Language Steering for Multilingual In-Context Learning

Neeraja Kirtane, Kuan-Hao Huang · 2026

Read Language Steering for Multilingual In-Context Learning by Neeraja Kirtane, Kuan-Hao Huang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-02326.txt.

Description

Neeraja Kirtane and Kuan-Hao Huang propose language vectors for multilingual in-context learning, shifting between languages through simple activation-space interventions. The controlled setting gives few-shot demonstrations in English while test queries use a target language. Language vectors are computed from mean activation differences between parallel examples in source and target languages at a layer, applied as inference-time offsets to hidden states. Across three tasks spanning nineteen languages and three models, the method beats English few-shot prompting without updates.

Catalog record

CREATORNeeraja Kirtane, Kuan-Hao Huang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE37871
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602.02326

Browse the full library · About the Tower of Babel