Natural Language Analysis, Machine Learning and more (Patreon)
Content
Note: You can safely ignore this post if you don't have background knowledge in programming and/or are not interested in back end development.
When we go silent for an abnormally long while, it usually means that we are trying out something new and unsteady. This time we attempted to integrate a self-learning chatbot in game which simulate real human interaction - you can vocally communicate with Erika, and receive proper feedback. Instead of fiddling around awkwardly with your device, this might be a better solution for VR environment control.
By deploying the solution provided by Dialogflow we successfully created an experimental Slack chatbot with limited functionality. Then we parsed JSON in Unreal to hook the api and direct the information flow to google cloud so the vocal input and output can be processed. However the efficiency of running JSON based application on C++ environment still has much to be desired - API.AI was devised for a limited range of approved online services for now, running the system on a 3rd party application has a negative impact on performance - both the training and responding take much longer than deemed acceptable. And it's also impossible to make the system work out of the box without connection to internet: the massive requirement of manual training will take a more than considerably amount of time and disk space.
tl;dr: We wasted a week on a feature that was still far from practical by now. However the setback would not impede the regular update schedule we have set - the preparation work for story mode has already started, and it will roll out with version 1.20.
As always, I will be around to answer for your messages and comments tomorrow.