Knowlaedge information analysis of Historical Document Understanding

在这个研究项目中，我们研究了以下四个研究课题：(1)构建历史文献理解知识；(2)利用我们的历史字符OCR词典研究历史字符识别；(3)开发一个历史字符识别词典；(4)开发一个计算机化历史字符词典原型系统。(1)我们开发了一个历史语料库，分析了包括1300份文件和243000个字符的许多贷款合同。从文档中提取公共表达式和频繁表达式。利用这些数据，我们利用n-gram方法开发了一个历史文献分析支持系统的原型。通过可用性测试，统计证实了该界面的有效性。(2)开发了一些基于神经网络的历史汉字识别方法。(3)除(1)中所述的数据库外，我们还制作了另一个《Kuzushi-ji Kaidoku Jiten》的历史字符数据库，该数据库由《Tokyoudo Syoten》出版，是标准词典之一。这本词典有25000个字符。(4)我们做了一些基础研究，制作了一个便携式电子词典的原型系统，可以根据语音、形状和笔画顺序进行检索。

In this research project, we studied the following four research topics : (1) Structuring the knowledge of Historical Document Understanding, (2) Study of the Historical Chara cter Recognition using our Historical Character OCR dictionary, (3) Developing a Dictionary for Historical Character Recognition, and (4) Developing a proto-type system of a computer ized Historical Character Dictionary (1) We developed a Historical Corpus Database analyzing many loan contracts consists of 1, 300 documents and 243, 000 characters. The common and frequent expressions are extracted from the documents. By using the data, we developed a proto-type of Historical Document Analysis Supporting System using n-gram method. Through an usability test, efficiency of the interface is confirmed statistically. (2) We developed some Historical Character Recognition methods using neural network. (3) Besides the database described in (1), we made another Historical Character Database of "Kuzushi-ji Kaidoku Jiten" published from Tokyoudo Syoten, which is one of the standard dictionary. The dictionary includes 25, 000 characters. (4) We have done some basic studies and have made a proto-type system of portable electrical dictionary, which could be searched by pronunciation, shape, and stroke order.