الفهرس Only 14 pages are availabe for public view
 |  | from | 116 |  |  |
| | | from | 116 | | |
Abstract
Human translators and machine translation systems are often faced with the task of transliterating ph”ases’like person names and locations. Transliteration is the process of replacing words in a source language with their approximate phonetic or spelling equivalents in a target language.
Transliterating names between languages that use similar alphabets and sound system is very simple procedure. In such case, a simple table look-up procedure may be used. However, the transliteration becomes more difficult when transliteration between languages with different sound and writing systems. Transliterating names from Arabic to English is a difficult task. This pr”cess’is referred to Romanization. For instance, vowels in Arabic come in two ’Inds, long and short. Short vowels are rarely written In Arabic text. There is no one-to-one correspondence between Arabic sounds and English sounds. For example, English letters ”P” and ”B” are both mapped Into Arabic letter “”. Arabic letter ”¿”, ”A” are mapped”into English “H”.
The aim of the thesis Is to develop an Intelligent Bi- directional transliteration system English/Arabic and Arabic/English. The scope of this study, focus on the process of per’on names.The thesis starts with a literature review of the different transliteration practices and standards. These practices and standards are scattered in different sources: papers, books and websites. This Is followed by the proposed techniques for the bidirectional transliteration system. These techniques are proved to be superior to existing methods.
The developd system is three tiers.’Graphical User Interface. Logic layer and a back-end that differs according to the transliteration direction. For the English/Arabic subsystem, ”he back-end consists of a database and an Artificial Neural Network (ANN). W’Ile for the Arabic/English subsystem, the back-end consists of a database and a lookup-table for Romanization rules.
The developd English/Arabic subsystem uses ART-1 to recognize Arabic names that are transliterated in different romanization ways differently by clustering them. Our evaluation consists of two sets of data consisting of 592 English/Arabic pair (3’0 unique names) person names. The training set consists of 85% o” the data and the testing set consists of the rest. The performance Is 84.33% on training set and 81.82” on testing set.
The developd Arabic/English subsystem ensures that there is no more than one diacritic mark on the same Arab’c letter and then It applies the proposed Romanization rules. romanization rules are included for each Arabic letter, Non-Arabic, Maghribi letters, vowels, d”phth’ngs and diacritic marks.