CARes Lab - Northwestern University Department of Computer Science
They Computational Model of Need works to guide a robot to provide social assistance to a user completing a specific task without impeding on the autonomy of the user.
When I joined the project, the Computational Model of Need already had models that predicted need based off of eye gaze and task progress. Rather than building off existing models, I created a new model that uses vocal cues to determine how much help is needed.
I specifically focused on concepts from Speech & Language Processing by Daniel Jurafsky & James H. Martin. I identified rules from request and complaint speech acts from the user that indicated the need for assistance. Similarly, I identified speech patterns that indicated that the user was making progress and did not require assistance.
Once identifying informal patterns in the user's language, I used deductive reasoning to develop a formal language model. I then implemented this model with Prolog. The model predicts the amount of help needed based on components of the user's logical statements.
It relies on the idea of a "need spectrum," where some components indicate more need than others. Once the user's need amount reaches a certain threshold, the robot provides assistance.