Flynn, Robert (2021) Augmenting the CoAST system with automated text simplification. Masters by Research thesis (MSc), Manchester Metropolitan University.
|
Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (1MB) | Preview |
Abstract
Proper comprehension of academic texts is important for students in higher education. The CoAST platform is a virtual learning environment that endeavours to improve reading comprehension by augmenting theoretically, and lexically, complex texts with helpful annotations provided by a teacher. This thesis extends the CoAST system, and introduces machine learning models that assist the teacher with identifying complex terminology, and writing annotations, by providing relevant definitions for a given word or phrase. A deep learning model is implemented to retrieve definitions for words, or phrases of a arbitrary length. This model surpasses previous work on the task of definition modelling, when evaluated on various automated benchmarks. We investigate the task of complex word identification, producing two convolutional based models that predict the complexity of words and two-word phrases in a context dependent manner. These models were submitted as part of the Lexical Complexity Prediction 2021 shared task, and showed results in a comparable range to that of other submissions. Both of these models are integrated into the CoAST system and evaluated through an online study. When selecting complex words from a document, the teacher’s selections, shared a sizeable overlap with the systems predictions. Results suggest that the technologies introduced in this work would benefit students, and teachers, using the CoAST system.
Impact and Reach
Statistics
Additional statistics for this dataset are available via IRStats2.