Deep Temporal Reasoning in Video Language Models: A Cross-Linguistic Evaluation of Action Duration and Completion through Perfect Times

TL;DR

This paper introduces the Perfect Times dataset, a multilingual benchmark for evaluating video-language models' understanding of action duration and completion, revealing current models' limitations in temporal reasoning.

Contribution

The work presents a novel quadrilingual dataset and evaluation framework that specifically tests models' grasp of temporal dynamics and perfectivity in videos, highlighting gaps in current models.

Findings

State-of-the-art models struggle with temporal reasoning in videos.

Models perform poorly on perfectivity and causality tasks.

The dataset sets a new standard for evaluating temporal understanding in VLMs.

Abstract

Human perception of events is intrinsically tied to distinguishing between completed (perfect and telic) and ongoing (durative) actions, a process mediated by both linguistic structure and visual cues. In this work, we introduce the \textbf{Perfect Times} dataset, a novel, quadrilingual (English, Italian, Russian, and Japanese) multiple-choice question-answering benchmark designed to assess video-language models (VLMs) on temporal reasoning. By pairing everyday activity videos with event completion labels and perfectivity-tailored distractors, our dataset probes whether models truly comprehend temporal dynamics or merely latch onto superficial markers. Experimental results indicate that state-of-the-art models, despite their success on text-based tasks, struggle to mirror human-like temporal and causal reasoning grounded in video. This study underscores the necessity of integrating deep multimodal cues to capture the nuances of action duration and completion within temporal and causal video dynamics, setting a new standard for evaluating and advancing temporal reasoning in VLMs.