Applications of Mathematical Science and Information Science (Machine Translation, etc)

作为数学科学和信息科学的应用，自然语言分析和计算机翻译的各种研究工作都是由我们团队的研究人员进行的。在各个研究领域，1996年取得了以下新成果：(1)在机器翻译系统方面，从初中英语教材《皇冠I》和《皇冠III》中增加了各种词汇和语法数据，这两本教材的特点是口语表达丰富。词汇和语法数据也来自Shogakukan的朋友英日/日英词典。除了MS-DOS(16位操作系统)版本的机器翻译系统，我们在1996年开始了针对Windows 95(32位操作系统)的新版本。同时，我们从用户界面的角度以及从翻译过程中查找速度的角度重新审视了词典的基本架构。新发现的特性之一是，字典页面的最有效记录长度在MS-DOS上是512位，在Windows 95上是1024位，而直到1996年，我们一直使用256位页面。在假名-汉字转换技术和自然语言人机接口领域，提出了新的方法，并通过实验证明是有效的。在以10余名学生为被试的语音识别和音位学实验中，通过对语音波形的频谱分析，发现类比推理在口音变异中起着重要的作用。另一方面，在普遍语法理论框架下，通过设置适当的参数，可以更简单地描述古英语中非常复杂的重音分布规律。考察了不同语言祈使句中主语的位置及其限制。极简主义原理被推广到渗流问题。

As Applications of mathematical science and information science, various research works on natural language analysis and translation by use of computers are carried out by the research members of our team. In each domain of the research subjects, the following new results are obtained in 1996.(1). As for the machine translation system, various lexical and grammatical data are added from "Crown I" and "Crown III", which are junior high school English textbooks characterised by their rich colloqual expressions. Lexical and grammatical data are also collected from Shogakukan's Friend English-Japanese/Japanese-English dictionary.Besides MS-DOS (16-bit OS) version of the machine translation system, we started a new version for Windows 95 (32-bit OS) in 1996. At the same time, we reexamined the fundamental architecture of the dictionary from the user interface point of view as well as from the view point of the look-up speed dureing the translation process. One of the newly discovered feature is that the most efficient record length of a dictionary page is 512 bits for MS-DOS and 1024 bits for Windows 95, while we had used 256-bit pages until 1996.(2). In the fields of Kana-Kanji conversion technology and human robot interface by natural language, new methods are invented and proved to be efficient by experiments.(3). In experiments about speech recognition and phonology with more than 10 students as subjects, voice waves are analyzed by means od spectrograph and the important role playd by analogical inference in accent variation was noticed. On the other hand, very complicated stress distribution rules in Old English was found to be more simply described within the framework of Universal Grammar theory by setting adequate parameters.(4). The subject position and its restriction in imperative sentences in various languages was investigated. The minimalist principle was extended to the percolation.