BiToD

BiToD是由香港科技大学人工智能研究中心（CAiRE）和阿里巴巴集团共同创建的双语多领域任务导向对话数据集。该数据集包含7,232个双语对话（144,000个话语），涵盖五个领域的七个服务，每个对话都标注了对话状态、言语行为和服务API调用。BiToD旨在为多语言国家和地区的虚拟助手开发提供支持，特别是在新加坡、香港、印度和瑞士等地。数据集的创建过程涉及从公开的香港旅游信息中收集知识库，并通过四个阶段的流程收集对话数据。BiToD的应用领域包括评估双语任务导向对话系统和跨语言迁移学习方法，旨在解决在低资源条件下的系统性能提升问题。

BiToD is a bilingual multi-domain task-oriented dialogue dataset co-created by the Center for AI Research (CAiRE) at the Hong Kong University of Science and Technology and Alibaba Group. This dataset contains 7,232 bilingual dialogues (144,000 utterances in total), covering seven services across five domains. Each dialogue is annotated with dialogue state, speech acts, and service API calls. BiToD aims to support the development of virtual assistants for multilingual countries and regions, particularly in Singapore, Hong Kong, India and Switzerland. The dataset creation process involved collecting a knowledge base from publicly available Hong Kong tourism information, and gathering dialogue data through a four-stage workflow. Application areas of BiToD include evaluating bilingual task-oriented dialogue systems and cross-lingual transfer learning methods, with the goal of improving system performance under low-resource conditions.