Grounding and Distinguishing Conceptual Vocabulary Through Similarity Learning in Embodied Simulations

TL;DR

This paper introduces a method that uses embodied simulation experiences to ground and distinguish conceptual vocabulary in object representations, leveraging similarity learning and affine transformations to analyze transformer-based language models.

Contribution

It presents a novel approach combining embodied experiences and similarity learning to ground word vectors, revealing properties of transformer embedding spaces and their relation to object and action concepts.

Findings

Grounding object token vectors improves verb and attribute vector grounding.

Transformer models' embedding spaces exhibit properties consistent with psycholinguistic theories.

The method effectively identifies correct concepts in the object embedding space.

Abstract

We present a novel method for using agent experiences gathered through an embodied simulation to ground contextualized word vectors to object representations. We use similarity learning to make comparisons between different object types based on their properties when interacted with, and to extract common features pertaining to the objects' behavior. We then use an affine transformation to calculate a projection matrix that transforms contextualized word vectors from different transformer-based language models into this learned space, and evaluate whether new test instances of transformed token vectors identify the correct concept in the object embedding space. Our results expose properties of the embedding spaces of four different transformer models and show that grounding object token vectors is usually more helpful to grounding verb and attribute token vectors than the reverse, which reflects earlier conclusions in the analogical reasoning and psycholinguistic literature.