We present CrossSum, a large-scale cross-lingual abstractive summarization
dataset comprising 1.7 million article-summary samples in 1500+ language pairs.
We create CrossSum by aligning identical articles written in different
languages via cross-lingual retrieval from a multilingual summarization
dataset. We propose a multi-stage data sampling algorithm to effectively train
a cross-lingual summarization model capable of summarizing an article in any
target language. We also propose LaSE, a new metric for automatically
evaluating model-generated summaries and showing a strong correlation with
ROUGE. Performance on ROUGE and LaSE indicate that pretrained models fine-tuned
on CrossSum consistently outperform baseline models, even when the source and
target language pairs are linguistically distant. To the best of our knowledge,
CrossSum is the largest cross-lingual summarization dataset and the first-ever
that does not rely solely on English as the pivot language. We are releasing
the dataset, alignment and training scripts, and the models to spur future
research on cross-lingual abstractive summarization. The resources can be found
at this https URL

通过跨语言检索，提出跨语言摘要数据集 CrossSum 及其训练模型，表现出优异的 ROUGE 和 LaSE 性能，成为目前最大的跨语言摘要数据集

CrossSum: 跨越 1500 多种语言对的英语中心的跨语言生成性文本摘要

CrossSum: Beyond English-Centric Cross-Lingual Abstractive Text  Summarization for 1500+ Language Pairs

Graph Convolutional Networks (GCNs) have proved to be a most powerful
architecture in aggregating local neighborhood information for individual graph
nodes. Low-rank proximities and node features are successfully leveraged in
existing GCNs, however, attributes that graph links may carry are commonly
ignored, as almost all of these models simplify graph links into binary or
scalar values describing node connectedness. In our paper instead, links are
reverted to hypostatic relationships between entities with descriptional
attributes. We propose GCN-LASE (GCN with Link Attributes and Sampling
Estimation), a novel GCN model taking both node and link attributes as inputs.
To adequately captures the interactions between link and node attributes, their
tensor product is used as neighbor features, based on which we define several
graph kernels and further develop according architectures for LASE. Besides, to
accelerate the training process, the sum of features in entire neighborhoods
are estimated through Monte Carlo method, with novel sampling strategies
designed for LASE to minimize the estimation variance. Our experiments show
that LASE outperforms strong baselines over various graph datasets, and further
experiments corroborate the informativeness of link attributes and our model's
ability of adequately leveraging them.

我们提出了 GCN-LASE 模型，该模型通过采用节点和链属性作为输入来充分捕捉链和节点属性之间的相互作用。在此基础上，我们定义了几个图核，并开发了相应的架构。此外，通过蒙特卡罗方法估计整个邻域中元素的特征总和，进一步设计了新的采样策略以最小化 LASE 的估计方差。我们的实验证明，LASE 在各种图形数据集上优于强基线，并进一步证实链接属性的信息性以及我们的模型适当地利用它们的能力。