Third-language Translators Doing Japanese-to-English TranslationHumans or machines, they demonstrate that you get what you pay for.

(January 28, 2019)

Executive Summary: Japanese-to-English translators with neither Japanese nor English as a native language continue to appear useful because of their low cost, but this is a delusion. And machine translation is still only an extremely poor replacement for a human translator producing high-quality translations. If a high-quality Japanese-to-English translation is necessary, it is still cheaper and quicker to have a human with native English ability translate from scratch.

If you are wondering what a third-language translator is, it is a person translating between two languages, neither of which is their native language. Consider the case of a native speaker of French translating a German text into Russian. You might think that this is a rare occurrence, and you would be correct for those languages. Recently, however, huge amounts of Japanese-to-English translation are done by a process similar to that. If you now imagine that the French translator has never lived and will never live for any significant time (or perhaps even visit) Germany or Russia, you have a situation even more similar to what goes on when a Japanese document is translated to English in places such as China.

The Often-violated Golden RuleThe Prevailing Wisdom Fails to Prevail

The prevailing wisdom among translators—and it prevails because it is true—is that a translator should translate into their her native language. A native speaker of French should not translate into Russian from German (or even from their native French into German). Claims of having more than one native language are almost always false or made by people without the ability to judge native skill. Although violators of this into-native-language golden rule of translation are not going to be rewarded with fame or fortune, because of the huge demand for quickly and cheaply (even if poorly) done Japanese-to-English translation, translators breaking the rule can be rewarded with meaningful income, at least in their country of residence.

The Old Approach Lives On

Setting aside third-language translators for a moment, the idea of having a person native to the source language (Japanese for Japanese-to-English translation) but with only non-native ability in the target language (English in Japanese-to-English translation) is nothing new; it has been the standard approach in Japanese-to-English translation for as long as English has been important to Japan for communicating with non-Japanese speakers. The reasons for this include cost, but there is also the long-standing misconception (by some Japanese people) that only a Japanese person could understand the Japanese-language source text. Many times, but not always, an unfortunate native English speaker (often not a translator) is given the job of repairing a poor Japanese-to-English translation, a process which is sometimes strangely characterized as "brushing up" the English produced by a Japanese native speaker.

Because the translation repair personnel are often not translators and sometimes not even able to read the original Japanese text, the repair work can turn a bad translation into a mistranslation, but one that reads deceptively well in English. Several more recent developments have encouraged even further transgressions of the into-native-language rule.

The Huge Demand and Tolerance for Poorly Done Japanese-to-English Translations

As an example, the putatively largest demand sector for Japanese-to-English translation in the US is civil (mostly patent) litigation involving Japanese companies. For people not familiar with the process, the rules for what is called discovery in civil procedure in the US enable a party to litigation to compel an adversary to disclose to them huge volumes of documents. If the disclosing party is a Japanese entity, most of these discovery documents will be in Japanese, leading to a large demand for document review and translation. Because the time and expense of translation (even if done by a large group of very inexpensive translators) would be prohibitive, the usual approach is that of performing a multi-stage document review (triaging) whereby the starting universe of documents—which could be many thousands of pages—is ultimately reduced to a more-reasonable volume to be translated, which can still be very large.

Enter Japanese-to-English Translators in China

Although translators in the US have traditionally done the majority of Japanese-to-English translation of discovery documents, recently, much of the translation of such documents is done in places like China, where both English and Japanese are foreign languages to the translators. They are third-language translators. Our experience with seeing these documents used in depositions demonstrates this trend beyond any reasonable doubt.

In their quest for profits and the need to provide unknowing clients (chiefly in the US) with Japanese-to-English translations that are “good enough,” numerous major US translation brokers serving law firms in the US have taken to using translators in China and even other translation brokers in China. In such cases, a letter certifying translation accuracy is often meaningless, since it is often created and signed by a non-translator working at the broker stating that the translation is accurate. Such a certification is nothing more than an act of faith, driven by the need for the translation broker to stay “afloat at the bottom” of the translation industry ocean by appeasing the client's need for assurance while providing cheap translations. The resulting misunderstandings and translation problems can be both comical and tragic, but sometimes go unnoticed until it is too late.

Chinese is similar to Japanese, right? Wrong.

Although it has borrowed a subset of its writing system from Chinese, the Japanese language itself bears little similarity to the Chinese language, and the two belong to totally difference language groups. The hurdles in asking a native speaker of Chinese to translate from Japanese to English arise because:

  • the Chinese-native translator, for numerous reasons that go beyond the scope of this article, is highly unlikely to ever have an opportunity to experience the Japanese language as it is actually used, outside of a textbook version of the language, and
  • the Chinese-native translator is highly unlikely to acquire native-level writing skills in the target language of English.

But the demands of the litigation process continue, and those demands are lately met by some of the worst Japanese-to-English translations we have seen. Clearly, the average quality of Japanese-to-English translation coming from the bulk translation brokers in China (fed to US clients by similar bulk translation brokers in the US that subcontract to brokers in China) is far below that producible by a qualified native English speaker translating from Japanese.

Enter machine translation, poised at achieving parity with mediocre human translation and capable of breathtaking mistranslation

A new trend in Japanese-to-English translation is the use of MT (machine translation). We see much hype recently about deep learning and artificial intelligence (AI) used in translation, although true (general) AI is not here yet and the AI said to be used in translation is just now poised to achieve parity with mediocre human translators.

Every indication we have seen from Japanese-to-English translations produced by machine translation systems is that machine translation is indeed a potential replacement for incompetent human translators of the type noted above. However, MT systems attempting to provide Japanese-to-English translations, similar to their human counterparts in places such as China, will continue to be hampered by having neither Japanese nor English as their native language. In fact, MT systems have no natural human language as their native language. Additionally, worse than even their incompetent human counterparts, in spite of the AI hype, MT systems cannot actually understand anything in the real world. They might appear to understand the subject matter, but what they are actually doing is attempting to mimic the behavior of a human translator, guessing what a translation should be, based on clues in a text, in combination with learning from a universe of translated samples, rather than from real-world knowledge or understanding. The result is a poor imitation of what a human translator with real-world knowledge and understanding of context can do.

All the above said, it is probably a safe bet that, just as Chinese translators have taken a large amount of discovery document translation work away from US-based human translators, MT, now arguably boasting parity with incompetent translators in China, will take (or is already taking) translation work away from even those third-language translators in China.

The Special and Limited Role of Third-language Translators

In spite of their obvious disadvantages, third-language translators of both the human and machine variety will surely continue to play a role in handling translation tasks for which they can provide translations that are “good enough," when good enough is not that good and the risks of mistranslation are not that high.

As noted above, one such translation task is providing quick and dirty translations of Japanese discovery documents for US litigation. But what happens when the “good enough” provided by third-language translators—human or machine—is not good enough?

Translation Repair is Costly and Often Risky

Translation consumers needing something better than a bad Japanese-to-English translation from China or a machine translation system sometimes attempt repair of the original poorly done translation. This approach is a faulted one for several reasons.

Correction of what is thought to be just bad English can be deceptive. A non-translator given the job of repairing a poorly done translation from either a human or an MT system can fix the English but, in the process, might mask translation problems.

Only a translator can reasonably be expected to succeed at repairing badly done translations. Two of the selling points of machine translation systems are speed and economy. But if the goal is not just a marginally acceptable translation, but rather a high-quality translation, machine translation continues to fail. And these failures are seldom repairable by the low-paid people who are asked for assistance in making a bad translation at least look better. The task almost always requires the intervention of a professional human translator, and that means an expenditure that, if billed correctly by translators who know their worth, can bring the overall cost to at least what the translation cost would have been by going to a professional human translator from the outset. Try as a translator might to repair a translation without retranslating things, it is very often impossible, thereby driving up the cost.

Our recent experience with having been handed machine translations to be used as a "reference" in translating a Japanese document for prosecution purposes in the US tells us that things have not changed. In a recent case, we needed to bill the job at the full translation fee as if they were totally new translations, because a totally new translation is what we provided. This is something that translation consumers need to consider when they find an important document among those good-enough translations that they want to be better than good enough. It will probably be necessary to commission the translation as a new task.