Cant Talk About This-NC is a dataset released by NVIDIA in 2024, focusing on dialogue topic control and safety alignment of language models. The related research paper is "CantTalkAboutThis: Aligning Language Models to Stay on Topic in Dialogues", which aims to enhance the ability of large language models to stay focused on the topic in task-oriented dialogues and improve their robustness against distractors.

The dataset contains 1,186 synthetic dialogue samples covering nine domains, including healthcare, banking, travel, education, finance, insurance, law, real estate, and computer troubleshooting. Each dialogue includes distractor turns designed to test the model's resistance to topic deviation. Fine-tuning on this dataset can significantly enhance the model's performance in instruction-following and safety tasks, enabling it to effectively identify sensitive topics and handle restricted content.

Dataset Composition

The dataset mainly includes the following fields:

• domain: The domain to which the dialogue belongs
• scenario: The specific scenario or task being discussed
• system_instruction: The dialogue strategy assigned to the model, typically containing complex instruction sets for permitted and prohibited topics
• conversation: The complete dialogue content, including the main topic and distractor turns
• distractors: A list of distractor turns, including the bot's turns in the dialogue and the user's distractor turns that should serve as responses to the bot's turns
• conversation_with_distractors: The complete dialogue including distractor turns

The dataset is divided into training and test sets, synthetically generated based on Llama-series models. It is used to evaluate the model's ability to resist topic deviation and maintain conversational topic stability. The license is CC-BY-NC-4.0, restricted to non-commercial research use only.

Citation

@inproceedings{sreedhar2024canttalkaboutthis,
  title={CantTalkAboutThis: Aligning Language Models to Stay on Topic in Dialogues},
  author={Sreedhar, Makesh and Rebedea, Traian and Ghosh, Shaona and Zeng, Jiaqi and Parisien, Christopher},
  booktitle={Findings of the Association for Computational Linguistics: EMNLP 2024},
  pages={12232--12252},
  year={2024},
  organization={Association for Computational Linguistics}
}