This thesis discusses a method of reducing syntactic ambiguities and dictionary size in machine translation by recognizing simple compound words before syntactic analysis phase. The recognition of the compound words is done by using rules, patterns, or a dictionary according to the kind of the compound word. Our experiment reports that this method which uses fewer entries in the dictionary can be at least as good as that depends on a complete compound word dictionary. This system is composed of five routines : (1) morphological analysis for simple words, (2) compound word recognition by dictionary look-up, (3) compound word recognition by applying rules, (4) disambiguation of sentences, and (5) compound word recognition by pattern matching. The system gets an english sentence and produces word lists which include recognized compound words. These results are then used as inputs to syntactic analysis phase.
