Korean machine translation is not a blanket replacement for human translation. Sure, machine output can be produced for free, and in many cases, it is good enough for its intended purpose. It has also improved greatly in recent years. However, without additional input, machine translation (MT) is often of dubious quality. For this reason, machine translation is frequently supplemented by MT engine training, edited by human translators, integrated into a hybrid Korean translation workflow, and even used as a tool to support high-quality human Korean translation.
This article discusses these approaches to Korean machine translation to help you understand your options and make good decisions about machine translation in your Korean translation projects.
Overview of Korean machine translation
Evolution of the technology
Machine translation between Korean and English has been available for a while. I remember purchasing a rules-based MT software package about twenty years ago. I hoped it would help me in my Korean translation work. Unfortunately, the output was horrible. In fact, the translations from the software were unusable in any practical sense.
However, machine translation technology has evolved. The machine translation engines (meaning, the software algorithms that produce machine translation) for the Korean language have improved dramatically. Thanks to Google Translate and other similar online tools (including Bing Microsoft Translator), Korean machine translation is now remarkably accessible and accurate. Today, millions of people use these tools to translate between English and Korean.
As a professional Korean translator, I see machine translation as both a threat and an opportunity. Of course, it is logical to ask: If people can get free Korean machine translation, why would they pay a professional to translate for them? However, free Korean machine translation is often flawed–sometimes critically–and is not a good solution in every translation scenario.
For this reason, machine translation also helps funnel work into the human translation ecosystem. Through the free translation tools, consumers of information now have access to vast new storehouses of knowledge in Korean. Users of Korean machine translation can find and identify content in Korean that they need to understand. Since machine translation cannot provide consistently dependable output, consumers often turn to human translators for high-quality translation of the most important content.
Thus, it is true that the number of words translated by machine is increasing exponentially. But the volume of work handled by professional human translators also continues to rise, albeit more slowly. So far, Korean translation performed by humans is still in high demand.
Predictions for the future
Still, the trajectory of Korean machine translation productivity has been steep. Many observers ask when human translators will become obsolete. This question has been around for decades, with the date of full computerization continuously seeming to be a few years in the future. Thus, it always appears we are getting closer to universally usable machine translation, but we never quite get there.
In spite of the improvements in Korean machine translation, there are probably limits to how “perfect” Korean translation can get in the foreseeable future. This is because the current technology does not involve any machine understanding of the underlying meanings. In many cases, the translations from the computer are serendipitously accurate; in others, they miss the point entirely or in part. And it’s not always obvious which translations are wrong and which are right without reviewing carefully. Furthermore, within my highest quality workflows, I have found that machine translation is helpful only on the margins. Frequently, it is little more than a distraction.
Fundamental limitations in current computer algorithms are the reason human translation will be with us for a while. Human translators will continue to work unaided by machine translation on some material. They will also edit machine translation output to improve MT translation quality. By integrating the human and machine translation workflows, translation professionals are finding ways to meet market needs at all levels of the quality, budget, and schedule spectrum.
Approaches to Korean machine translation
Not all content is important enough that its translation requires a human translator. At the same time, some human translators produce work that is barely distinguishable from machine translation. Further, free machine translation does not translate well enough for many purposes. For these reasons, the choice of whether to machine-translate a Korean text, or whether to have a professional Korean translator translate it is not always easy to make. It also does not have to be a binary decision. Today, the boundaries between man and machine are blurring. To maximize value, an astute buyer of Korean translation services must understand and be able to weigh the pros and cons of each approach.
1. Raw machine translation
Machine translation engines do not actually “understand” anything at all. Even with the latest MT technology iteration (referred to as NMT, or neural machine translation), the software doesn’t know what the words mean. It doesn’t conceptualize the grammar, and it doesn’t have any idea what the message of the text is. Instead, Korean machine translation simply provides a target-language output most closely related to the associations it has formed in its index for units of data in the source text. (“Source” refers to the language the MT engine is translating from. “Target” refers to the language it is translating into.)
Here are a couple of tips if you want your source text to translate well in a machine translation environment. Use sentence phrasings and terms that are most likely to be associated with the right target-language translations. Avoid figures of speech, complex sentence structures, and unusual and ambiguous words and phrases. These simple steps will help you machine-translate your message in a way that your readers will understand.
However, an MT user only sometimes has the luxury of crafting the message first before translation. In most cases, consumers of machine translation want to understand materials in another language that they cannot edit first. Unfortunately, content that hasn’t been “dumbed down” does not always translate well through a machine translation program.
For this reason, as a reader of machine translation, you must temper your expectations. Remember that machine translation output is only an approximation of the source text meaning. If you’re lucky, it may be an adequate translation. It may even look like an excellent translation, but be wrong. In other cases, grammatical errors and ambiguous phrasings will obscure the meaning.
Also, don’t forget that the companies that design these machine translation tools are continuously working to improve their engines. One way of doing this is to process and analyze the texts that are inputted into their engines. On a practical level, the concerns about confidentiality seem overblown. Text that enters a machine translation engine is chopped up and analyzed in ways that make it unidentifiable for later use.
However, some confidentiality regulations in today’s medical, legal, and corporate worlds do not take this into account. Many policy manuals prohibit the use of free machine translation tools because of the way the engines process the information. Fortunately, these same machine translation tools also provide paid versions of their services. The paid engines do not save the data or otherwise process it back into the index. Therefore, if confidentiality is an issue, you should purchase the paid version. Of course, this will increase your costs. However, if confidentiality is important for legal and other policy reasons, then make sure you follow the rules.
2. Trained machine translation
One reason tools like Google Translate fail to deliver good translations every time is that the machine translation engines are asked to translate many different kinds of content without any means of knowing or controlling the context of that information. Data from many sources and subjects interferes in the machine translation algorithms and degrades the quality of the output.
To overcome this problem, specialists in machine translation have developed techniques to create engines for specific types of content. “Engine training” is done using large amounts of previously human-translated content of similar subject matter and style. This data is fed into the translation engine, which then crunches the data. Such a specialized engine can then be used to produce relatively high-quality machine translation within that same specialized field. Getting such an engine set up can be expensive. However, with enough ongoing and focused content for translation, engine training can be an investment worth making.
3. Post-edited machine translation
Even with trained machine translation engines, the output is rarely perfect. Therefore, another workflow is often added to improve the machine translation further. Post-edited machine translation (PEMT) is a growing area within the translation industry, including within the field of Korean translation. With this approach, human translators take translations produced by either a non-trained Korean machine translation tool (such as Google Translate) or a trained proprietary Korean machine translation engine and edit it to a target level.
Post-edited machine translation has become an entry-level task for new professional human translators because of the low rates and quality targeted by the standard PEMT workflow. As a consumer, you can cut your costs for translation by utilizing such an approach. In fact, a good post-edit machine translation editor should be able to improve a machine translation significantly. However, the PEMT output is still often not dependable from a quality standpoint, especially if it isn’t produced from a trained engine.
In fact, low-quality providers of so-called human translation services occasionally provide translations that are difficult to distinguish from post-edited machine translation. This may explain why “human Korean translation” is often marketed at rates on par with PEMT. Indeed, the advent of the post-edited machine translation workflow makes it harder than ever for consumers of translation services to know what has gone into the work they pay for. When you buy cut-rate translation services, you must hold your language provider responsible for the agreed quality of work. If you can’t do this, then you are at risk of overpaying for inferior translation deliverables.
4. Hybrid Korean translation
Hybrid Korean translation represents a higher level of integration between machine and human translation. This approach doesn’t attempt to build an engine or merely “edit up” the machine translation output. Instead, the focus is on the translator’s skill in referring to the output from one or more machine translation engines within an advanced CAT-tool environment to produce a correct and acceptable human-directed final translation.
A Korean hybrid translation still isn’t intended to be a polished document. However, the translation service provider works with skilled human translators and provides assurances about the correctness of the output. Because adoption of machine translation is held back by both known issues (awkward and clearly wrong meanings) and unknown pitfalls (translations that look right but are actually wrong), hybrid translation can bridge the gap between translation mistrust and trust. This greatly reduces the potential for poor business decisions caused by faulty machine translation output.
The distinguishing factor between post-edited and hybrid machine translation is the extent to which the process is human-directed. Because of the human-led workflow, the hybrid translation provider stands behind the correctness of the translation. (Hybrid Korean translation is also a service that I offer.)
5. Human translation augmentation
Your last option for translating Korean content is, of course, to hire a professional human translator for the whole thing. I’ve discussed at length on this website about best-practice Korean translation qualities and work traits of the best Korean translators. Because best practices cost money, it is simply impossible to deliver the very best work cheaply.
Still, many online translation mills advertise cheap prices for human translation and use entry-level translators at rates that don’t incentivize good work. Keep in mind that sloppy human translation is not necessarily better than machine translation or PEMT. In fact, cheap “human” translation may be disguised PEMT.
On the other hand, don’t think that human translation means working with a pen and paper or a typewriter! Today, the very best human Korean translators employ advanced technologies, including machine translation to a limited extent. Machine translation is only one of the tools grouped under the broader category of computer-aided translation (CAT). These tools may include a machine translation component, but they also provide a workspace for efficient human translation. This workspace directs the translator’s attention to terminology, quality-assurance, and consistency elements that increase efficiency and quality.
The professional translation community is engaged in an ongoing and lively discussion about machine translation. A major area of focus is on how best to integrate machine translation into the workflow. Some prognosticators see great potential here for improving translation output quality and efficiency. On the other hand, I have found that machine translation is helpful in a hybrid translation approach. But in my experience, Korean machine translation is not useful on most projects where I am targeting my highest quality work.
Machine translation is not always helpful–and is sometimes even a hindrance–because of information overload and distraction. Machine translation produces attention overhead by putting more information in front of the translator. If the machine translation isn’t very good, it bogs down the translator’s work when aiming for the best possible output.
Let me explain my reasoning logically: Even if the machine suggests a great translation, I still have to go through the text word-for-word to understand the source first. I then have to compare my understanding against the machine translation output. If the machine translation is perfect, I can move on. But I have still already invested a significant effort in this validation step.
On the other hand, I usually see elements of the machine translation output that I can improve. Thinking through the editing options and then making those edits are yet two more drains on my time and effort. In a second-to-worst-case scenario, I realize right away that the machine-translated text is terrible. Then I toss the entire machine-translated sentence and retranslate the text from scratch. In this case, the machine-translated output has only weighed me down a bit. But in a worst-case scenario, I waste time trying to improve the translation little by little. Sometimes I only realize later that I could have done it faster without the machine translation. If I’m aiming for a “good enough” output (i.e. hybrid Korean translation described above), then this unfortunate outcome only happens rarely.
With hybrid machine translation, my goal is to be good enough, not perfect. However, when I’m aiming for my best, I run every word and phrase through my through processes to try to improve. This entire process can take significantly longer than a regular translation job. It can also result in inferior work and is especially prone to error.
Accordingly, I find that if the original machine translation is not already excellent (and it seldom is), the use of machine translation is mainly a distraction when I’m aiming for perfection. For this reason, I don’t use machine translation in my full-service workflow. I only use machine translation for my Korean hybrid translation approach, where it can significantly raise efficiency.
Korean machine translation is a wonderful tool that can be extremely useful if applied properly. This article has discussed five key ways that machine translation is being used in Korean translation workflows today. In some workflows, the technology dominates. But in others, human translators take advantage of machine translation to improve productivity and high quality. Following best practices in machine translation can help you get more value from your Korean translation projects.