Discovering Properties of Inflectional Morphology in Neural Emergent Communication

TL;DR

This paper investigates how neural agents develop morphological properties in emergent communication, revealing that phonological constraints promote concatenative morphology and attribute fusion, akin to natural languages.

Contribution

It introduces a novel setting simulating inflectional morphology in emergent communication and develops metrics to analyze morphological properties in neural emergent languages.

Findings

Phonological constraints encourage concatenative morphology.

Emergent languages tend to fuse grammatical attributes.

The setting enables meaningful comparison to natural language morphology.

Abstract

Emergent communication (EmCom) with deep neural network-based agents promises to yield insights into the nature of human language, but remains focused primarily on a few subfield-specific goals and metrics that prioritize communication schemes which represent attributes with unique characters one-to-one and compose them syntactically. We thus reinterpret a common EmCom setting, the attribute-value reconstruction game, by imposing a small-vocabulary constraint to simulate double articulation, and formulating a novel setting analogous to naturalistic inflectional morphology (enabling meaningful comparison to natural language communication schemes). We develop new metrics and explore variations of this game motivated by real properties of inflectional morphology: concatenativity and fusion. Through our experiments, we discover that simulated phonological constraints encourage concatenative morphology, and emergent languages replicate the tendency of natural languages to fuse grammatical attributes.