A dependency parser of Korean based on connectionist/symbolic techniques

Abstract

This paper suggests a hybrid method for dependency parser of Korean which integrates both connectionist and symbolic techniques. First, the dependency parser tries to identify syntactic dependencies between individual words, specifying the grammatical role of dependency in each case. In most normal cases, a symbolic approach can work well, but a connectionist approach is also used for handling the cases where case-markers are omitted in a sentence. All the plausible dependencies are kept in a Syntactic Dependency Graph (SDG). Then, structural disambiguation is carried out by a constraint satisfaction process, in which various constraints on dependency structure are propagated through SDG to filter out inconsistent dependencies. For an efficient control mechanism, a forward checking algorithm is used together with both variable-ordering and value-ordering heuristics. Finally, the experiment showed very promising results with parsing accuracy of above 90% for sentences of less than 10 Korean words.

This research was supported by KOSEF(Korea Science and Engineering Foundation), Korea-China cooperative project (93.5–95.5)