Study Reveals AI Models’ Bias Against Dialects

AI models show bias against dialects, attributing negative stereotypes.

Large language models exhibit bias against dialect speakers, attributing negative stereotypes to them. This conclusion was reached by researchers from Germany and the United States, reports DW. 

“I believe we see truly shocking epithets attributed to dialect speakers,” noted one of the study’s lead authors, Minh Duc Bui, in a comment to the publication. 

An analysis by Johannes Gutenberg University revealed that ten tested models, including ChatGPT-5 mini and Llama 3.1, described speakers of German dialects (Bavarian, Cologne) as “uneducated,” “farm workers,” and “prone to anger.”

The bias intensified when the AI was explicitly pointed to the dialect.

Other Instances 

Similar issues are observed globally by researchers. A study from the University of California, Berkeley in 2024 compared ChatGPT’s responses to various English dialects (Indian, Irish, Nigerian). 

It was found that the chatbot responded with more pronounced stereotypes, derogatory content, and a condescending tone compared to standard American or British English. 

Emma Harvey, a computer science graduate student at Cornell University, called the bias against dialects “significant and troubling.” 

In the summer of 2025, she and her colleagues also discovered that Amazon’s shopping assistant Rufus provided vague or even incorrect answers to people writing in African American English. If queries contained errors, the model responded rudely. 

Another striking example of neural network prejudice involved an Indian job applicant who turned to ChatGPT to check his resume in English. The chatbot ended up changing his surname to one associated with a higher caste. 

“The widespread adoption of language models threatens not just to preserve entrenched prejudices but to amplify them on a large scale. Instead of mitigating harm, technologies risk giving it a systemic character,” said Harvey.

However, the crisis is not limited to bias—some models simply do not recognize dialects. For instance, in July, the AI assistant of Derby City Council (England) failed to recognize a radio host’s dialect when she used words like mardy (“whiner”) and duck (“dear”) on air. 

What Can Be Done? 

The problem lies not in the AI models themselves but rather in how they are trained. Chatbots read vast amounts of text from the internet, which they then use to generate responses. 

“The main question is who writes this text. If it contains biases against dialect speakers, the AI will replicate them,” explained Carolin Holtermann from the University of Hamburg.

She emphasized, however, that the technology has an advantage: 

“Unlike humans, AI systems’ biases can be identified and ‘switched off.’ We can actively combat such manifestations.”

Some researchers propose creating customized models for specific dialects as an advantage. In August 2024, the company Acree AI already introduced the Arcee-Meraj model, which works with several Arabic dialects. 

According to Holtermann, the emergence of new and more adaptable LLM allows us to view AI “not as an enemy of dialects, but as an imperfect tool that can be improved.”

As reported in The Economist, journalists warned of the risks AI toys pose to children’s mental health.