🤖 AI Summary
This study addresses the inefficiency and error-proneness of manual grading of handwritten answer sheets, particularly in complex scenarios involving answer misalignment, strikeovers, or connected writing, where existing automated methods suffer from insufficient accuracy and compromised fairness. The work proposes the first application of general-purpose vision-language models (VLMs) to handwritten answer recognition, leveraging holistic page-level semantic understanding and prompt engineering guided by reference answers to achieve high-precision single-character judgment. A novel fairness-oriented evaluation framework is introduced, explicitly distinguishing between false negatives and false positives to significantly reduce false negative rates—errors disproportionately disadvantageous to students. Evaluated on 61 anonymized exam papers comprising 3,141 answer fields, the method achieves 98.4% overall accuracy with a false negative rate of only 0.58%; among all papers, only three received lower scores, and these discrepancies are detectable through a student self-review step.
📝 Abstract
Correcting handwritten exams by hand is time-consuming and error-prone, particularly for large cohorts, while fully digital exams tend to force a didactic narrowing towards closed question formats. A practical middle ground keeps paper-based, problem-oriented tasks but records the assessment-relevant answers as single capital letters in a table that a machine can read. The open question is whether this reading can be made accurate and, above all, fair enough for unsupervised grading. Earlier automated approaches reached only about 88%--91% recognition -- too low -- and failed on the cases that matter most: answers placed outside the cell, crossed out, or written in cursive. We show that general-purpose vision-language foundation models (VLMs), which interpret the page rather than match pixel templates, close this gap. On a benchmark of 61 anonymised exams (3141 answer positions) the best model reaches 98.4% accuracy, well above the previous baseline. Crucially, we centre the evaluation on fairness: we distinguish false negatives (a correct answer marked wrong, which disadvantages the student) from false positives, and a lightweight prompt that supplies the reference solution as context lowers the false-negative rate to 0.58%. Under an exemplary grading scheme only three of the 61 exams would be graded worse, all caught by a student self-review step. Fully automated, fairness-aware exam grading at scale is therefore defensible; we release the anonymised benchmark to support reproducibility.