Open Information Extraction for SOV Language Based on Entity-Predicate Pair Detection

Open IE usually has been studied for English which of one of subject-verb-object(SVO) languages where a relation between two entities tends to occur in order of entity-relational phrase-entity within a sentence. However, in SOV languages, two entities occur before the relational phrase so that the subject and the relation have a long distance. The conventional methods for Open IE mostly dealing with SVO languages have difficulties of extracting relations from SOV style sentences. In this paper, we propose a new method of extracting relations from SOV languages. Our approach tries to solve long distance problems by identifying an entity-predicate pair and recognizing a relation within a predicate. Furthermore, we propose a post-processing approach using a language model, so that the system can detect more fluent and precise relations. Experimental results on Korean corpus show that the proposed approach is effective in improving the performance of relation extraction.