A Principled Framework for Evaluating on Typologically Diverse Languages

TL;DR

This paper introduces a systematic framework for selecting typologically diverse languages in multilingual NLP evaluation, demonstrating that better sampling improves model generalizability across languages.

Contribution

The authors propose a new language sampling framework based on typological features, improving upon previous methods for selecting diverse languages in NLP evaluation.

Findings

Systematic sampling methods retrieve more typologically diverse languages.

Diverse language sampling enhances the evaluation of multilingual models.

Sampling impacts the perceived generalizability of NLP systems.

Abstract

Beyond individual languages, multilingual natural language processing (NLP) research increasingly aims to develop models that perform well across languages generally. However, evaluating these systems on all the world's languages is practically infeasible. To attain generalizability, representative language sampling is essential. Previous work argues that generalizable multilingual evaluation sets should contain languages with diverse typological properties. However, 'typologically diverse' language samples have been found to vary considerably in this regard, and popular sampling methods are flawed and inconsistent. We present a language sampling framework for selecting highly typologically diverse languages given a sampling frame, informed by language typology. We compare sampling methods with a range of metrics and find that our systematic methods consistently retrieve more typologically diverse language selections than previous methods in NLP. Moreover, we provide evidence that this affects generalizability in multilingual model evaluation, emphasizing the importance of diverse language sampling in NLP evaluation.