Machine Transliteration

Machine Transliteration Mohamed Abdel Fattah 1, Fuji Ren 1, 2, Shingo Kuroiwa 1 1. Faculty of Engineering, University of Tokushima 2-1 Minamijosanjima Tokushima, Japan 770-8506 2. School of Information Engineering, Beijing University of Posts & Telecommunications Beijing, 100088, China (mohafi, ren, kuroiwa) @is.tokushima-u.ac.jp Abstract. In the present study, we present different approaches for transliteration proper noun pair’s extraction from parallel corpora based on different similarity measures between the English and Romanized Arabic proper nouns under consideration. The strength of our new system is that it works well for low-frequency words. We evaluate the presented new approaches using an English- Arabic parallel corpus. Most of our results outperform previously published results in terms of precision, recall and F- Measure. 1 Introduction Recently, much research has been done on machine transliteration for many language pairs, such as English/Arabic [1] and English/Korean [2]. Most of the above approaches require a pronunciation dictionary for converting a source word into a sequence of pronunciations. However, words with unknown pronunciations may cause problems for transliteration. On the other hand, much research has focused on the study of automatic bilingual lexicon construction based on bilingual corpora. Proper names and corresponding transliterations can often be found in parallel corpora or topic-related bilingual comparable corpora. However, many methods dealt with this problem based on the frequencies of words appearing in corpora, an approach which cannot be effectively applied to low-frequency words. Fung, used different approaches to create translation pairs from parallel and comparable corpora [3]. We have exploited the pattern matching method of [3] to extract transliteration pairs from English – Arabic parallel corpus and we used it as a base line method. 2 The Arabic-English Parallel Corpus We have applied our transliteration techniques on the “Arabic English Parallel News Text Part 1”, Linguistic Data Consortium (LDC) catalog number LDC2004T18 and ISBN 1-58563-310-0. [4]. 3 Patten Matching Approach We treat the transliteration compilation problem as a pattern matching problem in [3] with little modification in the first step to decrease the computation time. In the first step of the algorithm, we did not tag the English half of the parallel text only but we also tagged the Arabic half in order to restrict the matching process on as few words as possible to decrease the computation time. We achieved accuracy and recall of 71.4% 66.5% respectively for the best matched pairs. We also achieved accuracy and recall of 73.8% and 68.2% respectively for the top three Arabic transliterations for an English proper noun respectively. We found that many mistaken transliterations resulted from insufficient data. 370 4. The Proposed English-Arabic Proper Noun Transliteration Pairs Creation Approach The system extracts all proper nouns from the English sentence using the CLAWS4 POS tagger. It also extracts all proper nouns from the associated Arabic sentence using the Buckwalter Arabic Morphological Analyzer Version 1.0. All the Arabic proper nouns are romanized using [5]. The similarity between every English and Romanized Arabic proper noun pair is measured. The English- Arabic proper noun pair which has …

Content truncated to 500 words