Photo by Charli, 420
by Yutong Liu
Read the Faculty Introduction
Abstract
This paper first analyzes the advantages and limitations of AI translation compared to human translation. Then it explores the foreground of the translation profession and discusses the reason why human translators could not be replaced. Finally, based on the advantages and disadvantages of AI translation, this essay highlights the essential skills that university students need to enhance in order to utilize AI tools correctly during their learning process, aiming to provide valuable insights and guidance for further emphasizing the university students' learning.
Introduction
“Can you please help me revise this interview? It is translated by ChatGPT, you just need to change it to make it sound more natural in Chinese.” This kind of request was frequently given to me when I worked as a student worker in the Center for Career Development (CCD). At some point, people have started to rely on AI for the initial translation, and then make adjustments to the output before finalizing the translation.
Back in the early 2000s, Google Translate was first launched. The way people translate and learn a different language has evolved from consulting dictionaries to searching the internet and utilizing translation tools. Since then, many people enthusiastically envisioned a bright future where machines would replace human translators. However, the translation generated by a machine translator was so stiff and simple that it was far from being a threat to human translators at that time.
In 2023, when ChatGPT and other AI tools emerged all of a sudden, the machine translation skill has progressed significantly, since the large language module has a huge database and enormous language source that it could analyze the text based on the context (Annie). Additionally, AI can adjust translations based on specific human instructions, improve quality, or meet certain special requirements for the style of the translated text. Therefore, there exists an ongoing discussion about whether the translation profession can be replaced by machines.
Before discussing the answer to this issue, let’s first return back to my work task in CCD that I mentioned in the beginning. I originally thought that the task of revising the translated text is really easy for a native Chinese speaker and I would only need to make adjustments to some sentences that I found unnatural. Nevertheless, it is much more difficult when I actually do it. When assessing the translation abilities of artificial intelligence, conducting experiments with random sentences or paragraphs might give the impression of good performance. However, the true challenge lies in long-text translation, which is a complex and comprehensive task. ChatGPT can exhibit inconsistent performance, sometimes resulting in translations with lots of hidden mistakes and inconsistencies. Finally, I found that even with the assistance of intelligent translation tools like ChatGPT, translators still need to exert considerable mental effort and time.
Based on my observation, the use of AI translation also commonly appears in students’ daily study, especially for those whose native language is different from the required language. When writing an assignment in English, I found lots of my classmates at NYUSH first wrote the essay in their native language. Then, students translate the whole essay into English and make some adjustments to form the final version of their assignment. What’s more, some even wrote their transcripts of presentations in their own native language and then translated it into English to present.[1] With the help of generative AI, it seems that the barrier between languages has become smaller. But is this really the case?
From my point of view, AI cannot fully replace human translators; however, it holds promising applications and potential value when utilized as an effective tool and aid in the translation process. I will illustrate this argument later in this article.
Advantages of AI in Translation
There is no doubt that AI is a really good tool when it comes to translation. AI has lots of advantages compared to human translators:
Firstly, one of the most notable benefits of AI translation is its speed. Studies have shown that AI translation systems can process and translate text at remarkable rates, significantly outperforming human translators in terms of efficiency and quantity. For instance, AI models like ChatGPT can translate thousands of words per minute, making them ideal for handling large volumes of content swiftly and accurately. A human might take a day or more to translate 5,000 words. AI, the future of localization, can deliver a full translation almost instantaneously (Crangasu).
In addition to speed, AI translation systems excel in multilingual capabilities. Whether it's traditional machine translation like Google Translate or large language models like ChatGPT, these systems can translate between dozens or even hundreds of languages. In contrast, human translators are typically proficient in translating only between their native language and one foreign language, with only a few capable of handling translations across multiple languages. Additionally, it takes way more time for a human to be proficient in a certain language than AI tools.
Furthermore, AI translation benefits from rapid knowledge accumulation and continuous improvement. Developers constantly refine and update the knowledge base of AI engines based on real-world usage, resulting in enhanced accuracy and functionality over time. This iterative process ensures that AI translation systems remain adaptive and effective in diverse linguistic contexts.
Moreover, AI translation accessibility facilitates cultural exchange and dissemination. With AI translation tools readily available, individuals can explore content in other languages independently, gaining insights into different cultures and technological developments. This approach allows users to identify areas of interest before seeking professional translation services for more accurate and detailed information. Previously, such tasks were labor-intensive and costly when handled manually.
Limitations of AI translation Compared to Human Translation
Despite the advantages above, we have to admit that there still exist some limitations of AI translation. Moving to the limitations of AI, this section will mainly focus on the translation between Chinese and English. Through discussing the limitations of translating these two languages, we can get some inspiration for using AI to translate various languages in the world. By analyzing and comparing the advantages and limitations of AI translation, it is helpful for us to conclude if AI can replace human translation or not.
The first limitation was widely discussed through the decades, which is the translation of literature and poems. One significant feature of literature is the use of literary devices, which is hard for AI to replicate. About ten years ago in the field of translation, there was a saying that “literary translation is the final fortress of human translation” (Yuan). At that time, many believed that AI was incapable of effectively translating literature. Nowadays, AI tools started to show their capability of translating literature by spreading the English version of online novels out of China. According to Shuyin Zhang, a translation professor of the Chinese University of Hong Kong, Shenzhen, the Chinese collection at the British Library now includes 16 works of Chinese online literature. In particular, genres like fantasy martial arts novels and urban romance novels are especially well-suited for AI-assisted translation. In fantasy martial arts novels, terms such as weapons, martial arts sects, and levels can be standardized, making them ideal for machine translation based on learned data and terminology. Similarly, urban romance novels, which are rooted in human emotional patterns, can also benefit from machine translation.
However, AI is not currently suitable in the translation of classical literature and poetry, and I believe that it will never be capable of translating these works. I will illustrate it through this following example. There is a very famous Chinese poem by Qingzhao Li that says, “寻寻觅觅，冷冷清清，凄凄惨惨戚戚.” This line alternates between flat and oblique tones, creating a rhythmic and melodious effect, conveying deep and intense emotions. When I input it to AI tools, it gave me a really stiff translation: “Searching, cold, sad and miserable.” This translation did not convey the beauty of this poem at all. Here is the version translated by a famous translator Yuanchong Xu （许渊冲）:
I look for what I miss,
I know not what it is,
I feel so sad, so drear,
So lonely, without cheer.
This version, while staying faithful to the original text, introduces a protagonist named "I." Not only does it achieve perfect rhyme, it also vividly conveys the author's feelings of melancholy and sorrow.
The second limitation of AI translation is that it struggles to capture human sentiments and lacks the ability to adapt to them in real-time. In translation, there may not always be a definitive or standard answer. A correct answer is often based on a specific situation and cares about the emotion and the actual need of the reader. For example, when translating 滴滴 to foreigners who may not be familiar with this app, translators can describe its technical nature by translating it as “DiDi, a Chinese ride-hailing app” or “Chinese Uber” in order to fit readers’ habits and needs. Therefore, the “correct” English translation of 滴滴 depends on the audience. It hinges on whether the audience understands Chinese and is familiar with Pinyin pronunciation. Translators should adjust the pronunciation accordingly based on the audience's background and knowledge.
No matter how much we train AI, one thing that is certain by now is that current AI cannot create emotions like humans. Meanwhile, a key attribute of human translators is their ability to empathize. Translators need to consider the audiences’ needs, which AI cannot fulfill if we simply provide it with the text that needs to be translated. If we can provide AI tools enough information about the audience, AI may perform well and appropriately. But when we take the time and actual case into consideration, the good performance of AI is still hard to achieve and remains a limitation of AI translation.
Another challenge is that AI translation lacks the ability to engage in social interactions. When I traveled to a foreign country, one thing that I really enjoyed was to speak to the native in English and get a closer feeling of the country’s culture. Even though my English was not that good, I always received some good communications if the people spoke English. However, when I traveled to a country that does not speak English, all of my communication relied on translation apps on smartphones. There's virtually no opportunity to engage in casual conversation with strangers.
The Future of Translation
Now we analyzed the advantages and limitations of AI translation. It is obvious that the appearance of AI has a big impact on the translation profession. The following paragraphs will discuss the future prospects of translation.
I really like a metaphor published in the Financial Times Chinese: Human translation can be likened to a gourmet chef, while AI translation resembles a sophisticated combination of McDonald's and pre-packaged ingredients with automated cooking machines (Yuan). McDonald’s provides convenient and affordable food which is a good option when you're in a hurry and not concerned about food quality or nutrition. However, whether it's McDonald's or using pre-packaged ingredients with automated cooking machines, they cannot cause other restaurants to shut down. In this case, highly skilled human translators are like Michelin three-star chefs, offering "luxury services" for clients with exceptionally high standards for translation quality (Yuan).
Therefore, highly skilled translators are scarce and remain their own value in translating high standards text. And for newcomers or translators with average skills, there are two options for them in the future: aim for the top-tier market by excelling in their craft, or combine translation with other skills to innovate and explore new specialized areas, such as developing prompts to assist AI in refining translations.
In the context of AI, we do not need to compete with AI translation in terms of speed or quantity, nor should we blindly trust it. As human translators, we should understand the characteristics of AI translation and develop a keen sense of appreciation and discernment for the domains it translates. This underscores the importance of translators' post-editing skills, which involve proofreading and editing text to ensure accuracy and quality. Ultimately, humans must continue learning and fostering creativity to remain masters of machines.
Lessons for College Students in the Era of Artificial Intelligence
As a college student, AI has a big impact on our daily lives and studies. It is really important to us to know how to better use AI tools. In terms of using AI translation, as is mentioned in the previous text, AI is fast and convenient in translations which could help us a lot when we are learning a new language and reading essays written in a different language. However, if we over rely on the translation of AI, we might lose our ability to learn and think in a new language. In this circumstance, we should use AI tools properly and keep learning in order to enhance our overall language ability.
AI also has a big impact on fields other than translation. In the past year, several Silicon Valley high-tech companies, including Google, have downsized thousands of engineering and technical staff partially because of ChatGPT and other AI tools. Similar to translators replaced by AI, these coders often perform standardized technical tasks that can be automated. Therefore, as college students, we should consider these influences of AI when we are learning knowledge to be better prepared for our future career.
When it comes to learning, it's essential to go beyond textbook knowledge and explore various fields. This approach helps cultivate a diverse skill set, allowing for better fit in career jobs. Furthermore, it is crucial to begin early career planning and apply specialized knowledge in practical ways. United Nations and European Union conference interpreter Maha El-Metwally said that:
When I deliver lectures and training sessions at universities, I see my students as future colleagues. They receive interpreter training at university, and my role is to share practical experience with them. Therefore, my lectures focus on aspects beyond classroom learning, such as essential business skills required before entering the profession. I also cover the practical skills needed in the interpreting field, such as terminology management, pre-translation preparation, and remote interpreting technology.
This quote means that there remains some difference between the knowledge used in a career and the knowledge taught in school. The improvement of one’s professional ability is the way to survive in the AI era.
Conclusion: Collaboration of AI and Human Translator
Overall, this essay mainly focuses on AI translation and begins by analyzing the advantages of AI in translation. AI excels in its fast speed, multilingual capabilities, rapid knowledge accumulation and continuous improvement, and its role of facilitating cultural exchange and dissemination. Then, the essay discusses the limitations of AI translation compared to human translators. Based on my research, AI translation has the following limitations: First, it is not suitable for classical literature and poem translation. Second, AI struggles to capture humanistic sentiments and lacks the ability to adapt in real-time. Third, AI lacks the ability to engage in social interactions. The paper then analyzes the future of translation in the era of artificial intelligence. This paper concludes with the lessons for college students in the terms of using AI and preparing for their future career in the era of AI.
The main point of this essay is that though AI has lots of advantages in translation, AI cannot replace human translators due to its current limitations. However, AI holds promising applications and potential value when utilized as an effective tool and aid in the translation process.
Works Cited
Crangasu, Adrian. “AI Translation vs. Human Translation: Pros and Cons.” Localization Services by BLEND, 11 Sept. 2023, www.getblend.com/blog/ai-translation-vs-human-translation-pros-and-cons/.
Duh, Kevin, et al. "Adaptation data selection using neural language models: Experiments in machine translation." Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics Volume 2: Short Papers. 2013.
Lucas-Cuesta, J.M., et al. “On the Dynamic Adaptation of Language Models Based on Dialogue Information.” Expert Systems with Applications, vol. 40, no. 4, Mar. 2013, pp. 1069–1085.
Manjavacas, Enrique, and Lauren Fonteyn. “Adapting vs. Pre-Training Language Models for Historical Languages.” Journal of Data Mining & Digital Humanities, vol. NLP4DH, no. Digital humanities in..., 2022.
Nicholas, Gabriel, and Aliya Bhatia. “Lost in Translation: Large Language Models in Non-English Content Analysis.” Center for Democracy and Technology, 7 July 2023, cdt.org/insights/lost-in-translation-large-language-models-in-non-english-content-analysis/. Accessed 18 Apr. 2024.
Pardo, Melissa. “Using Large Language Models and AI to Bridge Linguistic Differences.” Appen, 21 Feb. 2024, www.appen.com/blog/pulse-of-language-evolution. Accessed 18 Apr. 2024.
Syed, Bakhtiyar, et al. “Adapting Language Models for Non-Parallel Author-Stylized Rewriting.” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 34, no. 05, 3 Apr. 2020, pp. 9008–9015.
Weng, Rong. “AI Translation Is Difficult to Reflect Humanistic Feelings.” Chinese Social Science Net, 23 Jan. 2024, www.cssn.cn/skgz/bwyc/202401/t20240123_5730228.shtml.
Yuan, Yilin. “Translation Profession Is Dying in the Era of AI?” FT Chinese, 5 Mar. 2024, mp.weixin.qq.com/s/pogR3PowC4N_M199bW4xdQ.
Zhao, Xiao. “Xu Bin: Can AI translation eliminate cross-language communication barriers?.” Chinanews, 8 Mar. 2024, www.chinanews.com.cn/gn/2024/03-08/10176936.shtml.
[1] I am not here to discuss whether this behavior is beneficial or harmful. This is just an example of students using AI as a tool in study.