Multilingual SubEvent Relation Extraction: A Novel Dataset and Structure Induction Method

Findings of EMNLP 2022

Published December 11, 2022

Viet Dac Lai, Hieu Man, Linh Ngo, Franck Dernoncourt, Thien Huu Nguyen

Subevent Relation Extraction (SRE) is a task in Information Extraction that aims to recognize spatial and temporal containment relations be-tween event mentions in text. Recent methods have utilized pre-trained language models to represent input texts for SRE. However, a key issue in existing SRE methods is the employ-ment of sequential order of words in texts to feed into representation learning methods, thus unable to explicitly focus on important context words and their interactions to enhance repre-sentations. In this work, we introduce a new method for SRE that learns to induce effective graph structures for input texts to boost repre-sentation learning. Our method features a word alignment framework with dependency paths and optimal transport to identify important con-text words to form effective graph structures for SRE. In addition, to enable SRE research on non-English languages, we present a new multilingual SRE dataset for five typologically different languages. Extensive experiments re-veal the state-of-the-art performance for our method on different datasets and languages.