xstorycloze_gl

Dataset Summary

xstorycloze_gl is a Galician multiple-choice narrative understanding dataset translated from the English StoryCloze dataset.

Each instance contains a four-sentence story context, two possible endings, and a label indicating which ending is correct. The dataset is intended for evaluating reading comprehension, narrative coherence, and commonsense reasoning in Galician.

Dataset Description

xstorycloze_gl is based on the StoryCloze task, in which a model must choose the correct continuation of a short story. The dataset includes:

• 360 instances in the train split
• 1,511 instances in the test split

Each example consists of:

• a story stem divided into four sentences
• two candidate endings
• the index of the correct ending

The dataset is designed as an evaluation resource for language models and other NLP systems working on Galician narrative understanding.

Source

This dataset is a Galician translation/adaptation of the English StoryCloze dataset and is distributed by Proxecto Nós through Hugging Face.

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

This dataset is a Galician translation/adaptation of StoryCloze and is distributed for research and evaluation purposes. Users are free to share and adapt the material, provided that appropriate credit is given to the dataset creators and to the original source resource when relevant.

Supported Tasks and Leaderboards

This dataset is suitable for:

• multiple-choice question answering
• reading comprehension
• narrative understanding
• commonsense reasoning
• evaluation of language models in Galician

Dataset Structure

The dataset is distributed in CSV format.

Each row contains the following fields:

• InputStoryid: identifier of the story
• InputSentence1: first sentence of the story
• InputSentence2: second sentence of the story
• InputSentence3: third sentence of the story
• InputSentence4: fourth sentence of the story
• RandomFifthSentenceQuiz1: first possible continuation
• RandomFifthSentenceQuiz2: second possible continuation
• AnswerRightEnding: correct ending, either 1 or 2

Example

Example

InputStoryid	InputSentence1	InputSentence2	InputSentence3	InputSentence4	RandomFifthSentenceQuiz1	RandomFifthSentenceQuiz2	AnswerRightEnding
fec30953-c68e-4d9d-9698-384bfe8fe857	Fíxenme fan de Lei e Orde en 2011.	Estaba a recuperarme dun ataque cerebral.	Cando volvín a casa, tentei ver todos os episodios.	Custoume ver do tirón unha serie que leva 20 anos.	Creo que Lei e Orde é unha das peores series que se fixeron.	Ao final, vinos todos.	2
35146100-1a33-4b70-ab26-4469198f909a	Todo o mundo adoraba a Bob porque interpretaba a un famoso personaxe nunha película.	Bob odiábao porque non se parecía nada ao personaxe.	Jim pediulle a Bob que dixese a frase característica do seu personaxe.	Foi demasiado para Bob e enrabiouse.	Bob díxolle ao home que o deixase só.	Bob pediulle un té ao home.	1
96e98dae-7c2e-4130-b6c2-4912c0e58938	Lita quería ver o seu programa de televisión favorito.	Os seus fillos insistiron para que sentase a velo con eles.	Lita díxolles que tiña cousas que facer.	Os seus fillos arrastraron a cesta da roupa ata o salón.	Lita pediu unha pizza de salame para comer.	Lita dobrou a bogada mentres miraba a televisión.	2

Intended Uses

xstorycloze_gl is intended primarily as an evaluation dataset. Suitable use cases include:

• Reading comprehension, as both stories and answers are expressed in natural language
• Narrative understanding, since models must identify the most coherent continuation of a story
• Commonsense reasoning, because many stories depend on everyday world knowledge
• Causal and temporal reasoning, as story events often involve implicit cause-effect and temporal relations
• Multiple-choice evaluation, since each item requires choosing between two competing endings

Limitations

• The dataset is a translation of an English original, so some examples may reflect translation choices rather than native Galician narrative style.
• It is primarily intended for evaluation, not large-scale training.
• Because the task is multiple-choice, performance may be affected by answer style or distractor quality in addition to narrative understanding.
• Some examples may rely on culturally general or language-specific commonsense assumptions inherited from the source dataset.

Dataset Splits

Split	Number of instances
Train	360
Test	1,511

Usage

Example with datasets:

from datasets import load_dataset

ds = load_dataset("proxectonos/xstorycloze_gl")
print(ds["train"][0])
print(ds["test"][0])

Acknowledgements

This dataset was compiled within the Nós Project, funded by the Ministerio para la Transformación Digital y de la Función Pública - Funded by EU – NextGenerationEU within the framework of the project ILENIA with reference 2022/TL22/00215336.