VLSP 2025 Shared Task: Vietnamese Temporal Question Answering (TemporalQA)

Abstract

In 2025, the eleventh edition of the Vietnamese Language and Speech Processing workshop (VLSP 20251) launched its first shared task on Vietnamese Temporal Question Answering(TemporalQA). This task aims to benchmark systems in their ability to comprehend and performreasoning over temporal information expressed in Vietnamese. TemporalQA is composed of twosubtasks. The first, Date Arithmetic (date-arith), evaluates models on questions involving temporalcomputation, such as adding or subtracting time intervals from a reference date, requiring preciseinterpretation of temporal expressions within a given context. The second, Duration Question Answering (durationQA), assesses systems on estimating the duration of events or activities based oncontextual descriptions. By formalizing these two subtasks and releasing curated datasets, TemporalQA provides the first dedicated benchmark for temporal reasoning in Vietnamese, filling animportant resource gap and supporting future research on temporal understanding and reasoning forVietnamese NLP. The benchmark was hosted on the AIHUB platform2, where system rankings weredetermined using a private test set following a public testing phase. Results show that the leadingsystem for Subtask 1 reached an accuracy of 99%, while the best-performing system in Subtask 2achieved an F1-score of 81.89% and an Exact Match score of 47.52%.Keywords: Temporal Reasoning, Question Answering, TemporalQA, VLSP 2025, Vietnamese 10:13:44 query returned open=false