SGD-X

SGD-X数据集，全称为Schema Guided Dialogue eXtended，是由谷歌研究院创建的，旨在评估和提高对话系统在面对不同语言风格时的鲁棒性。该数据集基于原始的Schema-Guided Dialogue (SGD)数据集，为每个服务提供了5种风格各异的变体，共计4755条数据。这些变体通过众包方式由400多名作者提供，确保了语言描述的多样性。SGD-X数据集的应用领域主要集中在对话系统的鲁棒性测试，特别是在零/少量样本转移至未见服务时的表现，旨在解决对话模型在实际应用中遇到的语言多样性问题。

SGD-X dataset, whose full name is Schema Guided Dialogue eXtended, was developed by Google Research to evaluate and enhance the robustness of dialogue systems against diverse linguistic styles. Built upon the original Schema-Guided Dialogue (SGD) dataset, SGD-X provides 5 stylistically distinct variants for each service, with a total of 4755 data instances. These variants were contributed by over 400 authors via crowdsourcing, ensuring the diversity of linguistic descriptions. The SGD-X dataset is primarily utilized for robustness testing of dialogue systems, particularly their performance when transferring to unseen services under zero- or few-shot settings, aiming to resolve the language diversity issues encountered by dialogue models in real-world applications.