chentong00/propositionizer-wiki-data
收藏资源简介:
--- license: apache-2.0 task_categories: - text2text-generation size_categories: - 10K<n<100K --- ## Dataset Summary This is the training data of the model `Propositionizer-wiki`. We prompt GPT-4 to decompose a Wikipedia paragraph into a list of propositions. We propose this training data to explore the concept of propositions as retrieval units. The propositions are defined as follows: 1. Each proposition should correspond to a distinct piece of meaning in the text, where the composition of all propositions would represent the semantics of the entire text. 2. A proposition should be *minimal*, i.e. it cannot be further split into separate propositions. 3. A proposition should be *contextualized and self-contained* ([Choi et al. 2021](https://aclanthology.org/2021.tacl-1.27/)). A proposition should include all the necessary context from the text (e.g. coreference) to interpret its meaning. Check out more details in the paper. ## Dataset Structure Here we provide details about the structure of the dataset. * `sources` represents a Wikipedia paragraph. It is always in the format of "Title: {title}. Section: {section}. {content}". The title will not be empty, but the section can be empty. * `targets` are a list of propositions in a JSON-formatted string. Example: ``` { "sources": "Title: Leaning Tower of Pisa. Section: . Prior to restoration work performed between 1990 and 2001, the tower leaned at an angle of 5.5 degrees, but the tower now leans at about 3.99 degrees. This means the top of the Leaning Tower of Pisa is displaced horizontally 3.9 meters (12 ft 10 in) from the center." "targets": "[\"Prior to restoration work performed between 1990 and 2001, the Leaning Tower of Pisa leaned at an angle of 5.5 degrees.\", \"The Leaning Tower of Pisa now leans at about 3.99 degrees.\", \"The top of the Leaning Tower of Pisa is displaced horizontally 3.9 meters (12 ft 10 in) from the center.\"]" } ``` ## Citation ``` ```
许可证:apache-2.0 任务类别: - 文本到文本生成(text2text-generation) 规模类别: - 10K<n<100K --- ## 数据集摘要 本数据集为模型`Propositionizer-wiki`的训练数据。我们通过提示GPT-4将维基百科段落拆解为命题列表。我们提出该训练数据以探索以命题作为检索单元的概念。命题的定义如下: 1. 每个命题对应文本中一处独立的语义单元,所有命题的组合可完整表达整个文本的语义。 2. 命题应具备**极小性**,即无法进一步拆分为独立命题。 3. 命题应**情境化且自包含**([Choi et al. 2021](https://aclanthology.org/2021.tacl-1.27/):命题应包含文本中所有必要的上下文信息(例如指代消解内容)以确保其语义可被独立解读。 可查阅论文获取更多细节。 ## 数据集结构 以下为该数据集的结构详情: * `sources`字段对应一段维基百科段落,其格式固定为"标题:{标题}。章节:{章节}。{内容}"。其中标题不能为空,但章节可为空。 * `targets`字段为JSON格式字符串的命题列表。 示例: { "sources": "Title: Leaning Tower of Pisa. Section: . Prior to restoration work performed between 1990 and 2001, the tower leaned at an angle of 5.5 degrees, but the tower now leans at about 3.99 degrees. This means the top of the Leaning Tower of Pisa is displaced horizontally 3.9 meters (12 ft 10 in) from the center." "targets": "["Prior to restoration work performed between 1990 and 2001, the Leaning Tower of Pisa leaned at an angle of 5.5 degrees.", "The Leaning Tower of Pisa now leans at about 3.99 degrees.", "The top of the Leaning Tower of Pisa is displaced horizontally 3.9 meters (12 ft 10 in) from the center."]" } ## 引用
数据集概述
该数据集是模型Propositionizer-wiki的训练数据,用于探索将命题作为检索单元的概念。数据集通过提示GPT-4将维基百科段落分解为一系列命题。命题的定义如下:
- 每个命题应对应文本中一个独立的意义片段,所有命题的组合应代表整个文本的语义。
- 命题应是最小的,即不能再进一步分割为独立的命题。
- 命题应是情境化且自包含的,应包含文本中解释其意义所需的所有必要情境(例如指代)。
数据集结构
数据集结构如下:
sources表示一个维基百科段落,格式为"Title: {title}. Section: {section}. {content}"。标题不会为空,但章节可以为空。targets是一个JSON格式的字符串,表示一系列命题。
示例: json { "sources": "Title: Leaning Tower of Pisa. Section: . Prior to restoration work performed between 1990 and 2001, the tower leaned at an angle of 5.5 degrees, but the tower now leans at about 3.99 degrees. This means the top of the Leaning Tower of Pisa is displaced horizontally 3.9 meters (12 ft 10 in) from the center.", "targets": "["Prior to restoration work performed between 1990 and 2001, the Leaning Tower of Pisa leaned at an angle of 5.5 degrees.", "The Leaning Tower of Pisa now leans at about 3.99 degrees.", "The top of the Leaning Tower of Pisa is displaced horizontally 3.9 meters (12 ft 10 in) from the center."]" }




