Role Conversion Considering Its Context and Syntactic Property

Semantic Role Labeling (SRL) is to determine the relationship between predicates and their arguments in a sentence. In order to determine the semantic roles, a large amount of corpus with annotated semantic roles is required. Nowadays the most widely used semantic corpus is Proposition Bank (PropBank) which is semantically annotated over the predicate and argument structure. But the Korean version of the PropBank could not be widely used because the corpus has limitation in size and be different from its original English version in its usability. To solve these problems, we also used another semantic tagged corpus, built by Sejong Plan, which is nation-wide Korean corpus construction project. However, the task of corpus construction with semantic roles defined in PropBank and Sejong is much time-consuming and these corpora use their own role sets. They finally require a way of converting one role to other side role(s). In this paper, we propose a method for automatically converting the roles. First, we use similarity between a given noun argument word to find a new role and noun words appearing in the example sentences of candidate roles. Second, we extract suffix of the argument word and estimate closeness between the suffix and candidate roles. Finally, the predicate itself is used for selection,that is we calculate the closeness between the predicate and the candidate roles. With these, the role is decided among multiple candidate roles. In the experiment, we convert 491 arguments automatically and about 78% of them show the agreement with manually annotated arguments.