Narrative processing is an important skill to model bothfrom a cognitive science perspective and a computa-tional modeling perspective which applies to intelligentagents. Communication between humans often involvesstorytelling patterns that make the mundane exchange ofinformation more interesting and with proper emphasison important communicative goals. Current narrativegeneration models evaluate their generations basedon either a priori domain semantics (e.g. game statefor an in-game conversation with player agents) orgeneric text quality measures (e.g. coherence). However,in utilizing storytelling as a communicative tool forreal-world interactions, domain-specific approaches failto generalize and text quality measures fail to ensurethat the narrative is perceived as interesting. Hence, suchgeneration needs to consider the cognitive processesinvolved in the perception of narrative. Using theories ofcognitive interest, we present results of an investigationof whether word embeddings (e.g. GloVe (Pennington,Socher, & Manning, 2014)) could be used to model andestimate cognitive interestingness in stories.