Automated Short Answer Grading (ASAG) is a popular task in education technology that automates the assessment of student-written short answers by leveraging computational techniques. Although ASAG has been investigated for many years, recent advancement in machine learning and deep learning has ignited new research in this field. In the past, researchers have come up with different approaches that can be broadly classified into two categories—handcrafted feature-based traditional approaches and deep-learning-based approaches. In recent times, transformer-based word embedding has revolutionized all the fields of NLP applications, including ASAG. This chapter aims to compare the performance of various methodologies employed in ASAG with particular attention to comparing feature extraction techniques, machine learning and deep learning models employed, the dataset used, and the assessment measures applied. The review of the literature reveals that combining transformer-based text embedding with handcrafted features is more effective than using conventional handcrafted features and deep learning features in both regression and classification tasks for ASAG. This finding is validated through an empirical study performed to evaluate and compare the efficiency of different handcrafted and semantic features on the ASAG task with a common dataset and two baseline models.