English to Korean Statistical Transliteration for Information Retrieval
Jae Sung Lee · 2008
In Korean technical documents, many English words are transliterated into Korean in various ways. Most of these words are technical terms and proper nouns that are frequently used as query terms in information retrieval systems. As the communication with foreigners increases, an automatic transliteration system is needed to find the various transliterations for the cross lingual information systems, especially for the proper nouns and technical terms which are not registered in the dictionary. In this paper, we present a language independent Statistical Transliteration Model (STM) that learns rules automatically from word-aligned pairs in order to generate transliteration variations. For the transliteration from English to Korean, we compared two methods based on STM: the pivot method and the direct method. In the pivot method, the transliteration is done in two steps: converting English words into pronunciation symbols by using the STM and then converting these symbols into Korean words by using the Korean standard conversion rule. In the direct method, English words are directly converted to Korean words by using the STM without intermediate steps. After comparing the performance of the two methods, we propose a hybrid method that is more effective to generate various transliterations and consequently to retrieve more relevant documents.