We can also run our trained chatbot in interactive mode to have a real-time conversation with it:

chatbot.interactive_mode()

We can interact with the chatbot by typing text whenever prompted by the ==> symbol:

==> good morning
hello what can i help you with today
==> i'd like to book a table for eight
i'm on it
==> 
any preference on a type of cuisine
==> with italian cuisine
where should it be
==> in bombay
which price range are looking for
==> i am looking for a moderate restaurant
ok let me look into some options for you
==> 
api_call italian bombay eight moderate
==> instead could it be in a cheap price range
sure is there anything else to update
==> actually i would prefer in london
sure is there anything else to update
==> no
ok let me look into some options for you
==> 
api_call italian london eight cheap
==> 
.
.
.

As we try various types of interactions, we can see that our simple memory network has learned to produce an appropriate response to most of our questions, but may stumble when dealing with proper nouns or named entities, such as restaurants and the associated facts stored in the KB. Dealing with such entities can be considered a line of research in itself, and many subfields of NLP research need to come together to build chatbots for real-world deployment.