Tag
The paper proposes a distribution-aware method for identifying language-specific neurons in multilingual large language models by leveraging pairwise activation distribution overlaps, improving specificity in isolating causal effects across languages.
This paper evaluates multilingual LLMs in Urdu story generation, revealing significant linguistic, semantic, and cultural weaknesses, which highlights limitations for low-resource languages.
This study compares probing techniques for identifying latent language in multilingual LLMs, finding that different methods yield inconsistent results, indicating they expose distinct aspects of multilingual processing rather than a single internal lingua franca.
This research investigates whether multilingual large language models develop shared internal representations for mathematical reasoning across languages, introducing a novel Geometry-Invariant Sparse Autoencoder (GI-SAE) method. It finds that cross-language feature sharing is model-dependent and that geometric similarity does not consistently imply functional interchangeability.
Introduces SurakshaEval, a safety benchmark for LLMs covering ten Indian languages and English, with human-written prompts spanning seven harm types. Benchmarks multilingual LLMs and finds issues like over-refusal and missed implicit bias in Indic contexts.
This paper proposes a data-driven framework using embeddings from multilingual LLMs to detect lexical gaps between languages, achieving high accuracy in Korean-English pairs.
Google Research introduces LocQA, a 12-language dataset revealing that multilingual LLMs exhibit strong US-centric and population-based locale biases when answering ambiguous locale-dependent questions.
This paper evaluates the mathematical reasoning capabilities of large language models in Sinhala and Tamil, two low-resource South Asian languages, using a parallel dataset of independently authored problems. The study demonstrates that while basic arithmetic transfers well across languages, complex reasoning tasks show significant performance degradation in non-English languages, with implications for deploying AI tutoring tools in multilingual educational contexts.
This paper presents a systematic benchmark of token pruning—a compression technique that removes tokens and embeddings for irrelevant languages—applied to Korean-centric LLM tasks. The study evaluates popular multilingual models (Qwen3, Gemma-3, Llama-3, Aya) across different vocabulary configurations and finds that token pruning significantly improves generation stability and reduces memory footprint for domain-specific deployments.