Blogs

All across the world, people are using free artificial intelligence tools like ChatGPT and Google Translate to perform the role of an interpreter, allowing them to communicate seamlessly with people who don’t speak their primary language. For low-risk contexts—like a tourist trying to ask a shop clerk how much their souvenir costs, for example—these tools can be quite valuable.  But in high-stakes environments like hospitals and courts, free artificial intelligence interpreters are simply not enough. Large language models and machine translation tools may be capable of interpreting simple, straightforward conversations, but they’re far from error-free: They struggle with nuanced or ambiguous language and lack the cultural knowledge and emotional complexity that human interpreters bring to the job. While Google Translate might be good enough to help you ask a passerby for directions to the nearest train station, it’s absolutely not good enough for a nurse trying to help a patient understand their discharge instructions.  In this blog post, we’ll go over the different ways in which artificial intelligence interpreters are being used today, and why human interpreters are still an absolute must-have in high-stakes environments like legal proceedings, the healthcare system, and emergency services. We’ll explain some of the key limitations of artificial intelligence interpreters and the importance of intervention by human professionals in the interpretation process. Table of Contents What Are Free Artificial Intelligence Interpreters? High-Stakes Environments Where AI Interpreters Are Used Risks of Relying Solely on Free AI Interpreters in High-Stakes Environments Why Human Oversight is Essential Best Practices for Using AI Interpreters in Critical Environments Frequently Asked Questions About Free AI Interpreters Conclusion Artificial intelligence interpreters combine speech-to-text and machine translation technologies to provide real-time language assistance, much like a human interpreter would.  The process works quite similarly to consecutive interpreting: One person speaks in their preferred language, while the artificial intelligence interpreter transcribes what they’re saying and then translates that into the other person’s preferred language. The translation can then be read aloud using text-to-speech technology, or the person can simply read a written translation of what the other person said. Here are a few examples of artificial intelligence interpreters that are available for free: Like a human interpreter, these platforms allow for fairly seamless conversation between two individuals. For low-risk settings—especially settings where you wouldn’t normally have an interpreter with you, like tourism or casual conversation—these can be very helpful tools. But they are also prone to errors that a human interpreter typically wouldn’t make.  For example, if there’s an error in transcribing the speaker’s speech, the translation could be completely wrong. Additionally, machine translation tools are notoriously bad at translating language that is ambiguous, nuanced, or slang-heavy. In fact, recent research shows that medical translations done by ChatGPT and Google Translate yielded errors in spelling, grammar, and readability that caused significant misunderstandings. This means a human interpreter is still necessary when it comes to high-stakes environments where a person’s life could depend on the quality of the interpretation. In general, artificial intelligence interpreters are not being used for high-stakes interpretation—at least not on their own. There are some organizations testing out artificial intelligence tools to bridge language gaps in high-stakes settings like hospitals and legal proceedings, but it’s important to note that these organizations are also taking into account several ethical considerations as well, to ensure that they’re using the tools responsibly. Take the Seattle Children’s Hospital, for instance: They’re testing out artificial intelligence tools to help provide patients and their families with discharge instructions in their preferred language before they leave the hospital (a written translation is also mailed to families after the patient has left the hospital). However, they’re using an internally developed tool to keep patient information private and have created an AI ethics board to oversee the implementation of the technology. And perhaps most importantly, human translators will still be double-checking the translations to ensure that the information is translated appropriately. Artificial intelligence interpreters and translators can present all sorts of risks when such measures aren’t taken. Below, we’ll look at a few of the issues that can arise when you rely on free AI tools alone in high-stakes settings: 1. Accuracy Concerns: All kinds of inaccuracies can come up when using AI interpreters. If the speaker’s speech is mistranscribed, the machine translation will not be an accurate representation of what they meant to say (for example, the word “femur” might be mistranscribed as “lemur,” potentially leading to serious complications). Beyond mistranscription, machine translation tools struggle with accuracy, especially for languages for which less training data is available. 2. Regulatory and Compliance Risks: Using free artificial intelligence tools also raises concerns about potential violations of industry regulations, such as HIPAA in healthcare or legal misinterpretations in court. Take, for instance, the Utah lawyer who was sanctioned for using ChatGPT to help write a brief that cited hallucinated citations (i.e., citations that the model made up). 3. Lack of Contextual Understanding: AI's inability to fully grasp cultural, emotional, or situational nuances that human interpreters would understand. Human interpreters can take cultural and emotional cues and other factors into context, a skill that sets them apart from machine translation and AI. This allows them to better parse out ambiguous or unclear language, leading to more accurate communication between both parties. Using artificial intelligence interpreters without any human oversight can certainly speed things up—but it also lowers the overall quality. Instead of implementing artificial intelligence on its own, it’s important to have a human in the loop—humans can help vet translations and make sure that the final product is a high-quality translation. A hybrid approach—in which artificial intelligence is used as a tool for interpretation, rather than as an interpreter itself—maximizes efficiency and minimizes the risks we’ve outlined above. When it comes to high-stakes settings like emergency room visits or court proceedings, it’s important to have a human interpreter on hand—this is the best way to ensure high-quality language assistance for individuals with limited English proficiency. In these settings, inaccuracies caused by poor contextual understanding or grammatical errors can have grave consequences—a human must oversee these processes. By working with a trusted language service provider to contract interpreters, you can be sure that a human interpreter will always be available when you need one, whether by phone or video call. For low-risk contexts, like simple administrative tasks, AI interpreters can be a useful tool to boost efficiency, but it’s still a good idea to have a human reviewing the final product to make sure that everything is accurate and editing things accordingly. Not necessarily—artificial intelligence can be a powerful tool to boost efficiency. That said, AI output needs to be thoroughly reviewed by a human to ensure its accuracy. Free AI interpreters are typically not specialized to the unique needs of those working in high-stakes environments, like hospitals, law firms, and emergency response teams. Individuals working in these industries should look for AI tools that are trained on domain-specific data that reflects their specific needs. In general, no. In healthcare, sharing patient data with a third-party typically violates HIPAA, meaning that tools like ChatGPT and Google Translate are typically not compliant—AI interpreting tools need to store all patient data internally. In the legal field, it varies depending on how and what the tools are being used for. Attorneys, paralegals, and others working in the legal field should also review local regulations and confirm that their use of certain AI interpreters does not violate ethical guidelines. No. Human interpreters are especially important when it comes to high-stakes environments, as AI tools tend to make mistakes that humans wouldn’t. In settings where miscommunication can have a serious impact on an individual’s life, it’s absolutely critical to have a human interpreter. It may not always be possible to get a human interpreter on-site, especially in emergency settings. But language service providers like Avantpage offer on-demand virtual, remote, and over-the-phone interpreting services, which connect you with a qualified interpreter in mere seconds. Consult with a trusted language service provider to identify reliable alternatives to artificial intelligence interpreters. It may be tempting to use a free artificial intelligence interpreter when you don’t have a human interpreter available. But understand that this is a risky gambit. At best, it can lead to an awkward encounter; at worst, a patient could lose their life due to a simple miscommunication that a professional interpreter could have resolved. At Avantpage, we work with a team of professional, human interpreters to provide language assistance in high-stakes settings like emergency healthcare and legal proceedings. Through our in-house platform interpreting program, we can connect clients to human interpreters remotely in a matter of seconds. If you’re looking for human interpreters to enhance your language access measures, don’t hesitate to contact us today at (530) 750-2040 or [email protected].

Whether it’s being used to detect early signs of disease or bridge language gaps between patients and their doctors, artificial intelligence (AI) is shaking things up in the healthcare industry.  According to recent research from Deloitte, 75% of leading healthcare companies are “experimenting with or planning to scale generative AI across the enterprise,” and even more (92%) see its potential to improve efficiency in healthcare. But although AI can certainly enhance human-led procedures, it’s far from a replacement for human healthcare professionals. One key strength of AI is its ability to enhance language access measures by speeding up the translation process, helping patients with limited English proficiency (LEP) communicate with their doctors. AI in translation is a powerful tool, but as we’ve seen with machine translation, generative AI-powered translation requires a human in the loop who can ensure linguistic accuracy, cultural appropriateness, and legal compliance.  In this blog post, we’ll take a look at the best practices that hospitals and other institutions should follow when implementing AI in healthcare translation. We’ll also go over some real-world examples of how providers are implementing AI in translation while also maintaining human oversight to ensure accurate and culturally sensitive language access measures. Table of Contents Understanding Generative AI in Healthcare Translation How Are Hospitals & Healthcare Providers Using AI? Benefits of AI in Healthcare Applications Risks of AI in the Healthcare Industry Best Practices for Implementing AI in Healthcare Translation Future Outlook of AI in Translation and Healthcare Frequently Asked Questions About Generative AI in the Healthcare Industry Conclusion  Generative AI is a form of AI that can create text, images, and videos based on prompts written in natural language. Tools like ChatGPT and Dall-E are classic examples of generative AI platforms. If a user copies and pastes meeting minutes into ChatGPT and asks it to write up a brief, neatly organized summary of the meeting, ChatGPT will generate a summary based on those notes. In healthcare, generative AI tools can be used to increase administrative efficiency by speeding up tasks such as summarizing doctors’ notes or translating text into a patient’s native language. Other proposed uses for generative AI in healthcare include developing new drug candidates based on data on existing molecular structures and improving the quality of medical images like MRI scans. It’s important to note, however, that generative AI tools tend to “hallucinate,” making up false information that may appear to come out of nowhere. For this reason, human oversight is key whenever AI is used in the healthcare system. When it comes to AI in translation, for example, AI can play a helpful role in producing fast and mostly accurate translations of medical documents and patient records—it can even be used for real-time communication between LEP patients and their caretakers. But it’s not a replacement for translators and interpreters altogether. Humans who are familiar with both the source and target language need to review and vet AI translations to ensure that the AI output doesn’t contain any inaccuracies or hallucinations. Here are a few examples of hospitals and healthcare providers that are using AI to enhance patient care: 1. Partners Healthcare: When Partners Healthcare’s COVID-19 hotline was overwhelmed with callers during the pandemic, the organization implemented an AI chatbot to act as a screening tool. In addition to screening callers for COVID-19 symptoms, the chatbot could also answer most questions about the virus and advise patients on whether they should visit an urgent care or emergency room. 2. Johns Hopkins Medicine: Doctors at Johns Hopkins Medicine can use AI to draft responses to patient messages in the patient portal, editing the generative AI response as needed. The team at Johns Hopkins is also working on ways to summarize charts. 3. Mount Sinai: Since 2013, doctors at Mount Sinai have been using algorithms to identify patients in their system who are more likely to get sicker. By using this information, they’re able to more effectively treat them before their conditions worsen significantly. Using AI in translation can have several benefits for healthcare providers, such as: Despite the benefits, using only AI in translation carries several risks that healthcare providers must be aware of: Although the benefits of AI may seem tempting, the risks of using AI on its own outweigh the advantages. Healthcare providers can still take advantage of the benefits of AI in translation by following these best practices: As the technology behind AI improves, you can expect to see AI being used more and more in the healthcare setting. AI has especially great potential for automating translation and administrative tasks—it’s possible that this technology will eventually be used to automate the translation of electronic health records. However, as AI becomes more and more widespread in the healthcare system, ethical concerns are bound to arise. Many concerns around the ethics of AI have already come up—for example, the idea that inaccuracies due to the use of AI could negatively impact a patient’s health. Looking toward the future, it will become important to address these concerns and figure out a way to balance the efficiency of AI with human factors like empathy and judgment. AI is improving efficiency across healthcare by automating administrative tasks and supporting translation for patients with LEP. It can also be used to more efficiently analyze medical images and detect early stages of disease, among other proposed applications of the technology. Not on its own. While AI can speed up translation and improve access, human oversight is essential to ensure accuracy, cultural sensitivity, and compliance with healthcare regulations.  No. AI-powered translations can be fast, but they often contain errors or omissions. Providers should always involve professional medical translators to review and verify AI-generated text. Healthcare providers should use AI tools specifically designed for medical contexts and compliant with regulations like HIPAA. These tools, paired with the expertise of human translators, ensure safer and more accurate outcomes. AI can improve healthcare translation workflows by making things faster, more scalable, and more efficient. But as powerful as AI may be, it can’t replace the expertise, cultural awareness, and judgment of human translators. As with other kinds of translation technology, AI requires human oversight to ensure that it’s being used safely. The most effective approach is a hybrid one—leveraging AI to speed up translation while relying on human professionals to ensure accuracy, compliance, and patient-centered care. For healthcare providers, the next step is clear: Embrace AI as a valuable tool, but implement it thoughtfully. By following best practices—such as incorporating human oversight, using domain-specific datasets, and ensuring regulatory compliance—hospitals and providers can deliver better, more inclusive care to patients of all backgrounds.  At Avantpage, we enable healthcare providers to bridge language gaps between care teams and their LEP patients using technology-forward translation and interpretation workflows with a human in the loop. If you’re looking for a language access solution, Avantpage has you covered. Contact us today at [email protected] or (530) 750-2040, or fill out this form for a free quote.

Since the launch of ChatGPT in November 2022, generative artificial intelligence (AI) has become a game-changer across industries, from government and education to commercial and healthcare settings. But when it comes to translation, the question remains: when should you rely on Generative AI vs. human translations? In this article, we’ll guide you through a clear roadmap to help you evaluate your content, project requirements, quality expectations, brand considerations, and risk factors so you can choose the best approach for your needs. Table of Contents Deciding Between Generative AI vs. Human Translations Question 1: What Type of Content Needs to be Translated? Question 2: What Are the Project Requirements? Question 3: What Are the Quality Expectations? Question 4: Are There Any Brand Voice or Consistency Considerations? Question 5: What Risks Does This Project Present and How Can You Prevent Them? Frequently Asked Questions About AI vs. Human Translation Get Help in Choosing Between Generative AI vs. Human Translations The answer is complicated — it all depends on the type of content you need translated and the specific requirements you have for that translation. That’s why we’ve devised the following roadmap for you to determine whether you should request AI-powered translation, human translation, or a hybrid model for your language services. Here, we’ll take a look at the questions you should ask yourself before determining which approach to use in your translation needs.
The first question you’ll want to ask yourself is about the type of content that you need translated. Generative AI is particularly useful when translating simple language and repetitive texts; however, it still lacks the human touch necessary for more creative types of content, like pun-heavy marketing materials or culturally nuanced messaging. The first question you’ll want to ask yourself is about the type of content that you need translated. Generative AI is particularly useful when translating simple language and repetitive texts; however, it still lacks the human touch necessary for more creative types of content, like pun-heavy marketing materials or culturally nuanced messaging. To help you decide between Generative AI vs. human translation services, you’ll need to consider the scope of your project. Here are just a few factors to look at: If you have a tight deadline for a given project and the content type is suitable for Generative AI translation, Generative AI could be the right way to go. Generative AI tends to be much faster, whereas human translators need more time. That said, if the content type isn’t ideal for Generative AI translation, you may find that the review process ends up taking longer than desired. In cases where your content needs to be translated quickly by a human, consider requesting a rush translation from a trusted language service provider. Human-produced translations are going to cost you more than a Generative AI translation. Still, it’s important to consider the fact that human translations are typically higher quality, and mistakes resulting from a Generative AI translation could be costly. Consider the type of content you need translated first before determining what your budget constraints allow for.
Learn more about the benefits of human-in-the-loop translation projects.
Quality expectations vary across projects — choosing between Generative AI and human translation may also hinge on factors like the expected accuracy, cultural sensitivity requirements, and the project’s potential impact on organizational outcomes and performance. Generally speaking, human translators will be able to ensure more accurate results, as Generative  AI often makes errors in vocabulary and context. Additionally, critical content like legal or medical documents demands the precision that human translators provide.  Generative AI, while advanced, often lacks the contextual understanding required for such materials. Generative AI translates words, not meaning. Humans translate meaning and words. This is particularly important to keep in mind with texts that may have culturally sensitive information. Direct translations can sometimes sound blunt or insensitive in another language, which is why human translators are essential for capturing cultural nuances and conveying the intended meaning. Learn more about the localization process and why having a human translator involved is crucial to producing quality translations.  For projects where translation quality directly impacts outcomes — think educational materials that will assist a patient in their health decisions, or a child’s individualized educational plan that will determine their educational path — investing in human expertise is generally the more strategic choice.
Human translators are generally better at maintaining your brand voice and adhering to style guidelines. While Generative AI can produce content quickly, it often struggles with nuance, tone, or jargon-heavy content. Tips for maintaining consistency:
Because Generative AI-powered translation tools tend to struggle with accuracy more than human translators do, it’s important to consider the risks associated with your project — that is, what could go wrong, and how you can prevent the likelihood of that happening. While Generative AI can speed up the translation and localization process, high-stakes content like legal, medical, or regulatory documents requires human oversight. Mistakes in these cases can be costly or even dangerous. Risk mitigation strategies: If you determine that the overall risk is low and opt for machine translation post-editing services, you can mitigate risk even further by incorporating thorough review and quality assurance processes. Quality assurance tools can perform objective measures of the quality of a given translation, while human reviewers are an absolute must, as they can fill in the gaps in any AI translation. Artificial Intelligence (AI) is a broad field focused on creating machines that can perform tasks that typically require human intelligence. Generative AI is a specialized subset of AI that focuses on creating new content, like text, images, or code, based on what it has learned. Modern AI translation tools, such as Google Translate and DeepL, use Generative AI. Human translation is performed by trained linguists who understand context, cultural nuances, and tone, ensuring the message is accurate and appropriate for the target audience. AI translation uses machine learning algorithms to convert text quickly, but may miss subtleties, idioms, or cultural references. AI is unlikely to fully replace human translators. While AI can handle bulk, repetitive, or simple text efficiently, humans are essential for high-stakes, culturally sensitive, or creative content where accuracy, tone, and meaning matter most. AI has become highly accurate for straightforward, common language, but it can still make mistakes with industry-specific jargon, idioms, and context-dependent meanings. Accuracy improves when AI is paired with human review in a hybrid workflow. Not necessarily. Humans can make errors, but they excel at understanding context, cultural nuances, and brand voice, which AI often misses. For critical content, a human translator’s judgment is generally more reliable than AI alone.  AI works best for internal documents, repetitive content, data-heavy materials, and texts where a quick deliverable is crucial. Marketing content, terminology glossaries, legal contracts, medical documents, and culturally sensitive materials are better suited for human translation or a hybrid approach. Before you decide to opt for a Generative AI translation tool over a human translator, ask yourself the following questions: Whether or not you use Generative AI or human translation depends on a wide range of factors — and sometimes, the answer still isn’t clear. In those instances, you may find that a hybrid approach works best. At Avantpage, we recognize the value of both Generative AI and human capabilities. We’re an experienced language service provider that integrates AI features, Generative AI, human expertise, and hybrid methods to ensure a customized solution that works best for your project’s unique demands. If you’re looking to learn more about whether Generative AI translation, human translation, or a hybrid approach is best for your project, contact us online or call (530) 750-2040. We’ll help you determine and execute the most effective strategy for your project.

Up until relatively recently, free machine translation tools were notoriously low quality—you’ve probably seen your fair share of posts poking fun at machine translation “fails” on social media. However, machine translation has improved significantly over the last decade, and it has become a part of many translators’ repertoires.  That said, machine translation is merely a tool to help translators move the process along faster—machine translation is still far from perfect on its own. These tools can produce deceptively fluent and easy-to-read text, but when you look closely at their output, you’ll begin to find errors and inaccuracies that human translators typically won’t make: They struggle with ambiguous language, fail to take into account the cultural context of the content, and it can be challenging to get them to stick to a consistent style guide. That’s why it’s critical to have a human in the loop. Translation technology enables rapid-fire translation of long texts—in 2022, Google Translate processed 146 billion words a day, more than most human translators will translate for their entire career. But it still makes mistakes that humans need to fix.   “Human-in-the-loop” translation refers to the process of combining the efficiency of machine translation with the accuracy and cultural awareness of human translators. It’s an important approach to translation in an age where machine translation tools are as commonplace as they are today. In this blog post, we’ll take a look at the benefits of human-in-the-loop translation and why organizations in healthcare, government, and law need to make sure there’s a human in the loop, rather than utilizing fully automated translation services. Table of Contents What is Fully Automated Translation? What is Human-in-the-Loop Translation? Benefits of Human-in-the-Loop Translation How to Choose Between Human-in-the-Loop vs. AI Translations When Fully Automated Translations Work Best When Human Oversight is Non-Negotiable Building a Translation Strategy That Scales Frequently Asked Questions About Human-in-the-Loop vs. AI Translations Conclusion As the name suggests, fully automated translation is any translation process conducted solely using translation technology, with little to no human oversight. With fully automated translation, a user simply inputs text in one language, selects the target language, and receives an automatically generated translation in a matter of seconds. Using translation technology cuts costs and speeds up the translation process significantly—instead of paying somebody and waiting for them to translate a document in a matter of days, organizations can get a final translation in the blink of an eye, without having to pay much. But raw, unedited machine translation output is often rife with errors, especially for more complex documents. Because of this, fully automated translation tends to be best suited for repetitive texts that use relatively simple, unambiguous language. The quality of machine translations also varies by language. In a recent study, researchers found that Google Translate had a 94% accuracy rate on English to Spanish translations of emergency discharge instructions, but 55% for translations into Armenian. Still, when it comes to translating important medical documents, even a minor error can have serious consequences. That’s why it’s important to have a human in the loop to review automated translations and make sure they’re accurate. Like fully automated translation, human-in-the-loop approaches begin with a machine translation—but they don’t end there. After first running the document through a machine translation tool, human translators and editors review the content to make sure it’s accurate. These individuals review the translation and the original document, editing it to make sure that the final product is accurate and reads smoothly. They typically use other tools like translation memory and quality assurance tests to make sure that the translation is as accurate as possible. Human-in-the-loop workflows benefit from the speed and efficiency of machine translation, but they also cut out any errors introduced by the machine translation tool. This makes human-in-the-loop workflows particularly well-suited for important documents that need to be translated on a tight deadline without sacrificing quality, such as in healthcare and government settings. Human-in-the-loop translation has several advantages over fully automated translation. Here are a few key benefits to keep in mind: Human-in-the-loop and fully automated translation workflows can be used in different situations. While it’s good to have a human in the loop, it’s not always 100% necessary. Below, we’ll outline specific scenarios where one process works better than another. While human oversight is always a good safety net, it’s not always 100% necessary. That’s why it’s important to work with a trusted language service provider that can guide you through carefully balancing automation with human quality assurance. A language service provider can help you integrate human reviewers into existing translation workflows so that you’re not starting from scratch. Many modern translation management systems support hybrid models, allowing organizations to kick off projects with machine translation and then route the content to qualified human linguists for post-editing and quality control. With a flexible workflow in place, low-risk, high-volume content like general outreach emails or web FAQs might move through a light-touch workflow, while sensitive legal notices or discharge instructions follow a stricter protocol with multiple human checks. This kind of scalable, tiered approach ensures you’re getting the best of both worlds: the speed and cost-efficiency of automation and the reliability and nuance that only human reviewers can provide. Human-in-the-loop translation is a translation flow in which human translators, editors, and/or proofreaders review machine-translated content to make sure that it is accurate and culturally appropriate for the target audience. No. While machine translation alone may be useful for simple, low-risk content, healthcare and legal documents typically require multiple human checks to make sure that no information has been mistranslated. Yes—human-in-the-loop workflows still incorporate translation technology like machine translation and translation memory into the translation process, making the translation process go by much faster than it would without the technology. This process also lowers costs as translators do not need to spend as much time and effort on producing the final, translated document. Yes, there are several regulations that require organizations to use a human-in-the-loop workflow, particularly when it comes to sensitive documents that include patient health information. Organizations providing healthcare, legal, and government services should be especially careful to follow regulations relevant to their line of work, such as HIPAA, the ADA, and Section 1557 of the Affordable Care Act. Fully automated translation is not always accurate. Depending on your target language, it may have a high error rate, and depending on how sensitive a given document is, these errors may have life-or-death consequences. For example, if patient discharge papers are not translated with the utmost accuracy, the patient may not take proper care of themselves after discharge, leading to higher readmission rates and even more severe medical outcomes. Yes. Language service providers and human translators alike use translation management systems and computer-assisted translation tools to review and edit machine-translated text in a process known as machine translation post-editing services. While machine translation tools can be speedy and cost-effective, they’re still far from perfect. Fully automated translation workflows sacrifice quality and accuracy in return for a quicker and cheaper final product. Human-in-the-loop translation balances this trade-off, effectively improving the overall quality of machine translations while still delivering a final product on a quick turnaround. Accuracy isn’t just a technical matter—it’s a civil rights issue. When translations are inaccurate and error-ridden, individuals with limited English proficiency are unable to access important medical, legal, or government services. That’s why human-in-the-loop translation is critical, especially in these industries. At Avantpage, we combine cutting-edge translation tools like AvantMemory with the knowledge of expert linguists to ensure every word resonates clearly and accurately. Whether you're navigating compliance requirements or simply aiming for better multilingual communication, we’re here to support your goals with services like machine translation post-editing. Reach out at [email protected] or (530) 750-2040—or request a free quote to get started.

Since ChatGPT dropped in 2022, large language models (LLMs) have been all the rage. ChatGPT and similar tools like Claude and Gemini have been heralded as high-powered productivity tools that can streamline your workflow and make many of our day-to-day work tasks significantly easier. And while it may be true that they can make our jobs a little bit easier, that doesn’t mean they can make our jobs easy. In the field of translation, LLMs have been widely applied alongside machine translation (MT) tools to produce fluent and accurate translations from one language into the next. But organizations looking to translate their content should be careful when employing tools like ChatGPT to translate texts — though they may yield accurate literal translations, these translations may not be fully adequate for your needs as a business. In this blog post, we’ll take a look at five key areas in which AI tools like ChatGPT are still lacking when it comes to translations. From their struggle to understand context to their tendency to editorialize, these are some of the most common issues you’ll want to look out for if you plan on using any of these LLMs to translate your content. Although LLMs can produce pretty accurate literal translations, they’re still not great for situations where you need to translate culturally nuanced language like idioms and other common expressions. According to a 2024 study, AI still struggles to make sense of (and thus, translate accurately) culturally nuanced phrases and ambiguities that human linguists are better prepared to parse out. So while these tools may excel with simple texts like routine forms and paperwork, you’ll still need a human in the loop when translating more complex texts like marketing brochures and web pages. If you’ve ever tried to correct ChatGPT after it produces inaccurate output, you know just how hard it is to get LLMs to correct themselves. But making corrections and editing our work is a key part of a translator’s work — language service professionals catch errors in the moment and adjust accordingly all the time. In a 2023 study, researchers found that LLMs have trouble “self-correcting” — that is, refining future output based, either on their “inherent capabilities” or on feedback to previous responses. According to the study, LLMs struggle to “self-correct their responses without external feedback, and at times, their performance even degrades after self-correction.” That means that these tools may produce errors and then repeat those errors throughout a given text. If you’re using these tools for translation, you’ll need to make sure somebody knowledgeable about both the target and source language is carefully reviewing the output and editing it to minimize these errors. LLMs have raised a wide range of data privacy concerns, and organizations working in highly regulated industries like healthcare and law should be leery of using them for translation tasks. Different industries and locations have different standards of data privacy, so it’s important to be aware of what is and isn’t acceptable for your circumstances. For example, LLMs are generally not HIPAA-compliant — healthcare organizations using this tool to translate texts into another language must make sure to mask any patient health information (PHI) such as name, date of birth, social security number, etc. before putting it into the tool. That means healthcare organizations must be careful to either manually or through automations (AvantShield) scan the original text for any such information and remove it entirely from the text, to avoid issues with HIPAA compliance. In addition to data privacy, another common concern that experts have raised about AI is its implicit biases. All sorts of AI models have drawn criticism for their biases — take, for example, Amazon’s recruiting tool that showed a bias against women applicants in the search and hire process. Such biases may also come up in LLMs performing translation tasks. LLMs are particularly notorious for tone-shifting and editorializing in their translations. They may shift the overall tone of a text to more closely align with its own standards of tone — for example, rephrasing a common yet pejorative buzzword like “woke” as something more neutral, like “aware of social inequality.” The translation of this phrase would in turn lose the connotation understood by the original terminology, harming the overall quality of the translation. Although LLMs can produce highly accurate and fluent text in languages like English, Spanish and French, the same isn’t true for all languages. These tools are trained on massive amounts of text in various languages, primarily taken from the internet — there’s more available input for languages like English and Spanish than there is for Pashto, for example. Languages with more training data will be easier to translate between; on the other hand, LLMs will struggle to produce accurate translations in languages that don’t have a large digital presence — like indigenous languages of the Americas, for example. LLMs may be useful tools, but they’re still far from adequate replacements for human linguists. Instead, human linguists should take a smart, balanced approach to incorporating AI tools into their workflow. By correctly identifying the proper use scenarios of AI in the translation process, we can leverage AI tools in an effective way that speeds things up without sacrificing the overall quality. Here’s why:
 
LLMs struggle to understand cultural or historical context that may be relevant to certain texts, making transcreation and localization tasks nearly impossible with AI alone.
These tools aren’t great at correcting themselves, even when they receive external feedback.
Organizations working in highly regulated industries like healthcare need to be careful not to violate data privacy laws.
AI is known for holding implicit biases which may affect the quality and content of the final translation.
The overall quality of translation will vary by language — languages that do not have a significant digital presence have less training data to draw from and produce an accurate translation.
 
At Avantpage, we’re well aware of the limits of AI tools in the translation process. We have a technologically driven translation workflow, with human linguists involved for quality control. If you need translation or localization services, contact us today at [email protected] or (530) 750-2040.
 

Artificial intelligence (AI) is transforming many industries, including language services, like interpreting, and the healthcare sector. Working at the intersection of these areas, we see our clients struggling to figure out how best to benefit from exciting developments in automation and AI-powered technologies without compromising interpreting and translation quality. At Avantpage, we’re supporting healthcare organizations by implementing AI carefully, safely, and transparently to optimize business processes and translation workflows.   Tools like machine translation can greatly enhance the work of human translators and there are ways that human interpreters can leverage AI to support their job as well. But while Avantpage embraces AI in certain aspects of language access, we encourage our clients to carefully consider the implications of services that offer to entirely replace linguists with AI interpreting technology. These offers of a service that’s faster and cheaper are understandably appealing when you’re on a tight budget. But are there more risks than benefits in a highly regulated industry where patient experience and outcomes are the top priorities?   The National Council on Interpreting in Healthcare¹ recently published their “Guidance for Healthcare Organizations Evaluating the Potential Use of AI-generated Interpreting,” where they highlight many areas for organizations to ask deeper questions.   We consider some of these here, highlighting five core concerns for health plans and hospitals considering an exclusively AI-interpreting model.   A crucial part of language access planning is understanding which languages you need to provide. The core languages spoken in your patient communities should be easy to identify. But how will you offer support to a patient who needs an interpreter in a less common dialect or a language of lesser diffusion? It is important to remember that AI models are only as good as the data used to train them. If that data is limited (eg. for indigenous languages or other less common languages and dialects), you cannot rely on the same level of accuracy as a commonly spoken language such as Spanish. If you are using an AI interpreting model, does your provider have a backup solution for languages that are not available through AI? And on the topic of Spanish, can the AI tool pivot to provide a Spanish speaker from Puerto Rico versus one from Mexico, to match the needs of an individual caller? These are important questions to consider when considering an AI interpreting provider.  As mentioned above, it is important to consider whether an AI interpreting service has been trained using sufficient data to provide accurate output for your members or patients. But the quality assurance questions don’t end there.   Human interpreters are required to meet certain qualifying criteria. These include proof of subject-specific training, nationally recognized credentials, field experience, continuing education, and adherence to the National Code of Ethics. Years of experience (Avantpage requires a minimum of 3 years) have taught interpreters to self-monitor and to constantly look out for subtle misunderstandings between the parties. If a human interpreter identifies a communication issue, such as a cultural misunderstanding or a moment of unintelligible speech, they can intervene before it becomes a larger issue. How does an exclusively AI-generated interpreter guarantee the same level of nuanced expertise and service quality? Has the AI been trained with terminology that is specific to the healthcare industry? Does the service offer an opportunity to switch to a human interpreter if it becomes apparent that a quality issue is developing on the call? Distinct from quality assurance, quality control is about monitoring performance and improving a service to maximize quality and outcomes for patients. When it comes to written translation, tools with AI features learn and improve through detailed feedback from human linguists. If you are considering an exclusively AI interpreting model, find out how the tool will learn from mistakes. Will there be an opportunity for your staff, patients, or human interpreters to give feedback, and will that be incorporated into the model’s improvement? How does a user submit a complaint, how will errors be tracked, and who is ultimately responsible should a serious error occur? Will the service be spot-checked and monitored from the quality perspective, to be sure that patients and providers are receiving the level of quality they deserve? If you use a service like this, are you complying with federal regulations surrounding quality and AI, such as those outlined in Section 1557 of the ACA? Anyone who works in healthcare in the United States understands the importance of the Health Information Portability and Accountability Act (HIPAA) and, in the European Union, the General Data Protection Regulation (GDPR). When a health organization contracts with a language services provider such as Avantpage, that LSP must show that they offer translation and interpreting services that align with these regulations. This includes HIPAA training for staff and interpreters, secure and encrypted phone lines and servers, and many other data security measures that are monitored through annual audits. AI interpreting models work by recording conversations, converting them to text, translating them, and then voicing them in the target language. If an AI interpreting service is recording patient data, where is it being stored and for how long? What measures are in place to prevent a breach? Are patients aware that they are being recorded, and is there an option to opt out of this and be routed to a human interpreter on a non-recorded line?   Addressing the factors above is an important part of building patient trust, which is arguably one of the most crucial factors when it comes to providing quality care and improving health outcomes. You want your members and patients to feel that you care at every turn in their journey with your organization. As the National Center for Interpreting in Health Care stated in their guidance¹, “qualified human interpreters often play a crucial, but understudied role not only in ensuring effective communication but also in building empathy and trust in healthcare settings, aspects of care that AI will not be able to fully replicate.”   Where can you carefully implement AI within language services without eliminating the human approach? Can you be sure that your patients are not affected by unchecked racial biases, which can be unwittingly introduced in an AI interpreting tool by virtue of its limited training? If you do decide to offer AI interpreting services, is there an option for your patients to opt out and reach a human interpreter without any delay or other impact?  Avantpage is excited about the progress in this field and we have been incorporating AI-powered processes into our workflows for some time. By asking questions such as those suggested above, healthcare organizations can make informed decisions about the extent to which they are willing to replace human steps with AI. To learn more about how we provide meaningful language access by combining AI with human expertise, contact our team today.