Feasibility Analysis of Machine Learning-Based Autoscoring Feature for Korean Language Spoken by Foreigners
Aria Bisma Wahyutama, Mintae Hwang · 2023
This paper introduces a research concept and feasibility analysis for a Machine Learning (ML)-based autoscoring feature in a mobile application to evaluate foreigners' Korean language proficiency. The proposed concept involves a mobile application that records users' voices, transcribes them using a Speech-To-Text (STT) engine, and generates score predictions using an ML model. In the feasibility analysis, a dataset that consists of a sentence and score pair is collected using the STT engine and Sentence Transformer similarity check. An ML model with Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) is developed to train the gathered dataset to generate score prediction that resulted in 0.08054, 0.28380, and 0.20939 for MSE, RMSE, and MAE scores respectively, as performance evaluation. This paper also highlights various challenges and issues that may arise during the implementation of the proposed system.