A survey on extraction of causal relations from natural language text

Abstract As an essential component of human cognition, cause–effect relations appear frequently in text, and curating from text helps building causal networks for predictive tasks. Existing causality extraction techniques include knowledge-based, statistical machine learning (ML)-based, deep learning-based approaches. Each method has its advantages weaknesses. For example, knowledge-based methods are understandable but require extensive manual domain knowledge have poor cross-domain applicability. Statistical more automated because natural language processing (NLP) toolkits. However, feature engineering is labor-intensive, toolkits may lead to error propagation. In the past few years, attract substantial attention NLP researchers powerful representation ability rapid increase computational resources. Their limitations high costs a lack adequate annotated training data. this paper, we conduct comprehensive survey extraction. We initially introduce primary forms existing extraction: explicit intra-sentential causality, implicit inter-sentential causality. Next, list benchmark datasets modeling assessment relation Then, present structured overview three with their representative systems. Lastly, highlight open challenges potential directions.