hamishivi/ROCStories

---
license: cc-by-4.0
task_categories:
- text-generation
language:
- en
size_categories:
- 10K<n<100K
---

# ROCStories (prompt / continuation)

A reformatted version of the [ROCStories](https://cs.rochester.edu/nlp/rocstories/)
corpus, suitable for open-ended story-generation exercises.

Each example is a 5-sentence ROCStory, split into:

| field | description |
|---------------|------------------------------------------------------|
| `prompt` | the first sentence of the story |
| `continuation`| the remaining four sentences |
| `text` | the full (unmodified) 5-sentence story |

## Splits

| split | rows |
|--------------|---------|
| `train` | 70,676 |
| `validation` | 7,852 |
| `test` | 19,633 |

The `train` / `validation` split is a 90/10 shuffle (seed = 42) of the train
split from [mintujupally/ROCStories](https://huggingface.co/datasets/mintujupally/ROCStories);
`test` is mintujupally's test split unchanged.

## Provenance

This dataset was reconstructed to replace
`Ximing/ROCStories`, which is no longer available on the Hub, as the data
source for the CSED503 / CSE447 Natural Language Generation assignment.

The underlying stories come from the ROC Stories 2016 and 2017 releases
(Mostafazadeh et al.). Please cite the original corpus when using this data:

```
@inproceedings{mostafazadeh2016corpus,
title = {A Corpus and Cloze Evaluation for Deeper Understanding of Commonsense Stories},
author = {Mostafazadeh, Nasrin and Chambers, Nathanael and He, Xiaodong and Parikh, Devi and Batra, Dhruv and Vanderwende, Lucy and Kohli, Pushmeet and Allen, James},
booktitle = {Proceedings of NAACL-HLT},
year = {2016}
}
```