On the Multilingual Ability of Decoder-based Pre-trained Language Models: Finding and Controlling Language-Specific Neurons

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Kojima, Takeshi, Okimura, Itsuki, Iwasawa, Yusuke, Yanaka, Hitomi, Matsuo, Yutaka
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913297144479744
author Kojima, Takeshi
Okimura, Itsuki
Iwasawa, Yusuke
Yanaka, Hitomi
Matsuo, Yutaka
author_facet Kojima, Takeshi
Okimura, Itsuki
Iwasawa, Yusuke
Yanaka, Hitomi
Matsuo, Yutaka
contents Current decoder-based pre-trained language models (PLMs) successfully demonstrate multilingual capabilities. However, it is unclear how these models handle multilingualism. We analyze the neuron-level internal behavior of multilingual decoder-based PLMs, Specifically examining the existence of neurons that fire ``uniquely for each language'' within decoder-only multilingual PLMs. We analyze six languages: English, German, French, Spanish, Chinese, and Japanese, and show that language-specific neurons are unique, with a slight overlap (< 5%) between languages. These neurons are mainly distributed in the models' first and last few layers. This trend remains consistent across languages and models. Additionally, we tamper with less than 1% of the total neurons in each model during inference and demonstrate that tampering with a few language-specific neurons drastically changes the probability of target language occurrence in text generation.
format Preprint
id arxiv_https___arxiv_org_abs_2404_02431
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle On the Multilingual Ability of Decoder-based Pre-trained Language Models: Finding and Controlling Language-Specific Neurons
Kojima, Takeshi
Okimura, Itsuki
Iwasawa, Yusuke
Yanaka, Hitomi
Matsuo, Yutaka
Computation and Language
Current decoder-based pre-trained language models (PLMs) successfully demonstrate multilingual capabilities. However, it is unclear how these models handle multilingualism. We analyze the neuron-level internal behavior of multilingual decoder-based PLMs, Specifically examining the existence of neurons that fire ``uniquely for each language'' within decoder-only multilingual PLMs. We analyze six languages: English, German, French, Spanish, Chinese, and Japanese, and show that language-specific neurons are unique, with a slight overlap (< 5%) between languages. These neurons are mainly distributed in the models' first and last few layers. This trend remains consistent across languages and models. Additionally, we tamper with less than 1% of the total neurons in each model during inference and demonstrate that tampering with a few language-specific neurons drastically changes the probability of target language occurrence in text generation.
title On the Multilingual Ability of Decoder-based Pre-trained Language Models: Finding and Controlling Language-Specific Neurons
topic Computation and Language
url https://arxiv.org/abs/2404.02431