The Self-dialogue Corpus - a collection of self-dialogues across music, movies and sports
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.

The Self-dialogue Corpus

This is an early release of the Self-dialogue Corpus containing 24,165 conversations, or 3,653,313 words, across 23 topics. For more information on the data, please see our corpus paper or our submission to the Alexa Prize.

Using the data

  • corpus contains the raw CSVs from Amazon Mechanical Turk, sorted by individual tasks (topics);
  • blocked_workers.txt lists workers who did not comply with the requirements of the tasks, these are omitted by default;
  • is a preprocessing script which will format the CSVs into text (by default saved to dialogues), along with various options (see below).

Example usage: python This will by default read from corpus and write to dialogues.

Optional arguments:

  • --inDir Directory to read corpus from
  • --outDir Directory to write processed files
  • --output-naming whether to name output files with integers (integer) or by assignment_id (assignment_id);
  • --remove-punctuation removes punctuation from the output;
  • --set-case sets case of output to original, upper or lower;
  • --exclude-topic excludes any of the topics (or subdirectories of corpus), e.g. --exclude-topic music;
  • --include-only includes only the given topics, e.g. --include-only music.


For research using this data, please cite:

  title={Talking to myself: self-dialogues as data for conversational agents},
  author={Fainberg, Joachim and Krause, Ben and Dobre, Mihai and Damonte, Marco and Kahembwe, Emmanuel and Duma, Daniel and Webber, Bonnie and Fancellu, Federico},
  journal={arXiv preprint arXiv:1809.06641},
  title={Edina: Building an Open Domain Socialbot with Self-dialogues},
  author={Krause, Ben and Damonte, Marco and Dobre, Mihai and Duma, Daniel and Fainberg, Joachim and Fancellu, Federico and Kahembwe, Emmanuel and Cheng, Jianpeng and Webber, Bonnie},
  journal={Alexa Prize Proceedings},