(DDD) Deep Dungeons and Dragons


DDD is a model that aims to learn predictive ties between characters and their actions in narratives. Specifically, the focus is on characters and actions expressed in complex natural language. DDD uses neural language models to demonstrate that actions are better predicted when the character performing the action is also known, and character descriptions are likewise better modeled together with the associated actions.

DDD also introduces a new dataset of Role-Playing Game (RPG) transcripts collected from online play-by-post forums.

The DDD corpus and models are described in our NAACL 2018 work: "Deep Dungeons and Dragons: Learning Character-Action Interactions from Role-Playing-Game Transcripts". See our paper and poster.


The DDD Corpus contains collaboratively written role-playing game stories collected from the online forum roleplayerguild.com. The stories were preprocessed and filtered and the resulting corpus is available on this page. The unprocessed text source of the stories are also available.


You can download the corpus here ddd-v1.tar.gz [tar.gz 91MB]. See README.txt for detailed information about the data format.

If you would like to cite the corpus please use:

	author={Louis, Annie, and Sutton, Charles},
	title={{Deep Dungeons and Dragons: Learning Character-Action Interactions from Role-Playing Game Transcripts}},
	booktitle={The 16th Annual Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies},	

More Information

For more information on the corpus, please contact us: