Cross-Lingual Steering for Figurative Language Generation

Linfeng Liu, Tiffany Zhan, Louie Hong Yao, Saptarshi Ghosh, Tianyu Jiang · 2026

Read Cross-Lingual Steering for Figurative Language Generation by Linfeng Liu, Tiffany Zhan, Louie Hong Yao, Saptarshi Ghosh, Tianyu Jiang (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-2605-30443.txt.

Introduction

Linfeng Liu and four colleagues ask if vectors learned for steering in one language can shape generation in another, in a study of cross-lingual figurative steering. Within one language, steering lifts the share of the intended figure. Across 408 comparisons spanning four models, six languages, and four figures, steering lifts the figurative share from about 12 percent to about 20 percent. They split each language's steering vector into the mean of other languages' vectors plus a residual. The mean matches or beats its vector, while the residual steers worse.

Catalog record

CREATORLinfeng Liu, Tiffany Zhan, Louie Hong Yao, Saptarshi Ghosh, Tianyu Jiang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE128297
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.30443

Browse the full library · About the Tower of Babel