A Machine-Aided Approach to Generating Grammar Rules from Japanese Source Text for Use in Hybrid and Rule-based Machine Translation Systems

Abstract

Many automatic machine translation systems available today use a hybrid of pure statistical translation and rule-based grammatical translations. This is largely due to the shortcomings of each individual approach, requiring a large amount of time for linguistics experts to hand-code grammar rules for a rule-based system and requiring large amounts of source text to generate accurate statistical models. By automating a portion of the rule generation process, the creation of grammar rules could be made to be faster, more efficient and less costly. By doing statistical analysis on a bilingual corpus, common grammar rules can be inferred and exported to a hybrid system. The resulting rules then provide a base grammar for the system. This helps to reduce the time needed for experts to hand-code grammar rules and make a hybrid system more effective.