Journey to the Center of the Knowledge Neurons: Discoveries of Language-Independent Knowledge Neurons and Degenerate Knowledge Neurons
Yuheng Chen, Pengfei Cao, Yubo Chen, Kang Liu, Jun Zhao
TL;DR
The paper tackles how factual knowledge is stored in multilingual pre-trained language models by introducing Architecture-adapted Multilingual Integrated Gradients (AMIG) to precisely localize knowledge neurons. It uncovers Language-Independent Knowledge Neurons (LIKN) that preserve facts across languages and Degenerate Knowledge Neurons (DKN) that exhibit functional overlap, enhancing robustness and enabling internal fact-checking. Through experiments on m-BERT and m-GPT with the mLAMA dataset, AMIG achieves superior localization compared to baselines, LIKN enables effective cross-lingual editing, and DKN improves fact-checking performance, including in Chinese and autoregressive settings. The work provides a mechanistic view of knowledge storage in multilingual PLMs and offers practical tools for editing and validating factual knowledge, with code available online.
Abstract
Pre-trained language models (PLMs) contain vast amounts of factual knowledge, but how the knowledge is stored in the parameters remains unclear. This paper delves into the complex task of understanding how factual knowledge is stored in multilingual PLMs, and introduces the Architecture-adapted Multilingual Integrated Gradients method, which successfully localizes knowledge neurons more precisely compared to current methods, and is more universal across various architectures and languages. Moreover, we conduct an in-depth exploration of knowledge neurons, leading to the following two important discoveries: (1) The discovery of Language-Independent Knowledge Neurons, which store factual knowledge in a form that transcends language. We design cross-lingual knowledge editing experiments, demonstrating that the PLMs can accomplish this task based on language-independent neurons; (2) The discovery of Degenerate Knowledge Neurons, a novel type of neuron showing that different knowledge neurons can store the same fact. Its property of functional overlap endows the PLMs with a robust mastery of factual knowledge. We design fact-checking experiments, proving that the degenerate knowledge neurons can help the PLMs to detect wrong facts. Experiments corroborate these findings, shedding light on the mechanisms of factual knowledge storage in multilingual PLMs, and contribute valuable insights to the field. The code is available at https://github.com/heng840/AMIG.