Dataset Card for HuCOLA

Dataset Summary

This is the dataset card for the Hungarian Corpus of Linguistic Acceptability (HuCOLA), which is also part of the Hungarian Language Understanding Evaluation Benchmark Kit HuLU.

Supported Tasks and Leaderboards

Languages

The BCP-47 code for Hungarian, the only represented language in this dataset, is hu-HU.

Dataset Structure

Data Instances

For each instance, there is aN id, a sentence and a label.

An example:

{"Sent_id": "dev_0",
 "Sent": "A földek eláradtak.",
 "Label": "0"}

Data Fields

• Sent_id: unique id of the instances, an integer between 1 and 1000;
• Sent: a Hungarian sentence;
• label: '0' for wrong, '1' for good sentences.

Data Splits

HuCOLA has 3 splits: train, validation and test.

Dataset split	Number of sentences in the split	Proportion of the split
train	7276	80%
validation	900	10%
test	900	10%

The test data is distributed without the labels. To evaluate your model, please contact us, or check HuLU's website for an automatic evaluation (this feature is under construction at the moment). The evaluation metric is Matthew's correlation coefficient.

Dataset Creation

Source Data

Initial Data Collection and Normalization

The data was collected by two human annotators from 3 main linguistic books on Hungarian language:

• Kiefer Ferenc (ed.) (1992), Strukturális magyar nyelvtan 1. Mondattan. Budapest, Akadémiai Kiadó.
• Alberti, Gábor and Laczkó, Tibor (eds) (2018), Syntax of Hungarian Nouns and Noun Phrases. I., II. Comprehensive grammar resources. Amsterdam University Press, Amsterdam.
• Katalin É. Kiss and Veronika Hegedűs (eds) (2021), Postpositions and Postpositional Phrases. Amsterdam: Amsterdam University Press.

The process of collecting sentences partly followed the one described in Warstadt et. al (2018). The guideline of our process is available in the repository of HuCOLA.

Annotations

Annotation process

Each instance was annotated by 4 human annotators for its acceptability (see the annotation guidelines in the repository of HuCOLA).

Who are the annotators?

The annotators were native Hungarian speakers (of various ages, from 20 to 67) without any linguistic backround.

Additional Information

Licensing Information

HuCOLA is released under the CC-BY-SA 4.0 licence.

Citation Information

If you use this resource or any part of its documentation, please refer to:

Ligeti-Nagy, N., Ferenczi, G., Héja, E., Jelencsik-Mátyus, K., Laki, L. J., Vadász, N., Yang, Z. Gy. and Váradi, T. (2022) HuLU: magyar nyelvű benchmark adatbázis kiépítése a neurális nyelvmodellek kiértékelése céljából [HuLU: Hungarian benchmark dataset to evaluate neural language models]. XVIII. Magyar Számítógépes Nyelvészeti Konferencia. (in press)

@inproceedings{ligetinagy2022hulu,
  title={HuLU: magyar nyelvű benchmark adatbázis kiépítése a neurális nyelvmodellek kiértékelése céljából},
  author={Ligeti-Nagy, N. and Ferenczi, G. and Héja, E. and Jelencsik-Mátyus, K. and Laki, L. J. and Vadász, N. and Yang, Z. Gy. and Váradi, T.},
  booktitle={XVIII. Magyar Számítógépes Nyelvészeti Konferencia},
  year={2022}
}

Contributions

Thanks to lnnoemi for adding this dataset.