Discriminating Form and Meaning in Multilingual Models with Minimal-Pair ABX Tasks

TL;DR

This paper presents a set of training-free ABX discrimination tasks to evaluate how multilingual models encode language identity and semantic content, revealing how these representations evolve across layers and training stages.

Contribution

It introduces a novel, interpretable, zero-shot evaluation method for analyzing multilingual model representations based on ABX tasks, applicable without additional training.

Findings

Language discrimination declines over training and is concentrated in lower layers.

Meaning discrimination improves and stabilizes in deeper layers.

ABX tasks correlate with linguistic learning performance.

Abstract

We introduce a set of training-free ABX-style discrimination tasks to evaluate how multilingual language models represent language identity (form) and semantic content (meaning). Inspired from speech processing, these zero-shot tasks measure whether minimal differences in representation can be reliably detected. This offers a flexible and interpretable alternative to probing. Applied to XLM-R (Conneau et al, 2020) across pretraining checkpoints and layers, we find that language discrimination declines over training and becomes concentrated in lower layers, while meaning discrimination strengthens over time and stabilizes in deeper layers. We then explore probing tasks, showing some alignment between our metrics and linguistic learning performance. Our results position ABX tasks as a lightweight framework for analyzing the structure of multilingual representations.