Lost in Translation: The Hidden Flaws of AI in Bridging Cultures
With over 7,000 languages spoken worldwide, AI-powered translation tools like Google Translate and DeepL help us bridge communications gaps easily. But can AI truly understand culture?
While these tools excel at interpreting words, how proficient are they at understanding nuances, idioms, humor, and context?
As we rely more and more on AI for translation, we must ask: what gets lost in translation, and, most importantly, why does it matter? Let’s explore the strengths, limitations, and hidden pitfalls of AI-driven language translation.
AI is a field focused on developing computers that can learn and think like humans while processing data at a scale that exceeds human capacity. These technologies rely on machine learning and deep learning to generate complex content. This AI-generated content includes data analytics and predictions which can be extremely valuable in business.
For language translations, AI uses algorithms via neural networks – a type of program designed to mimic how human neurons work to make decisions. Neural networks analyze vast amounts of data in different languages, identifying patterns and relationships between sentences and words faster than humans can.
Some benefits of AI-powered translation include:
Speed: AI can process large quantities of data in an instant, even in real time.
Range: It can translate hundreds of languages, something the smartest human polyglot cannot do.
Cost–effectiveness: One tool can translate increasing amounts of content without the need for continued investment.
Continuous learning: These systems have a built-in feedback loop. They improve and adapt the more we use them.
The cultural blind spots of AI translation
Does AI currently have specific blind spots for translation tools on the market? While the benefits of AI translation can be attractive, we must also consider some of its drawbacks:
Accuracy: It can miss cultural backdrop, emotionality and intended meaning, especially if the original text uses idioms, metaphors or complex language. For example, it might translate a phrase like ‘let the cat out of the bag’ literally. The phrase may work in German (die Katze aus dem Sack lassen), but not in French. How many other translations would come out sounding odd–or potentially offensive–if translated literally?
Bias: Since AI translation relies on human-generated training data, it may pull from our own biases. This can lead to translations that favor or misrepresent certain groups. For instance, research has shown that AI interprets terms like ‘anxious’ as a primarily feminine trait. Or it defaults to language that depicts doctors as men, while nurses are women. This mirrors human gender bias, rather than producing neutral language. For that matter, almost all language translation tools default to male pronouns when referring to people.
Humor: AI-powered translation tools may not catch humor or sarcasm tied to the cultural context of the original language.
Formality: Different cultures have varying norms of formality and politeness, which AI might fail to capture. This can be particularly damaging in certain fields – like politics – where maintaining formality is crucial.
Additional ethical concerns include:
Climate change: Training and using AI models require a significant amount of electricity, contributing to harmful CO2 emissions.
To illustrate some of the cultural context that AI translation often misses, let’s look at a few real-world case studies:
In literature: In 2024, Dutch publishers Veen Bosch & Keuning announced plans to use AI to translate literature into English, sparking concerns regarding potential insensitive or inaccurate translations that could hamper the hard work of many authors.
In daily life: In late 2024, Meta introduced real-time AI translation as a new feature for their Ray-Ban smart glasses. With this, users can have conversations with Spanish, Italian or French speakers. During a demo, a reviewer having an AI-translated conversation with a Spanish speaker noticed that the technology struggled with slang, translating no manches (no way!) literally as ’no stain’.
In government: In 2019, an Afro-Indigenous Brazilian-born man profiled in the news using the pseudonym Carlos fled Brazil due to gang violence. He arrived in the USA, where he was detained at a Customs Enforcement center. The staff spoke only English and Spanish, while Carlos spoke Portuguese, which he could neither read nor write. The staff used an AI-powered voice translator that couldn’t understand Carlos’ regional accent. Multiple errors in AI-powered communication caused his detainment and asylum application to stretch on for over 6 months.
Can AI ever learn culture?
Efforts at training AI on cultural aspects are being made, but for all of us to manage our expectations of those outcomes, transparency needs to be at the forefront when discussing the limitations of AI.
To improve AI translation tools, we can prioritize feeding this technology cultural contexts and ensure training data reflects diverse perspectives. Over time, this should generate translations that are less biased and respectful of the original content. AI-human collaboration can involve translators, industry experts and cultural advisors in the process. As Michele Hutchison, a Booker Prize-winning translator, emphasized: “A translator translates more than just words; we build bridges between cultures, taking into account the target readership every step of the way.”
In short, while AI can be a valuable tool for reducing costs and speeding up work, it can’t fully replace human intelligence in translation services today.
AI and human synergy
The future isn’t just automation—it’s collaboration. AI has become useful for cutting costs and accelerating processes. But when it comes to translation services, we must consider shortcomings like the lack of cultural context, misrepresentation of certain groups and inaccuracy in conveying intent. For ethical progress, we are required to develop the cost-effectiveness of AI-powered tools while addressing concerns about recorded bias and ensuring that true meaning is preserved in our language translations.
Alison Maciejewski Cortez is Chilean-American, born and raised in California. She studied abroad in Spain, has lived in multiple countries, and now calls Mexico home. She believes that learning how to order a beer in a new language reveals a lot about local culture. Alison speaks English, Spanish, and Thai fluently and studies Czech and Turkish. Her tech copywriting business takes her around the world and she is excited to share language tips as part of the Lingoda team. Follow her culinary and cultural experiences on X.
