Will It Still Be True Tomorrow? Multilingual Evergreen Question Classification to Improve Trustworthy QA

TL;DR

This paper introduces EverGreenQA, a multilingual dataset for classifying questions as evergreen or mutable, and demonstrates how this classification can improve trustworthiness and interpretability in question answering systems.

Contribution

It presents the first multilingual dataset with evergreen labels, benchmarks LLMs on question temporality, and develops a state-of-the-art classifier for practical QA improvements.

Findings

LLMs encode question temporality explicitly and implicitly.

EG-E5 classifier achieves state-of-the-art performance.

Evergreen classification enhances QA dataset filtering and model explanations.

Abstract

Large Language Models (LLMs) often hallucinate in question answering (QA) tasks. A key yet underexplored factor contributing to this is the temporality of questions -- whether they are evergreen (answers remain stable over time) or mutable (answers change). In this work, we introduce EverGreenQA, the first multilingual QA dataset with evergreen labels, supporting both evaluation and training. Using EverGreenQA, we benchmark 12 modern LLMs to assess whether they encode question temporality explicitly (via verbalized judgments) or implicitly (via uncertainty signals). We also train EG-E5, a lightweight multilingual classifier that achieves SoTA performance on this task. Finally, we demonstrate the practical utility of evergreen classification across three applications: improving self-knowledge estimation, filtering QA datasets, and explaining GPT-4o retrieval behavior.