Rule extraction applied in language translation (REAL) translation

The most common approaches in Machine Translation are the rule-based and example-based approaches. The rule-based approach yields high quality results but it relies predetermined linguistic resources, which requires much human labor (Bond, et. Al., 1997). While the example-based approach, although an effective paradigm by itself, can only operate on domain-specific languages, and is highly data dependent (Bond, et. Al., 1997). Thus, an English-Filipino MT system, TWiRL (Ang, et. Al., 2005), was developed. TWiRL used the rule-based approach with an integration of machine learning of rules to allow flexibility in translation. However, the system itself contains limitations, most notably, it translates only a subset of the English language. This research recited some of the limitation of TWiRL. Also, an there is little exploration of rule-based with rule learning approaches in MT with Filipino as source and English as target languages, this research focused on translating Filipino texts to English. As a result, a system that is able to learn transfer rules by analyzing learning corpora, and use the rules to translate English to Filipino texts and vice versa was constructed. However, development of more accurate linguistic resources such as POS taggers and morphological analyzers are recommended by this research