A Language Modelling Approach to Authorship Identification for Online Examinations in Sinhala

Abstract

With the Covid-19 outbreak, e-learning has become the ‘new normal’ with many universities and institutions adopting online platforms to deliver their programs. One aspect of this that has posed many challenges is in conducting written examinations. This is mainly because it has become increasingly difficult to verify the identity of individuals sitting for an examination remotely. The primary objective of this research is to address this problem by developing a Language Model that can be used in authorship identification for online examinations conducted in Sinhala. Essentially, the idea is that by training a language model solely on the writings of a given author, it is possible to determine the likelihood (probability) of an entirely new piece of writing having been written by that author. It was found that a character-level language model can be used to identify the author of whose writings it was trained, using the concept of perplexity.