This thesis describes a Korean syntactic analyzer in which structural ambiguities are resolved by extended preprocessing rules. Extended preprocessing rules are composed of uni-eojul rules, combination rules, grammatical relation rules, and parsing bo...
This thesis describes a Korean syntactic analyzer in which structural ambiguities are resolved by extended preprocessing rules. Extended preprocessing rules are composed of uni-eojul rules, combination rules, grammatical relation rules, and parsing boundary rules.
Uni-eojul rules are ones to modify syntactic information of an eojul, or to constraint sentence components to combine with that eojul. Combination rules are ones to combine a sequence of sentence components such as combination of main verb and auxiliary verb, idioms. Grammatical relation rules are ones not to combine specific eojuls with other sentence components when dependency relation between those eojuls can be determined certainly. And, parsing boundary rules are ones to mark partial sentence whose parsing boundary is clear, and for components inside the boundary not to combine components outside the boundary until the partial sentence inside the boundary is parsed completely. Parsing information from grammatical relation rules and parsing boundary rules are used in syntactic analysis phase though all extended preprocessing rules are applied in preprocessing phase.
Currently, there are 15 uni-eojul rules, 30 combination rules, 18 grammatical relation rules, and 3 parsing boundary rules. Also, some noun phrase ambiguities are resolved with combination possibility between two nouns and with quantified noun phrase information, and etc. Experimental result of our Korean syntactic analyzer using extended preprocessing rules shows that labeled precision is 72.44%, labeled recall is 97.95%, and f-measure is 83.29%.
,韩语毕业论文,韩语论文网站 |