Project

Story Comprehension

Prashanth 

Groups

The ability to automatically understand and infer characters' goals and their emotional states is key towards better narrative comprehension. Reasoning about mental representations of various characters in a narrative has been referred to as Theory of Mind (ToM) reasoning. In this work, we propose an unsupervised neural network that exploits the personal stories on social media and incorporates commonsense knowledge about characters' motivations and reactions to generate interpretable trajectories of characters' mental states. We find that our model is capable of learning coherent mental representations from characters' actions and their affect states. We evaluate our model using a publicly available dataset for mental state tracking of characters in short commonsense stories. 

The ability to automatically understand and infer characters' goals and their emotional states is key towards better narrative comprehension. Reasoning about mental representations of various characters in a narrative has been referred to as Theory of Mind (ToM) reasoning. In this work, we propose an unsupervised neural network that exploits the personal stories on social media and incorporates commonsense knowledge about characters' motivations and reactions to generate interpretable trajectories of characters' mental states. We find that our model is capable of learning coherent mental representations from characters' actions and their affect states. We evaluate our model using a publicly available dataset for mental state tracking of characters in short commonsense stories.