PERSIS: A Natural-Language Analyzer for Persian

概要

論文の詳細を見る
A natural language anlyzing system, PERSIS, is described. This system takes Persian texts as input and produces dependency networks which represent their meanings on a certain level of detail. Parsing is based on a model of grammar implemented by more than 850 syntactic production rules which employ 42 different structural descriptors. For the representation of concepts encoded in the language, PERSIS utilizes, depending on the structural descriptors, 17 prototypes with up to 10 syntactic and semantic attributes. These attributes are filled under Control of the parser by interatively applying feature-integration rules to simpler attribute lists with the primary attributes of words being obtained from the dictionary. At each parsing step, feasibility is verified through feature-checks made on the extracted attributes. Finally, when the parsing is finished, dependency networks for the input sentences and phrases are decoded from the final attributes by the recursive calling of 17 feature-interpreting routinesassociated with the attribute prototypes. Experimental results and examples throughout the paper illustrate capabilities of the system in handling syntax and semantics. PERSIS, the first analyzer developed for Persian, is written in LISP and machine translation is hoped to be among its many applications.
一般社団法人情報処理学会の論文
1986-03-15