“Do you use AI?” Scriptis and new translation technologies
“Do you use AI?” Scriptis and new translation technologies
We are often asked whether we use AI. The answer is yes. Long before Chat GPT exploded into public consciousness, the language services industry incorporated AI into the translation and quality assurance process. Back in 2020, Scriptis published an article about neural machine translation (NMT), an important development in AI-driven translation. In fact, our company’s translation work in the early 2000s contributed to the development and testing of the large language models underlying this technology.
AI-based tools increase efficiency and ensure consistency. The strategic use of technology has allowed us to maintain competitive pricing despite the economic challenges of the past few years. Over time, our large-volume clients have benefitted from lowered costs and shorter turnaround times. The more they use our services, the more our systems learn about their needs.
The technologies have improved, but expert human control of the translation process remains essential to professional language services.
Powered by AI, driven by experts
Translation for documents such as engineering manuals or financial reports can be initiated using an automated translation engine. However, to ensure accuracy, a professional translator performs a comprehensive comparison of source and target text. This is followed with a bilingual review by a senior subject matter expert. For some types of highly technical high-risk content, we would only use full human translation followed by back translation or other additional validation.
A translation engine can learn specialized technical and brand terminology if it’s trained properly. But if used improperly, it can also “learn” errors and propagate these errors across multiple media. This is why the world needs industry-specific translators. And although technology costs have decreased, true human expertise has become more valuable. When pricing translation services, clients need to make informed decisions about the risks of AI-driven translation.
Two kinds of risk: negotiable and non-negotiable
The rate you pay for translation bears an inverse relationship to the level of human oversight of the process. Less human oversight risks more errors. In general, our clients in insurance, banking, pharmaceuticals, and manufacturing have a narrow margin for error and a very low tolerance for risk. They face two types of risk when purchasing business translation services.
“Quality” is a moving target
The first type of risk is obvious: translation errors could cause mistakes or accidents. Inaccuracies in a chemical data sheet, a machine manual, or an insurance policy could expose a translation client to financial losses. At the least, they would face the expense of fixing the mistake across multiple media. In the worst-case scenario, they might face legal action resulting from serious harm.
The above scenario applies to high-risk content that requires the highest level of quality. But translation “quality” should always be defined against the context of the client’s needs. Not all translation needs are high-risk. We sometimes get requests to translate what might be called “shelfware” – required translation that no-one intends to read. An example might be translation for legal discovery. International lawsuits could involve millions of words of “evidence” but only a few items of genuine interest.
In these types of projects, reducing the level and quality of human surveillance will reduce costs. Whether to pursue a low-cost, high-risk solution is the client’s decision. Scriptis can estimate the likelihood of error for different levels of human review, and our clients should discuss this information with counsel before deciding on a translation workflow.
More translation = better translation
For clients we’ve worked with for years, reducing human review for certain types of content has become less risky. This is because our client-specific termbases and translation memories improve over time. For example, we have accumulated deep resources for English into US Spanish and Canadian French. On the other hand, if a client has never translated a particular type of text or needs a language they haven’t requested before, we would recommend the full process to begin building the resources for future projects. Also, for some less common languages, adequate AI solutions are not yet available. For these projects, full human translation is always necessary.
Data security: a non-negotiable risk
You’ve probably used Google Translate to get the gist of a message in another language, knowing full well that Google Translate provides no confidentiality. For business content, data security is essential for reasons beyond confidentiality. Viruses, malware, ransomware, and other types of cyber-crime can ruin a business. Purchasing translation services from an unknown source can expose clients to an unacceptable level of this second type of risk.
An outside accreditation agency makes a yearly audit of the systems used by Scriptis for content transfer between our clients, ourselves, and our partner translators. We have obtained the certification ISO 27001 for our information security management system, demonstrating compliance with the highest international standards for protecting against theft and other cyber threats. Because a single click on a bad link can expose our systems to risk, everyone in our organization receives monthly security training and testing.
What is the link between data security and AI-powered translation?
The proliferation of free and low-cost tools for translation has leveled the playing field for translation services. Because Google Translate is essentially free, an unsophisticated translation buyer may wonder why they are paying for translation services at all. A supplier without a strong data security system may offer low rates, but this may expose their clients to an unacceptable level of risk. Our cost of doing business includes the resources for keeping personal and IP data safe. This increases our rates, but it adds an invaluable layer of protection for our clients.
Translation supply chains can be complex, and our security systems protect against the possibility of a weak link. For languages we don’t handle in-house, we only contract with trusted partners who have obtained the ISO 17100 designation for translation quality assurance and who agree to work within our secure project management portal. It may sound a bit dramatic, but an arms race exists in the world of cyber security. AI is escalating the situation by powering both the legitimate businesses and the criminal enterprises.
You get what you pay for
We’ve all seen AI perform beautifully, but we’ve also seen it go off the rails. The promise of simultaneous, free, and accurate translation is a long way from being realized. Even if it did exist, the customers we serve cannot accept the security risks of a cheap translation solution sourced through dubious channels.
Scriptis excels at translating high volumes of content on time and at scale, but we never lose sight of the human dimension of language services. An intelligent balance between technologists and linguists brings quality and efficiency to everything we do.