ارزیابی تطبیقی کیفیت خروجی ماشین های ترجمه برخط رایگان بین عربی و فارسی بر اساس مدل DQF-MQM (مقاله علمی وزارت علوم)
درجه علمی: نشریه علمی (وزارت علوم)
آرشیو
چکیده
نیاز بشر به اجرای فرآیند ترجمه با بازدهی هر چه بیشتر موجب تلاش وی برای دست یافتن به فناوری های پیشرفته ترجمه بوده است. بخش اعظم تلاش ها در این میدان صرف رسیدن به ترجمه ماشینی یا خودکار (بدون دخالت انسان) شده است که کیفیت ترجمه انسانی را ندارد، اما دارای مزیت های دیگری مانند سرعت و دسترسی بالا و هزینه پایین است. اوج این مزایا را می توان در ماشین های ترجمه برخط رایگان دید. بعضی از این ماشین ها (یعنی گوگل، بینگ، یاندکس، رورسو، مادرن ام تی و نیوترنس) از ترجمه عربی به فارسی و برعکس پشتیبانی می کنند. هدف این پژوهش مقایسه کیفیت خروجی این ماشین های ترجمه با همدیگر و یافتن بهترین گزینه برای ترجمه خودکار بین زبان های عربی و فارسی است. برای رسیدن به این هدف، ابتدا دو پیکره کوچک عربی و فارسی هر کدام شامل 60 جمله با انواع و موضوع های تصادفی از جملات موجود در دو کتاب فرهنگ بسامدی عربی و فارسی انتشارات راتلج انتخاب شد، سپس این جملات تک به تک در ماشین های ترجمه اشاره شده وارد شد و خروجی دریافت شده با روش ارزیابی انسانی بر اساس مدل تحلیل و طبقه بندی خطای DQF-MQM مورد بررسی قرار گرفت. ماشین های ترجمه به ترتیب از بیشترین به کمترین کیفیت خروجی از این قرار بودند: گوگل، بینگ، یاندکس، مادرن ام تی، رورسو، و نیوترنس. این نتیجه مطلق و همیشگی نیست، بلکه آماری و احتمالاتی است؛ ماشین های با رتبه پایین تر بعضی جملات را بهتر از ماشین های با رتبه بالاتر ترجمه می کنند.Comparative Quality Evaluation of the Output of Free Online Translation Machines between Arabic and Persian Based on the DQF-MQM Model
Man's need to translate with more efficiency has made him endeavor to achieve advanced translation technologies. Most of the efforts in this field have been devoted to achieving machine (automatic) translation (without human intervention), which, although it does not have the quality of human translation, has other advantages such as speed and high availability and low cost. The peak of these benefits can be seen in free online translation machines. Some of these machines (i.e. Google, Bing, Yandex, Reverso, ModernMT, and NiuTrans) support Arabic to Persian translation and vice versa. The purpose of this research is to compare the quality of Arabic<>Persian translations provided by these machines with each other. In order to achieve this goal, first, two small Arabic and Persian corpuses, each containing 60 sentences with random types and topics, were selected from the sentences in the two Arabic and Persian frequency dictionaries published by Routledge, then these sentences were entered one by one into the aforementioned translation machines. and the received output was scrutinized by human evaluation method based on the DQF-MQM error classification and analysis model. The translation machines in order from highest to lowest output quality are: Google, Bing, Yandex, ModernMT, Reverso, and NiuTrans. This is not an absolute and constant result, but a statistical and probabilistic one; lower-ranked machines translate some sentences better than the higher-ranked machines. Keywords: Translation Studies, Translation Technology, Machine Translation Evaluation, Google Translate, Bing Translator, Yandex Translate, Reverso, ModernMT, NiuTrans Introduction Translation technology has been an important branch of translation studies. In 1972, at the third conference of applied linguistics, James Holmes introduced the field of translation technology as a sub-branch of the "applied" branch of the emerging interdisciplinary science of "translation studies". He divided this field into three categories: theories of translation by humans, by machines, and by both (Machine-Aided Human Translation or Human-Aided Machine Translation). Most of the efforts in the field of translation technology have been focused on making the machine able to translate without human intervention. This type of translation is called “machine translation”. Machine translation will not be able to beat professional human translation in the field of quality, but it has other advantages such as high speed, low cost and easy access. The pinnacle of convenient access and low cost for translation services can be seen in free online translation machines. They can be accessed and used for free through any system with a browser and connection to the Internet; Some also have a specific smartphone application that provides additional features such as offline translation. The evaluation of the phenomenon of machine translation generally includes many topics; Different aspects of it can be examined in different ways in response to the different needs of the people involved (including: end user, developer, and investor). Our focus in this article is on evaluating the quality of the output or product of the translation machines. The questions of the research are: 1- Which free online translation machine do produce Arabic to Persian translation with better quality? 2- Which free online translation machine do produce Persian to Arabic translation with better quality? A brief and widely used definition of “translated text quality” is as follows: “A quality translation demonstrates accuracy and fluency required for the audience and purpose and complies with all other specifications negotiated between the requester and provider, taking into account end-user needs” Literature Review Several scientific studies have dealt with the subject of comparative evaluation of machine translation for Arabic-English or Persian-English language pairs, but no research in this field has been published for Arabic-Persian language pairs. These researches have generally selected a test suite first, then translated it by several translation machines and studied the output using one or more special methods of machine translation evaluation. Here we present the summary of most recent researches. Ben Milad (2022), Almahasees (2020) and Al-Shalabi (2017) tested several machine translations between Arabic and English with different methods and all concluded that Google Translate produces better quality translations, just Abu-Ayyash (2017) concluded that Google Translate and Bing Translator produce similar quality outputs. Research Methodology There are various methods to evaluate machine translation quality. They are divided into two main subcategories of human and automatic evaluation. In this research we use a standard and up to date human evaluation model called DQF-MQM, and especially a subset of it that is appropriate for machine translation quality evaluation and is as follows: Four high-level error types of Accuracy, Fluency, Locale Convention, and Terminology. Accuracy type is further subdivided into four granular error types of Addition, Omission, Mistranslation, and Untranslated. Fluency type is further subdivided into three granular types of: Grammatical, Grammatical Register and Spelling. We used the excel template on the formal website of DQF-MQM to evaluate 60 Arabic-Persian and 60 Persian-Arabic translated sentences done by 6 online free translation machines that support Arabic<>Persian translation. The sentences were selected from two Arabic and Persian Frequency dictionaries and had random modes, genres and subjects. Conclusion