A Parsing Method of Natural Language by Filtering Procedure

Hiroshi SAKAKI  Kazuo HASHIMOTO  Masami SUZUKI  Izuru NOGAITO  Takashi TANAKA  

IEICE TRANSACTIONS (1976-1990)   Vol.E69   No.10   pp.1114-1124
Publication Date: 1986/10/25
Online ISSN: 
Print ISSN: 0000-0000
Type of Manuscript: PAPER
Category: Artificial Intelligence

Full Text: PDF>>
Buy this Article

The parsing method which analyzes sentences into trees may be divided into two categories, the procedural method and the filtering method. The former composes a tree avoiding the occurrence of improper analysis in the course of the analysis procedure by strictly defining the order of application of transformation rules concerning tree structure. The latter obtains the proper tree by obtaining, first, many analysis results through unlimited application of existing rewriting rules and by, later, eliminating improper trees with a filter. The former includes the parsers in the ARIANE 78 at Grenoble University and also at the group led by Kyoto University. This paper proposes a parsing method using the above-mentioned filtering scheme. The method is based on extended LINGOL which is included in the latter category and uses forbidden trees and exclusive trees which respectively forbid and foster the sentence pattern contained in these trees. These trees, cooperating together, work to filter out spurious results. The paper also treats OR nodes for expression of multiple trees in a single tree, giving an algebraic means to allow coexistence of OR nodes and the above-mentioned forbidden and exclusive trees. The algorithm presented in this paper has proved to act well in practical computer experiment showing that it is extensible for practical use.