Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Chou, Cheng-Ting, Liu, George, Sun, Jessica, Blondin, Cole, Zhu, Kevin, Sharma, Vasu, O'Brien, Sean
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2507.13410
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908596455866368
author Chou, Cheng-Ting
Liu, George
Sun, Jessica
Blondin, Cole
Zhu, Kevin
Sharma, Vasu
O'Brien, Sean
author_facet Chou, Cheng-Ting
Liu, George
Sun, Jessica
Blondin, Cole
Zhu, Kevin
Sharma, Vasu
O'Brien, Sean
contents Deterministically controlling the target generation language of large multilingual language models (LLMs) remains a fundamental challenge, particularly in zero-shot settings where neither explicit language prompts nor fine-tuning are available. In this work, we investigate whether sparse autoencoder (SAE) features, previously shown to correlate with interpretable model behaviors, can be leveraged to steer the generated language of LLMs during inference. Leveraging pretrained SAEs on the residual streams of Gemma-2B and Gemma-9B, we identify features whose activations differ most significantly between English and four target languages: Chinese, Japanese, Spanish, and French. By modifying just a single SAE feature at one transformer layer, we achieve controlled language shifts with up to 90\% success, as measured by FastText language classification, while preserving semantic fidelity according to LaBSE (Language-Agnostic BERT Sentence Embedding) similarity. Our analysis reveals that language steering is most effective in mid-to-late transformer layers and is amplified by specific attention heads disproportionately associated with language-sensitive SAE features. These results demonstrate the promise of sparse feature steering as a lightweight and interpretable mechanism for controllable multilingual generation.
format Preprint
id arxiv_https___arxiv_org_abs_2507_13410
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Causal Language Control in Multilingual Transformers via Sparse Feature Steering
Chou, Cheng-Ting
Liu, George
Sun, Jessica
Blondin, Cole
Zhu, Kevin
Sharma, Vasu
O'Brien, Sean
Computation and Language
Artificial Intelligence
Deterministically controlling the target generation language of large multilingual language models (LLMs) remains a fundamental challenge, particularly in zero-shot settings where neither explicit language prompts nor fine-tuning are available. In this work, we investigate whether sparse autoencoder (SAE) features, previously shown to correlate with interpretable model behaviors, can be leveraged to steer the generated language of LLMs during inference. Leveraging pretrained SAEs on the residual streams of Gemma-2B and Gemma-9B, we identify features whose activations differ most significantly between English and four target languages: Chinese, Japanese, Spanish, and French. By modifying just a single SAE feature at one transformer layer, we achieve controlled language shifts with up to 90\% success, as measured by FastText language classification, while preserving semantic fidelity according to LaBSE (Language-Agnostic BERT Sentence Embedding) similarity. Our analysis reveals that language steering is most effective in mid-to-late transformer layers and is amplified by specific attention heads disproportionately associated with language-sensitive SAE features. These results demonstrate the promise of sparse feature steering as a lightweight and interpretable mechanism for controllable multilingual generation.
title Causal Language Control in Multilingual Transformers via Sparse Feature Steering
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2507.13410