. In the first stage of the pipeline, we generate an internal tree representation for the passage in three main steps: (1.1) the text is split into sentences and pre-processed. (1.2) Then, a constituency parse tree is generated for every pre-processed sentence in the text. (1.3) Finally, node replacement and information-hiding techniques are applied to the constituency parse trees in order to complete the internal tree representation. In the second stage, a knowledge graph is extracted from the resulting tree representation (2).