Quasi-symbolic Semantic Geometry over Transformer-based Variational AutoEncoder

TL;DR

This paper introduces a formal semantic geometry framework for Transformer-based language models, enabling better control and interpretability of sentence generation through a low-dimensional latent space.

Contribution

It proposes a novel method to incorporate formal semantic geometry into Transformer-based LMs using a Variational AutoEncoder and a new probing algorithm.

Findings

Enhanced control over sentence generation.

Improved interpretability of sentence representations.

Potential for more precise manipulation of language models.

Abstract

Formal/symbolic semantics can provide canonical, rigid controllability and interpretability to sentence representations due to their \textit{localisation} or \textit{composition} property. How can we deliver such property to the current distributional sentence representations to control and interpret the generation of language models (LMs)? In this work, we theoretically frame the sentence semantics as the composition of \textit{semantic role - word content} features and propose the formal semantic geometry. To inject such geometry into Transformer-based LMs (i.e. GPT2), we deploy Transformer-based Variational AutoEncoder with a supervision approach, where the sentence generation can be manipulated and explained over low-dimensional latent Gaussian space. In addition, we propose a new probing algorithm to guide the movement of sentence vectors over such geometry. Experimental results reveal that the formal semantic geometry can potentially deliver better control and interpretation to sentence generation.