Annotation and Quantitative Analysis of Speaker Information in Novel Conversation Sentences in Japanese

Annotation and Quantitative Analysis of Speaker Information in Novel Conversation Sentences in Japanese Makoto Yamazaki, Yumi Miyazaki, Wakako Kashino National Institute for Japanese Language and Linguistics 10-2, Midori-cho, Tachikawa City, Tokyo, Japan {yamazaki, y-miyazaki, waka}@ninjal.ac.jp Abstract This study undertook a quantitative lexicological analysis using attributed speaker information, and reports on the problems of creating standards when annotating speaker information (gender and age) of conversation sentences in novels. In this paper, we performed a comparative analysis of vocabulary use by gender and age of conversation sentences and descriptive part sentences, as well as on the differences between Japanese novels and translations of foreign novels. In addition, a comparison with other spoken language materials was made. Keywords: novels, conversation sentences，speaker information，gender, age，Balanced Corpus of Contemporary Written Japanese， characteristic words，descriptive part sentences. 1. Introduction Although conversation sentences in novels are not actual speech, they were once used for language studies. This was probably because real spoken language data was difficult to obtain, unlike today. Language researchers of course recognized that conversation sentences in novels are not actual speech. For example, Takasaki (1981: 97) stated that “conversation sentences are difficult to use as speech materials and seem to require careful handling.” Regarding conversation sentences in novels, Oishi (1987: 78) stated that “in summary, we conclude that conversation sentences are not useful as speech materials.” Oishi (1987: 78-79) argued, however, that conversation sentences in novels can be considered representative of spoken language and be used for data in studies that understand the fundamental attributes of conversation. We believe that conversation sentences in novels are effective for the investigation of the following research objectives: (1) How are conversation sentences in novels different from actual conversation sentences? Clarifying these differences will make it possible to clarify the characteristics of actual conversations and can thus promote the understanding of the diversity of spoken language. (2) A classical literary work such as The Tale of Genji and conversation sentences from a modern novel can be considered the same based on the structure of the novel. Therefore, diachronically studying conversation sentences and analyzing the changes make it possible to clarify historical changes in conversation sentences and the rhetoric used in them. (3) By using speaker information, we can pursue the usage of role language (Kinsui, 2000), because role language appears in conversation sentences in novels more often than actual spoken language. (4) Novels consist of conversation sentences and descriptive part sentences, and by analyzing the relationship between the two, we can expect that there are implications for literary studies and stylistic studies. 1 Whether a sample is a novel or not was determined by the NDC (book classification), which is meta-information of the sample. Specifically, the NDC regards texts that correspond to 9x3 (x=1 to 8) as novels. 2. Data The data used in this study were taken from novel samples1 in the Balanced Corpus of Contemporary Written Japanese (BCCWJ). The BCCWJ includes 5,138 novel samples; however, the portion used in this study consisted of 2,419 samples taken mainly from …

Content truncated to 500 words