Unveiling the Influence of Amplifying Language-Specific Neurons

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Rahmanisa, Inaya, Andrylie, Lyzander Marciano, Ihsani, Mahardika Krisna, Wicaksono, Alfan Farizki, Wibowo, Haryo Akbarianto, Aji, Alham Fikri
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866914160788373504
author Rahmanisa, Inaya
Andrylie, Lyzander Marciano
Ihsani, Mahardika Krisna
Wicaksono, Alfan Farizki
Wibowo, Haryo Akbarianto
Aji, Alham Fikri
author_facet Rahmanisa, Inaya
Andrylie, Lyzander Marciano
Ihsani, Mahardika Krisna
Wicaksono, Alfan Farizki
Wibowo, Haryo Akbarianto
Aji, Alham Fikri
contents Language-specific neurons in LLMs that strongly correlate with individual languages have been shown to influence model behavior by deactivating them. However, their role in amplification remains underexplored. This work investigates the effect of amplifying language-specific neurons through interventions across 18 languages, including low-resource ones, using three models primarily trained in different languages. We compare amplification factors by their effectiveness in steering to the target language using a proposed Language Steering Shift (LSS) evaluation score, then evaluate it on downstream tasks: commonsense reasoning (XCOPA, XWinograd), knowledge (Include), and translation (FLORES). The optimal amplification factors effectively steer output toward nearly all tested languages. Intervention using this factor on downstream tasks improves self-language performance in some cases but generally degrades cross-language results. These findings highlight the effect of language-specific neurons in multilingual behavior, where amplification can be beneficial especially for low-resource languages, but provides limited advantage for cross-lingual transfer.
format Preprint
id arxiv_https___arxiv_org_abs_2507_22581
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Unveiling the Influence of Amplifying Language-Specific Neurons
Rahmanisa, Inaya
Andrylie, Lyzander Marciano
Ihsani, Mahardika Krisna
Wicaksono, Alfan Farizki
Wibowo, Haryo Akbarianto
Aji, Alham Fikri
Computation and Language
Machine Learning
Language-specific neurons in LLMs that strongly correlate with individual languages have been shown to influence model behavior by deactivating them. However, their role in amplification remains underexplored. This work investigates the effect of amplifying language-specific neurons through interventions across 18 languages, including low-resource ones, using three models primarily trained in different languages. We compare amplification factors by their effectiveness in steering to the target language using a proposed Language Steering Shift (LSS) evaluation score, then evaluate it on downstream tasks: commonsense reasoning (XCOPA, XWinograd), knowledge (Include), and translation (FLORES). The optimal amplification factors effectively steer output toward nearly all tested languages. Intervention using this factor on downstream tasks improves self-language performance in some cases but generally degrades cross-language results. These findings highlight the effect of language-specific neurons in multilingual behavior, where amplification can be beneficial especially for low-resource languages, but provides limited advantage for cross-lingual transfer.
title Unveiling the Influence of Amplifying Language-Specific Neurons
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2507.22581