Constructing Corpus of Scientific Abstracts Annotated with Sentence Roles

We aim to facilitate the search for relevant academic papers according to various aspects such as the “objective” and “result”. We constructed an annotated corpus consisting of the abstracts of approximately 6300 papers written in Japanese. We labeled the sentences according to their roles such as “objective” and “result” to construct an automatic sentence role classifier in abstracts. In this paper, we describe the definition of the sentence roles (the coding frame) and report some statistical properties of the constructed annotated corpus.