MoMi (Model Mistakes) Dataset

Dataset for Generate, Filter, and Rank: Grammaticality Classification for Production-Ready NLG Systems. These are responses produced by our NLG model, tagged for grammaticality and correctness by annotators.

Dataset columns

• grammatical: whether text is grammatical (1) or not (0)
• correct: whether the response correctly expresses information provided in goal and scenario (1) or not (0)
• text: response produced by model
• tag: source of text
• goal: intent to be expressed in text
• scenario: input to model which produced the text

Citation

@inproceedings{Challah:2019,
  author={Challa, Ashwini and  Upasani, Kartikeya and Balakrishnan, Anusha and Subba, Rajen},
  title={Generate, Filter, and Rank: Grammaticality Classification for Production-Ready NLG Systems},
  booktitle={Proc. of NAACL Industry Track},
  year={2019}
}

License

MoMi is Creative Commons Attribution-NonCommercial 4.0 International Public licensed, as found in the LICENSE file.