Joint Detection and Coreference Resolution of Entities and Events with Document-level Context Aggregation

Abstract

Constructing knowledge graphs from unstructured text is an important task that is relevant to many domains. Most previous work focuses on extracting information from sentences or paragraphs, due to the difficulty of analyzing longer contexts. In this paper we propose a new jointly trained model that can be used for various information extraction tasks at the document level. The tasks performed in this paper are entity and event identification, typing, and coreference resolution. In order to improve entity and event extraction, we utilize contextaware representations aggregated from the detected mentions of the corresponding entities and event triggers across the entire document. By extending our system to documentlevel, we can improve our results by incorporating cross-sentence dependencies and additional contextual information that might not be available at the sentence level, which allows for more globally optimized predictions. We evaluate our system on documents from the ACE05-E + dataset and find significant improvement over the sentence-level state-of-theart on entity extraction and event detection. 1

Movie Scene Event Extraction with Graph Attention Network Based on Argument Correlation Information

“…However, as error propagation problems are present in these works, the accuracy of downstream subtasks will be impacted. Thus, more and more joint models have been proposed to solve this problem [ 18 , 19 , 20 ]. Chen et al proposed the dynamic multi-pooling CNN model to extract semantic features [ 9 ].…”

Neural Coreference Resolution for Turkish

“…Coreference resolution is the task of identifying text spans that refer to the same entities and grouping these spans into coreference chains (clusters). The task has an impact on the performance of various natural language applications, including information extraction (Kriman and Heng, 2021), question answering (Bhattacharjee et al, 2020), and text summarization (Li et al, 2021;Steinberger et al, 2007).…”

CorefPrompt: Prompt-based Event Coreference Resolution by Measuring Event Type and Argument Compatibilities

“…Early neural methods focus on obtaining trigger representations by various encoders and then manually constructing matching features (Krause et al, 2016;Nguyen et al, 2016), while recent studies integrate event compatibility into judgments using well-designed model structures (Huang et al, 2019;Lai et al, 2021;Lu and Ng, 2021a) or directly incorporate argument information into event modeling (Zeng et al, 2020;Tran et al, 2021), alleviating noise brought by wrongly extracted or empty event slots. Other work (Kriman and Ji, 2021;Xu et al, 2022) learns ECR-aware event representations through contrastive learning or multi-level modeling. However, at least two limitations exist in the above studies.…”

“…Especially for coreferential event pairs with unbalanced information, the learned embeddings may significantly differ if one event has rich arguments while the other has only a vague trigger. To alleviate this problem, Kenyon-Dean et al (2018) and Kriman and Ji (2021) constrain event modeling by attraction and repulsion losses, making coreferential events have similar representations. However, this approach still performs event modeling and coreference discrimination separately, and Xu et al (2022) find that appropriate tensor matching can achieve similar effects.…”