The dataset was created using Cornell Movies Dialog Corpus which contains a large metadata-rich collection of fictional conversations extracted from raw movie scripts. Dialogs and meta-data from the underlying Corpus were used to design a dataset that can be used to InstructGPT based models to learn movie scripts.
Example :
User: Assume RICK and ALICE are characters from a fantasy-horror movie, continue the conversation between them
RICK: I heard you screaming. Was it a bad one?
ALICE: It was bad.
RICK: Doesn't the dream master work for you anymore?
Assistant: Sure
ALICE: I can't find him.
RICK: Hey, since when do you play Thomas Edison? This looks like Sheila's.
ALICE: It is...was. It's a zapper, it might help me stay awake.
RICK: Yeah, or turn you into toast.
from datasets import load_dataset
dataset = load_dataset("shahules786/OA-cornell-movies-dialog")@InProceedings{Danescu-Niculescu-Mizil+Lee:11a,
author={Cristian Danescu-Niculescu-Mizil and Lillian Lee},
title={Chameleons in imagined conversations:
A new approach to understanding coordination of linguistic style in dialogs.},
booktitle={Proceedings of the Workshop on Cognitive Modeling and Computational Linguistics, ACL 2011},
year={2011}
}