Toodles to Google Translate: Why Human Interpreters Are Irreplaceable in the Republic of Korea Special Forces (and Other Military Forces)

“Sergeant K, we will miss your position next year, but you were irreplaceable.”

It was exactly 48 hours before my discharge at the Republic of Korea Special Forces. The biggest event of my last day of work as a military interpreter was having a farewell talk with my boss, the chief of Operation Plan Branch Colonel Lee. As a military interpreter, I worked between the Korean and US Special Forces, providing linguistic support, including interpretation (oral translation) and written translation to the armed forces. Although the warm words of goodbyes and sweet-memory sharing was great, the only thing that lingered on my mind for the whole meeting was the rather uninvited news that passed through my ears during our talk: that is, that a good number of military interpreter positions in the base would disappear soon. Most of the bases in South Korea were trying to cut back on the position, and AI was being considered as one of the most probable candidates, the argument being that AI translators would be both more efficient, and cheaper, especially as it is likely to become only more accurate and credible in the future. However, as I will argue in this paper, a certain balance between AI and human interpreters must be retained in the Korean Special Forces, as they must be elsewhere. While AI might be helpful to increase efficiency in low-stake areas or stages of interpretation and translation, human military interpreters cannot be replaced completely because of what we might call the “human aspect” of military interpretation—namely, cultural understanding, operational accountability, and communicational discernment. In the realm of military operations, where the stakes can literally be a matter of life and death, the clear communication between individuals is absolutely imperative.

This is not to say there are no areas where AI translation can be helpful. For example, AI technology can be useful in low-stake areas where efficiency is preferred over accuracy, such as preliminary document translation or situations requiring interpretation of static, procedural words or simple communicational texts. With the recent development of AI, the technology of online translation platforms such as Google Translate, Bing Microsoft Translator, or DeepL has rapidly improved, and big tech companies like Meta are even developing a new AI translator that can translate spoken languages like Hokkien (a dialect of southern Min Chinese) into English to further knock down language barriers (Tech Desk). In light of this trend, utilizing AI for certain areas of military interpreters’ duties can be useful. Especially, for written translation, AI could be helpful for increasing efficiency in preliminary document translation. Before going into actual translation word by word, using the AI to translate the whole document in one-click provides a holistic view of the contents in the target language (that is, the language that one is trying to translate into), which allows the interpreter to gain a preliminary insight into the overall content, and to discern which parts need more focus and attention. Also, employing AI can be useful for saving time in translating structural parts of the document that do not require deep contextual understanding or thorough comprehension of the meaning of the text, elements such as titles, table of contents, or even short captions under images. Moreover, AI can be helpful in simple translation of static, procedural texts or emails for communication, as these tasks are often low-stake, and require more efficiency than accuracy.

However, even in low-stake areas, using AI in Korean-English translation still has a high degree of inaccuracy, which is especially concerning in high-stakes situations. The grammatical discrepancy between the Korean and English language is one of the most prominent reasons. Unlike several European languages that share similar roots and origin such as Spanish and Portuguese, Dutch and German, or Czech and Slovak, Korean and English have fundamentally different grammatical structures. While the basic sentence structure of the English language is “subject+verb+object” (say, for example, “I eat apple”), the structure of Korean language is “subject+object+verb” (which would be “I apple eat” in English), which poses more limitation on AI interpretation. This idea that similarities between languages impact the accuracy of AI translation is supported by the medical research conducted by the Pediatric Intensive Care Unit at Nottingham Children’s Hospital in the UK. The research examines the accuracy of Google Translate in different languages in the context of medical communication. According to their research, Western European languages had the highest level of accuracy, at 74%, while the Asian languages–including Korean–were at a low 46%. This bolsters the idea that the relationship between two languages affects the accuracy of AI translation: the more the language is similar to the source language (the language that one is trying to translate from, which in the given research is English), the more accurate the translation is (Patil and Davies 1). This would be true as much in a military context as in a medical context. The problem is that the discrepancy between Korean and English language is simply too huge for an AI to be valuable when it comes to Korean- English interpretation.

Presumably, however, over time, something like Google Translate will improve its grammatical accuracy. But even then, so much of Korean-English translation requires cultural understanding that AI just simply doesn't have. While AI can be used to increase efficiency in low-stake, preliminary, or simple stages of interpretation and translation, it cannot completely replace military interpreters when it comes to higher-stakes communications, especially those that require cultural understanding between parties. In order to understand why cultural insight is essential in military interpretation, it is important to understand the nature of military interpretation. As Viktor Balabin, a Professor of Military Institute in Ukraine, points out in his scholarly article “Redefining Military Translation,” military interpretation is “a special type of mediated intercultural communication in the armed forces” (Balabin 46). In other words, translation is not simply the replacement of one word for another; instead, it requires a knowledge of the shared cultural understandings that the language comes from, since, after all, language is the utmost product of a culture. A simple example would be jokes, sayings, or slang. Understanding the cultural context is important to grasp the meaning of these elements as oftentimes they do not directly translate into another language. Different cultures have different ways of expressing ideas, even if the central message is identical. For instance, in Korea, there is a saying that goes like “handclap sounds only if two hands meet each other,” which makes no sense in English. An equivalent saying in English would be “it takes two to tango,” which means cooperation is necessary to achieve a common goal. In order to understand the cultural context of a statement, which is the essential first step to understand the meaning and to provide interpretation of a given statement, a human's irreplaceable cultural insight is crucial.

Another aspect of human language that AI can't capture because it requires cultural and contextual understanding is polysemy. As Hyegyung Kim and Soyoung Yoon of Konkuk University’s Translation Research Center have illustrated, polysemy is a linguistic phenomenon when one word or phrase contains many possible meanings depending on the context or the background (Kim and Yoon 20). For instance, “육회” (pronounced yook-hoe) in Korean could mean two completely different things: “six times” or “raw beef.” As polysemy could be interpreted into completely divergent meanings depending on the context, when misinterpreted, it could lead to serious misunderstandings. This is why human military interpreters cannot simply be replaced with AI: correct interpretation of polysemy requires one to have the human capability to understand the comprehensive cultural context of a given statement (Kim and Yoon 21).

This is particularly true when it comes to Korean-English interpretation as Korean is one of the languages that the occurrence of polysemy is the most frequent. The Korean language is composed of words that originate from both traditional Chinese characters and pure Korean language. As a significant amount of nouns in the Korean language comes from the former, and many of the letters in Chinese character are subject to polysemy, frequent occurrence of polysemy in the Korean language is inevitable. Moreover, interpreting polysemy becomes a bigger challenge in the Korean language because, unlike Chinese (Mandarin) or Thai, it does not have tonal difference, which is how languages with frequent occurrences of polysemy normally distinguish polysemic words or characters. As recent AI translation services can detect the tone difference, and as the tone difference is often visible in the written language, polysemy is not a serious problem for languages like Chinese and Thai. However, considering the frequent occurrence of polysemy and the absence of tone difference in the Korean language, the use of AI is much more limited in Korean-English interpretation—and human interpreters are needed to fill in the gap.

Furthermore, the wide range of adjectives in the Korean language also explains why AI cannot completely replace human interpreters and instead requires the subtle distinctions human interpreters can make through greater cultural understanding. The Korean language is one of the languages that have the most diverse collection of adjectives. For instance, even if it is the same blue color, Korean language has different words that each describe the subtle difference of tone or vividness in the color. “파랗다” (pronounced “parat-da”) means “just blue” and “푸르스름하다” (pronounced “pureuseureum-hada”) means “a bit bluish,” while “푸르죽죽하다” (pronounced “pureujukjuk-hada”) means “dully and sporadically blue.” As AI interpretation tends to focus on efficiency and generalize the meaning of these terms, it simply cannot notice and deliver this subtle difference. This means that even if we use AI interpretation for efficiency, it could not be as efficient as expected, or it could even be more inefficient as human interpreters must intervene in order to correct or add to the meaning. This is not to say that these complex adjectives are commonly used in military talks. Rather, it is to highlight the unique characteristic of the Korean language that further differentiates it from the English language, which demonstrates that the human capacity for cultural understanding is vital when it comes to Korean-English interpretation.

Besides cultural understanding, operational accountability is another “human aspect” that explains why human military interpreters are irreplaceable by an AI. Military interpreters live in the base with other enlisted soldiers, non-commissioned officers, and higher officers or leadership personnel, which allows them to naturally acquire knowledge and insights into the structure, organization, culture, nature, and characteristics of not only the base or unit they are part of, but also the military as a whole. The experiences and knowledge accumulated through being accountable to the base, to the military, and to the security of the nation are the things that an AI machine cannot simply obtain. Also, these experiences and knowledge are an essential part in military interpretation as it is the key to understanding insider jokes, politics, hierarchical culture, and the relationship between different military units and personnel. For instance, in my battalion of the Republic of Korea Special Forces, there was an inside joke between higher officers that goes like “the flying tiger had lost its wings,” which referred to the 3rd Special Forces Brigade losing its military power due to a recent reduction in weaponry funding. This joke was important because many officers used it as a witty way to not only lighten the mood of the meeting, but also touch upon an ongoing issue in order to naturally start the conversation all while not directly offending the relevant units. AI cannot simply understand and deliver the hidden meaning behind these insider jokes or sayings because they are part of the larger operational functions of the military that it simply cannot register and thus be accountable to.^[1]

Some might nevertheless still argue that, with AI’s rapid development, replacement of human interpreters with AI might be possible in the distant future. AI’s capabilities are improving constantly, and it might eventually be able to at least mimic human’s cultural understanding capability (if not accountability) sometime in the future—it might develop a technology to detect polysemy or translate culturally nuanced jokes or sayings. If this is realized and the concern for inaccuracy is eradicated, using AI for interpretation will be beneficial as it would drastically increase efficiency.

However, even if AI technology reaches the level that it can actually have the capability of cultural understanding, human interpreters are still irreplaceable because of their communicational discernment. Discernment in communications, which is an ability to be both versatile and deliberate when delivering words and intentions, is significant because of the fundamental nature of military talks. Conferences and meetings in military settings are very different from ordinary ones, as it is often very politically sensitive. Given that negotiations and compromises regarding weapons and assets from different units and organizations directly impact the security of the nation at stake, even a subtle change in tone made by one side could significantly influence the posture and stance of the other. As So Un Park, a professional freelance Korean-English interpreter, mentions in her book The Role of Interpreter, the interpreter’s most important role is to discern which words to “pass on” and which ones to not (Park 11). A major objective of military interpreters (especially ones that work for two friendly nations like South Korea and US) is to make the negotiation happen for the cooperation between two countries, one must use their discernment and judgment in order to frame and deliver words wisely so that they foster a friendly atmosphere, and, simply put, prevent things from heating-up too much. And this discernment in interpretation cannot simply be outsourced to an AI machine that operates merely based on coding and algorithm.

To conclude, incorporating AI in military interpretation clearly has benefits. It could help increase efficiency in low-stake, static, and simple areas or stages of interpretation. However, a complete replacement of human military interpreters by an AI is not only impossible but also unwise because accuracy is crucial when it comes to military interpretation, which requires human capabilities such as cultural understanding, operational accountability, and communicational discernment. And these human capabilities are particularly important in Korean-English military interpretation because of the discrepancy between the two languages and the unique characteristics of the Korean language. Thus, in order to kill the two birds of accuracy and efficiency with one stone, we must use AI and human interpreters in conjunction with each other, in which AI is used as a complementary tool to help increase efficiency for human interpretation. Like all technological developments, AI’s adoption in military interpretation is a double-edged sword: while proper usage could be beneficial, over—or misusage could bring about serious consequences. As such, finding the balance is the key. However, at the same time, we must remember that the degrees might vary. For interpretation in different settings or different languages, AI technology might be more or less helpful. After all, it is our own judgment to decide how to utilize the double-edged sword to “win the battle.”

Author

Kyeongmook Lee

Kyeongmook “K” Lee is an Economics and Finance major at the University of Notre Dame. He currently lives in New York, but he grew up most of his life in Seoul, South Korea. K’s dream is to become a researcher in the field of international security, exploring how recent geopolitical threats such as cybersecurity and pandemic impacts the decision-making process of different financial and economic actors such as private firms, nonprofit organizations, and governments in the context of security. In his research paper, K attempts to answer the question of why human interpreters cannot be completely replaced by an AI technology in military forces, utilizing his own experience of serving as a military interpreter at the Republic of Korea Special Forces before coming to Notre Dame. K has a special “shout-out” to Professor Nathaniel Myers, for his tremendous support and help throughout the writing process.