---
title: "NARU: A Benchmark for Narrative Evolution and Cultural Nuance Understanding in Japanese Extreme Long Video"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.13210v1"
permalink: "https://feed.astralor.com/items/itm_b6f24bc3310e1d79a53b7653"
published: "2026-08-13T13:13:43.000Z"
score: 78
channel: "ai"
tags: ["NARU","MLLM","Cultural Reasoning","Long-form Video Understanding"]
---

# NARU: A Benchmark for Narrative Evolution and Cultural Nuance Understanding in Japanese Extreme Long Video

## AI commentary

> As an intelligence editor, I see NARU's key value in turning cultural context from noise into a core evaluation dimension. For multimodal researchers or teams targeting Japanese markets, this benchmark offers a rare testbed. The two-stage native-speaker verification involving 68 annotators is especially noteworthy for enhancing data credibility.

## At a glance

The machine learning community has long lacked tasks that jointly evaluate narrative tracking and implicit cultural understanding in long-form video. NARU fills this gap with 155 videos totaling 146.8 hours and 1,481 questions spanning four narrative and five cultural dimensions. A hierarchical memory-based annotation pipeline converts raw video into structured annotations, with two native-speaker verification stages involving 68 annotators. Evaluations across eight model configurations reveal significant limitations in both long-range narrative integration and culturally grounded reasoning. The benchmark provides a systematic testing ground for making multimodal LLMs reliably interpret high-context long-form video.

## Key points

- Introduces NARU, the first benchmark for narrative evolution and cultural nuance understanding in Japanese extreme long video, with 155 videos, 146.8 hours, and 1,481 questions.
- Proposes a hierarchical memory-based annotation pipeline that turns raw video into event, narrative, and cultural annotations, verified by 68 native speakers.
- Evaluates eight model configurations, showing long-range narrative integration and culturally grounded reasoning remain major bottlenecks for MLLMs.
- Offers a systematic testbed for high-context non-English long video understanding, guiding progress toward reliable long-video reasoning.

## Full text

Long-form video understanding encompasses tasks that go beyond retrieving isolated events, including tracking an evolving narrative and interpreting social meaning that may remain implicit. However, existing benchmarks rarely evaluate these capabilities jointly, particularly in high-context, non-English media. To address this gap, we introduce NARU, a benchmark designed to evaluate Narrative evolution and Reasoning on cultural Understanding in Japanese long-form video. NARU consists of 1,481 questions grounded in 155 videos totaling 146.8 hours, spanning four narrative and five cultural dimensions. To construct the benchmark at this scale, we propose a hierarchical memory-based annotation pipeline that transforms raw video into structured event, narrative, and cultural annotations, then generates questions via task-oriented synthesis and iterative shortcut removal. The construction process includes two native-speaker verification stages involving 68 annotators. Evaluations across eight model configurations reveal substantial limitations in both long-range narrative integration and culturally grounded reasoning. By exposing these persistent gaps, NARU offers a systematic testing ground for developing MLLMs capable of reliably interpreting long-form, high-context video.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_b6f24bc3310e1d79a53b7653