Modified Dijkstra-like search algorithm for English to Arabic machine translation system

Modified Dijkstra Like Search Algorithm for English to Arabic Machine Translation System Ahmed Hatem, Amin Nassar Elect. & Comm. Eng. Dept. Faculty of Engineering. Cairo University, Giza, Egypt amhatem@gmail.com, aminassar@maktoob.com Abstract. In this paper we are introducing a modified Dijkstra’s shortest path algorithm used to detect the target language phrases. We list the indexes of the source sentence’s words which were found in the target language corpus and create a directed graph to detect the phrases that form a shortest path walk in the graph. This method is used in a hybrid English to Arabic MT system. The system combines between rule based and example based machine translation techniques. The system uses an English/Arabic dictionary, a stemmer, search and Arabic corpus without parallel corpus. The system was examined and was found that results were promising to be used for domain specific and scarce resources translation. Keywords: English to Arabic hybrid Machine translation. Directed Graph Decoder. 1 Introduction Machine translation (MT) systems are divided to Corpus based Machine Translation systems (CBMT) and Rule Based Machine Translation systems (RBMT). Both types need a lot of development effort and time to create a working system. Dologlou et al., introduced a monolingual MT system (METIS-I) that can be produced with less effort. The system used a tagged and lemmatized target language (TL) corpus without the Source Language (SL) corpus. The SL corpus was replaced by a bilingual lexicon [2]. The METIS-I system was then adapted by the European consortium which were formed of the “Institute for Language and Speech Processing (ILSP) in Athens”, “the Universitat Pompeu Fabra in Barcelona”, “the Institute of Applied Information Sciences (IAI) in Saarbrucken” and “the Centre for Computational Linguistics (CCL) of the K.U.Leuven” to create the METIS-II project. Vandeghinste et al., described the METIS-II project as a hybrid solution to provide MT systems for the languages with little resources. The consortium built separate systems that use the Dutch, Greek, German and Spanish languages as the SL and English as the TL corpus [3]. In this paper we are describing our hybrid English to Arabic MT system that is similar to the METIS-II system described by Dirix et al., [4]. 12th EAMT conference, 22-23 September 2008, Hamburg, Germany 66 2 Arabic language challenges Arabic language is a challenging language for MT systems development. The absence of short vowels “diacritics” that appears with the Arabic letters to disambiguate similar word forms is one of the Arabic writing challenges [5]. The Arabic language writing common mistakes that are generated from the letters’ written form on the keyboard also introduces another challenge. This challenge was solved by word normalization [6], [7]. The strong structure, high derivational nature of the Arabic language and the ability to add a large number of affixes to each word are a morphological challenge. Gender, number (single, double, plural), grammatical case and linked pronouns to the word itself are another grammatical challenge of the Arabic language [5]. The flexible order of words in the Arabic language sentence, the ambiguity …

Content truncated to 500 words