Papers/2609.10758
🧪 Test?View on arXiv

Multilingual in Name Only? Cultural and Linguistic Weaknesses of LLMs in Urdu

Not provided in the content

multilinguallow-resource languagescontent generationcultural relevance
2609.10758
Builder Relevance
60%
4h ago

Abstract

This work investigates the reliability of multilingual LLMs in generating coherent and culturally relevant stories in Urdu, a low-resource language.

Reality Card

Core Claim

Current multilingual LLMs exhibit significant weaknesses in grammar, semantics, coherence, and cultural relevance when generating content in Urdu.

Method / Result

Generated a corpus of 93 Urdu stories using three contemporary LLMs and annotated errors under a nine-label taxonomy.

Limitations

The study highlights that cultural and context errors largely remain unresolved despite few-shot prompting.

Paper to code

Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.

No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.
← Back to all papers