Ghina Khamis and Nour Al Oneis

Evaluation of Machine Translation Tools and Human Translation in Economic Translation Ghina Khamis and Nour Al Oneis Instructor: Dr Ghada Awada Computer-Assisted Translation TRNS301 Table of Contents : Abstract……………………………………………………………1 1- Commentary………………………………………….….....2 1.1 -Test Texts……………………………………….……..2 1.2-Evaluation……………………………………………....2 1.2.1- Semantics……………………………………..2 1.2.2- Morphology……………………………….….3 1.2.3- Syntax………………………………………...4 1.2.4- Pragmatics…………………………………....5 1.2.5- Comprehension…………………………….....6 2-Recommendations…………………………………………...7 3-Conclusion……………………………………………….......8 Terminology………………………………………………………9 References………………………………………………………..13 Abstract : In their evaluation of Human Translation and Machine Translation, Nour Al Oneis and Ghina Khamis (2013) studied and compared the differences between human’s and machine’s translation, defining the nature of translation and standards and criteria on which it is based, studying semantics, syntax, pragmatics, morphology and comprehension. This study aims to compare the effectiveness of the popular machine translation system (Google Translate) used to translate English sentences into Arabic relative to the effectiveness of English to Arabic human translation. This study will show that the Human Translation is better than Google machine translation system in terms of precision of translation from English to Arabic. This paper is set out to study the accuracy of this believe. 1-Commentary: 1.1 Test texts We performed the evaluation on two economical texts of the writer Alan Blanchette, the first entitled “Tax Avoidance” and the second “Money Laundering” , both published in 2010 and which has been reserved for testing. The texts have 504-636 words. All examples from the two texts are illustrated in that evaluation. 1.2- Evaluation: 1.2.1- Semantics: Semantic roles can determine how to properly express information in another language. Regarding the study of the meaning of linguistic expressions, it’s noteworthy to mention that many expressions and words differ in meaning between the human translation and the machine translation. Example: English العربية Human Machine catch مسك قبض Involves تشمل تتضمن wire transfers تحويلات برقية تحويلت سلكية changing the money's currency تغيير وضع العملة النقدية تغيير عملة النقود Layering involves sending the money التصنيف يتضمن ارسال الاموال التصفيف ينطوي على إرسال الأموال make it difficult to follow مما يجعل صعوبة في متابعته تجعل من الصعب متابعة But there were also similarities in the meaning in the two Translations, and it marked a kind of accuracy, and an example of this: Similarities English العربية Global financial markets الاسواق المالية العالمية Money laundering غسيل الاموال Bank-secrecy law قوانين السرية المصرفية Single scheme مخطط واحد Authorities سلطات Criminals مجرمون Embezzlers مختلسون High-value transactions معاملات ذات قيمة عالية High-value items سلع عالية القيمة The main stream economy التيار الرئيسي للاقتصاد Bank-reporting laws قوانين الابلاغ المصرفية foil money-laundering operations. احباط عمليات غسيل الاموال Drug traffickers مهربو المخدرات Financial transactions معاملات مالية Illegal money اموال غير شرعية 1.2.2- Morphology: Concerning the morphology, machine translation was ambiguous in handling word forms, achieving Number and Gender agreement, Identifying the case ending of words, and in the proper handling of pronouns, while human translation has proved its efficiency in giving the exact verb tense number, possession, and the proper gender agreement and comparison. For example, the following sentences show Morphological issues with agglutinative, fusion and complex word structure: Source Sentence: Banks usually lend money to persons who need it, for a specified interest. Google Translate: البنوك عادة إقراض المال إلى الأشخاص الذين يحتاجون إليها، لمصلحة معينة Human Translation : تقوم البنوك باعطاء القروض عادة للأشخاص الذين يحتاجون إليها، في مصلحة معينة At this stage we have to compare the outputs of Google Translate system with the Human Translation. In the first comparison we found that there is: البنوك“ Banks" الذيwho” and “يحتاجون need” are common with human translation but “إقراض المال” and “اعطاء القروض” differs. Source Sentence: The temporary closure of international banks because of the earthquake I Japan, also dented profits. Human translation: وتسبّب الزلزال المدمر الذي ضرب اليابان بالإقفال المؤقت للبنوك الدولية ، الأمر الذي أدّى إلى انخفاض هذه الأرباح Machine translation: إغلاق مؤقت ل بنوك دولي بسبب الزلزال أنا اليابان ، تراجع أيضا الأرباح. Source Sentence: Investors remain cautiously optimistic. Machine translation: ولكن لا يزال المستثمرون متفائلون بحذر . Human translation: إلا أن المستثمرين بقيوا متفائلبن بحذر 1.2.3- Syntax: Concerning syntax, a big difference in the combination of words appears, especially in dealing with Superlative Adjectives, and arranging Nouns and Adjectives and generating Nominal Chunks, hence, the human Translation syntax study is more accurate and follows the grammatical rules: Subject –Verb and we can also notice the proper use of words such determiner, modifier, complement. In the machine Translation a majority of phrases started with the verb or the verb phrase instead of the Noun, and the structure is most likely similar to the combination of words in the Arabic language, hence, the Arabic uses the: Verb – Subject – Object combinational structure while, in the English language, its : Subject-Verb + prepositional phrase. Phrase: NG (noun group), AdjG (adjective group), AdvG (adverb group), PG (prepositional group), PossG (possessive group). -In machine translation, the phrase started with a verb then subject: قد تتكون طبقات من عدة بنوك إلى بنك التحويلات - The usage of prepositions that differs in meaning, but not in structure: ومخطط واحد عادة ما ينطوي على تحويل الأموال من خلال العديد من البلدان- - والسبب في أنه من ضروري - الذي يغسل الأموال وكيف تفعل ذلك 3- the substitution of the preposition, but the reservation of the same structure and the same meaning :مما يجعل من الممكن إيداع مجهول الأموال "القذرة So the syntactic transfer rules that map parse tree for one language into one for another are: – English to Arabic: • NP → Adj Nom Þ NP → Nom ADJ • VP → V NP Þ VP → NP V • PP → P NP Þ PP → NP P 1.2.4- Pragmatics and Comprehension: Concerning Comprehension, the translation must transfer the meaning intended, which makes the comprehension part a major one. Yet, in machine translation, it is not always the case, it produces sometimes sentences, that unfortunately mean something different from the original source text. The majority of machine translation expressions have resulted into a literal and meaningless translation of the saying. For instance, the English say “A good workman is known by his chips”; which has the Arabic meaning as was literally mistranslated into “ومن المعروف ان العامل الجيد من قبل رقائقه”, by the Google translation; which is very literal and very far from the actual meaning of the saying. Source Sentence: The rise of global financial markets makes money laundering easier than ever. Machine Translation : صعود الأسواق المالية العالمية يجعل غسل الأموال أسهل من أي وقت مضى Human Translation : ارتفاع الاسواق العالمية يجعل غسل الاموال اسهل بكثير Source Sentence: In this article, we'll learn exactly what money laundering is and why it's necessary. Machine Translation: في هذه المقالة، وسوف تتعلم بالضبط ما هو غسيل الأموال والسبب في أنه من الضروري، الذي يغسل الأموال وكيف تفعل ذلك Human Translation: في هذه المقالة, سوف تتعلم بالتحديد ما هو غسيل الاموال و ما السبب في انه ضروري Source Sentence: At this stage, the launderer inserts the dirty money into a legitimate financial institution. Machine Translation : في هذه المرحلة، وغاسلي الأموال القذرة يدرج في مؤسسة مالية مشروعة. Human Translation : في هذه المرحلة, يضع غاسلوا الاموال اموالهم في مؤسسة مالية مشروعة Source Sentence: At the integration stage, the money re-enters the mainstream economy in legitimate-looking form. Machine Translation: في مرحلة التكامل، وإعادة الأموال يدخل-التيار الرئيسي للاقتصاد في شكل المشروعة المظهر Human Translation: في مرحلة التكامل, يتم اعادة ادخال الاموال للتيار الرئيسي للاقتصاد في شكل شرعي. According to these examples, we can discern the following differences in meanings: Differences English العربية Human Machine easier than ever اسهل بكثير أسهل من أي وقت the launderer inserts the dirty money , يضع غاسلوا الاموال اموالهم غاسلي الأموال القذرة يدرج the money re-enters , يتم اعادة ادخال الاموال وإعادة الأموال يدخل Hit تلقت تعرضت in legitimate-looking form في شكل شرعي. في شكل المشروعة المظهر We have found out that the overall translation precision for Google was 31,4% , if we considered that the total of words in the two articles is approximately 600 words (504-636 words), and that 188.4 words/ expressions/ morphemes were considered as inaccurate, So the degree of accuracy of the Machine Translation from English to Arabic is around 30%. 2- Recommendations: -It is necessary to improve Machine Translation either, by developing more sophisticated methods or by imposing certain restrictions on the input in order to get a good translation that is fluent, grammatically well structured and readable in the target language. - The translation machine systems (Google, Babylon, tarjim..) should be guiding International economics and should guarantee the fidel translation of documents for businesspeople, bankers, financier and other professionals. - It’s recommended improve new developed features for Translation machines, such as automatic multilingual dictionaries, that handle more support of basic research in computational linguistics. - It is necessary to provide Global Translation service for economic translation of the different types of documents (Accounting statements, balance sheets, auditor’s repots and economis statistics), and this through developed software that are specific to this domain - A Global translation service should provide high-quality translation for any information contained in economic articles, getting it across to the target audience, by including specialized terminology and a good numbering. (statistics sheet) 3-Conclusion: To sum up, English-to-Arabic machine translation has been a challenging research issue for many of the researchers in the field of Arabic Language Processing. In this study, we have concluded that the effectiveness of Machine Translation is limited, furthermore, human translation remains the most accurate and effective in translating technical texts that are related to different domain, as the economic domain . And the human Translation will always be distinct, to an important degree, from the ways in which the major Online Translating Machines as Google Translate, babyllon, etc.. translate a source text in a certain specific domain.. The human translation hence, stays in the first raw in being indispensable for translating data.