The Efficiency of Neural Machine Translation Models in ArabicTurkish Translation Types: Example of ChatGPT
Sezer YılmazArtificial intelligence applications, which are gaining currency day by day due to their new features, in parallel with the development of technology, have attracted attention with the development of neural machine translation models, such as ChatGPT in the field of translation, especially in recent times, and have been regarded as a valuable phase and promising development in the translation world. With the development of ChatGPT, which is a neural machine translation model, “translation proofreading” has gained popularity in the translation industry and some changes have been made in translators’ job descriptions. Also, the question that has been a topic of discussion in the translation industry since the first years of the development of machine translation models has gained currency again: “Can machine translations replace translators?” The purpose of this study is to describe the quality of Arabic-Turkish translation outputs and the distance covered by artificial intelligence in the context of translation as a result of these developments. Artificial intelligence-based machine translation models, their development stages, and their most recent stage were included in the study, and then the ChatGPT Arabic-Turkish translation outputs were analyzed comparatively. In this study, the language of the source text has been determined to be Arabic and the language of the target text to be Turkish, and comparative analyses were carried out within the framework of ChatGPT translation outputs and human translations. Plain text, literary text, and technical texts, which were translated from Arabic into Turkish, were re-translated from Arabic into Turkish using ChatGPT, and the ChatGPT translation outputs were examined in the context of semantic and linguistic equivalence in general. The success rate achieved in the translation of different text types in the Arabic-Turkish language pair were evaluated.
Arapça – Türkçe Çeviri Türlerinde Nöral Makine Çeviri Modellerinin Verimliliği: ChatGPT Örneği
Sezer YılmazTeknolojinin gelişimine paralel olarak gün geçtikçe yeni özellikleriyle gündeme gelen yapay zekâ uygulamaları, özellikle son dönemlerde çeviri alanında ChatGPT gibi nöral makine çeviri modellerinin geliştirilmesiyle dikkatleri üzerine çekmeyi başarmış, bu durum çeviri dünyasında kıymetli bir aşama ve umut verici bir gelişme olarak kabul görmüştür. Nöral makine çeviri modelleri arasında yer alan ChatGPT modelinin geliştirmesiyle çeviri sektöründe “çeviri editörlüğü” popülerlik kazanmış ve çevirmenlerin iş tanımlarında bir dizi değişiklikler yapılmıştır. Öte yandan makine çeviri modellerinin geliştirildiği ilk yıllardan beri çeviri sektöründe bir tartışma konusu olarak karşımıza çıkan; “makine çevirileri çevirmenin yerini alabilir mi?” sorusu yeniden gündeme gelmiştir. Bu gelişmeler neticesinde; Arapça-Türkçe çeviri çıktılarındaki kalitenin ve yapay zekânın çeviri bağlamında katettiği mesafenin betimlenmesi çalışmanın ereğini oluşturmuştur. Çalışmada sırasıyla, yapay zekâ tabanlı makine çeviri modellerine, gelişim evrelerine ve geldiği son aşamaya yer verilmiş, akabinde Arapça – Türkçe ChatGPT çeviri çıktılarının karşılaştırmalı analizleri yapılmıştır. Kaynak metin dili Arapça, hedef metin dili Türkçe olarak belirlenen bu çalışmada karşılaştırılmalı analizler ChatGPT çeviri çıktıları ile insan çevirileri çerçevesinde sürdürülmüştür. Arapçadan Türkçeye çevirisi yapılan; düz metin, yazınsal metin ve teknik metinler ChatGPT ile Arapçadan Türkçeye yeniden çevrilmiş, ChatGPT çeviri çıktıları; genel olarak semantik ve dilbilimsel eşdeğerlik bağlamında incelenmiştir. Bu inceleme neticesinde Arapça – Türkçe farklı metin türü çevirilerinde yakaladığı başarı grafiği değerlendirmiştir.
In the globalizing world, the major obstacle, especially in the field of communication, is language. Scientists have been carrying out studies systematically for a long time to eliminate this obstacle. Artificial intelligence technologies are the most important efforts to solve this problem. Fortunately, artificial intelligence studies have gained great momentum recently, and this is undoubtedly reflected in translation studies. Especially with the development of the ChatGPT model, new job opportunities have been introduced into the translation profession in the translation industry. On the other hand, this development has made the question of whether artificial intelligence can replace translators a current issue. The main purpose of this study, which we have approached as an answer to this question, is to determine the role of ChatGPT in Arabic-Turkish translations. In line with this purpose, by focusing on artificial intelligence-based translation types and development phases thereof, Arabic-Turkish ChatGPT translation outputs were examined comparatively. In the study, the source text was determined to be Arabic and the target text to be Turkish. Plain texts, literary texts, and technical texts were re-translated using ChatGPT in this language pair, and equivalence and linguistic analyses of the translation outputs were carried out.
While until recently, translations made using machine translation were incomprehensible, since these translations included serious errors, today, it is seen that more accurate translations are presented with artificial intelligence technologies, such as ChatGPT. The machine translations can generally be studied under 4 headings. These are rule-based models, statistical models, hybrid models, and neural models. When machine translation had a rule-based process form, it started to provide outputs in a hardware format with statistical, hybrid, and finally neural networks in line with the needs of the time. In this process, the file volumes and time problems of international companies, the inability of rule-based machine translations to capture context, and the resulting erroneous or poor translation outputs were among the leading reasons for the development of machine translation models. With neural machine translations, sourceoriented translation outputs have largely been replaced by target-oriented translation outputs. Therefore, the development of artificial intelligence and the integration of neural technology into translation have brought artificial intelligence translations into a different state today. As a result of the trial outputs in the application section, first of all, it can be said that ChatGPT is not an adequate translation tool for the Arabic-Turkish language pair. However, its contributions to the translation process and its success in some translation types should not be ignored. The differences between human translations and artificial intelligence translations are as follows: the error margin is lower in human translations, and the error margin is higher in artificial intelligence translations. However artificial intelligence translations are faster and also less costly. On the other hand, neural network models such as ChatGPT constantly renew themselves and offer better quality translations every day. In summary, some findings, conclusions, and determinations have been reached in this study that artificial intelligence translations cannot replace human translations, so there will be a need for an editor to check the artificial intelligence translation outputs for a long time. Therefore, regardless of which machine translation model is chosen, machine translation is not sufficient on its own, and even the neural machine translation model cannot offer “acceptable” translation in every field. However, it has been concluded in this study that the promise of ChatGPT for the future is a valuable development, and the development of ChatGPT will make the work of translators easier. In the application section of the study, a comparative analysis of ChatGPT outputs and “human translations” was carried out. The examples in the application section were randomly selected. Arabic simple, compound, and complex sentences were translated in Example 1, a legal text from the technical text group in Example 2, and literary texts consisting of poetry and proverbs in Example 3. The translation outputs were examined from many aspects such as grammar, semantics, and context. The results of the examination showed that the quality of ChatGPT translation outputs varied by the text type. Furthermore, the following findings, conclusions, and determinations were obtained as a result of the study: ChatGPT is very successful in plain text translations in the Arabic-Turkish language pair. In technical text translations, it is possible to create a good translation with minor corrections, even if ChatGPT presents erroneous translations. However, unfortunately, for literary texts, ChatGPT could not provide the translation quality output offered by human translators.
