English to Persian Transliteration

Persian is an Indo-European language written using Arabic script, and is an official language of Iran, Afghanistan, and Tajikistan. Transliteration of Persian to English—that is, the character-bycharacter mapping of a Persian word that is not readily available in a bilingual dictionary—is an unstudied problem. In this paper we make three novel contributions. First, we present performance comparisons of existing grapheme-based transliteration methods on English to Persian. Second, we discuss the difficulties in establishing a corpus for studying transliteration. Finally, we introduce a new model of Persian that takes into account the habit of shortening, or even omitting, runs of English vowels. This trait makes transliteration of Persian particularly difficult for phonetic based methods. This new model outperforms the existing grapheme based methods on Persian, exhibiting a 24% relative increase in transliteration accuracy measured using the top-5 criteria.