Chiba University
The National Institute for Japanese Language
The National Institute for Japanese Language
ASTEM
The University of Tokyo
National Institute of Information and Communications Technology
The National Institute for Japanese Language
In this paper, we describe the design and the implementation of an electronic dictionary for morphological analysis, UniDic, which aims particularly at application to Japanese corpus linguistics. It has been indispensable for the development of a large-scale corpus to utilize an automatic morphological analyzer on computer. The existing dictionaries for morphological analyzers, however, reveal lots of problems when used in corpus linguistics, such as unevenness in defining a unit and failure in handling allomorphs and orthographic variants. Our dictionary, in contrast, deals with the uniformity of units and the identity of indexes, which are important requirements for linguistic analysis of corpora. We adopt multi-level definition of word units, consisting of short-, middle-, and long-unit words, and structured representation of indexes, composed of lemma, word form, orthography, and pronunciation. We develop a database system that straight-forwardly implements this design of the dictionary and a friendly user-interface for dictionary builders to be capable of searching and registering entries with grasping the complex structure of the indexes. We also show how this structured representation benefits us in analyzing morphologically annotated corpora, presenting case studies that investigate the variation of word form in spoken language corpus and the variation of orthography in written language corpus.