The Causal Strength Bank: A New Benchmark for Causal Strength Classification

Xiaosong Yuan, Renchu Guan, Wanli Zuo, Yijia Zhang

Abstract: Causal relation extraction is essential in the causality discovery of natural language processing. The development of causal relation extraction from the model-driven is staggering, so we resort to the data-driven method. More causal information is necessary because most current datasets only label the locations of causal entities or events, which may restrict the learning capacity of models. In this paper, we introduce a novel benchmark causal strength classification and corresponding dataset, Causal Strength Bank (CSB), consisting of a Chinese dataset (C-CSB) and an English dataset (E-CSB) which merge causal strength, causal polarity, and causal entity. To ensure credibility, we select four canonical English datasets and clean Wikipedia passages for the Chinese corpus. The corpus is then annotated and cross-checked by professional annotators in two stages, ensuring the accuracy of CSB. We evaluate various baseline methods on CSB and show that causal strength information benefits causal relation extraction, demonstrating the value of the proposed dataset. Our dataset is available at https://github.com/yuanxs21/CSB-dataset .