Skip to main content

Fuzzy Semantic Retrieval Strategies for Automated Short-Answer Grading with Large Language Models in Language Learning

Computational Linguistics in the Netherlands Journal, vol. 15, 2026

Educational NLP
Adapts translation-memory fuzzy matching to automated grading, with an accuracy and recall trade-off by shot count.
Authors

Thomas Moerman

Jasper Degraeuwe

Arda Tezcan

Published

March 1, 2026

Fuzzy matching from translation memories, applied to automated grading
Fuzzy matching from translation memories, applied to automated grading
Accuracy per strategy, best shot count per topic
Accuracy per strategy, best shot count per topic
The accuracy and recall trade-off by shot count
The accuracy and recall trade-off by shot count

This work adapts translation-memory fuzzy matching to automated grading of short-answer language exercises. It evaluates ten retrieval strategies across 306 experiments and finds an accuracy and recall trade-off governed by the number of examples, with semantic retrieval generally beating non-semantic strategies.

Research theme: Educational NLP

Citation

BibTeX citation:
@article{moerman2026grading,
  author = {Moerman, Thomas and Degraeuwe, Jasper and Tezcan, Arda},
  title = {Fuzzy {Semantic} {Retrieval} {Strategies} for {Automated}
    {Short-Answer} {Grading} with {Large} {Language} {Models} in
    {Language} {Learning}},
  journal = {Computational Linguistics in the Netherlands Journal},
  date = {2026-03-01},
  langid = {en}
}
For attribution, please cite this work as:
Moerman, Thomas, Jasper Degraeuwe, and Arda Tezcan. 2026. “Fuzzy Semantic Retrieval Strategies for Automated Short-Answer Grading with Large Language Models in Language Learning.” Computational Linguistics in the Netherlands Journal, accepted, March 1.