Machine Translation (MT) is the application of computers that translates texts from one natural language (source language) to another (target language). The past research dealt with problems mostly related to translating modern Arabic into English. This system is considered as the first of its kind to address the problem of translating classical Arabic into English where it involves cultures knowledge of the two languages. The work is a rule-based machine translation system and consists of three main modules, i.e., analysis, transfer and generation modules. In the transfer module phase, this research has developed and extracted the logical structure from Arabic and English to synchronize the sentences at lower level such as phrases. The generation module then combines the words and phrases to decide the appropriate meaning of them based on the situation of the sentences. A prototype was developed to prove the translation techniques that have been discovered. The performance of the system has been evaluated by comparing it with human translation. The accuracy of the results is 83.5%. These results proved the viability of this approach for Arabic-English machine translation.
PDF Abstract XML References Citation
How to cite this article
Machine translation is an application of Natural Language Processing (NLP), commonly known as MT and it is a sub-field of computational linguistics that investigates the use of computer software to translate text or speech from one natural language (source language) to another (target language) (Abu-Shquier, 2009). MT is not simply replacing words with other words in target language but it needs complex system that contains linguistic rules such as morphology, semantics and syntax. Arabic natural language processing in general is still underdeveloped (Chaudhuri and Chaudhuri, 2006), because of the complexities of Arabic (Albared et al., 2010). Moreover, tools used for other languages are not easily adaptable to Arabic due to the language complexity at both the morphological and syntactic levels (Abderrahim and Reguig, 2008). As Semitic language, Arabic has a rich derivational and inflectional morphology.
There has been much work on Arabic-English MT Most of this work are on translation of Modern Standard Arabic (MSA). However, the need for such a system is important in religious activities, Islamic education and poetry. It is also to overcome the language barriers when things related to Islam are discussed or debated.
The most important problem of the non-speakers in Arabic is how to understand everything about Islam and they do not understand classical Arabic and they needed to translate it to the language they understand. The translation of classical Arabic is a difficult task because translating the meaning from classical Arabic into English requires special traits in the translator and requires knowledge of two languages and cultures. In fact, there are several obstacles facing the translator when translation classical Arabic language to English such as-religious terms, the cultural differences between civilizations and religions, the lack of full efficiency and finding the appropriate meaning in the target language.
A researcher now has access to many systems, both commercial and research, of varying levels of performance (Pan et al., 2012). All translation systems have the same goal: translate text in one language into text in a second language. Recently, most of the researches in machine translation focus on statistical approach and building a better probabilistic model. However, this approach is inadequate for translating classic Arabic to English due to the lack of bilingual parallel corpora. Moreover, Arabic presents an interesting problem for statistical models due to its rich and complex morphology (Frikha et al., 2007; Saif and Aziz, 2011).
However, in the literature, several works have been proposed to address the translation from Arabic to other languages and vice versa. Most of these works are based on linguistic approach. Mohammed and Aziz (2011) applied transfer based approach to develop MT system to translate Arabic interrogative sentence to English in agriculture domain. In addition, the work by Shaalan et al. (2004) developed a machine translation that translates Arabic noun phrases to English. They also proposed a rule-based approach to solve these problems. Salem (2009) Also used transfer approach to translate English Noun Phrase (NP) into Arabic, The system is implemented in Prolog and the parser is written in DCG formalism. Shquier and Sembok (2008) develop rules based model to handle agreement and word-ordering problem in the context of English to Arabic MT. Salem and Nolan (2009) developed a system called UniArab to translate from Arabic to English using the Role and Reference Grammar Linguistic Model. They used the logical structure to represent Arabic sentences. They show how the characteristics of Arabic language influence the progress of MT tool. Mohammed and Aziz (2011) introduced a MT system which translates full text (abstract) written in English into Arabic.
CLASSICAL ARABIC LANGUAGE
Arabic language is one of the world major languages and it is usually divided into three distinct forms, namely, Classical Arabic (CA), Modern Standard Arabic (MSA) and Colloquial Arabic. Arabic is a language with a derivational and inflectional rich morphology (Abderrahim and Reguig, 2008; Albared et al., 2011; Al-Salem and Aziz, 2011).
Classical Arabic is considered to be the most prestigious form of Arabic for all our modern writers in the sense that they try to follow the syntactic and the grammatical norms put down by classical grammarians such as Sibawaih and others (Al-Saidat and Al-Momani, 2010).
Many important documents were revealed in Classical Arabic which is the main reason why the language has preserved its purity throughout the centuries. Arabs consider Classical Arabic as an important part of their culture throughout Islamic history and it was the language used for all religious, cultural, administrative and scholarly purposes.
|Table 1:||Examples show difference of vocabularies|
|Table 2:||Examples show difference of Rhetoric|
The main difference between Modern Standard Arabic language and Classical Arabic language lies in the vocabulary. Table 1 examples describe some of these differences. For example, the word , in the first row, is not used nowadays but it is replaced with word and also the other examples in same table show different between classical Arabic and modern standard Arabic language.
In Table 2, we can say that the Rhetoric is the art and study of the use of language with persuasive effect that is used frequently in literary techniques, poetry, etc. Through the meaning of the word which is shown strong meaning in the classical Arabic language and evolved into the word in MSA.
MATERIALS AND METHODS
The CArabicMT system is based on the transfer-based architecture with three major stages: An analysis stage, a transfer stage and a generation stage. This is done through some of translation process as according to Shirko et al. (2010).
The following summarized the main processes of the Rule based approach in the CArabicMT system:
|•||Input the source text in Arabic language (classical Arabic)|
|•||Split the sentence (source text) into multiple sentences then split each sentence for many tokens (words)|
|•||Morphological analysis: The Arabic morphological analyzer analyzes each Arabic word and extracts its features|
|•||The syntactic parser builds a syntactic parse tree. The Parser starts to determine the structure of the sentence|
|•||Lexical transfer module which looks for an equivalent English meaning of each word node in Arabic parse tree|
|•||Structure transfer where the Arabic parse tree is transferred to the corresponding English parse tree|
|•||After the transfer, the system will generate (synthesis) English sentence. The morphological generation module constructs the inflected English word; The syntactic generation module, in this step, traverses the English parse tree to generate the final structure of the English sentence|
|Fig. 1:||The Architecture of classical Arabic-English MT|
The architecture of CArabicMT system is given in Fig. 1.
Transfer-based machine translation: In this section, we will describe in more details the transfer approach and its three major stages: analysis stage, transfer stage and generation stage (Xiang et al., 2011). Moreover, we will show, given examples from classic Arabic, how they process an Arabic sentence and translate it to its English equivalent.
Analysis stage: NLP systems and MT systems require identifying words in texts in order to determine their syntactic and semantic properties (Li et al., 2009).
First of all, a classic Arabic sentence is passed through a tokenization module, in which the sentence is divided into tokens. By token, we mean the smallest syntactic unit; it can be a word, a part of a word (Wang et al., 2008).
|Fig. 2:||Architecture of the parser|
|Table 3:||Example of the morphological analyzer output|
In analysis stage the information about source language only are needed.
Secondly, an Arabic morphological analyzer is used to extract stems of the Arabic words and to identify their morphological and syntactic properties such as gender, number, tense, etc.
Table 3 show an example of the morphological analysis output given the following:
Syntactical analysis (Parsing) is the computer process of analyzing a sequence of tokens to determine its grammatical structure with respect to a given formal grammar; the parsing transforms input text into a data structure, usually a tree (Al-Saidat and Al-Momani, 2010). In the parsing, the classical Arabic sentence is represented in a tree of phrases, parse tree, where each phrase. Figure 2 shows the parse tree of an Arabic sentence.
Transfer stage: The transfer stage is a system of rules that relate words and structures in one language to words and structures of another language (Shirko et al., 2010). Transfer starts with the output of the analysis phase and ends where the phase of generation starts (Shquier and Sembok, 2008). In the extent of this research; the translation actually occurs in the transfer phase. There are two types of transfer: First, lexical transfer, second, structural transfer.
The lexical transfer converts every word in the source sentence to corresponding target language representation (English word). The syntactic transfer convert the parse tree of the Arabic sentence to its equivalent in English. In addition, the bilingual dictionary is essentially in transfer method. Figure 3 illustrates an example of lexical transfer.
The structural transfer provides the rules for converting source language parser trees into equivalent target language trees (Trujillo, 1999). The transfer rules deals with the restructuring of the parse tree and reordering of words between the source and target language. Thus, this step provided the rules for convert Arabic parse tree into equivalent English tree. Figure 4 shows an example of the transfer step.
Generation: In this step, the parsing rules of the target Language is used to produce the target sentence. The generation stage is generally divided into two parts, syntactic generation and morphological generation. Syntactic generation: the English parse tree is traversed to produce the final structure of the English sentence.
|Fig. 3:||Example of lexical transfer|
|Fig. 4:||Example of the transfer process|
|Fig. 5:||Example in generation stage|
Morphological generation that generates inflected English word in its correct form based on a set of English grammar rules.
As an example, Fig. 5 shows how the generation rules added some English words (the, of, are) to produce the final structure of the English sentence:
|Transfer||:||Days tashreeq day eating and drinking|
|Output||:||The days of tashreeq are the days of eating and drinking|
RESULTS AND DISCUSSION
In general, the rationale of this experiment is to investigate whether a machine translation system, Google and CArabicMT are sufficiently robust to be translated classical Arabic to English by comparing their output with the human translation. The test set is drawn randomly from a classic Arabic book Sahih Muslim.
The result of evaluation of translation has performed by counting out of the total score of each system for all sentence, score is a number between 0 and 10 (Roshidul et al., 2011; Jiang and Wei, 2012). Then, the score is divided it by the number of sentences multiplied by 10 to find their percentage out (Table 4).
The score is given by human expert in translation and it measures the differences between the human translation, Google and CArabicMT system. The range of scores is between 0 and 10 determines the correctness of the translation or matching translation according to the magnitude of error in the structure or the meaning of the generated English sentence. While 0 indicates an absolutely incorrect translation, 10 indicate absolutely correct (matched) translation.
The result was 49% for Google and 83.5% for CArabicMT.
|Table 4:||Experiment results|
|Table 5:||Type of problem frequencies with classical Arabic to English MT|
|Table 6:||Shows part of the result produced by this experiment|
|Sentences have been tested by Google in December 2010|
However, Table 5 represents all types of the problem returned by each of the examined systems, Google and CArabicMT and their frequencies. If we examined the first row for the Synonyms of a word will find that this type of problem frequented 36 times with Google and only 2 times with our system. Therefore, as a total this type of problem frequented 38 times with two systems. Table 6 show some examples which explain in details the evolution process.
This study has been concentrated on issues in the design and implementation of a machine translation system which translates a sentence from classical Arabic language to English based on rule-based approach. There are several reasons that make transfer based is desired by MT community (Trujillo, 1999). These reasons showed that the transfer based approach is promising. So that we can improve the quality of machine translation output and increases its usefulness in classical Arabic. However, we have faced some difficulties when building the system, the first difficulty is; most of the existing Arabic database does not contain words in classical Arabic. However, due to a difference in the sentence structure between MSA and classical Arabic, it makes the rules which developed by them are not appropriate for the classical Arabic. Features classical Arabic is short sentence, word affecting and rhetoric miracle. Therefore, the translation classical Arabic directly is miracles even for specialists, also that Scholars of Arabic language stood to explain the vocabulary and interpretation of grammatical compositions and the development of its implications. Brevity sentences and abundance of meaning make the translation of classical Arabic more complexes. Therefore, the translation classical Arabic needs to have more sophisticated analysis and to go deep to the semantic and pragmatic levels to interpret its meaning. We think that will be one of the most challenges to the Arabic computational linguists in the future. However, the translation machines of the literature language such as poems and novels still in its fancy stage even in the good studied languages such as English. In future works, there are more activities are needed to enlarge and enrich our system, so to as better handle more complicated situations. All concerned with extending our paper work to be more powerful and applicable. In what follows, we presented some Proposals: Prepare and build a Private database of religious domain that containing the words written in classical Arabic and in the religious, development of this system to deal with formation. In addition translation of classical Arabic by Semantic translation that is the process of using semantic information to aid in the translation of classical Arabic sentence to create an equivalent meaning of this sentence. Here, we just reported our modest effort to open the research in this direction for further enhancement.
- Al-Salem, B. and M.J.A. Aziz, 2011. Statistical bayesian learning for automatic Arabic text categorization. J. Comput. Sci., 7: 39-45.
- Al-Saidat, E. and I. Al-Momani, 2010. Future markers in modern standard Arabic and Jordanian Arabic: A contrastive study. Eur. J. Soc. Sci., 12: 397-408.
- Mohammed, E. and M.J.A. Aziz, 2011. English to Arabic machine translation based on reordring algorithm. J. Comput. Sci., 7: 120-128.
- Saif, A.M. and M.J.A. Aziz, 2011. An automatic collocation extraction from Arabic corpus. J. Comput. Sci., 7: 6-11.
- Shaalan, K., A. Rafea, A.A. Moneim and H. Baraka, 2004. Machine translation of English noun phrases into Arabic. Int. J. Comput. Process. Languages, 17: 121-134.
- Albared, M., N. Omar, M.J.A. Aziz and M.Z.A. Nazri, 2010. Automatic part of speech tagging for Arabic: An experiment using bigram hidden markov model. Rough Set Knowledge Technol. LNCS, 6401: 361-370.
- Albared, M., N. Omar and M.J.A. Aziz, 2011. Developing a competitive HMM Arabic POS tagger using small training corpora. Intell. Inform. Database Syst. LNCS, 6591: 288-296.
- Abu Shquier, M.M. and T.M.T. Sembok, 2008. Word agreement and ordering in English-Arabic machine translation. Proceedings of the International Symposium on Information Technology Volume 1, August 26-28, 2008, IEEE Xplore Press, Kuala Lumpur, Malaysia, pp: 1-10.
- Salem, Y. and B. Nolan, 2009. Designing an XML lexicon architecture for Arabic machine translation. Proceedings of the 2nd International Conference on Arabic Language Resources and Tools, April, 22-23, 2009, Cairo, Egypt, pp: 221-229.
- Jiang, J. and W. Wei, 2012. Automated scoring research over 40 Years: Looking back and ahead. J. Artif. Intell., 5: 56-63.
- Roshidul, H., A.R.M. Shariff, B.S. Blackmore, I.B. Aris, A.R.B. Ramli and J. Hossen, 2011. Deploying natural language with topological relations for robotics behavior. J. Applied Sci., 11: 2993-3000.