PoC (Points of correspondence)

融合包含不同内容的句子是一项非凡的人类能力，有助于创建信息丰富且简洁的摘要。对于现代抽象摘要器来说，如此简单的人类任务仍然具有挑战性，大大限制了它们在现实世界场景中的适用性。_x000D_
_x000D_
我们提出了一个数据集，其中包含 1,599 个句子融合示例（取自 1,174 个文档），并带有细粒度的对应点注释。对应点 (PoC) 是将两个句子连接成连贯文本的连贯装置。对应点的类型由文本衔接理论描述，涵盖代词和名词引用、重复等。

Summarizing sentences with distinct content via fusion is an extraordinary human capability that enables the creation of informative and concise summaries. Yet this seemingly simple human task remains challenging for modern abstractive summarizers, greatly limiting their applicability in real-world scenarios.

We present a dataset containing 1,599 sentence fusion examples (derived from 1,174 documents) with fine-grained Point of Correspondence (PoC) annotations. A Point of Correspondence (PoC) is a cohesive device that connects two sentences into a coherent text. The types of PoCs are described by text cohesion theory, covering pronominal and nominal reference, repetition, and more.

搜集汇总

背景与挑战

背景概述

PoC数据集专注于句子融合任务，包含1,599个带有细粒度对应点注释的示例，对应点类型基于文本衔接理论。该数据集由Adobe Research和中佛罗里达大学于2020年发布，旨在提升抽象摘要器的现实适用性。