How Does a Multilingual LM Handle Multiple Languages?

Santhosh Kakarla, Gautama Shastry Bulusu Venkata, Aishwarya Gaddam · 2025

Read How Does a Multilingual LM Handle Multiple Languages? by Santhosh Kakarla, Gautama Shastry Bulusu Venkata, Aishwarya Gaddam (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-2502-04269.txt.

Description

Santhosh Kakarla and a team of co-authors test how well large language models handle many languages, in two thousand twenty-five. Such models are trained on text in many tongues and aim to bridge gaps between them. Their skill is strong for widely used languages, but weaker for languages with little training text.

The team probed two named models with word meaning, name finding, and sentence matching tasks. They also tested whether skill in a rich language carries over to a poorer one. Probes of the models' inner word forms added depth beyond plain scores. The authors aim to guide models that serve all language groups more fairly.

Catalog record

CREATORSanthosh Kakarla, Gautama Shastry Bulusu Venkata, Aishwarya Gaddam
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE25 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-04269

Browse the full library · About the Tower of Babel