Intelligent Assistive Evaluation System for English Comprehension Skills

Abstract

Student comprehension has many metrics as well as categories in which it is measured. Assessment of comprehension skills of the English language in young children is a human-resource-heavy task. It requires an ample amount of personnel to conduct, check, score, and mark. A huge number of interactive sessions are available for the improvement of comprehension among children in the basic categories of reading comprehension namely, literal, lexical, interpretive as well as evaluative. While all the categories are important at their own stances, the most important ones are literal and lexical considering they form the building blocks for language comprehension. Automated testing is being used to check and score language tests all over the world. Even so, there remains a digital interaction gap amongst the students and their evaluation steps that needs to be fulfilled in Pakistan. This raises a need for a system to effectively assist personnel in conducting and marking of such assessments. The essence of the proposed system is to evaluate comprehension skills via effective application of machine learning by compare machine-generated answers with student-given answers. This can digitize the process of testing comprehension in students using intelligent techniques and models. The proposed system aims to provide an evaluating system based on literal comprehension. The methodology adopted to tackle this is quantitative with a sample consisting of 87 students who answered the questions. In addition to this, 2 teachers who constructed the questions and then marked their answers as correct or incorrect. Both the teachers and the students were taken from a reputable school in Pakistan that deploys English as a second language. The proposed system takes the answers given by students and compares them to answers given by three intelligent network models, namely Bert, R Net and Bert-Pytorch. If a certain level of similarity is achieved then the answer is marked correct. The average of the scores obtained by marking comprehension questions through comparison with machine-given answers were found to be in line with the scores obtained through human marking. The system incorporates literal comprehension for 10 to 12-year-old children that fall in the 5th and 6th grades in Pakistani education systems. In future, this work can be extended to include inferential and interpretive forms of comprehension and may be extended to cater for higher age levels of the sample.