Automatic Short Answer Grading in College Mathematics Using In-Context Meta-learning: An Evaluation of the Transferability of Findings

Mathematics teachers use open-ended (OE) problems to inspire creativity, facilitate learning by self-explanation, and encourage transfer learning. While these types of problems are pedagogically valuable, student answers often exhibit a combination of language and mathematical expressions, and the variation in these responses can make it difficult and time-consuming to assess. There have been growing efforts to research automatic short-answer grading (ASAG) methods to support teachers with this crucial task with promising results at the K12 level. However, whether these findings transfer to other student groups or across content, specifically at the college level, remains an open question. We implement a machine learning model for ASAG developed on K12 content and student responses and evaluate the transferability of previous findings to those at the college level. Our results show that the transferability can vary significantly, buttressing the assertion that this line of investigation warrants future research.