Progress report:

For the third week, I worked on having the speech agent provide a guess to a clue provided by the child. To allow more flexibility in the clue the child provides, I worked on being able to allow the agent to provide an accurate guess, even if the clue provided is not an exact match to one of the properties returned from ConceptNet. I also worked question-answer generation models and tried using a Simple Wikipedia article to see what kinds of question answer pairs it produces.

Some problems I encountered:

The main issues that I encountered this week had to do with long runtime due to filtering and trying to detect similarity between the clue and the words being returned by ConceptNet. I also ran into some issues with training the model because I don’t have a GPU and I kept getting errors.