After discovering that Q & A pairs work the best when creating a chatbot with Dialogflow, I did some research on how this could be achieved automatically.
A lot of people have been working on this problem, including Google who has patented the idea of generating question and answer pairs from conversational texts.
From what I’ve read, it’s more or less achievable when you train a neural network and use a certain dataset, one that consists a number of questions to which the answer is a segment of the corresponding text. Priya Dwivedi and her team were able to do this using the Stanford Question Answering Dataset, for example.
I’ve given it some thought, and although it would be really awesome to automatically generate question and answer pairs about a historic person, setting up and testing a neural network is a bit outside of the scope for the Historic Voicebot project.
If I have time left after the rest of the project is done, I would love to dive deeper into this problem. For now however, I’ve manually created 20 question & answer pairs and added these to Dialogflow. As you can see, Ada Lovelace’s answers are currently a lot better than before!