Rethinking what Matters: Effective and Robust Multilingual Realignment for Low-Resource Languages

Quang Phuoc Nguyen, David Anugraha, Felix Gaschi, Jun Bin Cheng, En-Shiun Annie Lee · 2025

Read Rethinking what Matters: Effective and Robust Multilingual Realignment for Low-Resource Languages by Quang Phuoc Nguyen, David Anugraha, Felix Gaschi, Jun Bin Cheng, En-Shiun Annie Lee (2025) 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-2511-06497.txt.

Description

Quang Phuoc Nguyen and colleagues ask whether aligning every language in a multilingual model actually helps, or whether a chosen few do the same job. Realignment nudges word meanings across languages into line, yet results have been mixed, above all for low-resource languages far from English. Parallel data for alignment is also scarce or noisy for exactly those languages.

Controlled experiments show realignment can help low-resource languages a great deal. Carefully picked, linguistically diverse language subsets matched full multilingual matching and even beat it on unseen low-resource languages. Covering every case is not needed, so data collection costs can fall while the transfer stays efficient and robust.

Catalog record

CREATORQuang Phuoc Nguyen, David Anugraha, Felix Gaschi, Jun Bin Cheng, En-Shiun Annie Lee
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE64 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-06497

Browse the full library · About the Tower of Babel