Density Matrices for Metaphor Understanding

TL;DR

This paper explores using density matrices to model metaphorical language as a form of lexical ambiguity, demonstrating that while challenging, the method outperforms basic baselines and some neural models.

Contribution

It introduces a novel application of density matrices to represent metaphorical meaning as mixtures of word senses, advancing computational metaphor understanding.

Findings

Density matrix models outperform simple baselines.

Metaphor modeling is more complex than other lexical ambiguities.

Best models outperform some neural language models.

Abstract

In physics, density matrices are used to represent mixed states, i.e. probabilistic mixtures of pure states. This concept has previously been used to model lexical ambiguity. In this paper, we consider metaphor as a type of lexical ambiguity, and examine whether metaphorical meaning can be effectively modelled using mixtures of word senses. We find that modelling metaphor is significantly more difficult than other kinds of lexical ambiguity, but that our best-performing density matrix method outperforms simple baselines as well as some neural language models.