A grammar checker is a software that detects and corrects misused words on phrases. Such programs are now available in all major word processing systems. However, most grammar checkers show high performance only in the detection of isolated word error...
A grammar checker is a software that detects and corrects misused words on phrases. Such programs are now available in all major word processing systems. However, most grammar checkers show high performance only in the detection of isolated word errors; their utility for context dependent errors is quite low, which involve correctly spelled words used incorrectly within a particular sentence. The detection of context dependent errors is a relatively more complex process that requires grammar checkers to analyze surrounding contexts.
Korean grammar checkers typically detect context dependent errors by employing heuristic rules; these rules are appended by language experts and appended each time a new error pattern is detected. Such grammar checkers, unfortunately, are inconsistent with recall, the fraction of errors that are detected. In order to resolve this shortcoming, a new error decision rule generalization method that utilizes the existing KorLex thesaurus, the Korean version of Princeton Word, is proposed. The method extracts noun classes from KorLex and generalizes error decision rules from them using the TCM suggested by Abe and Li, and information theory based MDL. The grammar checker, in order to both detect and to correct errors, relies on confusion pairs. A confusion pair within a block of text is matched against decision rules to determine whether it constitutes an error or not. The grammar checker developed in this research produces decision rules using selectional restrictions represented by noun classes. These selectional restrictions, which are limits on the applicability of predicates to arguments, are used as decision rules to detect verb errors.
The process of producing error decision rules consists of two phases: acquisition of noun classes from the KorLex, and generalization of the error decision rules. The first step in acquiring noun classes is to extract nouns that are slot values of an argument for a given verb. The next step is to use MDL from KorLex to select, from among those nouns, noun classes as decision rules. Which noun classes can be a decision rules will depend on the types of error that exist between confusion pairs.
At run-time, the grammar checker, when it encounters one of the confusion pairs, retrieves the nouns from the surrounding texts and employs KorLex to ascertain the classes to which they belong. If noun belonging to a class built into the decision rules is found, the grammar checker displays a message indicating an error.
The proposed grammar checker’s performance was evaluated on the basis of 11 confusion pairs using recall and precision, measures that have their origins in the field of information retrieval. The precision of the proposed method, as compared with that of the typical grammar checker, was lower, but its recall was higher.
，韩语论文范文，韩语论文范文 |
