【摘要】
As the key technology leading the development of the Industry 4.0 Era, big data + artificial intelligence (AI) has set off revolutionary waves in many fields of research. The long-standing field of study on ancient Chinese characters is thus confronted with the new problem of applying this cutting-edge new technology.
The long history of ancient Chinese character studies supposedly dates back to the Western Han Dynasty, when Gongwang of Lu 鲁恭王 discovered many ancient documents in Confucius’ house, thus initiating a new round of academic enthusiasm in ancient Chinese characters. In later centuries, more forms of ancient inscriptions were found, including those on bronzeware, stone tablets, and oracle bones, as well as Dunhuang manuscripts, Qin and Han dynasty bamboo slips and silk scripts, and Chu bamboo manuscripts in the Warring States Period. Boosted by these discoveries and excavations, research on ancient Chinese characters has lasted for over 2000 years. For a very long time, the research methodology was predominantly evidential and textual, until Wang Guowei 王国维 proposed his “dual attestation” method, combining textual analysis with newly obtained archaeological artefacts. Wang’s new approach marked a new stage in the study of ancient Chinese characters. In modern times, however, while documents and artefacts continued to increase, the number of people proficient in ancient Chinese scripts decreased drastically. While professionals worked studiously on the interpretations of ancient characters, the discipline itself was gradually estranged from the rest of academia, although there were signs of change in the new century.
The rapid growth and popularity of computer technology since the end of the last century gave rise to a new discipline—computational linguistics—whose cross-disciplinarity kindled new interest in ancient Chinese character studies. Academics from Hong Kong and Taiwan took the lead in applying digital technologies to the study of ancient inscriptions. In mainland China, a significant breakthrough was made by scholars at the Center for the Study and Application of Chinese Character (CSACC), East China Normal University, which is listed by the Ministry of Education as a key research base in humanities and social sciences. With 20 years of devoted work, researchers at the CSACC actualized a paradigmatic shift in paleography studies, by combining grammatology and computer science, building a series of ancient scripts databases, and exploring ways to make their research results relevant to contemporary life. With the leadership of the CSACC, the World Association of Chinese Characters Studies (WACCS) and Ideographical Big Data R&D Center were established, gathering sinologists together across the world. However, due to difficulties in encoding ancient characters, their display and retrieval in virtual space remained a problem, seriously restricting the accuracy and applicability of the digitalization projects.
AI technology brought new hope to the digitation of ancient scripts. Image recognition, in particular, has developing very quickly in recent years, and the use of two-dimensional barcodes (QR codes) has become a daily practice. Chinese characters are basically two-dimensional images; as Zang Kehe 臧克和 argues, they naturally correspond to and therefore are isomorphic with the objects or concepts they signify. However, the complex systems of ancient scripts cannot be easily reduced to “two-dimensional” structures. Twenty years of hard work at the CSACC yielded no significant result, until the successful use of artificial neural network technology in image and object recognition brought new hope. The CSACC recently developed the first AI recognition tool for ancient scripts, the “Shang and Zhou Bronze Script Smart Mirror.” With the latest neural network technology, the software can recognize ancient bronze characters of the Shang and Zhou Dynasties, at an accuracy rate of 90%, expected to reach 95% soon.
This ground-breaking achievement will have a “new” impact on the discipline of ancient characters in the following three aspects.
Firstly, the “new tool” will enable new ways of applying and learning ancient scripts and thus enhance their contemporary relevance.
With the “Smart Mirror,” ancient characters could be read and interpreted within seconds and become understandable to the general public. People interested in ancient texts can access systematic and professional etymological knowledge easily. The cultural heritage long sealed therein can be gradually uncovered, and new uses could be explored in all industries, while ancient characters could enter public life and hopefully become part of contemporary culture.
Secondly, the “Smart Mirror” can launch a “new mode” of learning and studying ancient scripts and initiate a new stage of big data-supported research.
Various digital platforms, where glyphs, texts, scripts, and interpretations could be retrieved, used to work quite independently of each other and could not be efficiently connected or coordinated. Successful retrieval of information from those platforms is determined in the first place by recognition of the character to be searched. When we face an image of a character that we cannot recognize, we would not be able to retrieve any information from the databases, let alone a cross-database search. AI technologies such as machine learning cannot work to their full capacities due to the lack of a “link” among the databases. Now, the “Smart Mirror” becomes the link. With intelligent recognition of ancient scripts, the various databases can now be connected and coordinated and the whole system can be activated in a way that may mark a new era in the discipline.
Thirdly, with “AI + paleography,” a cross-disciplinary “new areas of study” can be opened up and new approaches explored in ideographic studies.
Successful experiences in intelligence recognition and big data handling regarding Shang and Zhou bronze scripts could be extended to the study of other types of ancient scripts, including oracle bone inscriptions, bamboo slip and silk scripts, and stone inscriptions, among others, and even to the study of the ancient ideographs of ethnic minorities in China as well as those overseas.
Articles in this Special Issue demonstrate this new perspective of studying ancient scripts with the help of big data and AI. As the latest results of cross-disciplinary research, of paleography and computational science, they cover a wide range of topics, including the building and application of databases, image and character recognition, the development of the “Shang and Zhou Bronze Script Smart Mirror,” intelligent recognition of oracle bone inscriptions, deep learning and bronze inscription image retrieval, and the recognition of Japanese quasi-characters. We hope with the help of cutting-edge technology, ancient scripts, along with the cultural heritage they carry, can find new life and new relevance, both for academia and for the general public.
Executive Editors Guo Rui, Liu Zhiji
CSACC, East China Normal University
人工智能与古文字研究的创新
大数据+人工智能（AI）是引领工业4.0时代发展的关键技术，在社会各领域掀起革命性的浪潮，而历史悠久的中国古文字学如何运用新技术实现顺应潮流的变革，正是古文字学发展所面临的新问题、新领域。
中国古文字研究的历史可谓源远流长，相传早在中国西汉时期，鲁恭王坏孔子宅发现了一批古文字文献，当时的很多学者就开始了专业的古文字研究，此后金石文字、甲骨文、敦煌文献、秦汉简帛、战国楚竹书等陆续发掘发现，犹如一剂剂的“强心针”支撑着传统的古文字学绵延两千年。在研究方法上一脉相承着汉学考据之法，后来王国维提炼为“二重证据法”，开辟了古文字学推动史学研究进步的新局面。进入现代社会，出土文献数量增加，精通古文字的人越来越少，古文字研究专注于一字一词的考释，与其他学科的联系日渐剥离，遂沦为“冷门绝学”。这个局面直至进入新世纪才逐渐发生改变。
上世纪末本世纪初，计算机技术飞速发展，极大刺激了计算机与语言学的对接，进而派生出一个新的学科——计算语言学。这一交叉学科的兴起牵动了古文字学界部分学者的神经，中国香港和中国台湾的学者先行实践，开发了一些古文字的数字化成果，在中国大陆则是教育部人文社会科学重点研究基地华东师范大学中国文字研究与应用中心的科研人员，经过二十年的埋头工作，把数据库技术与文字学结合起来，引领古文字学研究模式的变革，孵化出基于数据库的文字学研究方向，创建了古文字系列网络数据库，极大推动了古文字学研究成果的社会应用，使得古文字学的研究走出冷门、告别绝学，促成了“世界汉字学会”的建立，进而创建表意文字大数据研发中心，把世界上汉学研究者凝结在一起。然而古文字国际编码迟迟无法落地，使得古文字在虚拟空间的显示和检索成为无法解开的难题，严重制约了古文字数字化成果的准确性和应用性。
接下来就是人工智能技术上场，作为人工智能重要领域的图像识别，近年来取得了极大的发展，尤其是“二维码”编码识别方式的流行，让从事古文字数字化研究的学者们看到了希望。汉字体系的基本类型就属于物理空间上的二维结构，“而且天然地易于与其对应的事物实现‘对应’关联，也可以说是与物体的抽象图像存在‘同构’关系（臧克和语）。”但是古文字图形的复杂因素远远超过“二维”空间，经过二十多年的试验探索均告失败。人工神经网络技术在图像识别、物体辨识的成功验证，再一次打破了古文字识别困局。文字中心研发的全球首个古文字智能识别工具——“商周金文智能镜”，运用神经网络技术使得对商周金文字形识别准确率达到了90%以上，而且研发团队表示有望到达95%以上。
这一革命性的成果对于古文字领域的影响可以用三个“新”字来概括：
一是古文字学习运用的“新工具”，实现古文字文献的真实价值。
古文字智能识别工具可以将被大多数人视为艰涩难懂的“天书”的古文字文献变成大众能够读懂方便使用的文字；或者说，一般大众依靠该工具，可以瞬间获得古文字专家才具备的古文字知识系统。可以预期，古文字文献这种由于历史条件被长期尘封的宝贵传统文化资源，将会在社会各专业领域得到越来越多应用，而古文字本身也将在此过程中得到更好的传承。
二是创建古文字学习研究“新模式”，开启古文字大数据智能化研究大门。
此前，虽然基于基础研究以及数字化建设已经形成了整合古文字的各种信息的数字平台，但是这个平台的数据链却因为一个盲点的存在而无法系统衔接贯通：虽然人们已经可以利用古文字数字平台检索字形、文句、考释等信息，但是要获得检索的成功必须有一个前提，那就是必须预先确定要检索的是什么字。如果面对的是一个不识之字，便无从利用数据库来查找它的任何信息。而这一盲点，对于数字平台所能实现的各类古文字数据之间的数字联系而言，同样造成断裂环节。于是，机器学习之类人工智能进程便会因此面临难以逾越的障碍。因此古文字智能识别的意义，就在于它可以消除这个关键性盲点，通过字形识别来打通古文字各类数据关联对接，盘活数字化营造的古文字大数据系统，推动古文字研究大踏步迈向智能化时代。
三是构建“AI+古文字学”交叉研究新学科，开辟表意文字学习研究新局面。
模式认知的方便之处便是可以复制。“商周金文”智能识别将从商周金文大数据处理，智能识别技术两个重要的领域进行探索，其成功的经验，将会成为机器学习探索和智能化研究范式，从而推广到其他文字类型的智能识别，包括甲骨文、简帛文、石刻文、纸写文等，以及少数民族表意文字、域外表意古文字识别。
因此，我们汇集这一领域的国内外最新研究动态组成“大数据与古文字智能识别”专刊，所刊文章内容涉及：面向智能识别的古文字数据库建构理论，商周金文计算机识别的成功案例，甲骨文智能识别的实验探索，基于深度学习金文快速图像检索的模型与算法；以及现代汉字识别的重要因素探究，日语中类文字的输入问题。希望通过上述文章展现文字学在新时代的创新性发展，以启来者。
执行编辑郭瑞、刘志基
华东师范大学中国文字研究与应用中心
【关键词】
【作者简介】
