Every non-English ChatGPT conversation contains a tiny irony. The chatbot will reply. It will construct sentences, conjugate verbs, and even try idioms. However, there is something wrong: a stiffness in the syntax, a propensity to use American phrasing when the user didn’t ask for it, and a subtle flattening of tone that is instantly noticeable to anyone who speaks, say, Urdu, Tagalog, or even British English fluently. The machine is making an effort. It simply involves thinking in a different language first.
This is the silent crisis at the heart of what some researchers refer to as the bilingual ChatGPT era, a time when AI tools are functionally dominated by English, particularly American English, but technically multilingual. Roughly 93% of the training data used to feed large language models comes from English-language sources, according to research from Johns Hopkins University and an examination of OpenAI’s own documentation. As a result, the system performs well when translating into English but struggles when translating into other languages, especially those that use non-Latin scripts like Arabic, Tamil, or Korean. Researchers at the University of Oregon discovered that ChatGPT was more likely to completely fabricate information and performed measurably worse when answering factual questions or summarizing text in non-English languages.
It’s difficult to ignore the pattern. According to computer scientist Pascale Fung, who speaks seven languages and oversees the Center for AI Research at the Hong Kong University of Science and Technology, Indonesian store owners may use AI to list products in English, but the opposite almost never occurs. She has noticed that Americans are just not driven to communicate with people who speak different languages. What was already true is reinforced by the tools, and more.
The speed and invisibility of this moment set it apart from previous waves of English-language dominance, such as Hollywood, the internet, and global finance. Most users don’t complain when AI systems produce awkward results in minority languages, flatten regional idioms, or default to American spellings. They adapt. They go to English. They no longer anticipate that the tool will arrive at their location. According to research from the AI lab at UC Berkeley, AI models favor American spellings over their British counterparts by up to 43%, subtly pushing the written world toward a single dialect without a vote.
Some in the industry are sounding the alarm. California Senator Alex Padilla questioned OpenAI’s Sam Altman about the company’s efforts to bridge the language gap during a congressional hearing. Altman discussed obtaining new data sets and forming alliances with governments. However, OpenAI admitted in its own GPT-4 documentation that the majority of its training data and evaluation efforts were in English, using what the company referred to as “a US-centric point of view.” To put it more bluntly, any good Spanish performance was described as a bonus rather than a goal in a support forum post made by an OpenAI employee in response to a user seeking assistance in Spanish.
Contrary to popular belief, corporate training programs that use AI language tools report that they require more human instruction rather than less because employees find it difficult to understand and improve what the machine generates. The subtlety is lost. The cultural background disappears. Strategic communication, including persuasion, negotiation, and audience-specific writing, still requires a depth of understanding that no language model currently offers and may not be able to, according to applied linguists who work with multinational corporations.
In the meantime, scholars such as Jessica Forde of Brown University have identified a significant imbalance in the assessment of these systems. Developers test ChatGPT’s ability to produce humor or act like a lawyer in English. In Bengali or Swahili, no one is using the same benchmarks. The lack of ambition speaks louder than any press release because it just isn’t there yet.
Linguists and educators believe that the window of opportunity to do this correctly is closing. Millions of users develop habits around English-first workflows each month when AI tools function with uneven language capabilities. Dialects are eroding. Local expressions become unsteady. The bilingual ChatGPT era is coming, but it’s coming unevenly—generous to some languages, indifferent to most others, and moving quickly enough that the harm to linguistic diversity may already be hard to undo by the time someone develops a suitable solution.
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.
