Computerized Database of English Sentences and Its Statistic Analysis Pertaining to Instruction for the Development of Fundamental Reading Ability in Various Academic Fields

通过两年的研究，我们取得了以下两个主要成果：数据库的建设（1）开发了一个高速、高精度的字符识别系统。该系统是通过各种计算机程序和机器可读的字典，支持商业光学字符识别软件的缺陷建立的。我们的系统“识别”英文字符串不是从直接从图像扫描仪传输的数据，而是从通过其他软件获得的图像文件。使用该系统，我们可以获得大量的基础英语句子数据，更高的效率。(2)计算机网络系统的引入，使我们能够很容易地获得基础英语资料。(3)句子数据被正确地格式化成一个统一的数据库格式（1条记录/1句），以便于下面的计算处理。数据库的分析（1）数据库-句子可以进行“一致性处理”。同时，我们还对日本初中教材中出现的英语动词的使用模式进行了调查。(2)我们建立了一个复杂的自动标记分配程序TAGASS（TAG ASSigner），它为数据库中的每个单词-句子-标注可能的语法标记。(3)完成从标签候选者集合中仅拾取一个优选标签的程序。该程序是在概率统计算法下编写的，准确率达到80%以上。我们将在未来的研究中探讨该算法和基于语法的算法之间理想事务的可能性。

Throughout the research within two years we have obtained two main achievements as follows :Construction of the Database(1) We develop a high-speed and high-accuracy character recognition system. This system is established through various computer programs and machine-readable dictionaries which support the deficiencies of the commercial optical-character-recognition software. Our system "recognizes" English character strings not from the data transferred directly from the image-scanner, but from the image-datafiles obtained via other software. With this system, we can obtain a large amount of fundamental English sentence data with greater efficiency. (2) The introduction of the computer network system enables us to obtain quite easily the fundamental English data. (3) The sentence data are correctly formatted into a uniform databasestyle (1 record/1 sentence) suitable for the following computational processings.Analyzes of the Database(1) The "concordance processing" is possible for the database-sentences. At the same time, we investigate the pattern of usage of the English verbs which appears in the textbooks for Japanese junior-high schools. (2) We build up a complex automatic tag assignment program named TAGASS (TAG ASSigner) which annotates possible grammatical marker (s) to every word in the database-sentences. (3) The program which picks up only one preferable tag from the set of tag-candidates is completed. This program, which realizes more than 80% accuracy, is written under the probabilisticstatistic algorithm. We will investigate the possibillity of the ideal transaction between this algorithm and the grammar-based algorithm in future research.