By Konstantinos Menyktas

Artificial Intelligence (AI) and tools like ChatGPT have revolutionized language translation, offering new possibilities, particularly in the legal translation field. However, their limitations are significant and, if not properly acknowledged, can lead to costly errors, legal liabilities, and even false judgments in court or business agreements.

The Challenge of Context and Language Variability

The effectiveness of AI translation can vary widely depending on the language and context involved. Legal systems differ not only between countries like Greece, Cyprus, the US, the UK, Canada, Australia, Ireland, and France but also within the same linguistic framework. Each language has unique ways of expressing its legal system on paper, with specialized jargon, fixed expressions (expressions figées), terminology, and syntax.

For example, in Greece, legal documents are often written in a highly complex manner, using long sentences, outdated expressions from a conservative form of Modern Greek known as “katharevousa,” and sometimes contain oversights due to the haste in which they were drafted. These nuances can pose significant challenges for AI translation tools.

AI’s Limitations in Legal Translation

When tasked with post-editing a condominium purchase and sale agreement from Greek to English, the AI-generated translation was flawed. I remember results like this in the past with machine-translated documents despite the translation tool being fed with a glossary and Translation Memory. In the case of the AI-generated translation, while the result was generally accurate, many sentences were omitted, and the meaning was altered in some cases. The AI tended to use terms which do not correspond to Greek legal terminology, highlighting a critical gap in the AI’s understanding of the legal context.

Such errors occur because AI, like ChatGPT, generates text based on patterns observed in its training data. When these patterns don’t align with the specific legal context or language, the resulting translation can be inaccurate, posing serious risks.

The Myth of Cost Savings with AI Translation

It’s important to debunk the myth that AI-generated translations are cheaper. While the initial cost might seem lower, the reality is that potential errors in AI translations can be difficult and time-consuming to fix, leading to additional expenses.

AI-generated texts that haven’t been thoroughly proofread or reviewed can also damage your reputation, potentially resulting in legal costs and the loss of loyal customers. In many cases, it takes more time and effort to correct a poorly done AI translation than it would to start from scratch using AI as a tool alongside Translation Memories (TMs) and glossaries.

Incorrectly relying on AI can increase costs, especially when errors necessitate extensive corrections or when a poorly paid linguist provides only superficial proofreading. Instead of reducing expenses, improper use of AI can lead to more significant costs and a poorer overall result.

The Critical Role of Humans in Legal Translation

A major challenge in AI translation, particularly in the legal context, arises with less-represented languages, such as Greek. The disparities in translation quality between widely spoken languages and these less-represented languages need targeted strategies.

To improve the accuracy of AI translations in legal contexts, simplifying the text can be helpful. This involves using clear, basic language and avoiding idiomatic expressions or highly technical terms. However, human linguists are essential to ensure the accuracy of legal translations. Generative AI can sometimes produce incorrect information. In legal translation, even a small error can have serious consequences.

Complex legal matters, therefore, require the involvement of human linguists. Human expertise is crucial for understanding the implications of clause or word changes in legal documents, interpreting new legal developments, and considering industry-specific terminology and standard practices.

Liability in AI-Assisted Legal Translation

A critical question in AI-assisted legal translation is who bears the liability for errors? This includes tort liability, professional obligations, contractual breaches, personal data breaches, confidentiality breaches, and other ethical and legal standards. Is it the client, the translator, the reviewer, the proofreader, or the AI machine? One thing is certain: the AI itself cannot be held accountable, as explicitly stated in its usage policies. This places the burden of responsibility on human professionals, whose reputations and legal standing are at stake.

However, when language professionals use tools, including AI, as aids, the potential for improvement is vast, making the process more efficient, cost-effective, and secure.

Conclusion: AI as a Tool, Not a Replacement

As I mentioned in my earlier article, “Legal Translation & AI: Who’s Afraid of the Big Bad Legal Liability?”, it is only humans who can ensure the accuracy and quality of legal translations. While AI can speed up the translation process, making it more efficient and (in some cases) cost-effective, it cannot replace the critical thinking and judgment of human professionals. By leveraging AI as a tool alongside human expertise, we can strive for better, more reliable, and ethically sound translations, ultimately serving the needs of clients and maintaining the integrity of legal communication.