Dialog flow as a context graph
We wanted to protoype a radical idea for a dialog system
What if it was a bucket of skills (or "contextual context handlers"), that would individually define when they should be activated, and bubble up if their constraints are satisfied.
.. Like a ballpit!
We modeled dialog flow as an interaction between context properties and context states: Properties are data, states are processes (or "pipes"). A certain state is activated through a boolean combination of events (or "signals"). Signals are generated when a property value changes, or when another state is activated.
... Like a spiking neural network!
We successfully implemented the core logic of state activations.
We successfully implemented an awesome UI with flask and d3.js to visualize the context.
We successfully connected console input/output, Speech input/output through Watson, Face recognition with ROS2.
Where to go next?
Pass the fucking Turing test already.