What Kind of Language is Easy to Language-Model Under Curriculum Learning?

TL;DR

This paper investigates how curriculum learning influences the inductive biases of language models, revealing that starting with simpler sentences significantly affects their learning patterns and typological predictions.

Contribution

It introduces a simple curriculum learning variant to study its effect on language model biases and typological pattern reproduction.

Findings

Curriculum learning substantially impacts language model biases.

Starting with simpler sentences alters the typological patterns learned.

The study expands understanding of learning scenarios in language modeling.

Abstract

Many of the thousands of attested languages share common configurations of features, creating a spectrum from typologically very rare (e.g., object-verb-subject word order) or impossible languages to very common combinations of features (e.g., subject-object-verb word order). One central question is under what conditions such typological tendencies can be predicted, and specifically whether the learning bias of language models (LMs) is sufficient to reproduce such patterns. In this study, we add one dimensionality to such analysis -- the learning scenario for LMs -- to explore its interaction with the inductive bias of LMs. Specifically, as a first study, we examine the effect of curriculum learning (CL), as a developmentally motivated learning scenario, i.e., starting with simpler sentences rather than randomly-ordered input. We expand existing LM-based exploration (El-Naggar et al., 2025a,b) with a simple CL variant and find that CL substantially impacts the apparent inductive bias of LMs.