This repository contains the datasets and metric umplementation from our paper:
DETECT: Determining Ease and Textual Clarity of German Text Simplifications
Contains the instructions and resources for generating SimpEvalDE,
a German-specific dataset for text simplification quality evaluation on which DETECT was trained and evaluated.
SimpEvalDE includes synthetic scores for simplicity, meaning preservation,
and fluency, generated by three LLM-based judges.
Implements:
- LENS [Maddela et al., 2023]
- DETECT, our German-specific adaptation of LENS
Please refer to lens/README.md for detailed usage instructions,
model loading, and scoring examples.