Semi-Automatic Annotation Tool to Build Large Dependency Tree-Tagged Corpus

Semi-Automatic Annotation Tool to Build Large Dependency Tree- Tagged Corpus∗ Eun-Jin Parka, Jae-Hoon Kimb, Chang-Hyun Kima, and Young-Kill Kima a1 Dongsam-Dong, Youngdo-gu, Busan, 606-791, KOREA, b161 Gajeong-Dong, Yuseong-gu, Daejeon. 305-700, KOREA jhoon@mail.hhu.ac.kr, {ejpark,chkim,kimyk}@etri.re.kr Abstract. Corpora annotated with lots of linguistic information are required to develop robust and statistical natural language processing systems. Building such corpora, however, is an expensive, labor-intensive, and time-consuming work. To help the work, we design and implement an annotation tool for establishing a Korean dependency tree-tagged corpus. Compared with other annotation tools, our tool is characterized by the following features: independence of applications, localization of errors, powerful error checking, instant annotated information sharing, user-friendly. Using our tool, we have annotated 100,904 Korean sentences with dependency structures. The number of annotators is 33, the average annotation time is about 4 minutes per sentence, and the total period of the annotation is 5 months. We are confident that we can have accurate and consistent annotations as well as reduced labor and time. Keywords: Annotation (tagging) tool, Workbench, Error Editor, Corpus construction. 1. Introduction More recently, a large corpus annotated with linguistic information is used in natural language processing. By using this corpus, natural language processing systems have learn some linguistic phenomena automatically. Building such a corpus, however, is an expensive, labor- intensive and time-consuming work. Furthermore, maintaining consistency of a constructed corpus is difficult. Therefore, we need an annotation tool for improving annotation efficiency and maintaining consistency. To help such work, some annotation tools (Atalay, 2003; Lim, 2002; Morton, 2003; Day, 1997; Brants T. and Plaehn, 2000) have already been used. In this paper, we describe an annotation tool for building a Korean dependency tree-tagged corpus with linguistic information about the segmentation of word phrases (called eojeols in Korean), part- of-speech (POS) tags, boundaries of chunks, and dependency links and relations. We design an annotation tool so that an annotator can carefully investigate them and edit errors on them through a GUI. We also design it so that errors in low level processing like POS tagging might not be propagated to higher level processing step like parsing. Moreover, the tool is characterized by the following features; 1) It is independent of special applications like information extraction. 2) It focuses on localizing errors to modules as far as possible, such as morphological analyzers. It can make annotators find and pay attention to errors related to the modules easily. 3) It has an error checking function to make possible that errors can not be stored as it can be. 4) It promptly shares annotated information among annotators so that annotators can keep consistency with others’ annotation within a working group. 5) It has a user-friendly interface. This paper is organized as follows: In Section 2, we introduce other annotation tools for establishing corpora In Section 3, we describe the architecture of our annotation tool for ∗ Copyright 2007 by Eun-Jin Park, Jae-Hoon Kim, Chang-Hyun Kim, and Young-Kill Kim 385 building a Korean dependency tree-tagged corpus. In Section 4 and 5, …

Content truncated to 500 words