Artificial intelligence (AI) models are proving to be surprisingly fluent in Polish. A new study by The University of Maryland and Microsoft reveals that, out of 26 languages tested, Polish emerged as the most effective for prompting AI systems. This finding challenges conventional wisdom about language complexity and AI training data.
Researchers put several leading AI language models—including OpenAI, Google Gemini, Qwen, Llama, and DeepSeek—through their paces. These models were given identical tasks across all 26 languages to see which yielded the most accurate responses. The results were striking: Polish consistently outperformed the others, achieving an average accuracy of 88%.
Beyond Expected Performance
This unexpected result is particularly noteworthy because Polish has historically been considered one of the more challenging languages for humans to learn. Its complex grammar and unfamiliar phonemes pose a significant hurdle for native English speakers. Yet, when it comes to AI, language complexity doesn’t seem to be as defining a factor.
Interestingly, English, often seen as the dominant global language in technology, only ranked sixth. This suggests that raw data volume alone isn’t the sole determinant of AI language proficiency. Furthermore, Chinese, despite having a vast amount of online text data available for training, performed disappointingly, ranking near the bottom.
The top 10 most effective languages for conversational AI were:
- Polish (88%)
- French (87%)
- Italian (86%)
- Spanish (85%)
- Russian (84%)
- English (83.9%)
- Ukrainian (83.5%)
- Portuguese (82%)
- German (81%)
- Dutch (80%)
What This Means for AI and Language
This study highlights a few key takeaways:
- The impact of linguistic structure: Perhaps the structure of Polish grammar, or unique phonetic features, lend themselves better to certain types of AI processing. Further research is needed to pinpoint exactly why Polish excels.
- Data availability isn’t everything: While extensive training data is crucial, it doesn’t guarantee top performance. Other factors like linguistic complexity and model architecture play a role.
- A shift in language priorities: The dominance of English in AI might be challenged as researchers explore other languages with strong performance potential. This could lead to more inclusive and globally accessible AI technologies.
This research opens exciting avenues for exploring the intersection of language, cognition, and artificial intelligence. As AI continues to evolve, understanding which languages it finds most intuitive could shape its future development and impact on communication worldwide.
