Word Embeddings Are Capable of Capturing Rhythmic Similarity of Words

TL;DR

This paper explores how word embeddings like Word2Vec and GloVe can capture rhythmic similarity, revealing that they effectively group rhyming words and introducing a new metric for measuring rhythmic similarity.

Contribution

It demonstrates the capability of existing word embeddings to encode rhythmic features and proposes a novel metric for quantifying rhythmic similarity between words.

Findings

GloVe outperforms Word2Vec in capturing rhythmic similarity.

Rhyming words have higher vector similarity than non-rhyming words.

A new metric for rhythmic similarity is introduced.

Abstract

Word embedding systems such as Word2Vec and GloVe are well-known in deep learning approaches to NLP. This is largely due to their ability to capture semantic relationships between words. In this work we investigated their usefulness in capturing rhythmic similarity of words instead. The results show that vectors these embeddings assign to rhyming words are more similar to each other, compared to the other words. It is also revealed that GloVe performs relatively better than Word2Vec in this regard. We also proposed a first of its kind metric for quantifying rhythmic similarity of a pair of words.