HeadlineCause: A Dataset of News Headlines for Detecting Causalities

TL;DR

HeadlineCause is a multilingual dataset of news headline pairs designed to facilitate the detection of implicit causal relations, supporting research in causal reasoning with models tested on causality detection and effects prediction.

Contribution

The paper introduces HeadlineCause, a novel dataset of over 14,000 news headline pairs in English and Russian for implicit causality detection, along with baseline models demonstrating its utility.

Findings

XLM-RoBERTa effectively detects causality in headline pairs.

GPT-2 can predict possible effects from causal headlines.

Dataset covers a wide range of causal and refutational relations.

Abstract

Detecting implicit causal relations in texts is a task that requires both common sense and world knowledge. Existing datasets are focused either on commonsense causal reasoning or explicit causal relations. In this work, we present HeadlineCause, a dataset for detecting implicit causal relations between pairs of news headlines. The dataset includes over 5000 headline pairs from English news and over 9000 headline pairs from Russian news labeled through crowdsourcing. The pairs vary from totally unrelated or belonging to the same general topic to the ones including causation and refutation relations. We also present a set of models and experiments that demonstrates the dataset validity, including a multilingual XLM-RoBERTa based model for causality detection and a GPT-2 based model for possible effects prediction.