Color Overmodification Emerges from Data-Driven Learning and Pragmatic Reasoning

TL;DR

This paper investigates how environmental factors influence overmodification in referential language, using neural networks to simulate language learning and pragmatic reasoning, revealing that infrequent or salient features increase overmodification likelihood.

Contribution

It introduces a neural network-based approach to study pragmatic language use, systematically varying training environments to understand overmodification patterns, which was previously difficult with human subjects.

Findings

Overmodification increases with infrequent environmental features.

Salient features also lead to higher overmodification.

Results align with probabilistic pragmatic communication models.

Abstract

Speakers' referential expressions often depart from communicative ideals in ways that help illuminate the nature of pragmatic language use. Patterns of overmodification, in which a speaker uses a modifier that is redundant given their communicative goal, have proven especially informative in this regard. It seems likely that these patterns are shaped by the environment a speaker is exposed to in complex ways. Unfortunately, systematically manipulating these factors during human language acquisition is impossible. In this paper, we propose to address this limitation by adopting neural networks (NN) as learning agents. By systematically varying the environments in which these agents are trained, while keeping the NN architecture constant, we show that overmodification is more likely with environmental features that are infrequent or salient. We show that these findings emerge naturally in the context of a probabilistic model of pragmatic communication.