Modelling Lexical Ambiguity with Density Matrices
This addresses the challenge of handling unrelated word senses in natural language processing, offering an incremental improvement over prior methods.
The paper tackled the problem of lexical ambiguity, specifically homonymy, by using density matrices to encode probability distributions over word senses within a compositional distributional model. Their best neural model outperformed existing vector-based models and strong sentence encoders on compositional datasets.
Words can have multiple senses. Compositional distributional models of meaning have been argued to deal well with finer shades of meaning variation known as polysemy, but are not so well equipped to handle word senses that are etymologically unrelated, or homonymy. Moving from vectors to density matrices allows us to encode a probability distribution over different senses of a word, and can also be accommodated within a compositional distributional model of meaning. In this paper we present three new neural models for learning density matrices from a corpus, and test their ability to discriminate between word senses on a range of compositional datasets. When paired with a particular composition method, our best model outperforms existing vector-based compositional models as well as strong sentence encoders.