REVIEW 14 cited by
Language-Specific Neurons: The Key to Multilingual Capabilities in Large Language Models
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Large language models (LLMs) demonstrate remarkable multilingual capabilities without being pre-trained on specially curated multilingual parallel corpora. It remains a challenging problem to explain the underlying mechanisms by which LLMs process multilingual texts. In this paper, we delve into the composition of Transformer architectures in LLMs to pinpoint language-specific regions. Specially, we propose a novel detection method, language activation probability entropy (LAPE), to identify language-specific neurons within LLMs. Based on LAPE, we conduct comprehensive experiments on several representative LLMs, such as LLaMA-2, BLOOM, and Mistral. Our findings indicate that LLMs' proficiency in processing a particular language is predominantly due to a small subset of neurons, primarily situated in the models' top and bottom layers. Furthermore, we showcase the feasibility to "steer" the output language of LLMs by selectively activating or deactivating language-specific neurons. Our research provides important evidence to the understanding and exploration of the multilingual capabilities of LLMs.
Forward citations
Cited by 14 Pith papers
-
One Anchor for All: Unified Multilingual and Multimodal Safety Alignment for LVLMs
A small set of model neurons shared across languages and modalities can transfer English-only safety training to multilingual and multimodal refusal behavior.
-
Targeted Interpretable Safety Neuron Enhancement for Multilingual Vision-Language Large Models
Precise Shield identifies safety neurons in VLLMs via activation contrasts and aligns only them with gradient masking, boosting safety, preserving generalization, and enabling zero-shot cross-lingual and cross-modal transfer.
-
Understanding and Mitigating Cross-lingual Privacy Leakage via Language-specific and Universal Privacy Neurons
Cross-lingual privacy leakage in LLMs is driven by a mix of language-universal and language-specific neurons, and deactivating those neurons lowers measured leakage by 23.3% to 31.6%.
-
Disentangling Language and Culture for Evaluating Multilingual Large Language Models
A new dual-axis evaluation framework shows multilingual LLMs answer culture-specific questions best when the question language matches the cultural context, with partial neuron-level evidence for the effect.
-
Paths Not Taken: Understanding and Mending the Multilingual Factual Recall Pipeline
LLMs recall facts through an English-centric internal path and then translate the answer; injecting a translation vector and a recall vector raises accuracy by over 35 percentage points in the weakest language.
-
Pierce the Mists, Greet the Sky: Decipher Knowledge Overshadowing via Knowledge Circuit Analysis
PhantomCircuit traces knowledge overshadowing to attention circuits during training and prunes circuit edges to recover the overshadowed answer.
-
Sparse Neuron Ablation Triggers Catastrophic Collapse of the Language Core in Large Vision-Language Models
Ablating just four neurons in LLaVA-1.5-7b's language-model down-projection layer triggers complete output collapse, with critical neurons concentrated in the language backbone.
-
What Language(s) Does Aya-23 Think In? How Multilinguality Affects Internal Language Representations
Aya-23-8B appears to activate multiple related languages internally and concentrate code-mixing neurons in final layers, but the paper's own limitations undercut the claim that these are language-specific neurons.
-
Detecting and Pruning Prominent but Detrimental Neurons in Large Language Models
Pruning the most attribution-prominent MLP neurons in a single layer, chosen via a 10-sample validation sweep, consistently improves multiple-choice accuracy across four instruction-tuned LLMs.
-
Pruning General Large Language Models into Customized Expert Models
Cus-Prun identifies and removes neurons that are irrelevant to a user's target language, domain, and task, producing specialized expert models without post-training.
-
How Syntax Specialization Emerges in Language Models
Syntactic specialization in language models emerges gradually during training, concentrates in particular layers, and appears to stabilize after roughly 16 million tokens.
-
Understanding How Value Neurons Shape the Generation of Specified Values in LLMs
ValueLocate identifies value neurons via activation probability differences between opposing value prompts, and amplifying or suppressing these neurons alters G-EVAL value scores in four LLMs.
-
Unveiling Instruction-Specific Neurons & Experts: An Analytical Framework for LLM's Instruction-Following Capabilities
Activation-frequency analysis identifies sparse units in LLMs that respond to instructions; same-category instructions share more of these units than different-category ones, and fine-tuning measurably changes the sets.
-
Neural Incompatibility: The Unbridgeable Gap of Cross-Scale Parametric Knowledge Transfer in Large Language Models
Directly transferring parameters between differently-sized language models is unreliable; the paper proposes a pre-alignment method (LaTen) and explains the failure via 'Neural Incompatibility'.
Discussion (0). Sign in to comment.