TyDiQA-GoldP

TyDiQA 是 Typologically Diverse Question Answering (TyDiWA) 数据集的黄金通道版本，该数据集是信息搜索问答的基准，涵盖九种语言。黄金段落版本是主要任务的简化版本，仅使用黄金段落作为上下文，排除无法回答的问题。因此，它与 XQuAD 和 MLQA 相似，但更具挑战性，因为在没有看到答案的情况下编写了问题，与 XQuAD 和 MLQA 相比，词汇重叠分别减少了 3 倍和 2 倍。

TyDiQA is the gold passage variant of the Typologically Diverse Question Answering (TyDiQA) dataset, which serves as a benchmark for information-seeking question answering and covers nine languages. The gold passage variant is a simplified version of the core task, which exclusively uses gold paragraphs as context and filters out unanswerable questions. As such, it is similar to XQuAD and MLQA but more challenging, as questions were formulated without prior access to their corresponding answers. Compared with XQuAD and MLQA, its lexical overlap with the context is reduced by a factor of 3 and 2 respectively.

搜集汇总

背景与挑战

背景概述

TyDiQA-GoldP是TyDiQA数据集的黄金段落版本，作为一个信息搜索问答基准覆盖九种语言。该版本简化了任务，仅使用黄金段落作为上下文并排除无法回答的问题，相较于XQuAD和MLQA，其词汇重叠更少，因此更具挑战性。