Advancing Event Causality Identification via Heuristic Semantic Consistency Inquiry Network

Anonymous

Abstract: Event Causality Identification (ECI) focuses on extracting causal relations between events in texts. Existing methods primarily utilize causal features and external knowledge to identify causality. However, such approaches fall short in two dimensions: (1) the causal features between events in a text often lack explicit clues, and (2) external knowledge may introduce bias, while specific problems require specific analyses. In light of these issues, we introduce a novel Semantic Consistency Inquiry (SemCI) to the ECI task and propose the Heuristic Semantic Consistency Discriminator (HSemCD), a model that is both straightforward and effective. HSemCD utilizes a Cloze Analyzer to facilitate a gap-filling game, aiming to help uncover the semantic dependency in the context. Subsequently, it assesses the semantic consistency between the fill-in token and the given sentence to detect the existence of causality. Through this assessment, HSemCD reveals the causal relations between events indirectly. Comprehensive experiments validate the effectiveness of HSemCD, which surpasses previous state-of-the-art methods on three widely used benchmarks.

Paper Type: long

Research Area: Semantics: Sentence-level Semantics, Textual Inference and Other areas

Contribution Types: Model analysis & interpretability, Data analysis

Languages Studied: English