UppsalaNLP/sou-corpus

---
license: cc-by-4.0
language:
- sv
task_categories:
- text-generation
- token-classification
tags:
- swedish
- government-reports
- dependency-parsing
- universal-dependencies
- nlp
size_categories:
- 100K<n<1M
source_datasets:
- original
---

# SOU Corpus - Swedish Government Official Reports

Cleaned and dependency-parsed Swedish Government Official Reports (Statens offentliga utredningar) from 1994-2020.

## Dataset Description

This dataset contains sentence-segmented and dependency-parsed text from Swedish Government Official Reports. The original documents were cleaned, processed, and annotated with Universal Dependencies-style parsing.

### Fields

- **document_id**: Original document identifier (can be linked to Riksdagen open data)
- **text_type**: Type of text section
- `full_text`: Main report body
- `summary_swedish`: Standard Swedish summary
- `summary_simple_swedish`: Simple Swedish (lättläst) summary
- `summary_english`: English summary
- **section**: Section header from the document
- **text**: Plain text sentence
- **parsed**: Dependency-parsed sentence (token//POS//deprel//head format)

### Parsed Format

Each token in the `parsed` field follows the format:
```
word//POS_TAG//DEPENDENCY_RELATION//HEAD_INDEX
```

Example:
```
Sverige//PM|NOM//nsubj//3 är//VB|PRS|AKT//cop//3 ett//DT|NEU|SIN|IND//det//3 land//NN|NEU|SIN|IND|NOM//ROOT//3
```

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("UppsalaNLP/sou-corpus")

# Access train/test splits
train = dataset["train"]
test = dataset["test"]

# Example
print(train[0]["text"])
print(train[0]["section"])
print(train[0]["text_type"])
```

### Extract Tokens

```python
def parse_tokens(parsed_str):
tokens = []
for t in parsed_str.split(' '):
parts = t.split('//')
if len(parts) >= 4:
tokens.append({
'word': parts[0],
'pos': parts[1],
'deprel': parts[2],
'head': int(parts[3]) if parts[3].isdigit() else 0
})
return tokens

tokens = parse_tokens(train[0]["parsed"])
```

## Source

Documents obtained from [Riksdagens öppna data](http://data.riksdagen.se). Original document URLs follow the pattern:
`https://data.riksdagen.se/dokument/{document_id}.html`

## Citation

```bibtex
@inproceedings{durlich-etal-2022-cause,
title = "Cause and Effect in Governmental Reports: Two Data Sets for Causality Detection in Swedish",
author = "D{\"u}rlich, Luise and Reimann, Sebastian and Finnveden, Gustav and Nivre, Joakim and Stymne, Sara",
booktitle = "Proceedings of the First Workshop on Natural Language Processing for Political Sciences",
month = jun,
year = "2022",
address = "Marseilles, France"
}
```

## License

This dataset is licensed under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).

## Links

- [Uppsala NLP](https://huggingface.co/UppsalaNLP)
- [GitHub Repository](https://github.com/UppsalaNLP/SOU-corpus)
- [Riksdagen Open Data](http://data.riksdagen.se)