The role of machine learning in personalised instructional sequencing for language learning

The origins of personalised instructional sequencing can be dated back to the times of the Ancient Greeks to the times of Alexander The Great's tutor, Aristotle. However, over the centuries the demand for education and growth of students has been disproportionately greater than the number of teachers in training. Therefore, there has been a longstanding interest in finding a way to scale education without negatively affecting learning outcomes. This interest was fuelled further with the advent of computers and artificial intelligence, where a plethora of systems and models were built to bring technology driven personalised instructional sequencing to the world. Unfortunately, results were far from groundbreaking and many challenges still remain. In my thesis, I investigate three aspects of personalised instructional sequencing: the personalised instructional sequencing mechanism, the student knowledge representation, and human forgetting. While I do not cover the entirety of personalised instructional sequencing, I cover what I consider the foundational components. I link psychological theory to model selection and design in each of my systems and present experiments to illustrate their impact. I show how reinforcement learning can be used for vocabulary learning. I also present a model that uses neural collaborative filtering to learn student knowledge representations. Lastly, I present a state-of-the-art model to predict the probability of vocabulary word recall for students learning English as a second language. The system's novelty lies in the use of word complexity to adapt the forgetting curve as well as its incorporation of psychological theory to select an appropriate model.