Ready to Risk Your Reputation on AI Translation Services?

Ready to Risk Your Company’s Reputation on AI Translation Services?

a man sits in front of a computer looking anguished due to poor translation services
Want to stay connected?

We periodically share news and updates around translation, language and culture. Rest assured we’ll never share your contact information with anyone!

    FacebookTwitterLinkedIn

If connecting with a global audience is part of your company’s strategy, then you know how important it is to do that in local languages using translation services. These days, you can just pop your content into ChatGPT, and you’re good to go – right?

Not so fast.

While AI-powered translation tools like Google Translate, Microsoft Translator, and ChatGPT are cheap and quick, relying on these tools to translate your business content is risky. These tools can jeopardize your company’s reputation, credibility, and success.

In fact, Google Translate generated incorrect Chinese translations nearly 20% of the time in medical settings. Worst of all, 8% of these error cases could have caused “clinically significant harm.”  The quality of artificial intelligence is always improving. Still, as you consider your options, it’s crucial to understand the risks, limits, and best use cases for AI translation today. 

 

The Risks of AI Translation for Your Business

AI translation services enable companies to translate high volumes of content with quick turnarounds and at low rates. But before you start putting all your content through AI, you must know the risks involved. It’s not an “add water and stir” tool and still requires experienced human input at multiple points throughout the process. Here’s why!

1. AI Translation Services May Make Egregious Mistakes

The most apparent concern with AI translation is the potential for inaccuracies and poor quality. While AI translation tools have improved over the years, they still struggle to match the accuracy and nuance of human translators.  AI may mistranslate context-specific expressions, idioms, and cultural nuances, leading to misunderstandings and misinterpretations. When it doesn’t know an answer, it “hallucinates” – meaning it simply makes up a filler response. As you can imagine, hallucinated content within a translation can lead to massive complications. For businesses, inaccuracies in translation can result in communication breakdowns, loss of credibility, and damage to brand reputation.

Medicine

We’ve already mentioned how using Google Translate in hospital settings can lead to “clinically significant harm” in 2% of cases in Spanish and 8% of cases in Chinese. As you can imagine, this error rate is even more troubling for non-Western and/or less common language pairs. 

Immigration

AI translation is also jeopardizing asylum cases within the immigration system. Often, these individuals don’t have access to professional translation and use AI to prepare their applications. Errors such as translated proper names, mixed-up pronouns, and scrambled information are widespread. Not to mention that AI isn’t able to translate with empathy, as many asylum seekers have to retell traumatic events. Estimates suggest that 40% of Afghan asylum cases have encountered roadblocks due to AI translation errors. 

Security

Errors in AI translation can also lead to unjust and/or life-threatening situations with the police. Recently, Facebook flagged a picture of a man by a bulldozer with the Arabic caption “يصبحهم,” or “good morning.” Unfortunately, AI translated this as “hurt them” or “attack them,” leading the police to arrest the man. As you can imagine, misunderstandings that arise from police or other security forces using AI can be severe, including unjust arrests, deportations, and police searches

2. AI Translation Doesn’t Have the Same Quality Level for All Languages 

The quality of an AI translation will depend significantly on the language. Some languages have more complex grammar, syntax, and morphology than others. AI translation systems may struggle to accurately capture the nuances of languages with complex structures, resulting in lower translation quality. Studies on the wide range of translation quality for different languages demonstrate this. For example, translations for languages such as Farsi and Armenian had error rates above 30% Some language pairs, such as English to French and English to Spanish document translation, have more training data and resources for AI systems and can perform better. Languages of lesser diffusion and less commonly paired languages have fewer parallel corpora and suffer from higher error rates. 

meme that reads: would you mind translating that into spanish?

3. AI Translation Doesn’t Pick Up on Cultural Context and Nuance

Embedded within language is cultural context. AI translation can miss out on vital cultural cues that can leave a translation sounding tone-deaf. Certain texts may include references to historical events, important movements, famous figures and cultural values that are tricky to capture in translation.  Regional dialect or slang can be wholly passed over, as in the example where a woman described her abusive father as “mi jefe” (common lingo in her region), which AI translated as “my boss.” As a result, her asylum case was denied due to inconsistencies in her account of who her abuser was. In particular, AI translation is not a good choice when translating expressions and culturally polemical texts. Errors, unnatural language, and more can get in the way of your message and even offend certain populations. Recently, when the German embassy to Afghanistan tweeted in Pashto about the Taliban’s ban on women working, the completely illegible AI translation caused a cultural backlash in online spaces.

4. AI Translation Doesn’t Understand Industry-Specific Terms and Proper Nouns. 

What’s more, AI translation is a generalized tool, and it may struggle with specialized or technical content. Languages with limited resources in specific domains may have lower-quality translations for specialized topics. At a basic level, AI can’t always identify proper nouns. For example, a Spanish government official named “Dolores del Campo” had her name translated by AI as “pain of the field.” In another case, AI translated the Brazilian city “Belo Horizonte” as “beautiful horizon.” Obviously, this can quickly become a problem for essential documents. 

meme that reads: Vitamin C is Spanish for vitamin yes

5. AI Translation May Be Biased in Important Ways

AI translation may also expose your company to bias. Gender and race bias in AI translation is well-documented, with tools often reinforcing stereotypes.  For example, AI tools like ChatGPT may assume gender stereotypes for occupations (e.g., that a doctor is a “he” and a nurse is a “she”). There’s also evidence that some tools “confer a higher respect to men.” They also fail to capture gender-neutral singular pronouns (such as “they”). Creating gender-inclusive content can be very complex, and translating it can be much more so as it requires a high level of nuance and cultural insight.

6. AI Translation May Not Properly Punctuate or Format the Document.

Although formatting may not be the most urgent error that AI commits, it can have serious repercussions. Formatting and punctuation concerns may include:

  • Naming conventions 
  • Date/time
  • Numbers/prices/currency
  • Sizes/measurements
  • Telephone/address format
  • Special characters

Just consider that the Norway Olympic team used Google Translate to make an order at a Korean supermarket. Their AI-generated order for 1,500 eggs instead became an order for 15,000 eggs because it overlooked Korea’s unique counting system. While this turned out to be a funny anecdote, imagine a business contract getting incorrectly translated with an extra zero. That could have significant consequences for the companies involved in the error.

7. AI Translation Software Doesn’t Always Have Clear IP and Confidentiality Clauses

Finally, perhaps the most significant liability of AI translation is the lack of clarity regarding IP, data privacy, etc. For example, these tools: 

  • May infringe on copyright during the AI model training process. Currently, there are a lot of lawsuits about AI models using unlicensed works (books, artwork, etc.) by scraping them online without permission. 
  • May have unclear IP clauses about who owns the translated materials. Who owns AI-generated works: the user, the AI company, none of the above? Courts are still working out this central issue
  • May have unclear confidentiality about how the translation is captured or used by the model. If you run confidential or sensitive company information through an AI tool, you may expose it to third parties. The AI company may also sell your data or texts to third parties without your knowledge. 

Your business should be hypervigilant about the implications of using AI translation for your IP, data privacy and legal liability. Depending on the nature of your content, the risks may not be worth the benefits. 

a meme that reads: I have not been entirely successful

How Humans and AI Work Together

Translation technologies have long been incorporated into professional processes. Technologies such as CAT (computer-assisted translation) tools, translation management systems (TMS), translation memories, and custom glossaries are frequently used by professional translators and translation agencies to make the translation process more efficient, consistent and tailored.  For example, machine translation post-editing (MTPE) involves starting with a machine-translated text and then having an experienced professional editor do a side-by-side edit and proofreading step. AI models can also be leveraged to evaluate the quality of machine-translated text, indicating to the translator where they should spend the most effort. Called “human-in-the-loop” workflows, these methods combine human and machine collaboration. It’s important to note that these tools are always used under expert supervision. In the case of AI, expert understanding and oversight are especially vital to guarantee a high level of quality.   

 

Balancing the Teeter-Totter of Human and AI Collaboration

At Multilingual Connections, here are some ways we blend human and AI translation while maintaining our commitment to quality, accuracy, and ethics.

1. Understand AI as Augmenting Translation Professionals

As the saying goes, “A tool is only as good as its user.” For this reason, we place our trust in our professional team to:

  • Screen for trustworthy tools according to their language and project requirements.
  • Understand the confidentiality clauses for these tools to prevent IP and data leaks.
  • Know how and when to leverage technologies during the translation process.
  • Incorporate human-first processes to revise, correct, add, format, and make necessary annotations.

Ultimately, AI will augment our professional linguists and project managers, but only because they’re already experts in their fields and understand where and how to use AI effectively. We don’t envision a future where non-professionals could apply these tools with excellent quality and accuracy. 

2. Emphasize the Value of the Human-Centered Expertise

Our professional linguists are at the heart of creating first-rate translations. We believe they should run the show and prioritize their expertise, needs and concerns.  Our subject matter experts know the best use cases for AI and other technologies. In many cases, muddling the translation process with AI doesn’t make sense. For certain languages, AI churns out nonsense. For others, it just doesn’t have the capacity to get  optimal results. AI falls flat when it comes to high-stakes or sensitive texts involving cultural nuance, regional dialect, industry-specific terms, a unique brand voice, etc. We trust our professionals to determine precisely when AI should or shouldn’t be incorporated into the translation process to ensure optimal results. 

3. Leverage QA Protocols to Revise Processes

AI is highly dynamic and evolves daily. That’s why our teams routinely analyze our tools and workflows to ensure they’re up to speed.  We have rigorous QA protocols to ensure top quality for our human-in-the-loop workflows. We also revise these processes as technology advances so that we’re at the cutting edge of human and AI collaboration without sacrificing our company values.   

 

Entrust Your Translation Project to Our Tech-Savvy Experts

At the end of the day, it’s essential to understand the risks, limits and best use cases for AI translation. At Multilingual Connections, we don’t shy away from technology. Instead, we thoughtfully integrate it to elevate our professional expertise.  As our industry masters blending AI and human collaboration, it’s worth having an expert partner at your side. Our world-class translators are integral to achieving top-tier outcomes that ensure quality, accuracy, efficiency and more. Contact Multilingual Connections today for more details about our human-in-the-loop workflows.

 

Check Also:


Want to stay connected?

We periodically share news and updates around translation, language and culture. Rest assured we’ll never share your contact information with anyone!

Leave a Reply

Your email address will not be published. Required fields are marked *