WISTERIA: Weak Implicit Signal-based Temporal Relation Extraction with Attention

2026-03-24

AI Summary

Existing approaches to temporal relation extraction often rely on globally salient tokens or explicit temporal markers, limiting their ability to capture event-pair-specific implicit temporal cues. This work proposes a localized and interpretable modeling framework that leverages multi-head attention combined with event-pair-conditioned Top-K pooling to focus on the most relevant contextual tokens for each event pair. For the first time, this method enables the use of arbitrary implicit linguistic signals as evidence for temporal relations. The approach achieves competitive performance on standard benchmarks such as TimeBank-Dense and MATRES, while simultaneously generating human-interpretable evidence that aligns with the underlying temporal language cues.

Technology Category

• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Reasoning under Uncertainty: Relational Probabilistic Models
• Data Mining & Knowledge Management: Mining of Spatial, Temporal or Spatio-Temporal Data

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

Abstract

Temporal Relation Extraction (TRE) requires identifying how two events or temporal expressions are related in time. Existing attention-based models often highlight globally salient tokens but overlook the pair-specific cues that actually determine the temporal relation. We propose WISTERIA (Weak Implicit Signal-based Temporal Relation Extraction with Attention), a framework that examines whether the top-K attention components conditioned on each event pair truly encode interpretable evidence for temporal classification. Unlike prior works assuming explicit markers such as before, after, or when, WISTERIA considers signals as any lexical, syntactic, or morphological element implicitly expressing temporal order. By combining multi-head attention with pair-conditioned top-K pooling, the model isolates the most informative contextual tokens for each pair. We conduct extensive experiments on TimeBank-Dense, MATRES, TDDMan, and TDDAuto, including linguistic analyses of top-K tokens. Results show that WISTERIA achieves competitive accuracy and reveals pair-level rationales aligned with temporal linguistic cues, offering a localized and interpretable view of temporal reasoning.

Problem

• Temporal Relation Extraction
• Attention Mechanism
• Implicit Temporal Signals
• Pair-specific Cues
• Interpretability

Innovation

• Temporal Relation Extraction
• Weak Implicit Signals
• Pair-Conditioned Attention
• Interpretable Reasoning
• Top-K Pooling