Assessing the Syntactic Capabilities of Transformer-based Multilingual Language Models

TL;DR

This paper investigates the syntactic understanding of multilingual Transformer models like BERT and RoBERTa across English and Spanish, comparing monolingual and multilingual versions to understand their generalization abilities.

Contribution

It introduces SyntaxGymES, a new Spanish syntactic test suite, and evaluates the models' syntactic capabilities across languages and model types.

Findings

Multilingual models show varied syntactic generalization across languages.

Monolingual models outperform multilingual ones on specific syntactic tasks.

Spanish syntactic tests reveal unique challenges for language models.

Abstract

Multilingual Transformer-based language models, usually pretrained on more than 100 languages, have been shown to achieve outstanding results in a wide range of cross-lingual transfer tasks. However, it remains unknown whether the optimization for different languages conditions the capacity of the models to generalize over syntactic structures, and how languages with syntactic phenomena of different complexity are affected. In this work, we explore the syntactic generalization capabilities of the monolingual and multilingual versions of BERT and RoBERTa. More specifically, we evaluate the syntactic generalization potential of the models on English and Spanish tests, comparing the syntactic abilities of monolingual and multilingual models on the same language (English), and of multilingual models on two different languages (English and Spanish). For English, we use the available SyntaxGym test suite; for Spanish, we introduce SyntaxGymES, a novel ensemble of targeted syntactic tests in Spanish, designed to evaluate the syntactic generalization capabilities of language models through the SyntaxGym online platform.