metadata
license: mit
language:
- de
base_model:
- benjamin/roberta-base-wechsel-german
tags:
- simplification
🧭 DETECT: Determining Ease and Textual Clarity of German Text Simplifications
This repository contains the trained checkpoint for DETECT, an automated German Automatic Text Simplification (ATS) quality evaluation metric introduced in
“DETECT: Determining Ease and Textual Clarity of German Text Simplifications”.
DETECT provides fine-grained scoring across simplicity, meaning preservation, and fluency, along with a composite total score. Further information about the metric can be found in the description of the GitHub repository or in our accompanying paper.
🔎 Note
- This repository hosts a checkpoint file only.
- You must load it through the DETECT codebase (see usage below).
- It is not directly compatible with
AutoModel.from_pretrained().- The model supports reference-based text simplification evaluation only — it does not provide reference-free evaluation.
⚙️ Usage
Clone and install the DETECT codebase:
git clone https://github.com/ZurichNLP/DETECT.git
cd DETECT/detect
pip install -e .
Then, in Python:
# Initialize model
detect = DETECT("ZurichNLP/DETECT/best-LENS_multi_wechsel_reducedhs-epoch=04.ckpt", rescale=True)
complex = [
"Sie sind kulturell den Küstenbewohnern von Papua-Neuguinea verwandt."
]
simple = [
"Sie sind kulturell den Menschen in Papua-Neuguinea ähnlich."
]
references = [[
"Sie sind kulturell den Küstenbewohnern von Papua-Neuguinea ähnlich.",
"Sie ähneln den Menschen aus Papua-Neuguinea, die an der Küste leben."
]]
scores = detect.score(complex, simple, references, batch_size=8, devices=[0])
print(scores)
# [{'simplicity': 78.6, 'meaning_preservation': 80.1, 'fluency': 77.3, 'total': 78.3}]
Citation
If you use DETECT, please cite:
@inproceedings{korobeynikova2026detect,
title={DETECT: Determining Ease and Textual Clarity of German Text Simplifications},
author={Korobeynikova, Maria and Battisti, Alessia and Fischer, Lukas and Gao, Yingqiang},
booktitle={Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)},
pages={2852--2882},
year={2026}
}