Can Uniform Meaning Representation Help GPT-4 Translate from Indigenous Languages?

Shira Wein · 2025

Read Can Uniform Meaning Representation Help GPT-4 Translate from Indigenous Languages? by Shira Wein (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-2502-08900.txt.

Description

Shira Wein tests whether meaning diagrams can help GPT-4 translate from Indigenous languages. AI models do well in major languages but stumble in low-resource ones with little training text. Her 2025 paper uses Uniform Meaning Representation, a scheme that maps a sentence's meaning in a form meant to work across languages.

Wein tried translation from three languages: Navajo, Arapaho, and Kukama. GPT-4 worked with and without sample translations, and with and without the meaning diagrams added to its prompt. In most test cases, adding the diagrams raised translation quality by a clear margin. The result hints that structured meaning notes could support tools for languages the big models barely know.

Catalog record

CREATORShira Wein
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE22 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-08900

Browse the full library · About the Tower of Babel