Question answering with Nengo

Eric Hunsberger


Given a list of facts, we wish to design a system that will answer questions about these facts. For this specific model, I will focus on the following:

1. Build a cognitive model of how people might perform this task, using working memory. This means that information should be stored in the activity of neurons only; no synaptic weights will be changed during the task. 2. Design the model so that information representation is flexible. This means the system can represent and deal with facts like "The spoon is in John" (assuming he ate it) as easily as it can with facts like "John is in the kitchen". 3. Ignore sentence parsing: I will assume that the facts and questions arrive to the system already parsed. 4. Ignore logical inference: There are many inferences that can be made about facts. For example, if John is in the kitchen and John has the spoon, then the spoon is also in the kitchen. I assume that this inference will be handled by a separate system.





Question answering result

Appendix A: Code

The Nengo code for the model can be found in the attached Python file.