الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 192 |  |  |
| | | from | 192 | | |
Abstract
Due to the increasing availability of large amounts of machine readable textual material, a number of research groups invesigate the possibilites for example-based approach to machine translation. In example-based machine translation EBMT system, a set of sentences in the source language and their corresponding translations in the target language, are given those examples are used to translate other, similar source-language sentences into the target language. The basic premise is that, if a previously translated sentence soccurs again the same translation is likely to be correct again.
The biggest problem with the EBMT system is that, it needs huge amounts of pre-translated text to make areasonable general-purpose translator. Through the research, we used two techniques to make the examples more effective. One is aligning texts at a finer granularity level than the sentence level, called chunks. It provides more flexibility, since different combinations of chunks can be extracted using the same corpus. second technique is generalizing the examples, where matching is based on word category so the same can be used to translate many sentences.
In this search, we aim to invsetigate EBMT in translating texts from english to arabic. The research includes the following activities :
Building proceudres that gathers information, The results we have obtained show, the automatic aligned corpus can be as good as manually constructed ones.
Suggesting different approches to EBMT, testing them to investigate their applicability in arabic language.
The research proposed three engines to EMBT, the first is pure EBMT, it uses no grammer nor analysis rules. it requires no additional knowledge beyond the parallel text forming the example bbase. But, it requires huge amount of examples to give satisfactory results, because matching is based on the chunks`s words. The second engine, requires less examples because matching is based on hte chunk`s templates of the source sentence. but, it requires new resources in analyzing the english sentence and regenerating the arabic sentence. The third engine, is proposed to be the most practical EBMT engine. it optimizes the use of examples, because, it searches for all possible english chunks` templates can participate in the english sentence.
As compared to other knowledge based machine translation system, the proposed engines overcome the effort required to built the knowledge base with tens of thousands of domain concepts. also, the engines algorithms are very simple, and don`t require extensive analysis nor understanding of the input english text.