The research paper takes a stab at augmenting word embeddings with a knowledge graph for the task of language modeling. The main problem it tries to address is that word embeddings are bad at capturing information about specific entities such as people and places.