BURCHARK语料库

BURCHARK语料库是由赫瑞瓦特大学交互实验室创建的一个自由可用的人-人对话数据集，专注于通过显式定义进行视觉基础词汇意义的交互学习。该数据集包含177个对话，涉及学习者通过与导师的互动学习视觉属性词汇（如“burchak”代表正方形）。数据收集使用了一种新颖的字符逐字符变体的DiET聊天工具，该工具能够以精细的粒度记录文本交互，包括所有字符级别的时间信息。BURCHARK语料库旨在为训练多模态对话代理提供资源，这些代理能够在自然、自发的对话中从人类伙伴那里积极学习视觉概念，解决机器人、家庭自动化设备等在操作中如何学习和适应用户使用的语言的问题。

The BURCHARK Corpus is a freely available human-human dialogue dataset created by the Interaction Lab at Heriot-Watt University, focusing on interactive learning of visually grounded lexical meanings through explicit definition-based interactions. This dataset contains 177 dialogues, where learners interact with tutors to learn visual attribute lexicons (e.g., "burchak" denotes a square). Data collection was conducted using a novel character-level variant of the DiET chat tool, which enables fine-grained recording of text interactions including all character-level temporal information. The BURCHARK Corpus aims to provide a resource for training multimodal dialogue agents that can actively learn visual concepts from human partners during natural, spontaneous conversations, addressing the challenge of how robots, home automation devices, and other similar systems can learn and adapt to the language used by their users during operation.