Are "Undocumented Workers" the Same as "Illegal Aliens"? Disentangling Denotation and Connotation in Vector Spaces

TL;DR

This paper introduces an adversarial neural network to disentangle denotation and connotation in pretrained language models, improving interpretability and enhancing information retrieval by capturing semantic nuances.

Contribution

It proposes a novel method to separate denotation and connotation in vector space representations, addressing a key challenge in NLP and semantic theory.

Findings

Words with same denotation but different connotations are closer in denotation space.

Disentangled representations improve viewpoint diversity in document rankings.

The approach enhances interpretability of semantic representations.

Abstract

In politics, neologisms are frequently invented for partisan objectives. For example, "undocumented workers" and "illegal aliens" refer to the same group of people (i.e., they have the same denotation), but they carry clearly different connotations. Examples like these have traditionally posed a challenge to reference-based semantic theories and led to increasing acceptance of alternative theories (e.g., Two-Factor Semantics) among philosophers and cognitive scientists. In NLP, however, popular pretrained models encode both denotation and connotation as one entangled representation. In this study, we propose an adversarial neural network that decomposes a pretrained representation as independent denotation and connotation representations. For intrinsic interpretability, we show that words with the same denotation but different connotations (e.g., "immigrants" vs. "aliens", "estate tax" vs. "death tax") move closer to each other in denotation space while moving further apart in connotation space. For extrinsic application, we train an information retrieval system with our disentangled representations and show that the denotation vectors improve the viewpoint diversity of document rankings.