ConTempo: A Unified Temporally Contrastive Framework for Temporal Relation Extraction

TL;DR

ConTempo is a temporally contrastive learning framework that teaches temporal relation extraction (TRE) models the symmetric and antisymmetric properties of temporal relations, instead of treating each relation as an independent classification label. We embed it in a unified framework compatible with all three main branches of TRE research, achieving state-of-the-art performance on MATRES and TBD. Along the way, we identified and corrected a large number of annotation errors in the MATRES test set, after which ConTempo’s gains become even more apparent.

How to Cite

@inproceedings{niu-etal-2024-contempo,
    title = "{C}on{T}empo: A Unified Temporally Contrastive Framework for Temporal Relation Extraction",
    author = "Niu, Jingcheng  and
      Liao, Saifei  and
      Ng, Victoria  and
      De Montigny, Simon  and
      Penn, Gerald",
    editor = "Ku, Lun-Wei  and
      Martins, Andre  and
      Srikumar, Vivek",
    booktitle = "Findings of the Association for Computational Linguistics ACL 2024",
    month = aug,
    year = "2024",
    address = "Bangkok, Thailand and virtual meeting",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.findings-acl.89",
    pages = "1521--1533",
}