Explainable Deep Neural Network Models

English speaking ability assessment has generally been dependent on the knowledge and experience of individual judges, leading to generalization issues. Therefore, in this research and development, the framework for English speaking proficiency determination can be prebuilt into the structure of the neural network to explain the decision-making process. The proposed framework will allow the AI to grow through collaboration with people by allowing experienced human judges to make final decisions on cases. Priority is made according to decisions regarding the AI’s judgments as less reliable.

Dialog-Based Speaking Assessment System

Ultimately, it is envisioned that the system being developed will allow English learners to carry out natural conversations with human performance judges, assuming that they will conduct speaking proficiency assessments while speaking with AI agents via their laptops, tablets, or virtual reality/augmented reality (VR/AR). The Institute of Perceptual Information Systems, Waseda University, is building modules for each of their specialist areas: language understanding, generation, and interactive control.