AI 说明文字在重构新加坡历史图像记忆中的论述作用 = The discursive role of AI-generated captions in reconstructing the memory of historical images of Singapore

Summarise with VIRAT
Authors: Bei Zhen 张梖真 Chong
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

随着人工智能生成技术在文化机构与数码档案中的应用,图像说明文字的生产方式正由人工撰写转向算法生成。这一转变不仅改变图像信息的传达方式,也使说明文字逐渐成为引导观看与建构历史意义的重要中介。 本文以二十世纪初新加坡历史照片为研究对象,选取新加坡国家档案馆的十张影像,在“无语境提示”与“有语境提示”两种生成条件下,比较人工智能生成的四种说明风格(策展式、旅游宣传式、新闻纪实式与标语式),并借助文图学分析框架,从语气、情感取向、叙事结构与历史语境建构四个维度分析。 研究发现,AI 生成说明文字(AI-generated captions)呈现语义趋同与结构稳定的特征。无语境条件下,文本多采用中性词汇与场景描述,弱化社会冲突与历史张力,使复杂的历史情境转化为易于理解的视觉场景。有语境条件下,说明文字虽引入制度性与历史性词汇,但整体叙事仍维持低风险与去冲突的表达模式。不同风格虽具修辞差异,却共享相似的叙述骨架,优先追求可读性与稳定性。 本文指出,AI 生成说明文字通过分类逻辑与语言筛选机制重新分配历史可见性,形成“算法观看”的结构,使历史意义由人与图像之间的诠释互动转向生成机制所组织的语言结果,并揭示其对历史观看与文化记忆建构的影响。 关键词:文图学、AI 生成说明文字、算法观看、历史影像、文化记忆、语义治理 With the application of AI-generated technologies in cultural institutions and digital archives, the production of image captions is shifting from manual writing to algorithmic generation. This transformation not only changes how visual information is conveyed but also makes captions an important medium that guides viewing and constructs historical meaning. This study examines historical photographs of early twentieth-century Singapore, selecting ten images from the National Archives of Singapore (NAS). Under two prompting conditions—non-contextualised and contextualised—it compares four styles of AI-generated captions: curatorial, tourism promotional, reportage, and slogan. Using a Text–Image Studies framework, the analysis focuses on four dimensions: tone, affective orientation, narrative structure, and historical contextualisation. The findings show that AI-generated captions display semantic convergence and structural stability. Under non-contextualised conditions, captions rely on neutral vocabulary and scene-based descriptions, weakening social conflict and historical tension and transforming complex historical situations into easily understandable visual scenes. Under contextualised conditions, captions introduce institutional and historical terminology, yet the overall narrative still maintains a low-risk and deconflicted mode of expression. Although stylistic variations produce rhetorical differences, the captions share a similar narrative framework that prioritises readability and stability. The study argues that AI-generated captions redistribute historical visibility through classificatory logic and linguistic filtering, forming a structure of “algorithmic viewing”. Within this structure, historical meaning shifts from interpretive interaction between viewer and image to language outputs organised by generative mechanisms, revealing their influence on historical perception and cultural memory. Keywords: Text and Image Studies, AI-generated captions, algorithmic viewing, historical images, cultural memory, semantic governance

Fields of Study & Subjects

Visual Culture and Art Theory Memory Trauma and Commemoration Photography and Visual Culture

Publication details

Journal / source
DR-NTU (Nanyang Technological University)
Publication date
1 January 2026
Language
ZH