Colors in Context: A Pragmatic Neural Model for Grounded Language Understanding

TL;DR

This paper introduces a neural pragmatic model for interpreting color descriptions in grounded communication, improving accuracy by combining speaker and listener perspectives, especially in challenging cases.

Contribution

It presents a novel recursive pragmatic reasoning framework that enhances neural models for grounded language understanding, validated on a new color reference corpus.

Findings

Pragmatic reasoning improves interpretation accuracy in difficult cases.

Combining speaker and listener models yields better results than individual classifiers.

The model reproduces human pragmatic behaviors in color reference tasks.

Abstract

We present a model of pragmatic referring expression interpretation in a grounded communication task (identifying colors from descriptions) that draws upon predictions from two recurrent neural network classifiers, a speaker and a listener, unified by a recursive pragmatic reasoning framework. Experiments show that this combined pragmatic model interprets color descriptions more accurately than the classifiers from which it is built, and that much of this improvement results from combining the speaker and listener perspectives. We observe that pragmatic reasoning helps primarily in the hardest cases: when the model must distinguish very similar colors, or when few utterances adequately express the target color. Our findings make use of a newly-collected corpus of human utterances in color reference games, which exhibit a variety of pragmatic behaviors. We also show that the embedded speaker model reproduces many of these pragmatic behaviors.