This is actually something I have been researching a bit on my own.
Most movie scripts can be structurally analysed by using writing theory such as Dramatica. Dramatica is based upon a hierarchy of concepts, which can be topic modeled. The hierarchy of topic models would seem to work very well with the capsule neural networks.
I have been working with computational creativity problems in a narrative generation. The state of the art methods use Partial Order Causal Link Planners, but they depend on propositional logic. Alonzo Church presented the Superman dilemma (Louis Lane does not know that Clark Kent is Superman, but Superman knows, that he is Clark Kent) and invented Intensional Logic as a solution; the basic idea is, that if we do not know the context of the narrative, the meaning is always in superposition and can only be understood through entangled meanings from the background story. So, in a sense, propositional logic is limited by classic information theory constraints, while Church's logic can take a quantum information-theoretic approach. I do not believe that classic information theory can resolve narrative analysis problems. So, basically, the meaning of a narrative collapses (the superposition gets resolved) by using the hierarchical narrative structure and what we know beforehand.
So my intuition would be the following:
We can use Dramatica and potentially other narrative theories (hierarchical metamemetics, reverse SCARF, etc.) to create a hierarchical network like ImageNet, but for narratives.
We can build conceptual topic models. Dramatica has a hierarchy of 4-16-64-64 concepts and annotated data exists already.
When using hundreds of topic models, there will be a lot of false positives. However, the superposition of the topic models can be collapsed by using the hierarchical levels and some other dramatic analytics.
By using the capsule neural networks, we might be able to build a system, which could determine a narrative interpretation of the full story, which would make the most sense by using the concept hierarchy.
I tried to prove my intuition, but, unfortunately, Dramatica only has 300 movies analysed, and I was able to find scripts of only 10 of them; not enough data.
However, there are other hierarchical ontologies out there and other narrative structures; could the same intuition be used for political news for example?