Here is a snapshot of how RoboJango conversates.  I am not using a standard DNN, rather something that I have been contemplating for a long, long time.  It will mimic a self organizing DNN, when all the assoications are in place among the objects, and their firing functiions are weighted properly. 

The tokenized responses are loaded.  These are tokenized prototype patterns of what may be spoken, and dynamic replacement amongst the Adjectives, Verbs, is performed to mix it up.  When I pose a Question or Command, or Statement, the tokens are matched to the Java Objects and the Collection of them comes into being, with associations linked between them. modeling a semantic net.  Whats neat is that OOP allows such beautiful modeling of the world, that concepts can be organized into heirarchies that model our perceptions and some links are static. like truths, and some are fluid, like emotions.  It appears to be working lol

After the response is generated dynamically, other responses may follow, as it has knowledge of the all the words that are in the sentence and can further elaborate on them, as it holds them in context.  Next, is to dynamically generate new Java objects on the fly and self program them with properties and associations, so his brain becomes self learning.  That will be fun!  Comments are welcome.