Reproduce “Enhancing Multi-Domain Automatic Short Answer Grading through an Explainable Neuro-Symbolic Pipeline” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Felix Künnecke, Anna Filighera, Colin Leong, Tim Steuer
A neuro-symbolic pipeline leverages weak supervision to automatically generate justification cue annotations for short answers, enabling a system that provides both highly accurate grades and interpretable explanations. It achieves state-of-the-art performance on a bilingual dataset, reducing RMSE by up to 0.3 compared to prior baselines.
Key insights
- Weak supervision effectively overcomes the data scarcity challenge for justification cue annotations, making fine-grained explainability practical without extensive manual labeling.
- The neuro-symbolic architecture successfully combines the pattern recognition strengths of neural networks with the transparent reasoning of symbolic models, providing both high accuracy and inherent explainability.
- Backpropagating the loss from the final grade prediction to the justification cue detection model implicitly refines cue identification, optimizing the entire ASAG process for accurate outcomes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.