Machine Translation – benefit or risk?

The world is no longer on the verge of a digital revolution – it is underway. Increasingly, we no longer hear about AI itself, but its updates. Since its development, ChatGPT is no longer the only product on the market that is available to users. However, it affects the entire labour market, without exception.

Problems of AI with the law and use in cases

ChatGPT has already been charged with tricking lawyers into including fictitious legal research in a Manhattan court filing. In this case, it is not ChatGPT itself that will suffer the consequences, but the Manhattan lawyers who used it to file a lawsuit against the airline. Effect? The lawsuit included references to previous cases that were entirely created by the chatbot. Artificial intelligence has offered several cases related to air accidents that lawyers have not been able to find with the traditional methods used in the firm. They were also unaware that ChatGPT could fabricate air accident cases. After a careful analysis of the documentation, the judge drew attention to the gibberish, from which nothing resulted, and the presented facts had nothing to do with reality.

Translation errors in working with AI

Not only lawyers, but also translators face a major challenge related to the use of AI. It is already becoming increasingly common for clients to order the MTPE (Machine Translation Post Edition) service, under the mistaken belief that this is merely uploading a file into a secure plug-in to examine the document for consistency. Meanwhile, it is the translator who thoroughly acquaints himself with the content of the document and often thoroughly edits it. Often the correction of such a text resembles the creation of a new one. The portal sprachprofis.de has created a table in which it summarizes the weakest points of AI when generating a translation. They mentioned, among others, the illusory impression of the right content in the translation, which is not consistent with the source text, the wrong interpretation of the source text, maladjustment or inconsistency in the preservation of the form, literal translation, inconsistency in the use of terminology, errors resulting from the context, typos and incorrect order of the sentence, double words (the so-called neural babble), the problem with semantics, which leads to logical errors and a number of inconsistencies that result from the spelling with a capital or lowercase letter, the use of concepts and terms, the order of the sentence, etc.

Safety when translating

AI disadvantages can also be observed in terms of safety. All related AI products are used on a mass scale, also in the EU institutions. All documentation that contains not only general data, but also sensitive data, has been and continues to be translated. Statoil, whose sensitive data was leaked to the network, has become acutely aware of how detrimental this affects the company’s credibility and stock quotes. The company used the popular free tool Translated.com to translate terminations, downsizing and outsourcing plans, passwords, code information, and contracts. The data was searched by Prof. Lise Lyngsnes Randeberg, who entered the name of the company Statoil in the search engine, and to his surprise, gained access to all the documents of that company.

BASF solutions

However, there is an opportunity to avoid severe consequences and use the assets that AI can offer. This solution was offered by BASF, which created its own machine for translating internal documentation, with limited access. The company has already noticed that employees use free tools, e.g. when writing emails, preparing offers, creating a business plan and presentations. Therefore, the company has planned training on the use of such tools and has created its own internal closed-loop tool, making data leakage practically impossible.

Summary

It is not without reason that translations are called the art of translation, and much more time is devoted to this process than to generating the text. AI is certainly a tool that makes the translator’s work much more effective, and the translators themselves are tasked with choosing the most appropriate version of the translation from a multitude of possibilities. However, each sentence and information should be treated with caution, as both the source text and the translation may contain logical and linguistic errors. It is also worth remembering that our inner intuition of checking information and the ability to analyze text is a skill that AI may never achieve at the required expert level.