Human VS. AI Translation Accuracy: A Comparative Study of English-Arabic legal Contract Translation

Faculty of languages Journal Issue 32 December 2025 Faculty of languages Journal 1 (23 ) Human VS. AI Translation Accuracy: A Comparative Study of English-Arabic legal Contract Translation Boreiya, Mohammed Department of English, Omar Al-Mukhtar university, Albeida, Libya mohamm[email protected]m ABSTRACT As the title indicates, this study aims to conduct a comparative analysis to evaluate the accuracy of English-Arabic legal contracts produced by expert human translators and artificial intelligence tools; namely, DeepSeek and Qwen Chat. This study employs a prescriptive, qualitative approach. 10 different legal texts were selected randomly from legal contracts and agreements and translated by both professional human translators and AI tools. The findings of this study indicated that the performance of DeepSeek and Qwen Chat in translating legal contracts into Arabic showed promise in producing fairly acceptable translations for legal texts; however, they are not totally accurate and need to be revised by professional human translators. While expert human translators have shown high-quality translations of legal precise terminology and complicated structures that meets the requirements of Arabic legal standards and Libyan context. Keywords: Human translation, AI translation, Legal contracts, Translation accuracy. صخلملا نراقم ليلتح ءارجإ لىإ ةسارلدا هذه فدهت ,ناونعلا يرشي امك لىإ ةييزلنجلإا ةغللا نم ةينوناقلا دوقعلا ةجمرت ةقد مييقلت ينجمترم ةطساوب ةجمترلا تناكأ ءاوس ,ةيبرعلا عانطصلاا ءكالذا تيادأ ةطساوب اهدليوت مت تيلا كلت وأ ,ينيشرب DeepSeek وQwen Chat تدمتعا . ةسارلدا هذه لا جهنلما مييوقلتا فيصو،عيولنا دقو رايتخا مت10 صوصن ةينوناق ئياوشع كلشب ةينوناق تايقافتاو دوقع نم،ةفلتمخ ينفترمح ينيشرب ينجمترم ةطساوب اهتجمرت تمتوتاودأو قيبطت ءادأ نأ ةسارلدا هذه جئاتن تراشأ .عيانطصلاا ءكالذا DeepSeek وQwen Chat ةينوناقلا دوقعلا ةجمرت في جئاتن رهظأ،ةدعاو تاجمرت ميدقتب كلذو دح لىإ ةلوبقمصوصنلل ام ،ةينوناقلا عمو،كلذ تسيل هيف،ةقيقد مزلتسي امم في .ينفترمح ينجمترم لبق نم اهتعجارم،لباقلما صوصنلل ةدولجا ةليعا تاجمرت نويشربلا نوجمترلما رهظأ،ةينوناقلا مستت تابيكترلاو تاحلطصلما في ةقلداب،ةدقعلما بييللا نيوناقلا قايسلاو ةيبرعلا ةينوناقلا يرياعلما تابلطتم بيلي امب . ةيحاتفلما تامكللا : ةجمترلا،ةيشربلا ةينوناقلا دوقعلا ,عيانطصلاا ءكالذاب ةجمترلاةجمترلا ةقد , . Boreiya, Human VS. AI Translation Accuracy Faculty of languages Journal 2 (23 ) Introduction In recent times, with the evolution of Machine Translation (MT), which originated in the 1940s, the burden of rendering the message to the audience partially shifted from human translators to artificial intelligence (AI) tools. Li (2024) stated that "with the continuous development of machine translation technology, artificial intelligence technology has been widely applied in various fields" (p.711). Starting from the notion of context and its pivotal role in delivering the accurate rendition of the meaning and by the emergence of AI tools in translation field, this investigation highlights the accuracy of both human translators (HT) and AI tools in the legal context. Khasanova (2023) claimed that context is of great importance for communication since it assist the audience understand the message properly. In order to grasp the meaning of a word, one should pay special attention to the surrounding context within the language, since the meaning of a word may rely on the words that surround it (Nida,1964). Halliday (1999) was the first to raise the issue of different contexts in translation. Objective and Research Questions of the Study The present study aimed to: ● Conduct a comparative analysis to evaluate the accuracy of translations produced by both human translators and AI in translating legal contracts from English into Arabic, concentrating primarily on the performance of "DeepSeek" and "Qwen Chat" compared to human translations mostly utilized in Libyan official documentation. ● Identify the strengths and limitations of AI tools and human translators in translating legal contracts from English into Arabic. The study seeks to answer the following questions: RQ1) To what extent can the accuracy of AI, specifically (DeepSeek) and (Qwen Chat), be assured in translating English legal contracts into Arabic? RQ2) To what extent can AI take the place of human translators? RQ3) what are the strength and limitations of both AI and human translators in legal translation? Faculty of languages Journal Issue 32 December 2025 Faculty of languages Journal 3 (23 ) Thus, the present study would be crucial since it enhances the comprehension of AI translation within legal context, more especially English-to Arabic legal translation, an area that remains underinvestigated. The findings of this study are of great importance since more and more people are utilizing AI tools to assist the translation process. Statement of the Problem Legal translation is seen as a serious issue due to the required precision, sophisticated technical terminology, variations in legal systems across countries, and the impact of cultural and legal context. Legal translation has the power to influence legal rights and obligations. Thus, accuracy is crucial. This type of translation requires translation expertise and precise conveyance to assure an accurate reproduction of the source language message since any grave mistake can lead to legal consequences. Literature Review Artificial Intelligence and its Use in the Field of Translation Al-Khalifa (2023) defines artificial intelligence as the replication of human behaviors and functions by machines, relying on a collection of techniques and algorithms that aid in decision-making and problemsolving. Artificial intelligence AI is the ability of computers to imitate human intelligence and may also be defined as a technology that machines use to mimic many intricate human skills (Sheikh et al., 2023). Nowadays, a large number of people communicate across language barriers by utilizing smartphones and online machine translation apps, which helps to bridge the gaps among linguistic systems and cultures. Artificial intelligence has revolutionized the translation industry with its creative ways to overcome linguistic obstacles and promote international communication (Seyidov, 2024). According to Balayev and Alizade (2016), artificial intelligence AI has transformed translation, enabling organizations and individuals to communicate across linguistic boundaries with unprecedented ease and effectiveness. Boreiya, Human VS. AI Translation Accuracy Faculty of languages Journal 4 (23 ) Artificial Intelligence AI Tools 1. DeepSeek DeepSeek is a Chines firm launched in 2023 by Liang Wenfeng, the founder of High-Flyer, an AI-powered quantitative hedge fund. The firm focuses on creating open-source artificial intelligence models. Its mobile app, which carries the same name, distinguishes itself from other AI tools such as OpenAI's "ChatGPT" in its ability to clarify the logic behind its responses prior to presenting them to users. This model enables the mobile automated chat application, which gained global attention with the online interface as a considerably less expensive alternative to ChatGPT. DeepSeek model is seen as a powerful competitor to the latest releases from OpenAI and Meta. 2. Quen Chat Qwen Chat is an AI tool driven by the Qwen series of models. It is available to everyone and free to use. Qwen is a huge language model created by the Chinese company Alibaba cloud. In July 2024, it was regarded as the top Chinese language model in many criteria. With deep research, tasks that used to take a long time can now be performed in a matter of minutes. Legal Translation Legal translation is not an easy task. Translating legal documents is a tough job that requires both linguistic accuracy and deep understanding of legal concepts, especially when translating between two languages that differ syntactically and culturally, such as English and Arabic (Altakhaineh et al,. 2025). Legal translation is defined by its peculiarity; it requires deep understanding of the law and legal jargon in both source and target languages (Ben Zayed, 2024). It involves the translation of contracts, patents, court documents, international treaties and agreements, personal documents such as (certificates and passports), commercial records, and official correspondences. Legal translators must have a strong background in the legal language in order to deliver an accurate version and avoid ambiguity and misinterpretation. Faculty of languages Journal Issue 32 December 2025 Faculty of languages Journal 5 (23 ) Newmark (1982) stated that, legal texts have a dual purpose as directive and imperatives. In (1988), he added this observation by suggesting that legal texts may also have an expressive purpose. Legal translation is differentiated from other types of translation in that the message is expressed utilizing codes that represent certain legal concepts (Farghal & Shunnaq, 1992). Characteristics of Legal Language Legal language is unique due to its specialized nature, which is defined by its clarity, generality, and precision. The following characteristics are also considered as issues of legal translation. Firstly, legal language sometimes integrates words from other languages. According to Rahim (2024), "legal language sometimes incorporates words from other languages" (p.200). Such as the legal term "force majeure" which derives from the French civil law. The clause "force majeure" refers to a superior force that prevents one of the contracting parties from performing their contractual obligations. It can be translated as " ةرهاقلا ةوقلا" or " ةرهاقلا ةوقلا طشر". Secondly, legal context tends to use of archaism. Some lawyers and specialists tend to utilize archaism (Old words not commonly used) instead of new ones. For example, they tend to utilize (peruse) instead of (read), (inquire) rather than (ask), etc. The utilize of archaism (old words) is intentional (Alcaraz & Hughes, 2002). The reason behind this is to provide a sense of formality to the language to which they are associated. For example: The word thereafter which can be translated into Arabic as ادعاصف نلآا نم Hereinafter referred to as___. ـب دقعلا اذه في دعب اميف هليإ راشلما ____ . The parties hereto agree as follow. لي ام على دقعلا اذه بجومب ينفرطلا قافتا مت . Boreiya, Human VS. AI Translation Accuracy Faculty of languages Journal 6 (23 ) Thirdly, the use of the modal "shall". In legal context, "shall" expresses obligation or responsibility rather than futurity. Sabra (1995), asserts that any legal verb preceded by "shall" is commonly translated into Arabic in the present form. For example: First party shall pay the brokerage fees not less than four months. رهشأ ةعبرأ زواجتت لا ةدم في ةطاسولا باعتأ لولأا فرطلا عفدي . Ultimately, legal language is basically concerned with reference accuracy, which leads to lexical repetition and, as a result, functional redundancy. Words like "the above mentioned", "first party", "second party", "lease", "lessee" and "lessor". When translating legal documents into the target language, it is advised to preserve the same level of redundancy as the original one. Therefore, legal translators should emphasize that the suggested version is as obvious as the original. For instance, the lessee shall pay to the lessor at the office of the lessor. رجؤلما بتكم في رجؤلما لىإ رجأتسلما عفدي . Challenges of Legal Translation 1. Variations between Legal Systems Each country possesses a distinct legal system, such as Islamic law (Sharia), common law, civil law, or a combination of civil and common law. In addition, some Islamic countries adopt civil law influenced by Islamic law, which means a combination of civil law and "Sharia". All of these systems have their own terminology. English legal systems may be based on civil law, common law, or a combination of both systems. For example, UK rely on common law, whereas Scotland based on mixed law. It is worth mentioning that the English legal systems within on language differ from country to another, for instance in the UK and US, the term lawyer, attorney, solicitor, barrister, advocate, counsel, and counselor are utilized differently. These terms are commonly translated into Arabic as مامح"" ; however, they are not synonyms, and incorrect utilize of either term may lead to legal consequences. Faculty of languages Journal Issue 32 December 2025 Faculty of languages Journal 7 (23 ) 2. Specialized Terminology Legal terminologies have accurate meaning that differ from their typical use. For instance, the English legal term "consideration" in contract law does not mean "reflection" or "thinking" as in general English, but rather "something of value exchanged between parts" it can be rendered into Arabic as " دقاعلتا لباقم". 3. Neutrality and accuracy Legal translator must remain faithful to the original text and avoid adding any interpretations. For example, “The dispute shall be settled by arbitration", the appropriate translation into Arabic is " نع عانزلا لصفي ميكحلتا قيرط". Adding any extra words such as " لولدا ميكحلتا" would distort the meaning. 4. Cultural differences Law is linked to culture, some legal concepts are rooted in history and culture, making them difficult to translate. For instance, the word " صاصق" in Islamic criminal law has no accurate equivalent in English and is often paraphrased into English As "retribution" or "equal retaliation". 5. Lack of Standardized Terminology no single dictionary can possibly covers all legal contexts, therefore translator must look up relevant precedents and legal texts. For example, the English legal term "tort" can be rendered into Arabic as " ةييرصقلتا ةليوؤسلما" or "رضرلا". Both are correct, but rely on the legal context. Previous Studies Altakhaineh et al. (2025), in their article, investigated the accuracy of AI generated tools in translating legal texts from English into Arabic, concentrating on the performance of ChatGPT compared to human translations. This investigation employs a comparative research design to evaluate legal texts, including agreements and contracts translated by AI and expert human translators. This study revealed that AI tools Boreiya, Human VS. AI Translation Accuracy Faculty of languages Journal 8 (23 ) can provide relevant translations for simple documents. However, they frequently fail to convey the exact legal terminology and complicated structures needed for effective legal communications. The results also showed that human translations outperformed AI tools versions across all parameters. Showing high level of accuracy, proficiency, and clarity in rendering legal terminology and complicated structures that meet the formal legal standards and Arabic legal context. Rahim (2024), conducted a quantitative analysis, examining the Arabic translations generated by (Google Translate) for the English legal articles. The main goal of this study is to examine whether Google Translate can effectively substitute human translators for legal contracts. Six segments of various legal contracts were selected and translated into Arabic utilizing the commonly used free web-based tool (Google Translate), and the translations were evaluated for lexical and syntactic accuracy. The results obtained show that, Google Translate is faster than professional translators in producing EnglishArabic translations, but struggles with sophisticated legal terms and grammar. Polysemy, homonym, legal doublets, and adverbs are examples of linguistic errors, while syntactic errors include morphological parsing, concord, and modality. Moneus and Sahari (2024), applied a comparative analysis of the translation variations of legal texts from English into Arabic, evaluating the performance of ChatGPT, Bing Chat, and ChatSonic in terms of accuracy and fluency compared to human translators versions. The study also explored the worries about decreasing need for human translators as AI advances, as well as evaluating the possibility of relying solely on AI tools in legal translation and analyzing the strength and weakness of both approaches. To achieve this goal, a selection of legal articles from various legal contracts was chosen. These pieces distributed to human legal translators and translated by both human translators and AI tools to examine the differences between them. The results demonstrated that human translators can be distinguished from AI translation due to the professional translator's broad practical experience and legal background. A comparative study between human translation HT and artificial intelligence AI translation by Abadich (2024), aimed to compare Faculty of languages Journal Issue 32 December 2025 Faculty of languages Journal 9 (23 ) between AI based machine translation and HT in terms of efficacy. Focusing primary on the variations between them when translating English into Arabic and vice versa. This investigation also aimed to highlight the strength and weakness of both broaches in order to offer a better grasp of the variations between HT and AI-driven translation. The researcher employs a quantitative approach, where both AI-driven machine translation and HT were tested. The chosen texts include cultural and complex linguistic expressions that require deep human understanding. This approach applied a direct comparison of the performance of both AI-driven machine and HT in handling such texts, in order to identify their strength and limitations. The results obtained showed that the human translation remains the preferred choice when handling complex texts that require strong cultural background, while AI-driven machine translation characterized by its extreme speed, making it an appropriate choice for translations that don't require high level of precision. Ben Zayed (2024), conducted a comparative approach to evaluate the accuracy and reliability of (Google Translate) and (ChatGPT4) in translating legal documents between English and Arabic. The translations were examined based on multiple criteria, including linguistic accuracy, cultural nuances, legal terminology precision, and maintenance of the original intend. To accomplish this task, the researcher selected five samples of different legal documents, translated by both tools (Google Translate and ChatGPT), then carried out a detailed comparison of the translated documents to find out their quality. The findings revealed that (Google Translate) excels in accurately translating precise legal terminology. ChatGPT-4, by contrast, despite possessing a broader conversational capacity, did not perform the same level of accuracy in translating specialized legal document within the tested samples. Khoshafah (2023, as cited in Zayed & Nuirat, 2024) discussed the accuracy of ChatGPT as translation tool, and the findings obtained revealed that ChatGPT can be trusted as it able to handle many intricate and language pairs. Nevertheless, the findings also showed that ChatGPT cannot be suitable for translating specialized content that requires high level of accuracy and domain-specific knowledge, Boreiya, Human VS. AI Translation Accuracy Faculty of languages Journal 16 (23 ) In this Article, DeepSeek mistranslated the legal term "Agent" as "ةلكاو". It shifted the obligation from the agent (the individual) to the agency (the institution), creating inaccuracy in legal responsibility. The obligation is not the Agency itself; the correct subject that accurately reflects Agent is "ليكولا" . in addition, both AI translation tools, DeepSeek and Qwen Chat translated the phrase "fair competition" as "ةلداعلا ةسفانلما", which is acceptable; however, Libyan commercial legal texts typically prefer the phrase "ةفيشرلا ةسفانلما" as an appropriate rendition for the phrase "fair competition". In contrast, human translators proposed a fully acceptable translation for this Article as "ةفيشرلا ةسفانلما دعاوق ةعاارمب ليكولا متزلي", balancing between accuracy and readability as well as adherence to Libyan legal context. Utilizing the phrase متزلي " ليكولا " instead of " ةلكاولا متزلت" as mistranslated by DeepSeek. Also, "ةفيشرلا ةسفانلما" instead of ةسفانلما "ةلداعلا" as rendered by both DeepSeek and Qwen Chat. Article 5 Source Text (ST) The two parties entered into a contract last year. DeepSeek Translation .ضيالما ماعلا دقع في نافرطلا لخد Qwen Chat Translation .ضيالما ماعلا دقع في نافرطلا لخد Human Translation (HT) َادقع نافرطلا مربأ .ضيالما ماعلا By looking at the above Article, it is obvious that both DeepSeek and Faculty of languages Journal Issue 32 December 2025 Faculty of languages Journal 17 (23 ) Qwen Chat mistranslated it by providing a literal translation as لخد" "دقع في نافرطلا, which sounds odd in the Arabic legal context. Both DeepSeek and Qwen Chat produced literal translations and failed to reflect the Arabic legal tone. Whereas, human translators, on the other hand, delivered a readable, accurate, and natural version as نافرطلا مربأ ماعلا َادقع " ضيالما " , which meets the Libyan legal style and legal standards in the Arabic framework. In short, both AI translation tools (DeepSeek & Qwen Chat) maintain the general meaning but utilize a literal and less professional style. Human translation (HT) is all in all superior since it applies the accurate legal term "مربأ", which instantly matches the contractual context. Article 6 Source Text (ST) The validity for signing this contract is ten 10 banking days as from the date of this contract. DeepSeek Translation ( ةشرع ةدلم دقعلا اذه على عيقولتا ةلهم يسرت10 ) مايأ .هيخرات نم أدبت ةيفصرم لمع Qwen Chat Translation ( ةشرع دقعلا اذه على عيقولتا ةيحلاص غلبت10 مايأ ) .دقعلا اذه ماربإ خيرات نم ارابتعا ةيكنب Human Translation (HT) ةشرع ةدلم دقعلا اذه عيقوت يسري10 نم ةيمسر لمع مايأ .دقعلا اذه ريرتح خيرات Both DeepSeek and Qwen Chat translations suffer from non-standard legal terminology and sophisticated structures, such as the legal term Boreiya, Human VS. AI Translation Accuracy Faculty of languages Journal 18 (23 ) "Banking days" translated literally by both AI translation tools as " مايأ ةيفصرم لمع" and "ةيكنب مايأ", which are inappropriate translations; its suitable conveyance into Arabic could be "ةيمسر لمع مايأ". Moreover, the phrase "The validity of signing" as rendered by Qwen Chat as غلبت "عيقولتا ةيحلاص", sounds a bit like a report, not a legal contract. It could be rendered as " دقعلا اذه عيقوت يسري". By way of contrast, human translators effectively conveyed the phrase "The validity for signing this contract" into Arabic as " اذه عيقوت يسري" دقعلا instead of "دقعلا اذه على عيقولتا ةيحلاص غلبت" or يسرت" " اذه على عيقولتا ةلهم دقعلا as translated by DeepSeek and Qwen Chat. Also, the phrase "banking days" is rendered into Arabic as " ةيمسر لمع مايأ" instead of "ةيفصرم لمع مايأ" and "ةيكنب مايأ" as translated by AI translation tools. Human translation (HT) combined accuracy, clarity, readability, and adherence to Libyan and Arabic legal context. Article 7 Source Text (ST) Whereas First Party through arrangements with Spring Oil Company has the right and privilege to occupy and use a plot of land in ____ area (designed as____) of ____ square Kilometers for setting up camp facilities for its activities in this area. DeepSeek Translation فرطلا عتمتي ,ليوأ غنيبرس ةكشر عم تابيترت بجومبو في ضرأ ةعطق مادختساو لغش في زايتملاا و قلحاب لولأا ةقطنم ____ ب ةفّرعلما( ____ ةحاسم _____ ترموليك .ةقطنلما في هتطشنلأ ركسعم قفارم ةماقإ فدهب )عبرم Faculty of languages Journal Issue 32 December 2025 Faculty of languages Journal 19 (23 ) Qwen Chat Translation ةكشر عم تابيترت للاخ نم ,لولأا فرطلا نأ ينح في مادختساو للاتحاب زايتملااو قلحاب عتمتي ,ليوأ غقنيبرس ةقطنم في ضرأ ةعطق ____ ـك ةنيعلما ( _____ ةحاسمب ____ ةلوازلم ركسعم قفارم ءاشنلإ )عبرم ترموليك .ةقطنلما في هتطشنا Human Translation (HT) عم تابيترت بجومب ,هليإ لآ دق لولأا فرطلا نأ ثيحو ضرأ ةعطق للاغتسا في زايتملااو قلحا ,ليوأ غنيبرس ةكشر ةقطنمب ةنئكا _____ ـب ةددحلما( ______ ةحاسمب _____ ركسعم قفارم ةماقإ ضرغل كلذو ,)عبرم ترموليك .ةقطنلما في هتاطاشن ةلوازلم In the above Article, as we have seen, DeepSeek translation succeeded in delivering the basic meaning of this article, utilizing terms like "بجومبو", مادختساو لغش" " , and "لولأا فرطلا عتمتي". Such terms reflect the legal tone of the text. Nevertheless, the phrase "designed as" was inappropriately rendered by DeepSeek into Arabic as "ـب ةفّرعلما", which is less formal than the standard "ـب ةددحلما" in Arabic legal context. Overall, in translating this Article, DeepSeek provided the legal sense but lacks accuracy and readability. On the other hand, Qwen Chat appropriately rendered the phrase "designed as" into Arabic as "ةنيعلما" instead of "ةفّرعلما" as conveyed by DeepSeek. However, it employed the expression "نأ ينح في" which is legally inaccurate, instead of the contract standard ثيحو" "نأ . More critically, Qwen Chat mistranslated the term "occupy" as "للاتحا", a term that carries negative connotations of forcible seizures. Such mistakes undermine its accuracy in Boreiya, Human VS. AI Translation Accuracy Faculty of languages Journal 20 (23 ) rendering legal texts. By way of contrast, the Human translation (HT) version flows more properly and excels in its professional formulation and adherence to Arabic legal context. It delivered appropriate translations, beginning with "نأ ثيحو" instead of "نأ ينح في" , utilizing the phrase " ةقطنمب ةنئكالا", instead of "ةقطنم في", "ـب ةددحلما" instead of "ـب ةفّرعلما", and "للاغتسا" instead of "للاتحا", all of which align with Libyan legal context. In other words, HT displayed a high quality version in terms of readability, accuracy, and adherence to Arabic legal standard and Libyan legal context. Article 8 Source Text (ST) Seller expressly warrants that property, improvements, building or structures, the roof, plumping, heating/ or ventilation systems are in good and working order. This clause shall survive closing of title. DeepSeek Translation وأ نيابلماو ةيراقعلا تانيسحلتا نأ ًةحاصر عئالبا نمضي وأ /و ةئفدلتا ةمظنأو ةيحصلا ةمظنلأاو فقسلاو تآشنلما اذه لظي .حيحص كلشب لمعتو ةديج ةلابح نوكت ةيوهلتا .ةيكللما لاقتنا دعب لوعفلما يراس دنلبا Qwen Chat Translation وأ نيابلماو ,راقعلا تانيستح نأ نمضيو ًةحاصر عئالبا رقُي ,ةيوهلتا وأ/و ةئفدلتا مظنو ,ةكابسلا مظنو ,حطسلاو ,كلايلها ًايراس دنلبا اذه لظيو .لمعلل ةلحاصو ةديج ةلاح في نوكت .ةيكللما لقن مامتإ دعب Human Translation (HT) تانيسحلتاو عيلبا لمح راقعلا نأ ًةحاصر عئالبا نمضي ةكابسلاو ,حطسلاو , تآشنلما وأ نيابلماو ,راقعلا على ةقحللما Faculty of languages Journal Issue 32 December 2025 Faculty of languages Journal 21 (23 ) ةلاح في اهعيجم ,ةيوهلتا و ةئفدلتا ةمظنأو ,)ةيحصلا داولما( خيرات تىح ةرقفلا هذله ةينوناقلا راثلآا يسرت نأ على ,ةديج .يترشلما لىإ ةلمكالا ةيكللما لاقتناو ئياهلنا ليجستلا By looking at the translation generated by DeepSeek, It can be seen that, DeepSeek produced acceptable translations for phrases such as "Seller expressly warrants" as عئالبا نمضي" " " "ةحاصر , "plumping" as ةمظنأ" "ةيحص, and "building or structures" as تآشنلما وأ نيابلما" " , reflecting the nature of legal language. Nevertheless, DeepSeek failed to deliver the exact legal meaning of the phrase "This clause shall survive closing of title", and inappropriately rendered it into Arabic as يراس دنلبا اذه لظي " ةيكللما لاقتنا دعب لوعفلما", which sounds odd and vague in Libyan legal context. Since its appropriate translation into Arabic could be قىبت نأ على لىإ ةلمكالا "ةيكللما لاقتناو ئياهلنا ليجستلا مامتإ تىح ةذفانو ةيراس تانامضلا هذه يترشلما", or as proposed by human translators (HT) as راثلآا يسرت نأ على" يترشلما لىإ ةلمكالا ةيكللما لاقتناو ئياهلنا ليجستلا خيرات تىح ةرقفلا هذله ةينوناقلا". Furthermore, Qwen Chat improperly rendered the phrase "Seller expressly warrants" into Arabic as "نمضيو ًةحاصر عئالبا رقُي" which "acknowledges" rather than "warrants". In contract language, the term "acknowledge" is weaker than "warrant"; thus, this choice may weakens the binding legal obligation. Moreover, it mistranslated the phrase "plumping" as "ةكابسلا مظن", which lacks precision. Human translation, in contrast, distinguished as the most professional version in terms of legal accuracy, readability and adherence to legal standards in Arabic legal framework. It accurately conveyed phrases like "Seller expressly warrants" as " ًةحاصر عئالبا نمضي" instead of رقُي" "نمضيو ًةحاصر عئالبا and "plumping" as "ةيحص داوم" instead of ةكابس" مظن" , as render by Qwen Chat. Crucially, human translators provided an Boreiya, Human VS. AI Translation Accuracy Faculty of languages Journal 22 (23 ) appropriate rendition for the phrase "This clause shall survive closing of title" as ليجستلا خيرات تىح ةرقفلا هذله ةينوناقلا راثلآا يسرت نأ على " لاقتناو ئياهلنا يترشلما لىإ ةلمكالا ةيكللما" , which is precise, readable, and fully aligned with Libyan legal formulation standards. Article 9 Source Text (ST) The Lessor hereby leases to the Lessee, and the Lessee herby accepts the lease of, the building, land, and appurtenances commonly known as____ and located at____ (hereinafter referred to as the "Building"). DeepSeek Translation رجأتسلما لبقيو ,رجأتسلما لىا رجؤلما رجؤي ,اذه بجومب مساب ةفورعلما تاقحللماو ضرلأاو نىبلما راجئتسا _____ في ةعقاولاو _____ .)"نىبلما" ـب دعب اميف اهليإ راشي( Qwen Chat Translation ,رجأتسلما فرطلل اذه بجومب انه رجؤلما فرطلا رجؤي تاقحللماو ضرلأاو نىبلما ,رايجلإاب رجأتسلما فرطلا لبقيو ةفرعُملا _____ في ةدوجولماو ______ اميف اهليإ راشُيو( .)"نىبلما" ـب دعب Human Translation (HT) رجأتسلما لبقو ,رجأتسملل دقعلا اذه بجومب رجؤلما رجأ تاقحللماو ضرلأاو نىبلما رجأتسي نأ دقعلا اذه بجومب مساب اهيلع فراعتلما ___ في ةنئكالاو _____ اهليإ راشيو ( .)"ءانلبا" ةملكب دعب اميف In the above Article, as we have seen, DeepSeek conveyed the overall meaning of the Article, properly captured the mutual obligations and consent highlighted in this Article. Nevertheless, the wording style is Faculty of languages Journal Issue 32 December 2025 Faculty of languages Journal 23 (23 ) somewhat informal and less aligned with legal formulations of Arabic contracts. For instance, utilizing of the term "لبقيو" in the present tense which is not aligned with contract drafting, it would rendered into Arabic as "لبقو" in the past tense. In addition, DeepSeek, provided expressions such as " اذه بجومب" which can be conveyed as اذه بجومب" "دقعلا, also "ةفورعلما تاقحللما" could be rendered as "اهيلع فراعتلما تاقحللما" maintaining the legal tone of the text. As with DeepSeek, Qwen Chat, to some extent, deliver an acceptable rendition for the above "Article"; however, utilizing inappropriate expressions such as " رجؤلما فرطلا رجؤي" انه which can be conveyed as ا رجأ" رجأتسملل دقعلا اذه بجومب رجؤلم " , also رجأتسلما "لبقيو" and " ةفرعُملا تاقحللما" that could be translated into Arabic legal context as "رجأتسلما لبقو" and "مساب اهيلع فراعتلما تاقحللماو" . In comparison, human translators (HT) produced the most legally accurate version of the above Article. In their translation, Human Translators utilized standard legal expressions such as "دقعلا اذه بجومب" instead of "اذه بجومب" as employed by both DeepSeek and Qwen Chat, also "رجأتسلما لبقو" instead of "رجأتسلما لبقيو", "مسأب اهيلع فراعتلما تاقحللما" instead of "ةفّرعلما تاقحللما" or "ةفورعلا تاقحللما" , and "في ةنئكالا" instead of "في ةدوجولما". Findings and Discussion The findings obtained from this investigation revealed that, the performance of AI models; namely, DeepSeek and Qwen Chat, in translating English legal contracts into Arabic showed promise in producing acceptable translations for legal texts in certain instances. However, they are not totally accurate in translating complex structures where recognizing and respecting the target culture are essential. Thus, DeepSeek and Qwen Chat translations need to be revised by professional human translators. This finding aligned with Boreiya, Human VS. AI Translation Accuracy Faculty of languages Journal 24 (23 ) the finding revealed by studies conducted by Altakhaineh et al,. (2025), Rahi (2024), and Moneus and Sahari (2024), since they emphasized that AI modules delivered acceptable renditions for simple legal texts whereas failed to produce an appropriate translation for sophisticated legal terminology. However, this finding contradicts the finding obtained by Ben Zayed (2024), since it demonstrated that AI-powered tools such as Google translate and ChatGPT exceled in appropriately rendering precise and intricate legal terminology. The findings of this study also demonstrated that, expert human translators have shown high-quality translations of precise terminology and complicated structures that meet the requirements of Arabic legal standards and the Libyan legal formulations. This result is in harmony with a study presented by Abadich (2025), which indicated that the Human Translation (HT) remains the preferred chose when dealing with intricate legal terminology. Ultimately, the human mind is indispensable, and artificial intelligence tools are merely assistive aids. This finding is consistent with results revealed by a study conducted by Khoshafah (2023, as cited in Zayed & Nuirat 2024), where they emphasized that AI generated translation should be postedited by professional human translators as a necessary step. Ultimately, the vast majority of errors committed by AI tools can be classified into two categories: cultural and contextual errors in terms of cultural references and context sensitivity, and linguistic errors represented in the appropriate usage of synonyms and legal doubts. Conclusion In conclusion, this research has been conducted to evaluate the accuracy of AI and human translators in translating English legal contracts into Arabic. Based on the findings presented above, translations generated by DeepSeek are not totally appropriate, however they can be so in case of some amendments are made to them by qualified human translators. Thus, AI translation tools, including DeepSeek and Qwen Chat are a promising complement to human translators; they do not yet supplant the necessity for expert human translators, especially in the legal domain. Human translators, in contrast, have shown a high level of accuracy, readability, and quality required in legal context, maintaining the Faculty of languages Journal Issue 32 December 2025 Faculty of languages Journal 25 (23 ) Libyan legal tone and Arabic standards. Furthermore, showing a more in-depth awareness of legal culture background and textual nuances. Recommendations Based on the results of this investigation, we may conclude with the following recommendations: ➢ As AI improves, human translators who are skilled in integrating Artificial intelligence in translation will have a better chance than those who are unfamiliar with translation technology. ➢ AI is helpful; nevertheless, it’s not perfect and can sometimes make mistakes or even make things up. ➢ More research is needed to be conducted in order to update and enrich the literature of AI translation. References Abadich, Y. (2024). Exploring the comparative efficacy of human translators and artificial intelligence in translation processes: ArabicEnglish, vice versa [Master’s thesis, Cadi Ayyad University]. ResearchGate. Alcaraz, E., & Hughes, B. (2002). Legal translate explained. Routledge. Al-Jarf, R. (2025). AI translation of full-text Arabic research articles: The case of educational polysemes. journal of Computer Science and Technology Studies, 7(1), 311. Al-Khalifa, H. (2023). Mukadima fi el dakaa el istiaahi el tawlidi [Introduction to Generative AI]. Iwane. Altakhaineh, A. R. M., Alghathian, G. A., & Jarrah, M. M. (2025). A comparative study of accuracy in human vs. AI translation of legal documents into Arabic. International Journal of Language & Law, 14,