A Non-morphological Entity Boundary Detection Approach for Korean Text

Abstract

Even though being able to automatically annotate text with DBpedia URIs is a crucial step towards interconnecting the Web in any language, the problem of detecting and annotating DBpedia URIs within non-English text has not been widely tackled. In particular, no attempts have been made to automatically annotate DBpedia URIs within Korean text. Unlike previous Korean named entity recognition research, we tackle entity boundary detection as a separate problem. We describe an entity boundary detection approach for Korean text utilizing Support Vector Machines that does not require morphological annotations. We compare performance of this approach against several rule-based methods, including one that utilizes automatically annotated morphological annotations. Based on these results, we argue that several characteristics of the language makes entity boundary detection non-trivial enough that machine learning methods are preferable over rule-based methods, even with morphological annotations.