ShAnEL-2: A Multilingual Benchmarking Dataset for Short-Answer Language Learning Exercises
LREC 2026
Educational NLP
A multilingual dataset of 1,185 learner responses with teacher corrections, plus a Gemma 3 benchmark.
Presents ShAnEL-2, a multilingual dataset of 1,185 authentic learner responses across English, Spanish, and Dutch with expert teacher corrections. It benchmarks Gemma 3 in three setups and shows that few-shot retrieval-augmented generation reaches the best accuracy and recall.
Research theme: Educational NLP
Citation
BibTeX citation:
@inproceedings{degraeuwe2026,
author = {Degraeuwe, Jasper and Moerman, Thomas},
title = {ShAnEL-2: {A} {Multilingual} {Benchmarking} {Dataset} for
{Short-Answer} {Language} {Learning} {Exercises}},
booktitle = {Proceedings of the Fifteenth Language Resources and
Evaluation Conference (LREC)},
date = {2026-05-01},
langid = {en}
}
For attribution, please cite this work as:
Degraeuwe, Jasper, and Thomas Moerman. 2026. “ShAnEL-2: A
Multilingual Benchmarking Dataset for Short-Answer Language Learning
Exercises.” Proceedings of the Fifteenth Language Resources
and Evaluation Conference (LREC), accepted, May 1.