Make Oracle Bones Great Again：川普遇上甲骨文AI

Abstract

中文摘要

本研究的目的在於回應古文字學習門檻過高、教材缺乏互動性與趣味性的問題，嘗試運用人工智慧技術，降低學習者進入古文字世界的障礙，並探索如何以數位化與互動化的方式推動教育普及與文化傳承。研究團隊希望能透過影像辨識、語音合成與生成式內容創作，使古文字不再只是專業領域的研究對象，而能成為一般學習者易於接觸且富有吸引力的學習資源。 在研究過程與方法上，我們採用模組化設計，將系統劃分為七個步驟，涵蓋從環境建置到成果展示的完整流程。首先，建立深度學習與生成式 AI 的基礎環境，並進行 API 連結與硬體檢測，以確保系統可順利運行。接著，透過影像辨識技術，將甲骨文、金文等古文字字形與現代漢字比對，並輔以語音生成模組，提供字形演變與語義的語音導讀。為了提升趣味性，研究亦引入生成式 AI，創造古文字相關的文本、笑話與插畫，並進一步利用影音生成與合成技術，製作成數位繪本與短片。整體方法結合了深度學習影像辨識、TTS 聲音克隆、文本創作、插畫生成與影音後製，展現跨領域的整合應用能力。 研究結果顯示，本系統成功將古文字轉化為兼具「可視、可聽、可互動」的多模態學習資源，能幫助學習者更快理解字形與意義，並藉由趣味化內容提升學習動機與記憶效果。此成果不僅適用於中文母語者，也能作為僑生與外籍學習者的輔助資源，未來更可應用於教育機構、博物館及語言學習平台。整體而言，本研究不僅是一項技術實驗，更是一項結合教育、文化與科技的應用實踐，展現人工智慧於人文教育與文化傳播上的潛力。

Abstract The purpose of this study is to address the challenges of high entry barriers, lack of interactivity, and limited engagement in ancient character learning. By applying artificial intelligence technologies, the research seeks to lower the obstacles for learners entering the world of ancient characters and to explore how digitalization and interactivity can promote educational accessibility and cultural transmission. The research team aims to employ image recognition, speech synthesis, and generative content creation so that ancient characters are no longer confined to the domain of specialists but can become accessible and appealing learning resources for general learners. In terms of process and methods, the study adopts a modular design, dividing the system into seven steps that cover the complete workflow from environment setup to final presentation. First, a foundation for deep learning and generative AI is established, along with API integration and hardware testing, to ensure smooth system operation. Next, through image recognition technologies, ancient scripts such as oracle bone inscriptions and bronze inscriptions are compared with modern Chinese characters, supplemented by a speech synthesis module that provides audio narration on character evolution and meaning. To enhance engagement, the study incorporates generative AI to create texts, jokes, and illustrations related to ancient characters, and further applies video generation and editing techniques to produce digital picture books and short films. The overall methodology integrates deep learning-based image recognition, TTS voice cloning, text generation, illustration creation, and multimedia post-production, showcasing the capacity for interdisciplinary applications. The results indicate that the system successfully transforms ancient characters into multimodal learning resources that are “visual, audible, and interactive,” helping learners more quickly grasp both form and meaning while boosting motivation and retention through engaging content. These outcomes are valuable not only for native Chinese speakers but also for overseas Chinese and foreign learners, and the system holds potential for application in educational institutions, museums, and language learning platforms. Overall, this research is not only a technical experiment but also an applied practice that integrates education, culture, and technology, demonstrating the potential of artificial intelligence in the fields of humanities education and cultural dissemination.