Paper Accepted in the EMNLP 21 Main Conference: A Semantic Feature-Wise Transformation Relation Network for Automatic Short Answer Grading

 Congratulations! A paper by Zhaohui Li, Yajur Tomar, and Rebecca J. Passonneau,  "A Semantic Feature-Wise Transformation Relation Network for Automatic Short Answer Grading", has been accepted to the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP 2021) Main conference. 

This paper is about the Automatic short answer grading (ASAG), the task of assessing students’ short natural language responses to objective questions. In this paper, we propose a Semantic Feature-wise transformation Relation Network (SFRN) that exploits the multiple components of ASAG datasets more effectively. As shown in Figure 2, a neural network is applied to capture relational knowledge among the questions (Q), reference answers or rubrics (R), and labeled student answers (A). A relation network learns vector representations for the elements of QRA triples, then combines the learned representations using learned semantic feature-wise transformations.

In addition, we apply translation-based data augmentation to address the two problems of limited training data, and high data skew for multi-class ASAG tasks. Our model has up to 11% performance improvement over state-of-the-art results on the benchmark SemEval-2013 datasets. 

What is more exciting is that this work will be used in a real world education project which is a collaboration with Sadhana Puntambekar (University of Wisconsin, Education) and ChanMin Kim (PSU, Education). We are creating a new collaborative "Writer's Notebook" system to help middle school students learn physics and improve their scientific writing skills. This SFRN algorithm will be used for automated analysis of the student’s answer and essay and providing feedback to the teachers and students. 

EMNLP 2021 will be held on 7th – 11th November 2021 online and in the Barceló Bávaro Convention Centre, Punta Cana, Dominican Republic. The codes, paper, and dataset will be ready by then.