Dependency parsing for Indonesian

Abstract

Dependency parsing has gained many focus lately for its many advantages over constituency-based. Dependency parsing uses dependency grammar. The observation which drives dependency grammar is a simple one: In a sentence, all but one word depends on other words. The one word that does not depend on any other is called the root of the sentence. A word depends on another either if it is a complement or a modifier of the latter. This paper tried to build dependency grammar for Indonesian. It presents a parsing algorithm of Covington, M. (1990) in Prolog using GULP, based on dependency grammar. It parses simple declarative sentences of Indonesian without handling subordinate and coordinate clause. GULP (Graph Unification Logic Programming) is an extension to Prolog that facilitates the implementation of unification-based grammar. Input of the parser is list of words (sentence) and the output is dependency tree. The system contains of lexicon, dependency rule and parsing algorithm. There are about 30 dependency rules that applied to this parser, some of them are adapted from Stanford Dependency Label because Indonesian has some similarities with English. Some Indonesian dependency rules have typical which is the root of the sentence is the main verb, and for verb phrase or noun phrase, the head locates in front of its dependents. The parser is tested with 20 sentences and resulted in valid dependency trees. It needs to be improved in parsing complex sentences with subordinate clauses and coordinate clauses. This limitation can be overcome by adding new rules to the dependency grammar and enriching feature structures.