The phenomenon of Arabic replies appearing within English-language ChatGPT conversations is causing confusion among a number of users in the United States, after they noticed the sudden inclusion of Arabic words in the replies.
Recently, social media users have been sharing screenshots showing that the chatbot sometimes adds Arabic texts or changes some words and numbers into non-English formats, without the user requesting it.
One user on Reddit said that it happened to him more than once, both on his phone and his work computer, noting that ChatGPT once presented him with food recipes written in Arabic even though he does not live in an Arab country.
Other users also reported the appearance of different languages within the same replies, such as Armenian, Hebrew, Spanish, Chinese and Russian, along with some numbers changing to the Arabic version.
Although some have interpreted this as a kind of "AI hallucination," a condition in which the system produces inaccurate or illogical answers, experts point out that it is related to the way language models work.
ChatGPT, a large language model developed by OpenAI, relies on dividing text into small units called "symbols," which may represent whole words, parts of words, or even passages from different languages.
During the generation of responses, the model selects the most likely symbols according to the context, which may sometimes result in the appearance of words from other languages if they are statistically appropriate or more efficient in expression.
This does not mean that artificial intelligence deliberately changes languages, but rather that it builds its sentences based on probabilities derived from training data.
ChatGPT has been widely used by hundreds of millions around the world since its launch in 2022, for tasks such as writing, translating, explaining concepts and producing content.
Despite competition from other systems such as Google’s Gemini, xAI’s Grok and Anthropic’s Cloude, ChatGPT remains one of the most widely used generative AI tools globally.
OpenAI has previously acknowledged technical issues related to generating incomprehensible text in earlier updates, but has not issued any recent official explanations regarding the phenomenon of language mixing or the unexpected appearance of Arabic in English responses.
Some users believe that the words that appear in other languages are not entirely random, but may sometimes have meanings close to the English words they replaced.
Experts confirm that understanding this phenomenon requires understanding how texts are divided into "symbols" within linguistic models, where a single word may be broken down into several parts.
For example, a word like "understanding" can be broken down into multiple parts such as "under", "stand", and "ing".
Accordingly, the model selects the most probable sequence to complete the sentence, even if this sometimes results in the insertion of words from different languages.
While the debate continues about the cause of this phenomenon, some users see it as unusual compared to previous versions, prompting them to question its nature and causes.
