Autobiographical

Once again apologies for readers and subscribers. Here I continue on what is a key issue for my profession. I also mention again the offer for investors in Kazakhstan and art fans.

The risks posed by AI today for professional translators, agencies and clients mirror similar issues raised by previous disruptive technologies: machine translation and translation memory software. I will consider here the evolution of the industry in one country, the challenges that it created and the opportunities going forward for everyone ready to forsake short-term profit for a sustainable, long-term business.

Back in the early 2000s in Russia translation software providers, such as Trados, for example, sought to persuade companies to acquire their software by offering free training to the translators that they employed. Naturally, the sale was predicated on two differing strategies depending on the audience.

The providers promised the companies more consistency, greater accuracy, and a subsequent reduction in translation costs, as it would be possible, as it was euphemistically put,  to “optimise” the number of translators employed, supplemented where necessary by assistants. Thanks to the in-house translation memory, subsequent bilingual deliverables would take less time and require less experienced staff.

At the same time, the providers sought to convince translators that the software would free them from having repetitive translations, standardise their translations, and improve consistency, thereby enabling them to take more time on their work thanks to the software aid.

In reality, texts still differed significantly, but the translators found that they were expected to translate far more than in the past, as it was assumed by management that the translation memory made their work easier. 

However, if you are able to translate rapidly, (on the order of 5,000 – 10,000 words a day), the use of translation software inevitably slows you down due to the need to laboriously enter the text into the program. Later on, the software providers targeted translation agencies, citing consistency, but above all efficiency savings at the translator’s expense. I myself opted not to choose the software option and would then be required by agencies that operated such systems to agree to lower pay for repetitive text (down to zero) and a reduction of the actual paid word count by over 50% in the case of so-called “fuzzy” text, even where there were material differences between instances of “repeat text”, which required careful consideration. Shortly afterwards I stopped working for them.

It struck me that there was no tangible long-term benefit for the translator. A number of the agencies spearheading the use of such software closed down, as the final product was far from perfect, due in part no doubt to the quality of the work of the translators who were persuaded to agree to extremely low rates. 

And then along came machine translation, soon superseded by the allegedly superior AI versions. The translator is reduced to the role of a glorified editor of machines spurting out gibberish which in a number of cases has to be “corrected”, in other words re-translated by a translator paid for editing, which is generally half the rate for translation.However, the problems of relying on software have been building up in industry over the past two decades. I will provide here a couple of real-life examples of where it can all go wrong. A couple of colleagues used at one point Trados software as demanded by an advisory firm, both while employed in-house and subsequently as freelancers. Initially, they would be asked to translate audit reports on listed companies from Russian into English using the software. After a couple of years, the reports were no longer assigned to the translators, as they were handled in-house by the respective industry departments, primarily by non-native speakers, in particular audit assistants newly hired employees whose English was far from perfect and who were usually under significant time pressure. On a couple of occasions, I was contacted by a concerned audit partner who requested a translation review of the current version of audit reports on two clients in Russian and English.

I did a painstaking check that took in each case as long as if I had translated the texts myself from scratch. In both instances the two allegedly identical Russian and English versions differed by 20 to 30 per cent. Here the issue was resolved positively, as a conscientious partner had had the good sense to contact a professional translator. 

However, if such checks aren’t requested, a number of risks can arise. In the case of an audit and advisory firm, this concerns firstly reputation risk and fees. The client expects professional work, notably that deliverables will be accurate and consistent, and chooses a provider that is respected by industry competitors, shareholders and potential investors and stock markets. Any shoddy work is likely to lead the client to look elsewhere, notify others of the experience, including the media, and also demand a refund. Concerns about translation quality may trigger worries about the underlying audit work.

However, the problem doesn’t stop there. Let us assume that a client is seeking investors or plans to raise capital. Conflicting versions of financial statements and audit reports will not inspire confidence. Even worse, an investor or lender might at some stage sue the client on the grounds that it was misled by information that differed from one version to another. The client in turn might well take the audit firm to court. 

Such reputational and litigation risks are directly attributable to the use of translation software where the required oversight is missing. And translation error can prove costly.

In 2014 I was appointed as an expert in a dispute involving claims of more than USD 100 million. I was asked as the expert translation witness to provide an expert opinion (expert review) on the meaning of certain terms where three different Russian words had been translated as value, even though the common translations were price, value and amount. Some form of software had clearly been used to ensure this consistency! These errors were of particular concern when the text concerned the share price on a particular date rather than the nominal value of the shares. As a result of the expert review, the court proceedings were cancelled at the last minute.. I had effectively saved the client USD 100 million.

So it can actually pay in the long run to turn to professional translators rather than relying on software or machine translation. Other risks related to use of the latter are also now emerging. As in AI, machines must be trained on an initial corpus, which is obtained from an unknown source by a law firm or agency. There is no clarity as to the actual source, and downloading a company’s data or translator’s data in the initial data used by the machine is potentially unlawful. Litigation could also arise due to the use of language which is indicative of set phrases or branding terminology used by specific companies.

However, the key issue today is one of liability. Translators take out limited liability insurance, which as a rule covers them for translation, and not editing. So what happens if there are errors after a post-machine translation edit? And how is the liability assigned if a number of translators were used in a legal contract/case or valuation assignment (a vendor due diligence, for example)? The deadlines for such work are often tight, and different texts are distributed between a pool of translators. 

If the translator cannot be held liable, then does this mean that the agency is to blame? Or what if the errors were due to the initial data used to train the machine? Or is the law firm to blame, if some amendments were stylistic changes made by a lawyer who didn’t like the way something had been translated or edited?  And when were the errors initially made in court cases which can go on for years?

Clearly, there is a problem in the use of any type of translation software if the professional translator is accorded a nominal role, paid a minimum amount, and expected to work rapidly on assignments where it is virtually impossible to obtain the necessary insurance and where in actual fact the agencies should be responsible for such cover.

Notwithstanding the risks, there is an opportunity: audit, advisory and law firms and professional agencies will eventually be required to draw on the expertise of professional translators to translate the initial texts and the revised versions and pay them equitably. This should not be a problem, all the more so as in a number of jurisdictions such work is tax-deductible. There is clearly also a need for companies to request translation reviews by professional translators of previous texts, unless they want to scrimp today while potentially facing litigation tomorrow.

Other opportunities will follow. Let me provide one more example. There are areas where AI and machine translation cannot help: intuitive understanding of the underlying risks in proposals and deliverables. Seminars and conferences on the potential of professional translators to enhance the quality of texts, promptly identify and notify clients of potential pitfalls in the original text, including excessive assurance included in the deliverable that could trigger lawsuits from their counterparties, and proposals on how to avoid such risks, would boost the relationship between industry professionals and their clientele. Better networking between translators in different languages for the delivery of multiple language services to public sectors, where perception and concern over the impact of the text on the final recipient and non-disclosure are prioritised, for example, the National Health Service in the UK for medical reports.

I have been lucky to have good agencies and have built up a good relationship with them. They know that they can count on me to deliver a professional translation every time and to adhere to non-disclosure undertakings, to point out where there are errors and assist where necessary round the clock.

Here I turn to potential investors or art lovers. I sponsored an artist from Kazakhstan for many years in the 1990s. His name was Manas Kisamedinov. He became a close friend. Unfortunately he passed away at the pinnacle of his career from a drug overdose, dying far too early, as did his late a father, the famous graphic artist Makum Kisamedinov. In return I would receive works of art. I show here and in the next article paintings that appeared in previous blog entries. More paintings and graphic works will appear in later entries. As time passes, I am now ready to pass them on for the right price naturally.

Painting by Kazakh artist Manas Kisamedinov, 1992.