dstc
v1
该数据曾经托管在 http://camdial.org/~mh521/dstc/ 上
查看版本下的所有内容。
原始挑战结果:
对话状态跟踪挑战 2 和 3 (DSTC2&3) 是研究挑战,重点是提高跟踪语音对话系统状态的技术水平。状态跟踪,有时称为信念跟踪,是指随着对话的进展准确估计用户的目标。准确的状态跟踪是可取的,因为它提供了对语音识别中的错误的稳健性,并有助于减少对话等时间过程中语言固有的歧义。
在这些挑战中,参与者获得了带有标签的对话语料库来开发状态跟踪算法。然后,跟踪器在一组常见的保留对话上进行评估,这些对话在一周内发布,未标记。
该语料库是使用 Amazon Mechanical Turk 收集的,由两个领域的对话组成:餐厅信息和旅游信息。旅游信息包含餐厅信息,包括酒吧、咖啡馆等以及多个新时段。使用该数据进行了两轮评估:
用于训练的对话框有完整的标签;用户转录、用户对话行为语义和对话状态都被注释。 (因此,该语料库也适合口语理解研究。)
欲了解更多详细信息,请参阅手册。
@inproceedings { henderson2014second ,
title = { The second dialog state tracking challenge } ,
author = { Henderson, Matthew and Thomson, Blaise and Williams, Jason D } ,
booktitle = { Proceedings of the 15th annual meeting of the special interest group on discourse and dialogue (SIGDIAL) } ,
pages = { 263--272 } ,
year = { 2014 }
}
@inproceedings { henderson2014third ,
title = { The third dialog state tracking challenge } ,
author = { Henderson, Matthew and Thomson, Blaise and Williams, Jason D } ,
booktitle = { 2014 IEEE Spoken Language Technology Workshop (SLT) } ,
pages = { 324--329 } ,
year = { 2014 } ,
organization = { IEEE }
}
@article { williams2014dialog ,
title = { The dialog state tracking challenge series } ,
author = { Williams, Jason D and Henderson, Matthew and Raux, Antoine and Thomson, Blaise and Black, Alan and Ramachandran, Deepak } ,
journal = { AI Magazine } ,
volume = { 35 } ,
number = { 4 } ,
pages = { 121--124 } ,
year = { 2014 }
}
@article { williams2016dialog ,
title = { The dialog state tracking challenge series: A review } ,
author = { Williams, Jason and Raux, Antoine and Henderson, Matthew } ,
journal = { Dialogue & Discourse } ,
volume = { 7 } ,
number = { 3 } ,
pages = { 4--33 } ,
year = { 2016 }
}