Home  | Publications | KMN+25

MEXA: Multilingual Evaluation of English-Centric LLMs via Cross-Lingual Alignment

MCML Authors

Link to Profile Hinrich Schütze PI Matchmaking

Hinrich Schütze

Prof. Dr.

Principal Investigator

Computational Linguistics

Abstract

English-centric large language models (LLMs) often show strong multilingual capabilities. However, the multilingual performance of these models remains unclear and is not thoroughly evaluated for many languages. Most benchmarks for multilinguality focus on classic NLP tasks, or cover a minimal number of languages. We introduce MEXA, a method for assessing the multilingual capabilities of pre-trained English-centric LLMs using parallel sentences, which are available for more languages than existing downstream tasks. MEXA leverages the fact that English-centric LLMs use English as a kind of pivot language in their intermediate layers. It computes the alignment between English and non-English languages using parallel sentences to evaluate the transfer of language understanding from English to other languages. This alignment can be used to estimate model performance in other languages. We conduct studies using various parallel datasets (FLORES-200 and Bible), models (Llama family, Gemma family, Mistral, and OLMo), and established downstream tasks (Belebele, m-MMLU, and m-ARC). We explore different methods to compute embeddings in decoder-only models. Our results show that MEXA, in its default settings, achieves a statistically significant average Pearson correlation of 0.90 with three established downstream tasks across nine models and two parallel datasets. This suggests that MEXA is a reliable method for estimating the multilingual capabilities of English-centric LLMs, providing a clearer understanding of their multilingual potential and the inner workings of LLMs.

inproceedings

Findings @ACL 2025

Findings at the 63rd Annual Meeting of the Association for Computational Linguistics. Vienna, Austria, Jul 27-Aug 01, 2025.
Conference logo
A* Conference

Authors

A. H. Kargaran • A. Modarressi • N. Nikeghbal • J. Diesner • F. Yvon • H. Schütze

Links

URL

Research Area

 B2 | Natural Language Processing

BibTeXKey: KMN+25

Back to Top