README.md

---
language:
- da
pretty_name: DaLA - Danish Linguistic Acceptability Dataset
tags:
- linguistic-acceptability
- nlp
- danish
- benchmark
- text-classification
- minimal-pairs
task_categories:
- text-classification
license: cc-by-4.0
dataset_info:
  features:
  - name: text
    dtype: string
  - name: corruption_type
    dtype: string
  - name: label_da
    dtype: string
  - name: label
    dtype: int64
  splits:
  - name: train
    num_bytes: 677973
    num_examples: 4592
  - name: validation
    num_bytes: 55377
    num_examples: 386
  - name: test
    num_bytes: 398975
    num_examples: 2678
  - name: full_train
    num_bytes: 796707
    num_examples: 5352
  download_size: 1048229
  dataset_size: 1929032
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
  - split: full_train
    path: data/full_train-*
size_categories:
- 1K<n<10K
---

# DaLA: Danish Linguistic Acceptability Evaluation Dataset

**NOTE: This is a variant of [DaLA Standard](https://huggingface.co/datasets/giannor/dala) with labels in Danish language instead of English (as in the original one), the data is the same. The following information are the same contained in the original repository**

---

**DaLA** ([paper][1]) is a benchmark dataset for **linguistic acceptability judgment** in Danish, designed to evaluate how well NLP models, especially large language models (LLMs), understand grammaticality in real-world Danish sentences. The dataset extends previous resources by introducing a broader and more realistic set of error types and providing data splits suitable for evaluation via few-shot or finetuning.

---

## 🔗 Links
- DaLA variants are linked and described below
- [Paper][1]
- [GitHub Repository](https://github.com/N-essuno/DaLA) (code, data generation scripts)

---

## 📖 Overview

In linguistic acceptability tasks, models must distinguish between **grammatically acceptable** and **unacceptable** sentences. The DaLA dataset was created by:

- Analyzing real-world Danish writing errors.
- Designing **14 distinct corruption functions** that reflect common Danish mistakes (e.g., pronoun confusion, suffix errors, interchange of determiners).
- Applying a single corruption to each correct Danish sentence creating an incorrect counterpart, resulting in **minimal pairs** of sentences that differ by only one error.

The dataset includes:
- The original correct sentences (*acceptable*).
- The corrupted sentences (*unacceptable*).
- A binary acceptability label.
- A corruption type identifier.

---

## 📦 Dataset Variants and Splits

There are three variants of the DaLA dataset, each with different sizes and proportions:

| Split Variant | Description | Size (approx.) | Link |
|------------------|-------------|----------------|----------------|
| `dala` | Standard benchmark with proportions comparable to prior Danish acceptability datasets | 3,328 samples | [DaLA Standard](https://huggingface.co/datasets/giannor/dala) |
| `dala_medium` | Expanded version using more available samples | ~6,056 samples | [DaLA Medium](https://huggingface.co/datasets/giannor/dala_medium) |
| `dala_large` | Largest version with the full expanded dataset | ~7,656 samples | [DaLA Large](https://huggingface.co/datasets/giannor/dala_large) |

Each variant includes train, validation, and test splits.

---

## 🧠 Tasks & Usage

DaLA is primarily intended for:

✔ **Model evaluation and benchmarking**: Assessing model competence in grammatical judgment
✔ **Minimal-pair evaluation**: Error type discrimination and fine-grained analysis

You can load the dataset using the Hugging Face `datasets` library as follows:

```python
from datasets import load_dataset

# Standard split
dataset = load_dataset("giannor/dala")

# Medium or large variants
dataset_medium = load_dataset("giannor/dala_medium")
dataset_large = load_dataset("giannor/dala_large")
```

## 📊 Baselines & Model Performance

In the corresponding paper, DaLA was used to benchmark a variety of open-source LLMs and model types. Across many models, performance on DaLA was **lower** than on previous Danish acceptability benchmarks, highlighting DaLA’s **greater difficulty and discriminatory power**. ([DaLA paper][1])

---

## 📄 Citation

If you use this dataset in your work, please cite the following paper:

```bibtex


@misc
{barmina2025daladanishlinguisticacceptability,
title={DaLA: Danish Linguistic Acceptability Evaluation Guided by Real World Errors},
author={Gianluca Barmina and Nathalie Carmen Hau Norman and Peter Schneider-Kamp and Lukas Galke},
year={2025},
eprint={2512.04799},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2512.04799},
}
```

---

## ⚖️ License

This dataset is shared under the **CC BY 4.0** license.


[1]: https://arxiv.org/abs/2512.04799 "DaLA: Danish Linguistic Acceptability Evaluation Guided by Real World Errors"