dataset_info:
features:
- name: correct
dtype: string
- name: wrong
dtype: string
- name: category
dtype: int64
splits:
- name: train
num_bytes: 199862
num_examples: 840
download_size: 124946
dataset_size: 199862
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
Dataset Card for Norwegian Comma Benchmark (NCB)
Sentence-level evaluation of the comma placement abilities of large language models in Norwegian sentences.
Dataset Description
The NCB corpus (version 0.1) is a collection of 840 human-written Norwegian sentence pairs. The sentences are manually collected from publicly available sources such as articles and governmental reports. The sentences aim to be representative of Norwegian non-fiction, in particular governmental prose. Each sentence pair tests one Norwegian comma rule: one sentence is correctly punctuated, while the other contains faulty comma usage. The dataset tests both rules where there should be a comma and rules where there should not be a comma.
- Point of Contact: Hans Christian Farsethås and Joakim Tjøstheim
- Language: Norwegian (bokmål)
- License: The data is distributed under a Creative Commons Attribution-NonCommercial licence (CC BY-NC 4.0). The licence is motivated by the need to block the possibility of third parties redistributing the original reviews for commercial purposes. Note that machine learned models, extracted lexicons, embeddings, and similar resources that are created on the basis of NoReC are not considered to contain the original data and so can be freely used also for commercial purposes despite the non-commercial condition.
Uses
The data is intended to be used for testing and finetuning models for Norwegian sentence-level comma usage.
Dataset Structure
Data Instances
Each data instance contains the following features: correct, wrong and category. An instance from the dataset that require a comma looks like the following:
{"correct": "Eva klatrer, og Ola går på ski.", "wrong": "Eva klatrer og Ola går på ski.", "category": 1}
An instance with a sentence that is correct without a comma looks like the following:
{"correct": "Husleien går opp og har nå doblet seg på ti år.", "wrong": "Husleien går opp, og har nå doblet seg på ti år.", "category": 1}
Data Fields
- 'Correct': a string containing a Norwegian sentence with correct use of comma.
- 'Wrong': a string containig the same Norwegian sentence with wrong use of comma.
- 'Category': an integer indicating the rule tested by the pair of correct and wrong sentences.
Distribution
The dataset includes 840 instances, each consisting of a pair of sentences: one correctly punctuated and the other incorrectly punctuated. Of these, 600 instances feature sentences that are grammatically correct with the inclusion of commas, while 240 instances contain sentences that are correct without any commas. Among the 600 sentences that require commas, the majority need just one comma, whereas 207 instances require the inclusion of two commas. Five sentences utilize a comma as a decimal separator in addition to the grammatical commas.
Norwegian comma rules
The model tests the following Norwegian comma rules:
1. Det skal alltid være komma mellom helsetninger som er bundet sammen med sideordnende konjunksjoner
2. Det skal alltid være komma mellom leddsetninger som er bundet sammen med sideordnende konjunksjoner
3. Det skal alltid være komma etter en leddsetning som står først i en helsetning
4. Det skal alltid være komma etter en innskutt leddsetning
501. Det skal alltid være komma før og etter apposisjoner som ikke befinner seg i slutten av, men inni en helsetning
502. Det skal alltid være komma før og etter tillegg som ikke befinner seg i slutten av, men inni en helsetning
503. Det skal alltid være komma før og etter innskudd
601. Det skal alltid være komma før apposisjoner som står til slutt i en helsetning
602. Det skal alltid være komma før tillegg som står til slutt i en helsetning
7. Det skal ikke være komma når det til ett subjekt står to eller flere predikater forbundet med konjunksjon
8. Det skal ikke være komma etter preposisjonsstyrte infinitiver og andre ikke-setningsformede ledd
9. Det skal ikke være komma etter ufullstendige leddsetninger
10. Det skal ikke være komma mellom leddsetninger hvor en leddsetning står som siste ledd i en annen leddsetning
11. Det skal alltid være komma i en oppramsing dersom det ikke står noen konjunksjon
Dataset Creation
Curation Rationale
The sentences in the dataset are primarily curated to benchmark the performance of language models on Norwegian comma usage.
Data Preparation
The sentences in the dataset are collected from publicly available documents. After collection, they are proofread and lightly edited to remove ambiguity and ensure they align with the specific rule being tested.
The dataset contains 60 sentence pairs for each rule. Thus, it is representative of the rules rather than of language in actual use. The unacceptable version of the sentences are made by manually removing or adding commas.
Generalizability
The dataset aims to be representative of Norwegian non-fiction, with an emphasis on governmental prose. The rules tested are the main syntactic comma rules. Other forms of comma usage, such as commas indicating sentence rhythm and cases where correct placement depends on the writer's intended meaning, are not tested.
Personal and Sensitive Information
The data does not contain information considered personal or sensitive.
Recommendations
Results obtained on this data might not generalize to texts from other domains or genres. Any biases in the sentiments expressed by the original review authors may carry over to models trained on this data.
Dataset Creators and Curators
The Norwegian Comma Benchmark was created by Hans Christian Farsethås and Joakim Tjøstheim.
To cite this work
The dataset is introduced as part of NorEval in the article NorEval: A Norwegian Language Understanding and Generation Evaluation Benchmark.
@article{mikhailov2025noreval,
title={NorEval: A Norwegian Language Understanding and Generation Evaluation Benchmark},
author={Mikhailov, Vladislav and Enstad, Tita and Samuel, David and Farseth{\aa}s, Hans Christian and Kutuzov, Andrey and Velldal, Erik and {\O}vrelid, Lilja},
journal={arXiv preprint arXiv:2504.07749},
year={2025}
}