An Approach toward Register Classification of Book Samples in the Balanced Corpus of Contemporary Written Japanese

An Approach toward Register Classification of Book Samples in the Balanced Corpus of Contemporary Written Japanese Wakako Kashino ab and Manabu Okumura b aDepartment of Corpus Studies, National Institute for Japanese Language and Linguistics, 10-2 Midoricho, Tachikawa City, Tokyo, 190-8561, Japan waka@ninjal.ac.jp bPrecision and Intelligence Laboratory, Tokyo Institute of Technology, 4259 Nagatsuta, Midori-ku, Yokohama, 226-8503, Japan oku@ lr.pi.titech.ac.jp Abstract. Japanese books are usually classified into ten genres by Nippon Decimal Classification (NDC) based on their subject. However, this classification is sometimes insufficient for corpus studies which describe characteristics of the texts in the book. Here, we propose a method of classifying text samples taken from Japanese books into some registers and text types. Firstly, we discuss useful criteria to describe various characteristics of the texts and propose a two-step approach for stable annotation. We then apply our method to 161 book samples from the prerelease version of the Balanced Corpus of Contemporary Written Japanese (BCCWJ), a balanced Japanese corpus comprising 100 million words developed by National Institute for Japanese Language and Linguistics. Finally, we evaluate our method in terms of stability of annotation using kappa coefficients and correlation coefficients. Keywords: text type, register, corpus, kappa coefficient 1 Introduction As the volume and the variations of texts compiled in language corpora are greatly increasing in recent years, providing an appropriate and informative classification code is becoming more and more important. Among many classification measures that have been proposed and employed so far, register classification, or text type annotation, has attracted a particular attention of many researchers (Biber, 1988; Tambouratzis et al., 2001; Portele, 2002; Gamon 2004; Sato, Matsuyoshi and Kondoh 2008; Biber and Conrad, 2009; among others). With appropriate classification codes, users will be able to extract on a particular set of the texts according to their objectives. For example, in many cases, usage of a word tends to depend strongly on the register and/or genre, and therefore, it is important for language researchers or learners to refer to register information of the text. In addition, appropriate register information is also crucial for creating automatic text classification systems as a source of baselines. Hence, we propose a new method of text type annotation for corpus texts taken from books. The rest of this paper is organized as follows. Section 2 briefly reviews the style classification methods in the literature. Section 3 introduces the proposed method. Section 4 explains how our method is practically applied to a Japanese corpus called BCCWJ. Sections 5 and 6 give evaluation results and concluding remarks, respectively. 2 Style Classification EAGLES preliminary recommendations for text typology annotation of corpora (EAGLES, 1996) makes use of three major external (E) and two major internal (I) criteria for text classification: PACLIC 24 Proceedings 433 E.1.origin — matters concerning the origin of the text that are thought to affect its structure or content. E.2.state — matters concerning the appearance of the text, its layout and relation to non-textual matter, at the point when it is selected for the corpus. E.3.aims — …

Content truncated to 500 words