ConvBank

该数据集包含了1500条由自动语音识别技术生成的文本语句，这些语句源自于一个人机口语对话系统，并带有SCUD标注。该语料库遵循UD 2.0标准，并在口语对话语境下进行了一些扩展。这个数据集由经验丰富的标注员进行了标注，显示出高度的标注一致性。规模上，该数据集包含了1500个句子，其任务是针对口语对话系统进行依存关系解析。This dataset comprises 1500 textual utterances generated by automatic speech recognition (ASR) technology, which originate from a human-machine spoken dialogue system and are annotated with SCUD tags. This corpus adheres to the UD 2.0 standard, with appropriate extensions made for spoken dialogue contexts. Annotated by experienced annotators, this dataset exhibits high annotation consistency. With a total of 1500 sentences, the core task of this dataset is dependency parsing for spoken dialogue systems.

搜集汇总

背景与挑战

背景概述

ConvBank是一个人类-机器对话依赖解析数据集，用于训练和评估口语对话系统的依赖解析模型。该数据集基于EMNLP 2019论文，提供预训练模型和代码，支持在通用依赖英语网络树库上预训练后微调，适用于解析口语对话数据。