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

TL;DR

This paper investigates how decoder-based multilingual pre-trained language models internally represent different languages, identifying language-specific neurons mainly in early and late layers, and shows that altering these neurons significantly impacts language generation.

Contribution

The study reveals the existence and distribution of language-specific neurons in decoder-based PLMs and demonstrates their critical role in language generation through targeted neuron tampering.

Findings

Language-specific neurons are mostly in early and late layers.

Less than 5% neuron overlap between languages.

Tampering with language-specific neurons alters language output significantly.

Abstract

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.