---
configs:
- config_name: wiktionary
  data_files:
    - split: train
      path: "wiktionary/train.csv"
    - split: validation
      path: "wiktionary/valid.csv"
- config_name: web
  data_files:
    - split: train
      path: "web/train.csv"
    - split: validation
      path: "web/valid.csv"
license: mit
---

# CompoundPiece

Dataset of compound words for the paper [CompoundPiece: Evaluating and Improving Decompounding Performance of Language Models](https://arxiv.org/abs/2305.14214).

Load the balanced dataset of hyphens and non-hyphenated words scraped from the web (used as pretraining data):

```python
load_dataset("benjamin/compoundpiece", "web")
```

Load the dataset of compound and non-compound words (used for fine-tuning):

```python
load_dataset("benjamin/compoundpiece", "wiktionary")
```

# Citation

```
@article{minixhofer2023compoundpiece,
  title={CompoundPiece: Evaluating and Improving Decompounding Performance of Language Models},
  author={Minixhofer, Benjamin and Pfeiffer, Jonas and Vuli{\'c}, Ivan},
  journal={arXiv preprint arXiv:2305.14214},
  year={2023}
}
```

# License

MIT