Companies are not doing adequate cost/benefit analysis on business translations, resulting in a great deal of value left on the table. Most businesses seem to be only assessing the costs and benefits of translations within the media sector, which lacks technical terminology and uses an elementary-grade reading level. However, legal translations, especially corporate and patent documents, include substantial complex technical terminology and are targeted at much more sophisticated readers. Therefore, the value of business law translations should be assessed independently of media and localization translation, using a process that centers on the role of the professional.
Effect of Translation
Translations used by business professionals can significantly impact their company’s performance, but quantifying their effect can be challenging. After all, the translation is just one part of the “machine” that is a business operation, which makes it hard to separate its value. As it turns out, the machine metaphor is highly accurate and the effect of translation looks a lot like modern air conditioning and heating systems, which are affected by a value known as the Coefficient of Performance. In a heat pump air conditioner system, the coefficient of performance is a multiplier that tells the user how much heating or cooling is generated from a specific wattage input. For example, with a coefficient of performance of “5,” the heat pump takes 1,000 watts of electricity and converts it to 5,000 watts of heating. If the coefficient of performance is “1,” then you would need to use 5,000 watts of electricity to generate 5,000 watts of heating, which is the result you’d get from buying a bunch of space heaters. Essentially, the coefficient of performance takes your existing input and multiplies it to result in a much better output.
In business translations, the coefficient of performance likewise takes inputs in the form of business spending, which could be anything from advertisements to lawyers, and multiplies it to generate economic value. For example, imagine a local office is illegally selling EAR-regulated materials used in China to Iran or North Korea, and these transactions are not formally accounted for. Local staff misses the issue because they don’t really understand the EAR regulation framework or don’t consider it very important. In this particular case, the company would be fined $500,000 when caught by US authorities (though damages in some cases have gone into the billions). Therefore, when caught early, the value of an attorney’s compliance review to the company is $500,000. However, the attorney has to rely on translations, and there are several different options for obtaining this service. One option is to have the translation done by a lawyer-linguist, which will give the attorney a 100% chance of catching the issue. Another is to hire ILR4-graded ex-FBI linguists from a premium boutique translation agency, yet this only gives the attorney a 30% chance of catching the issue (the above statistics and cases are adapted from several real-life cases and federal government reports, discussed below). In this case, we could say the coefficient of performance for the translation is 1.00 or .300 if using batting-average type numbers. So, the overall marginal value of using the European Union’s lawyer-linguist qualification grade above the FBI’s ILR4 grade is $350,000. That’s not to knock on the FBI’s linguists, however, who would still save the company $150,000 over what a low-cost translation agency would.
Thus, when engaging a translator, a cost-benefit analysis should be done to multiply the coefficient of performance against the other inputs the company is putting into action. Neglecting to do so can lead to extraordinary costs. For example, I am aware of a billionaire who asked his law firm to hire the cheapest translators possible to work on translations used in a business crime trial in the United States. The law firm subsequently looked up translation agencies on Google and asked for a bunch of quotes. Who they ended up hiring turned out to be an agency where translators just copied and pasted machine-translated text while pretending it was human-made. Where is this billionaire now? In a federal medium security prison and probably sick with chronic COVID. If serious money—or your personal liberty—is at stake, then saving a trivial amount of money on translations by opting for the cheapest people possible will undoubtedly do more harm than good.
Value-blindness Causes Fraud
Ignoring cost/benefit analysis and focusing solely on costs is not a phenomenon isolated to translation but affects a wide variety of business services integral to an operation not typically included in a company’s cost of goods sold analysis. However, translation is unique in that big translation companies routinely hire untrained casual workers, and outright faking work is something that’s very easy for unethical linguists to get away with. For example, Multilingual Magazine reported that the British court system prosecuted a court interpreter for having impersonators of a verified translator serve in court. A “top five” UK legal translation company did the initial vetting and hiring, but zero subsequent quality control or auditing. It was only after UK court staff realized that the person’s voice was different at different times that the courts became aware of the fraud. This is a serious matter because the affected community was not only deprived of their legal rights, but it also created a perception of racial discrimination against this community that echoed all the way back to their countries of origin. The case is very similar to USA v. Anwari, where prosecutors discovered that a language services company had been staffing “fake” interpreters. Can we expect the involved language services vendor to be more careful in the future? In my view, that’s highly unlikely. If they didn’t oversee their staff a decade ago, a case like this probably wouldn’t cause them to start.
This is a bizarre and dangerous business model and a stark difference from how corporations manage the quality of legal services. Unlike interpreters, “imposter” attorneys are a very rare phenomenon among US corporations, and the relatively high level of assurance derives from how legal work is structured and delegated. A typical corporation has a legal department that employs corporate counsel, often a general counsel. These are not only licensed practitioners, but even newer hires can typically demonstrate expertise in their field of practice. They delegate work to outside law firms, whose attorneys are all licensed and whose partners generally have board-certified qualifications. The outside lawyers are usually better than the in-house counsel in their particular fields, but the in-house counsel can tell that these people are indeed qualified.
Compare this to the disasters described in cases like USA v. Anwari or USA v. Meng Wanzhou. In these cases, the corporations lacked in-house teams of translators specializing in the field. In general, even if a corporation has in-house linguist staff, they are almost always uncertified translators with no industry recognition. Even in a large corporation or law firm, it’s normal for someone with no industry certification or expertise to go out and hire a translation company—the one named in the “imposter” interpreter case in the UK has many big law firm clients—and expect those translation companies to be staffing expert linguists. That is a big lie, however, as virtually no major translation company has certified linguists among its employees. It’s extremely rare that even companies with hundreds of millions of dollars in revenue would have project managers who can speak the languages they’re working with—not even people with very minimal training. Thus, these sorts of scams are very easy to pull off for government, law firm, or major corporation projects.
The main motivator that drives these changes in a corporation for its legal work, and not linguistic work, is really due to the relatively high quality of leadership in non-translation fields in corporations. A general counsel knows and understands legal work very well and attaches a great deal of value to that work. The law firms general counsel hire also understand legal work very well and greatly value that work. Most corporations and governments, on the other hand, aren’t employing people who value linguistic work and the translation companies they hire don’t employ people who value or really even understand linguistic work either. If companies do hire linguists, they tend to hire non-certified linguists to cut costs, that is, people with relatively little recognition from industry peers. Thus, when they go about having translation work done, nobody is thinking about the value of what is being purchased. The supply chain management system is focused pretty squarely on minimizing prices, which is how we wound up with bizarre situations like court interpreters in New York being paid less than typists. This is pretty demoralizing for the actual language service workers and causes them to feel that their work has no value or meaning other insofar as it fulfills a client’s order. Thus, translators faking work by using AI, and interpreters faking work by using imposters, are extremely rampant.
The 70% Rule
I propose that, for a typical business translation project, reliance on “typical” translation service approaches will result in the entire project failing about 70% of the time. This number appears quite frequently in major corporate and legal project data. For example, a survey of Netflix subtitling found that 70% of viewers of subtitled programs decided to turn the program off due to the low quality of the subtitles, whereas shows in native languages are shut off due to incomprehensibility issues less than 0.1% of the time. About 70% of China initiative cases failed due to bungling of evidentiary issues, whereas fewer than 1% of federal prosecutions where foreign languages are uninvolved fail. The China Initiative is a fairly interesting case in that its linguists are generally uncertified translators vetted and verified internally by the agency, but they tend to be the bottom end of what the federal system gets and are at the bottom of the pay scale, with the use of part-time hourly linguists also being very common. Native Chinese criminal defense attorneys in the US defending against China Initiative prosecutions described the FBI linguists used to me as “flagrantly incompetent,” to the point of putting innocent people in jail — which is what the DOJ statistics later indicate. However, incompetent is still better than fake, so this is about twice as good as what a typical translation intermediary would offer, keeping in mind the UK courts’ fraud case would never happen at the FBI.
There is good reason to think that the FBI would deliberately hire poorly qualified linguists: to save money. You can see the FBI doing this in routine equipment purchases. In a May 2022 scandal, the FBI and DHS confirmed that they were purchasing Chinese-made drones to defend American national security from foreign threats. This was shortly before the Russian military praised the effectiveness of these same drones from the same manufacturer for use in the Ukraine war. Thus, the China Initiative—a huge catastrophe—provides a great benchmark for the coefficient of performance provided by mid-tier uncertified linguists: wiping out 70% of the value otherwise expected. Big law firms in the US probably see closer to a 90% wipe when typical translation intermediary linguists (who love fraud) are used for legal and financial matters, and 70% for media and marketing-related translations, where the failure rate will be much lower due to the relatively low challenge of these fields. While a significant amount of data analysis and research is needed to quantify the value of these coefficients with high specificity, these numbers are a reasonably reliable estimate of how a translation will affect a company.
Translation Pricing Rules of Thumb
An easy rule of thumb can be applied when determining what an appropriate budget for a legal translation should be for typical matters. Generally, a company should assess the expected legal costs for a matter (or part of a matter) where legal translation services will be applied; using about 10% of the budget for service level (non-supply chain) costs is an appropriate rule of thumb in most cases to avoid the investment in legal counsel from going to waste; or 20% where PRC-based (less expensive) legal counsel are used. General counsel are pretty good at determining how much legal spending is appropriate, and the translation requirements are usually closely tied to the legal activity. The term service level refers to the level within the overall organization where services are provided, and translation companies usually do not provide translation services, they outsource the work (the full-time employees are all brokers). So, at the top (multi-lingual vendor) level, supply chain costs typically account for 1/3 of costs, another 1/3 goes to middle-level (single-language vendor) supply chain costs, and only 1/3 or so is spent at the service level, which would triple the needed budget to keep pace with legal requirements.
Here is an example of how this plays out in reality: suppose a law firm has an associate spend one hour summarizing the contents of ten corporate e-mails produced by a document review attorney over the course of a one-week review, with a 0.5% relevance rate and speed of 40 documents per hour (this is for an M&A warranties and representation fraud review, where the responsiveness rate is a lot lower than litigation). The cost here is $5,500 for the review and $250 for the associate, or $5,750 overall. The e-mails average 250 unique words each, or 2,500 words overall. This comes out to $2.30 per word of e-mails found or about $0.23 per word of the documents. On the $1 million company, about $80,000 is clawed back during the second M&A closing, and all for just $5,000. Not a bad investment. Incidentally, these are realistic service-level prices. What most law firms usually do is pay that rate to a multi-lingual vendor. So, in reality, the translation team ultimately hired is getting maybe $0.06 per word—and that means using artificial intelligence to fake it and act like a human translated it. In that case, the $80,000 is effectively flushed down the toilet.
Another relevant scenario is in the litigation context, where massive numbers of responsive but minimally important documents are called in for translation and also are being produced at the lowest cost realistically possible, say $3 for each bilingual document reviewed or about $0.012 per word out of 30 million words. That’s still $360,000. In this case, if you drop 30 million words on a translation vendor, the only way to fill the project would be to haphazardly contact massive numbers of minimally trained casual workers. The problem with this approach is that around the year 2020, Google Translate actually began to surpass the translation competence of minimally trained, casual workers. You could spend $3 million to translate that volume, or even just $1 million to go with some MTPE workflow. In either case, the output results from such a huge volume would still be worse than a machine.
In this case, since each document in the massive list of documents will have been coded as being relatively low-impact or less important, using a trained machine translation engine might be a more logical result. There are two main components of this approach: first, extracting common words, phrases, and duplicate segments, and, second, training a machine translation engine to follow a custom approach defined by expert translators.
The extraction approach has a huge impact. Statistically, a 30 million word litigation case will have documents that, by word count, are 40% comprised of repeating words and phrases used within the industry, and when reduced to list form the total is usually less than 3,000 items. Thus, 1.6 million words done by 80 people can be reduced to 15,000 words done by a single expert. Second, the machine translation engine can be trained to follow repeating patterns with the unique 60% word count; about 40% of this word count will actually be fundamental English grammar, with the remaining 20% being language structures particular to the related parties. Finally, the law firm would review the documents for any problems or illogical output (which it should do with all translations regardless) and flag problems for retranslation. Using hourly billings, this approach can reduce costs by over 90% with improved quality.
Nonetheless, big translation companies basically never follow this approach, and for several reasons. An overarching reason is that by having a human do the same repetitive tasks over and over again, the company can charge a lot more money. Additionally, translation companies are ineligible for ISO certification in legal translation; thus, their service warranty actually requires them to follow an approach used for non-legal materials. Finally, big translation companies do not have the capability to hire and work with experts; the management process focuses on media localization, which forces the production of bell-curve-shaped knowledge competencies that favor “layperson” level knowledge, not professional expertise, therefore they also generally lack these capabilities. An expert-driven firm like CBL Translations, however, does have these capabilities.
