TermExtract: Accuracy of Compound Noun Detection in Japanese

摘要

Term recognition in the Japanese language is known as one of the challenging problem in natural language processing and information retrieval. We often use morphological analyzers to process Japanese documents. These tools usually do not recognize compound nouns. These nouns are combinations of single nouns expressing different meaning compared to basic nouns. Morphological analyzers usually separate compound nouns into single nouns. Therefore reconstructing compound nouns is essential to preserve text semantics. There is a tool called TermExtract to do the aforementioned reconstruction. In this study we evaluate its accuracy. To identify terms created by TermExtract, online resources are utilized. They are the ALC online dictionary, Wiki-pedia and Google phrase search service. Experiments are conducted with abstracts of scientific documents from the NTCIR-1 collection. We found that TermExtract is able to reconstruct 36.23% of all compound nouns in the corpus. Most of these nouns belong to scientific terminology.