Give your name as Nuñez when you call the automated service line of a large US bank. Not the anglicized version, but the original, with the tilde, pronounced like a native speaker from Medellín or Monterrey would. There’s a good probability it won’t be detected by the system. You might be prompted to repeat yourself. It might produce a transcript that says “Nunez,” completely removing the diacritical mark—a little omission that, in other circumstances, completely alters the meaning of the word.
After that, it either stalls, cycles back to the main menu, or routes you improperly. You put it in words. The system continues to falter. Now, you’ve been attempting to identify yourself to a machine for four minutes. It’s not an edge case. It’s Tuesday for tens of millions of Americans who speak Spanish.
Although the numbers supporting this are consistent, they are not as striking as you might think. Hispanic names and Spanish-English code-switching have error rates of about 30%, according to research on AI speech recognition and natural language processing performance across languages. This type of failure receives little attention because it doesn’t result in a single viral incident but rather an accumulating pattern of minor frictions dispersed over a population of about 62 million people.
The technical rationale is simple: Spanish data is often machine-translated from English rather than derived from native-speaker material, and about 90% of generative AI training data is in English. The model ultimately learns a type of corporate, standardized Spanish that flattens everything into a single generic accent, treating the Spanish of Madrid, Mexico City, and Buenos Aires as interchangeable, which is almost comically incorrect to anyone who has spent time in those locations.
The depth of the gap is shown by the individual phonetic failures. The tilde, which is a little mark placed above the ñ in words like “or,” “−,” and many other Spanish surnames, is not ornamental. The word is altered when it is removed. In transcription or text synthesis, AI systems frequently remove it, and depending on the context, the resulting inaccuracies might range from slightly perplexing to truly embarrassing.
What happens when someone code-switches in the middle of a sentence, switching between Spanish and English as bilingual speakers normally do in casual conversation, is equally illuminating. One language at a time, the model’s acoustic baseline becomes unstable. As a result, Spanish words are often assigned English phonetics in the transcript, which distorts both the meaning and the sound in a way that no native speaker would.
This has long been known to engineers working on automated customer support systems, and instead of solving it, they have mostly responded by routing around it. To approximate what the NLP was unable to detect, Levenshtein distance algorithms—an older, blunter technique that basically counts the number of character changes between two strings—are brought back into use. Sometimes it works.
However, it’s a workaround rather than a solution, and it highlights an unsettling difference between the real state of these systems and what their marketing claims they should be.
Tracking this issue over time makes it difficult to avoid feeling that it has been handled as a rare edge case when it is anything but. In the US market, sixty-two million people is not a rounding error. A consumer economy of $2.5 trillion a year is not a niche market. This kind of failure was always going to result from the choice, whether deliberate or not, to develop AI language systems on English-dominant data while characterizing them as multilingual.
It remains to be seen if the companies developing these technologies will make significant investments in real dialect variety and native-speaker training data, or if they will continue to close the gap using outdated algorithms and hopeful product copy. However, there are no unanswered questions about the frictions. Four minutes at a time, they occur every day.
London Bilingualism's content on health, medicine, and weight loss is solely meant for general educational and informational purposes. This website does not offer any diagnosis, treatment recommendations, or medical advice.
We consistently compile and disseminate the most recent information, findings, and advancements from the medical, health, and weight loss sectors. When content contains opinions, commentary, or viewpoints from professionals, industry leaders, or other people, it is published exactly as it is and reflects those people's opinions rather than London Bilingualism's editorial stance.
We strongly advise all readers to consult a qualified medical professional before acting on any medical, health, dietary, or pharmaceutical information found on this website. Since every person's health situation is different, only a qualified healthcare provider who is familiar with your medical history can offer you advice that is suitable for you.
In a similar vein, any legal, regulatory, or compliance-related information found on this platform is provided solely for informational purposes and should not be used without first obtaining independent legal counsel from a licensed attorney.
You understand and agree that London Bilingualism, its editors, contributors, and affiliated parties are not responsible for any decisions made using the information on this website.
