Chaining Event Spans for Temporal Relation Grounding

2025-06-17

Conference of the European Chapter of the Association for Computational Linguistics

AI Summary

This work addresses spurious correlation issues arising from answer overlap in Temporal Reading Comprehension (TRC) and Temporal Relation Extraction (TRE). We propose a fine-grained temporal relation modeling framework. Our method introduces: (1) a novel time-span prediction module that explicitly models start/end intervals between events; and (2) a two-stage inductive reasoning mechanism that constructs interpretable timelines by chaining multiple questions referencing the same event. The proposed Timeline Reasoning Network (TRN) jointly leverages semantic and syntactic representations with cross-question temporal chain reasoning. TRN achieves state-of-the-art performance on four benchmarks—TORQUE, TB-dense, TRC, and TRE—demonstrating significant gains over prior methods. It robustly disambiguates semantically similar but temporally inverse expressions (e.g., “before” vs. “after”), thereby enhancing both robustness and interpretability in temporal QA and relation extraction.

Technology Category

• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Reasoning under Uncertainty: Relational Probabilistic Models
• Planning, Routing, and Scheduling: Temporal Planning

Application Category

• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Accurately understanding temporal relations between events is a critical building block of diverse tasks, such as temporal reading comprehension (TRC) and relation extraction (TRE). For example in TRC, we need to understand the temporal semantic differences between the following two questions that are lexically near-identical: “What finished right before the decision?” or “What finished right after the decision?”. To discern the two questions, existing solutions have relied on answer overlaps as a proxy label to contrast similar and dissimilar questions. However, we claim that answer overlap can lead to unreliable results, due to spurious overlaps of two dissimilar questions with coincidentally identical answers. To address the issue, we propose a novel approach that elicits proper reasoning behaviors through a module for predicting time spans of events. We introduce the Timeline Reasoning Network (TRN) operating in a two-step inductive reasoning process: In the first step model initially answers each question with semantic and syntactic information. The next step chains multiple questions on the same event to predict a timeline, which is then used to ground the answers. Results on the TORQUE and TB-dense, TRC, and TRE tasks respectively, demonstrate that TRN outperforms previous methods by effectively resolving the spurious overlaps using the predicted timeline.

Problem

• Understanding temporal relations between events accurately
• Resolving unreliable results from answer overlaps in questions
• Predicting event time spans for proper reasoning behaviors

Innovation

• Predicts event time spans for reasoning
• Uses Timeline Reasoning Network (TRN)
• Chains questions to predict timelines