In this directory, we provide our code which we used to reproduce the German text simplification models. We have reproduced the following models:
| model name | reference | code | comment |
|---|---|---|---|
| hda_LS | Siegel et al. (2019) | https://github.com/rstodden/easy-to-understand_language | We have slightly updated the original code. |
| Sockeye-APA-LHA | Spring et al. (2021) | https://github.com/ZurichNLP/RANLP2021-German-ATS | We haven't changed the original code, please follow the instructions of the original authors. |
| trimmed_mbart_sents_apa | Stodden et al. (2023) | reproduction-based-on-checkpoints.ipynb | model loaded from Huggingface checkpoint |
| trimmed_mbart_sents_apa_web | Stodden et al. (2023) | reproduction-based-on-checkpoints.ipynb | model loaded from Huggingface checkpoint |
| BLOOM zero-shot | Ryan et al. (2023) | reproduction_bloom_by_ryan-eta-al-2023.ipynb | We have slightly updated the original code. |
| BLOOM 10-random-shot | Ryan et al. (2023) | reproduction_bloom_by_ryan-eta-al-2023.ipynb | We have slightly updated the original code. |
| BLOOM 10-similarity-shot | Ryan et al. (2023) | reproduction_bloom_by_ryan-eta-al-2023.ipynb | We have slightly updated the original code. |
| customer-decoder-ats | Anschütz et al. (2023) | reproduction-based-on-checkpoints.ipynb | model loaded from Huggingface checkpoint |
| mT5 models | Stodden et al. (2024) | mt5-models-loop.py | mT5 models fine-tuned using the provided code |
Additionally, we have fine-tuned mT5 on the simple-german-corpus and on DEplain-APA. The corresponding code is also linked in this repository.