Development of the methods of Exploratory Data Analysis for Japanese linguistics

本研究的目的如下：1.研究目的。通过重新审视过去许多统计研究的结果和知识，验证“探索性数据分析(EDA)”在日本语言学中的有效性。清楚地展示EDA的哪些技术可以用于调查和研究什么样的日本语言。为了对EDA在日本语言学中的具体方向进行描述，我们选择了一些过去的统计日语研究，并用EDA对其进行了分析。此外，我们建立了初中教科书或报纸专栏的语料库，并用EDA对其进行了分析。在此基础上，我们阐明了EDA对日语语言学的有效性和必要性，并为日语语言学的EDA方向制定了指南。在这本指南中，介绍了被认为在日语语言学中有效的十种EDA技术，并用具体的应用实例进行了说明。这10种技术是：茎和叶显示、数字汇总和平行框图、数据平滑、数据重新表达、抗性线、游动框图、根图、相对于已识别的分布分析的双向分析(中值抛光)和Logit变换。人们认为，使用这些EDA技术的日语语言学可以发现隐藏在日语各种数据中的新的字符和结构。

The purposes of this research were as follows.1. To verify that "Exploratory Data Analysis (EDA)" is effective in Japanese linguistics by reexamining the result and knowledge of many past statistical researches.2. To show clearly which techniques of EDA can use for investigation and research of what kind of Japanese linguistics.3. To make the guide in which the concrete directions of EDA in Japanese linguistics were described.For these purposes, we chose some past statistical Japanese researches, and analyzed them by EDA. Moreover, we created the corpora of junior high school textbooks or newspaper columns, and analyzed them by EDA. Based on these analyses, we clarified the validity and the necessity for EDA to Japanese linguistics, and drew up the guide of the directions of EDA for Japanese linguistics. In this guide, ten sorts of techniques of EDA considered to be effective in Japanese linguistics are introduced and explained with the concrete application examples to research. The ten sorts of techniques are as follows; stem and leaf display, numerical summary and parallel boxplots, data smoothing, re-expression of data, resistant line, wandering boxplot, rootogram, two-way analysis (median polish), relative to an identified distribution analysis, and logit transformation. It is thought that the Japanese linguistics using these techniques of EDA makes it possible to discover new character and structures hidden in various data of the Japanese language.