AI vs. human translations
Editor’s note: Andrea Doebler is the linguist and assistant at SyncScript. This is an edited version of an article that originally appeared under the title “The trained linguist’s edge.”
In the last few years, we have seen the rapid growth and slight fall of AI technology as a one-stop solution. What was once touted as an impossibly simple and efficient fix for any issue previously requiring work by an actual human has revealed itself to be, well, inaccurate. From the AI Summary at the top of your Google search results to the eloquent answers provided by ChatGPT, each solution produced by AI technology comes with the same disclaimer: “Generative AI is experimental. X program can make mistakes.” This isn’t to say that AI technology doesn’t have a place as a tool in a larger process, but it’s hard to trust a machine with the important stuff when it tells you directly that it might be giving you the wrong answer!
Streamlining and increasing productivity while lessening the workload are huge potential perks with AI tech, but understanding capabilities and limitations is essential to maintaining data integrity. Many opt for AI tech when transcribing into simple Word document formats in English, with the perks of lower cost and faster turnaround time outweighing the potential negatives. But what happens when you throw some survey programming language into the mix, or when a project requires translation (or both)?
The more complex and technical the topic, the less accurate the AI transcript. Without a linguist at the helm for translation, there is no way to be sure that the machine generated result is even correct. When transcripts are the foundation for drawing insights in any qualitative research project, the resulting translation must be correct to maintain data integrity.
So, what does this mean when it comes to considering AI tech for market research translations?
Comparing CAT tools and AI translation
Let’s pivot and talk about computer assisted translation (CAT) tools. CAT tools have been around for decades, serving as an indispensable asset to professional translators. These tools do not replace human expertise but rather enhance a translator’s efficiency, accuracy and consistency. CAT tools work by segmenting text into manageable chunks – typically sentence by sentence – and storing previous translations in a translation memory. This means that when a similar phrase or sentence appears in future projects, the system can suggest previously used translations, ensuring consistency across large documents or ongoing projects.
Additionally, CAT tools often include features such as terminology databases, spell checkers and quality assurance checks to catch inconsistencies or errors. They also allow for document formatting retention, meaning a complex layout – like a legal contract or a technical manual – remains intact while the translation occurs. Unlike AI-powered machine translation, CAT tools rely on human input at every stage, with the translator making the final decision on how a sentence is structured and ensuring the translation remains accurate and contextually appropriate. We can infer and determine the place of AI tech in translation by looking at the success of CAT tools in this realm over the decades – as a tool for linguists to streamline the process and not a replacement for trained professionals.
The linguist’s advantage
Even when dealing with plain text, AI translation tools have notable weak spots that can compromise accuracy. Typos in written responses, mixed-language sentences (e.g., a survey response that starts in English but switches to Spanish) and idioms that don’t have direct equivalents often trip up automated systems. AI models generally translate words and phrases based on patterns in their training data, but they lack the ability to infer meaning beyond statistical likelihood or matches in past examples.
This is where human translators have a distinct advantage. A trained linguist can recognize when a response contains a misspelled word and determine the intended meaning, something AI might translate incorrectly or ignore entirely. Similarly, they can interpret hybrid language responses with proper context, understanding why a speaker might have code-switched mid-sentence and ensuring the translation reflects that nuance.
Idioms and cultural expressions pose another major challenge for AI. A phrase like “raining cats and dogs” might be rendered into Spanish as “lloviendo gatos y perros” – a literal translation that makes no sense to a native speaker. A human translator, however, would recognize the need to substitute the equivalent Spanish idiom: “llueve a cántaros” – “it’s raining pitchers.” The ability to draw from real-world experience, cultural awareness and linguistic intuition makes human translators irreplaceable, particularly for specialized content where precision and tone matter.
The challenges of AI translation
Many translation projects involve more than just converting text from one language to another; they also require maintaining document integrity and preserving context. When it comes to surveys, for example, AI tools may not recognize that listed multiple choice answers correspond directly to the question in terms of context. These formatting intricacies complicate the process and perplex AI tools, resulting in inaccurate translations. Specialized formats, such as Excel grids, database exports or surveys embedded with programming scripts, are particularly challenging for AI translation tools. These tools may inadvertently alter or misinterpret coding elements, disrupt formulas or remove crucial formatting, leading to incorrect data outputs or broken functionality.
For example, an AI tool might translate a survey question but mistakenly alter a variable name or a placeholder meant for dynamic content. In a best-case scenario, this results in confusion. In a worst-case scenario, it could render an entire survey unusable and create hours of extra work for the unlucky team member tasked with repairing it. Linguists armed with CAT tools, however, are experts at working with structured documents and preserving non-translatable elements while ensuring the content is accurately translated and formatted correctly.
The importance of human transcription
AI translation technology is a powerful tool but it still lacks the depth of understanding, adaptability and critical thinking that human translators bring to the table. Instead of viewing AI as a replacement for trained linguists, it’s best seen as supplementary, like CAT tools – useful for speed and convenience but far from infallible when it comes to accuracy and nuance.
While new automated processes may be accurate enough for some simple documents and straightforward transcriptions, intervention from a professional is vital when it comes to specialized formats and market research translations. After all, would you build your entire project on a foundation that carries the disclaimer, “This is experimental and may contain mistakes?” Trained linguists armed with advanced tools are the only way to ensure that your key insights aren’t lost in translation.