---
pretty_name: Norwegian punctuation restoration
language:
- nob
license: cc-by-sa-4.0
license_name: CC BY-SA 4.0
tasks:
- instruction_following
creation_approach: semi-synthetic
task_categories:
- text-generation
- text2text-generation
---

# Dataset Card for Norwegian punctuation restoration

<!-- START-SHORT DESCRIPTION -->
Punctuation and capitalization corrections for Norwegian Wikipedia passages.
<!-- END-SHORT DESCRIPTION -->

## Dataset Description

<!-- START-DESC-STATS -->
- **Number of samples**: 4.50K
- **Number of tokens (Llama 3)**: 1.63M
- **Average conversation length in tokens (min, max)**: 362.28 (164, 1.37K)
- **Average number of turns (min, max)**: 2.0 (2, 2)
<!-- END-DESC-STATS -->

This subset contains 4,496 examples for restoring punctuation and capitalization
without changing the words. It uses only the source training split. Validation and
test examples and the model's reasoning traces are not included.

## Dataset Structure

<!-- START-SAMPLE -->
```py
{
  "id": "reasoning-norwegian_00000",
  "messages": [
    {
      "role": "user",
      "content": "Gjennomgå og korriger følgende tekst for feil i tegnsetting (feilplasserte, manglende eller overflødige tegn) samt bruk [...]"
    },
    {
      "role": "assistant",
      "content": "I sin nåværende form er stedsnavnet Dunbar avledet fra sin gæliske ekvivalent, moderne skotsk-gæliske i betydningen «bor[...]"
    }
  ],
  "source": "reasoning-norwegian",
  "added": "2026-09-03",
  "created": "2025-01-01, 2025-02-28",
  "language": [
    "nob"
  ],
  "task": "instruction_following",
  "token_count": 337
}
```

### Data Fields

An entry in the dataset consists of the following fields:

- `id` (`str`): A unique identifier for each sample.
- `messages` (`list[dict]`): The conversation turns — see the [message format](../../README.md#data-fields) section in the main README.
- `source` (`str`): The source dataset (see [Source Data](#source-data)).
- `added` (`str`): The date when the sample was added to this collection.
- `created` (`str`): The approximate date range when the source data was created, as `"YYYY-MM-DD, YYYY-MM-DD"` (start, end). Year-level granularity is used when the exact date is unknown.
- `language` (`list[str]`): The language(s) of the sample as ISO 639-3 codes (e.g. `["nob"]`, `["nno"]`, `["nob", "eng"]`).
- `task` (`str`): The task type of the sample.
- `token_count` (`int`): The total number of tokens across all messages, computed using the Llama 3 tokenizer.
<!-- END-SAMPLE -->

### Dataset Statistics

<!-- START-DATASET PLOTS -->
<div style="display: flex; gap: 20px;">
<p align="center">
<img src="./images/dist_token_count.png" width="400" />
</p>
<p align="center">
<img src="./images/dist_turns.png" width="400" />
</p>
</div>
<!-- END-DATASET PLOTS -->

## Additional Information

### Source Data

The data comes from
[`pere/reasoning_norwegian`](https://huggingface.co/datasets/pere/reasoning_norwegian)
at revision `214ce4bd6dc06156b7f930a8992cedd5a1cf3e1e`. Norwegian Wikipedia
paragraphs were given punctuation and capitalization errors, then sent to
DeepSeek-R1 for correction. The source includes the original and corrupted text, the
generated response and reasoning trace, and the Wikipedia URL and paragraph number.

### Data Preparation

The fixed correction instruction and the corrupted text form the user message. The
source `text_result` field becomes the assistant response. Empty responses are removed,
as are responses that change the normalized sequence of letters and numbers. This
leaves 4,496 of the 6,245 training examples. Validation and test rows are not used.

The reasoning traces are omitted. `attribution.jsonl` keeps the Wikipedia URL and
paragraph number for each retained example.

### Limitations

The responses are synthetic. The filter checks that words were not added, removed, or
changed, but it does not prove that every punctuation and capitalization choice is
correct. The source does not name the exact DeepSeek-R1 checkpoint or publish its
generation code. It also does not provide Wikipedia page revision IDs, so the linked
pages may have changed.

### License Information

The source is based on Norwegian Wikipedia but does not keep page revision dates. Since
the dataset was published after Wikipedia moved to CC BY-SA 4.0, this subset is
released under
[CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/). DeepSeek-R1 is
released under the [MIT License](https://github.com/deepseek-ai/DeepSeek-R1/blob/main/LICENSE).