Optimizing Rules in English to Finnish Machine Translation

Rule-based machine translation requires several sets of rules in various phases of procession. The analysis of source language requires rule sets for morphological disambiguation and syntactic mapping. In this report we exclude these rules and we concentrate on rules, which produce the correct translation in target language. Three types of rule files are needed, (a) rules for isolating MWEs. (b) rules for semantic disambiguation, and (c) rules for adding various types of tags, which will then be converted into surface form clitics and added to the word. Because the number of rules needed for translating unrestricted text tends to grow so that processing speed will be affected, it is important to find ways for downsizing the rule files. This paper will discuss some of such ways using the type (c) rules as examples.