Multi-Relational Graph Transformer for Automatic Short Answer Grading

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2001 - 2012 July 10-15, 2022 ©2022 Association for Computational Linguistics Multi-Relational Graph Transformer for Automatic Short Answer Grading † Rajat Agarwal Varun Khurana∗ Karish Grover∗ Mukesh Mohania Vikram Goyal Indraprastha Institute of Information Technology, Delhi {rajata, varun19124, karish19471, mukesh, vikram}@iiitd.ac.in Abstract The recent transition to the online educational domain has increased the need for Automatic Short Answer Grading (ASAG). ASAG auto- matically evaluates a student’s response against a (given) correct response and thus has been a prevalent semantic matching task. Most ex- isting methods utilize sequential context to compare two sentences and ignore the struc- tural context of the sentence; therefore, these methods may not result in the desired perfor- mance. In this paper, we overcome this prob- lem by proposing a Multi-Relational Graph Transformer, MitiGaTe, to prepare token rep- resentations considering the structural con- text. Abstract Meaning Representation (AMR) graph is created by parsing the text response and then segregated into multiple subgraphs, each corresponding to a particular relationship in AMR. A Graph Transformer is used to pre- pare relation-specific token embeddings within each subgraph, then aggregated to obtain a sub- graph representation. Finally, we compare the correct answer and the student response sub- graph representations to yield a final score. Ex- perimental results on Mohler’s dataset show that our system outperforms the existing state- of-the-art methods. We have released our im- plementation1, as we believe that our model can be useful for many future applications. 1 Introduction Grading student work is critical for assessing their course understanding. However, answer grading can become monotonous and tedious for teachers. Automatic Short Answer Grading (ASAG) task is to assign ordinal scores to student answers, given † This research work is supported by Extramarks Edu- cation India Pvt. Ltd. (an education technology company). Last Author acknowledges the support of Infosys Center of AI IIITD and iHuB Anubhuti. ∗ Equal contribution 1https://github.com/kvarun07/asag-gt Refining the solution All stages are influenced except setting the program requirements. If a test fails, it can change the whole design, implementation, etc of a program. Directly: Refining, coding. Because Refining is right before the Testing Phase and Coding is right after the Testing Phase. Indirectly: Production, Maintenance. Refining the solution, Production and Maintenance are all influenced by the Testing stage. What stages in the software life cycle are influenced by the testing stage? The testing stage can influence both the coding stage (phase 5) and the solution refinement stage (phase 7) Student A Score: 5/5 Student C Score: 3/5 Student D Score: 1.5/5 Model Answer Student B Score: 4/5 Question Figure 1: A motivating example for using multiple rela- tions in automatic short answer grading. some ‘model’ answer by an academician or instruc- tor. Figure 1 presents a sample question, model answer, and student answers from an undergradu- ate computer science course (Mohler et al., 2011). One of the early approaches for solving the ASAG task has been …

Content truncated to 500 words