This page is about my work as Research Assistant in Knowledge Media Institute (part of the Open University).
Following some of the topics first met in my Master’s Thesis, I keep working on the conversational domain.
The main goal is to enable a voice interface on the robots in order to be able to receive commands and perform some actions.
The main challenges are related to:
- find and use a relevant dataset to train the NLU module
- perform the grounding of the objects mentioned
- analyze the feasibility of the actions required with respect to the abilities of the considered robot
- start the actions or suggest a simpler one in case the robot cannot perform the desired one
The work involves dealing with different datasets (HuRIC, FrameNet) and analyzing how a Deep Learning model can be interpreted (in terms of looking where the attention of the neural network is captured).