Rule-based Indonesian Open Information Extraction

Abstract

Open Information Extraction (Open IE) is a paradigm that tries to extract as much information as possible, with less restriction on the information type to be extracted. It extracts relation tuples, in which a relation tuple consists of a relation tuple trigger and several relation arguments. Previous studies on developing Open IE systems have mainly been for English. Recently, several works have also been carried out in other languages, but there is no study on Open IE for Indonesian. In this paper, we investigate several rule-based methods for building an Open IE system for Indonesian. We use lexical and syntactic features that were obtained from an Indonesian language processing tool and compare the extraction results against the standard English Open IE systems. The experimental results for English-Indonesian parallel sentences show that the POSTag+Noun Phrase-based rules have better performance. At the same time, the dependency relation-based performance depends on the dependency parser performance, which still needs improvement since we use a small size dataset on training the parser. However, both approaches show good performance in identifying the relation tuple trigger, with the recall score being 0.96 for the POSTag+Noun Phrase-based rules and 0.6 for the POSTag+Dependency relation based-rules.