Learning Translation Rules for a Bidirectional English-Filipino Machine Translator

Learning Translation Rules for a Bidirectional English-Filipino Machine Translator Michelle Wendy Tan, Bryan Anthony Hong, Danniel Liwanag Alcantara, Amiel Perez, and Lawrence Tan 1 2401 Taft Avenue, Manila, Philippines 1004 {mwendygt, bashx5, dlalcantara_bscs, rebirthproject, zymeth02}@yahoo.com Abstract. Filipino is a changing language that poses several challenges. Our goal is to develop a bidirectional English-Filipino Machine Translation (MT) system using a hybrid approach to learn rules from examples. The first phase was an English to Filipino MT system that required several language resources. The problem lies on its dependency over the annotated grammar which is currently unavailable for Filipino, which makes reverse translation impossible. Phase 2 addresses this limitation by using information taken from English and Filipino POS Taggers. The seed rules are generated by aligning the POS tags from English and Filipino examples, including their constraints. To perform compositionality, the system deduces the constituent labels by using the longest adjacent POS tags found in both the English and Filipino rule. The system groups together similar rules and generalizes it to encompass a wider range of unseen examples. Keywords: Rule Generation, Compositionality, Rule Generalization, Bilingual Corpus, Example- based, Rule-based, Machine Learning, Machine Translation, Natural Language Processing 1 Introduction Filipino is the national language of the Philippines. The 1987 Constitution of the Philippines stated that “The national language of the Philippines is Filipino. As it evolves, it shall be further developed and enriched on the basis of existing Philippine and other languages.” This characteristic poses problems for developers of Filipino machine translators. Our goal is to develop a bidirectional English Filipino machine translation system. Our approach is to use a hybrid of rule-based and example based paradigms by learning rules from examples. Several researches have incorporated machine learning such as those by [1], [2], which automatically generates these rules from bilingual corpus. This technique is most applicable for languages with limited resources such as Filipino. The first phase of development was an attempt to build an English to Filipino Machine Translator based on the Seeded Version Space Learning presented by [1]. Using the same approach, several issues were realized for the development of the second phase, the Filipino to English translation. 2 Phase 1: English to Filipino Translator Phase 1 focused on the development of an English to Filipino Translator. It combines rule-based and example-based paradigm. Since the goal was to develop a bidirectional MT system, the hope was to use the same architecture for the development of the reverse translation. 386 2.1 Phase 1 Learning Module Architecture Phase 1 has two modules. The training module learns rules based on examples found in the bilingual corpus. In turn, the translation module uses the rules to produce the Filipino sentence based on the English input. Refer to [3] for the system’s architecture for the learning phase. The learning module begins by reading each sentence pair in the bilingual corpus through the Sentence Tokenizer. Only the English sentence is passed to the Lexical and Morphological Analyzer. These are passed to the English Parser where it …

Content truncated to 500 words