Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Only One Relation Possible? Modeling the Ambiguity in Event Temporal Relation Extraction

Yutong Hu, Quzhe Huang, Yansong Feng

TL;DR

This work tackles the ambiguity in event temporal relation extraction by reframing Vague as a multi-label prediction problem. It introduces METRE, an encoder–classifier architecture that predicts a distribution over well-defined relations and uses an adaptive threshold to declare Vague when multiple relations are plausible or when confidence is low. A dynamic Confusion Set, built from top predicted relations and confusion mappings, guides training to reveal latent relations behind Vague, improving both well-defined relation accuracy and minority-class performance. Across TB-Dense, MATRES, and UDS-T, METRE yields consistent gains, enhances interpretability of Vague, and remains effective in low-resource settings by exploiting the information embedded in Vague.

Abstract

Event Temporal Relation Extraction (ETRE) aims to identify the temporal relationship between two events, which plays an important role in natural language understanding. Most previous works follow a single-label classification style, classifying an event pair into either a specific temporal relation (e.g., \textit{Before}, \textit{After}), or a special label \textit{Vague} when there may be multiple possible temporal relations between the pair. In our work, instead of directly making predictions on \textit{Vague}, we propose a multi-label classification solution for ETRE (METRE) to infer the possibility of each temporal relation independently, where we treat \textit{Vague} as the cases when there is more than one possible relation between two events. We design a speculation mechanism to explore the possible relations hidden behind \textit{Vague}, which enables the latent information to be used efficiently. Experiments on TB-Dense, MATRES and UDS-T show that our method can effectively utilize the \textit{Vague} instances to improve the recognition for specific temporal relations and outperforms most state-of-the-art methods.

Only One Relation Possible? Modeling the Ambiguity in Event Temporal Relation Extraction

TL;DR

This work tackles the ambiguity in event temporal relation extraction by reframing Vague as a multi-label prediction problem. It introduces METRE, an encoder–classifier architecture that predicts a distribution over well-defined relations and uses an adaptive threshold to declare Vague when multiple relations are plausible or when confidence is low. A dynamic Confusion Set, built from top predicted relations and confusion mappings, guides training to reveal latent relations behind Vague, improving both well-defined relation accuracy and minority-class performance. Across TB-Dense, MATRES, and UDS-T, METRE yields consistent gains, enhances interpretability of Vague, and remains effective in low-resource settings by exploiting the information embedded in Vague.

Abstract

Event Temporal Relation Extraction (ETRE) aims to identify the temporal relationship between two events, which plays an important role in natural language understanding. Most previous works follow a single-label classification style, classifying an event pair into either a specific temporal relation (e.g., \textit{Before}, \textit{After}), or a special label \textit{Vague} when there may be multiple possible temporal relations between the pair. In our work, instead of directly making predictions on \textit{Vague}, we propose a multi-label classification solution for ETRE (METRE) to infer the possibility of each temporal relation independently, where we treat \textit{Vague} as the cases when there is more than one possible relation between two events. We design a speculation mechanism to explore the possible relations hidden behind \textit{Vague}, which enables the latent information to be used efficiently. Experiments on TB-Dense, MATRES and UDS-T show that our method can effectively utilize the \textit{Vague} instances to improve the recognition for specific temporal relations and outperforms most state-of-the-art methods.
Paper Structure (29 sections, 8 equations, 3 figures, 9 tables)

This paper contains 29 sections, 8 equations, 3 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Our Approach
  4. Encoder Module
  5. Multi-Label Classifier
  6. Training Process
  7. Construction of Confusion Set
  8. Top2 Relations.
  9. Top1's Confusion relation.
  10. Training Objective
  11. Experimental Setup
  12. Dataset
  13. TB-Dense
  14. MATRES
  15. UDS-T
  16. ...and 14 more sections

Figures (3)

  • Figure 1: The overview of METRE. $h_{e_1,e_2}$ is the representation of the event pair $(e_1,e_2)$.
  • Figure 2: Comparison between majority class and minority class. "CS Rate" indicates the class's proportion in $CS$ as a confusion relation, and "Dataset" indicates their proportion in the dataset.
  • Figure 3: Results of few shot learning on TB-Dense.