The legal translation industry is currently grappling with a cognitive bias known as anchoring bias. This phenomenon has led to the widespread use of unreliable legal translation dictionaries and parallel corpora among industry translators; a practice that encourages translators to recycle incorrect past answers without even considering whether they were wrong. In this article I’ll explain why anchoring bias is such a disaster for translators and how the legal translation process can avoid being negatively affected by anchoring.
Dictionary Anchoring Problems
A particularly obvious place where anchoring causes problems for translation is the use of the humble bilingual dictionary. It’s no secret that most translators and other “bilingual” professionals across the globe today are not all that fluent in their second language. Conversational, yes, but they are also often extremely limited by the range of topics about which they can converse. How do bilingual professionals compensate for these issues? Generally, they heavily rely on dictionaries and similar resources. Reliance on these tools, however, also tends to be very blind; unlike legal manuals or medical references, dictionaries do not include any citations to evidence to support their assertions. Nevertheless, even doctors and lawyers often quickly forget their evidence-based training and start consuming dictionary entries whenever attempting a translation, not unlike how InfoWars readers consume fake news. The practice cuts costs for translators because they need not learn anything, but results in inaccurate translations.
Rather than a flaw on the part of the professionals, this bizarre behavior is primarily a product of the educational system. Foreign language educators are still actively conditioning learners to blindly accept anything written in a bilingual dictionary, despite numerous thought leaders openly opposing the practice. In practice, if you take a doctor or lawyer out of professional practice and put them into translation, the evidence-based approaches they learned in medical school and law school go out the window and they instead default to what they learned in foreign language education. Thus, it takes a lot of discipline to realize bilingual dictionaries are written by people who never intended for them to be treated as authoritative, despite this being what people actually wind up doing.
Nonetheless, even if a translator realizes that bilingual dictionaries should not be blindly relied upon, their brainwashing effect will still likely take hold of the translator due to anchoring bias. This is due to both short-term memory anchoring and a phenomenon called fossilization. Short-term memory based anchoring is the typical result of a bad translation practice; when a new word is encountered in a translation text, a translator will typically consult a bilingual dictionary before referring to authentic texts. For example, a science translator working with a study that cites several English-language articles packed with all the relevant vocabulary, is still more likely to refer to a bilingual dictionary first. Even if the translator does look at prior research, they will generally miss or ignore existing articles.
Fossilization anchoring, on the other hand, is much more deadly and has resulted in the Chinglish phenomenon—it is capable of crippling translation careers. Fossilization tends to be most serious in East Asian countries where memorization has long been the prevailing educational strategy. Professors studying the issue in China characterize fossilization as an observed plateau in an individual’s second language competence, in which the incorrect use of language becomes habit. Fossilization has a major effect on word usage in Chinese to English translation, since memorization has taught learners to follow perfect grammatical forms but incorrectly assume that pairs of Chinese-English words are identical in all contexts. Students often come away from the memorization process with a false belief in the absolute equivalence of the word pairs, a belief unsupported by scientists and linguistic evidence.
As a result, most translators working between Chinese and English have an ability to instantly recall an answer that is incorrect most of the time. Moreover, following that instant recall, the translator tends to latch on to the incorrect answer and is less willing to consider alternatives that may be correct. The translation process used is also extremely fast, results in highly consistent results, and accounts for the rock-bottom translation rates earned by translators in China. However, this process is seriously harmful to any business relying on it.
Advanced Anchoring Techniques
An alternative to the bilingual dictionaries much loved by translators, at least according to those in China who I’ve spoken to, is parallel corpora. These parallel corpora are essentially databases that contain completed translations and show existing solutions. A problem with parallel corpora is that their translations are unverified and lack any evidential basis. Worse yet, the same translators completing the translations in the parallel corpora also cheat and frequently label machine translations as human translations to cut costs. Thus, when a translator uses a parallel corpora, a common result is that they end up relying on someone else’s flawed answer, oftentimes itself a machine translation. Even in the best circumstances, translators uploading translations to parallel corpora tend to be low-paid and low-skilled since parallel corpora companies generally do not generate enough revenue to use specialist skilled translators for that particular purpose. As a result, your anchoring point will be that of a mediocre translator, and using such resources will lock the translator’s performance into a cycle of mediocrity.
Finally, a common and highly pernicious anchor point used by translators today is machine translation. Yes, Google Translate is successfully rewriting the brains of translators who use it! A translator who routinely works with machine translation will actually anchor on to the translation result even if they are revising it. As a result, common machine translation users wind up having their translation output conform to the machine translation result. Marketers know this, and routinely set up clever experiments where they document low-cost translators’ opinions of their machine translator’s output — proudly proclaiming that these translators agree that their machine’s output is highly accurate. In reality, these translators have simply been brainwashed through their addiction to machine translation tools! Translators doing post-editing work can avoid this problem and still work with machine translation, especially if they learn to treat the machine output as a foreign language or, more accurately, an “interlanguage.”
Highly specialized fields are extremely susceptible to anchoring. Previously, I helped revise translations related to PRC eminent domain law done by an in-house team. In my professional experience, I had worked on eminent domain and condemnation actions and done work on the City Hall side. The in-house team, using conventional resources, produced a terminology describing eminent domain actions using a vocabulary totally unlike anything I’d ever encountered in condemnations before. Moreover, the vocabulary was totally alien to what condemnation statutes across the world actually say, and using such a translation would have been highly detrimental to the client. How did the eminent domain translators reach this result? In short, their entire translation practice revolved around pre-packaged answers that were simply accumulations of prior biases, rather than facts and reality. If you look back at the long chain of people copying other people over several decades, you will ultimately find that the first translator to come up with the translation, too, was simply copying other people.
Breaking Free of the Anchor
There are two ways to break free of the anchor; the first is to rely on genuine language acquisition models and the second is to apply evidence-based translation methodologies. Genuine language acquisition models for translators would mean that translators would learn their second language by acquiring it like a native speaker would, not by memorizing or imitating artificial patterns. For example, consider how US-educated JDs from China learn legal English versus how legal translators in China learn legal English. The legal translator in China looks closely at dictionaries showing word equivalences or uses published books that have a parallel corpora showing equivalence between sample Chinese sentences and its English translations. The US-educated JD holder, on the other hand, will have read a case book entirely in English, listened to law lectures entirely in English, and collaborated with colleagues, again, entirely in English. These people tend to learn superior second language skills, far superior to legal translators. In fact, legal translators in China are poorly trusted due to their lack of competence. The difference between the two cohorts is that the translators learned their second language by defining it in terms of their first language. The lawyers learned their second language as an independent body of knowledge. Translators can also acquire language in a general manner by using the same methods native-speaking professionals use to acquire language, as the mainstream method is actually pseudoscience.
Second, apply evidence-based translation methodologies to outsmart the anchoring bias. What most translators tend to do, which I see using screen sharing technology in workshops, is immediately resort to bilingual dictionaries for a quick answer when given any translation question. Translators following this same approach need to stop doing that and instead look at evidence of how people working in the specific field actually communicate. For instance, if you are a Chinese-English translator working on semiconductor intellectual property right infringement, your best source of knowledge about how semiconductor engineers communicate in the United States would be patents and US research publications. Most people don’t want to do this because reading all that technical information could take hours, whereas a looking up a word in a dictionary takes seconds. They think this is more efficient, but it’s not, because the dictionary produces zero value whereas reading the industry material has a lot of value. A wrong answer has no value, no matter whether you can swindle a client into paying for it, and clients that rely on this defective work will ultimately fail in their enterprises.
