Representations of language in a model of visually grounded speech signal

TL;DR

This paper introduces a visually grounded speech perception model that maps spoken language and images into a shared semantic space, revealing hierarchical semantic and form representations within the model.

Contribution

It presents a novel multi-layer recurrent highway network that captures both form and meaning in speech, with detailed analysis of how these representations evolve across layers.

Findings

Semantic encoding becomes richer in higher layers

Form encoding increases then plateaus or decreases

Model effectively links speech and visual semantics

Abstract

We present a visually grounded model of speech perception which projects spoken utterances and images to a joint semantic space. We use a multi-layer recurrent highway network to model the temporal nature of spoken speech, and show that it learns to extract both form and meaning-based linguistic knowledge from the input signal. We carry out an in-depth analysis of the representations used by different components of the trained model and show that encoding of semantic aspects tends to become richer as we go up the hierarchy of layers, whereas encoding of form-related aspects of the language input tends to initially increase and then plateau or decrease.

Figures

Full text

See pages 1-last of paper.pdf